Just published @ScienceMagazine
Higher medical accuracy at lower cost
A 3-agent AI for sequential diagnosis nearly doubled accuracy compared with physicians for >300 @NEJM CPC cases and >300 ER cases AND substantially lowered cost, also exceeded performance of many AI models, physicians plus AI @HarshaNori
@erichorvitz @Dominic1King @mustafasuleyman
Peter Lee (computer scientist)
Peter Lee (born November 30, 1960) is an American computer scientist. He was President and head of Microsoft Research. Previously, he was the head of the Transformational Convergence Technology Office of the Defense Advanced Research Projects Agency and the chair of the Computer Science Department at Carnegie Mellon University. His research focuses on software security and reliability. More on Wikipedia
-
Peter Lee reposted
Eric Topol
@EricTopol
-
Peter Lee
@peteratmsr
Wow this is impressive. One of those tech bio advances that is both expected and yet still very surprising when you see it working for real.
Frank Noe
@FrankNoeBerlin
Dynamics: folding & binding of an intrinsically disordered peptide to its protein receptor via an intermediate state.
A 5 milliseconds simulation would take >10 GPU years with MD simulation.
At 5 microsecond steps, BioEmu2 needs 40 minutes on the same GPU – a >100,000 x speedup! -
Peter Lee reposted
Ezra Klein
@ezraklein
Bill Gates on why we should tax A.I. work: “If you’ve let go of a worker and hired a robot to do that job, why are you so incentivizing the trade-off against the human labor that you don’t ask the robot to also pay into the pension fund? What is it? Is it that the pro-robot union has gotten the tax laws to say: No, if you’re not flesh, then let pensions go bankrupt?”
This media couldn't be displayed. See it on the original post. -
Peter Lee
@peteratmsr
For the past 16 years, I'm pretty sure I've had the best job in all of tech. But an opportunity has come up that forced me to ask the question, "Considering my stage of life and career, will I regret it if I pass on this?" The answer I came up with was "Yes," and so I've taken a big leap and decided to retire from Microsoft.
Working at Microsoft, having the privilege of leading @MSFTResearch, and most recently serving as President of Microsoft Science, has been an incredible experience, full of personal growth and the satisfaction of positive global impact on people and on Microsoft. My colleagues, every day, created a working environment that felt like living in the inevitable future. In my view, @satyanadella, @kevin_scott, @judsonalthoff, and the entire senior leadership of Microsoft have created the modern-day equivalent of Camelot, seeking to align Microsoft's business priorities with the goal of global empowerment and prosperity for all. I've been proud to be a part of it all.
As for what I'll be doing next, what I can say at this point is that it is a once-in-a-lifetime opportunity to join an elite team, fulfilling a dream I've had since childhood. It's a major detour from what I've been doing over the past 2+ decades, but rest assured that I'll be able to continue my participation with @MayoClinic, @KPMedSchool, @seattlechildren, and other great organizations that seek to improve and advance healthcare delivery and medical science. -
Peter Lee reposted
Gianrico Farrugia
@GFarrugiaMD
Today, Mayo Clinic announced a generous gift from the Penske Foundation to support Bold. Forward. Unbound. in Rochester. In recognition of the Penske Foundation's generosity and enduring partnership, Mayo will name the Penske Arrival Building. Read more: newsnetwork.mayoclinic.org/discussion/mayo-clinic-receive…
-
Peter Lee reposted
Nicolò Fusi
@nfusi
Today we’re introducing Project Quine.
Quine combines a world model of biology with a research harness that connects it to literature, computation, and the wet lab, allowing the system to explore hypotheses, gather experimental evidence, and adapt what it does next.
This has been a long-term effort by many people at MSR, and I'm immensely proud of the team
www.microsoft.com/en-us/research/blog/introducing-quine-a… -
Peter Lee reposted
Microsoft Research
@MSFTResearch
Custom-made molecules are advancing medicine, materials, and agriculture, but producing them is slow and expensive. A new Nature paper highlights RetroChimera, a predictive model that helps accelerate chemical synthesis, helping researchers explore a wide range of molecules.:
-
Peter Lee reposted
Tristan Harris
@tristanharris
Proud to report that tomorrow Sept 15th "The AI Doc" documentary film will be streaming on Netflix.
Feeling confused, anxious, or overwhelmed by the recent AI headlines, from rogue OpenAI agents carrying out cyberattack to prominent AI researchers resigning to warn about existential risks to humanity?
To understand how we got here, and what we can do, watch “The AI Doc” which is now streaming on Netflix starting Sept 15th as well as other major streaming platforms. Interviews 40+ experts on AI, including 3 out of the 5 majors CEOs of frontier AI companies.
Host a screening, share it with your friends. Clarity is agency. #theaidoc #theaidocfilm
cc @Liv_Boeree @DKokotajlo @PeterDiamandis @JeffLadish @aza @SnehaRevanur @ajeya_cotra @hendrycks @NPCollapse @davidevanharris @janleike @JasonGMatheny @Yoshua_Bengio @ESYudkowsky @harari_yuval @reidhoffman @rajiinio @RameshMedias @nitashatiku @ilyasut @ShaneLegg @demishassabis @sama @sanmikoyejo @_KarenHao @DarioAmodei @peteratmsr -
Peter Lee reposted
Christina Farr
@chrissyfarr
@peteratmsr is back on Lifers, this time with @CALonghurst, who is the new CEO of Seattle Children's.
We got into why ambient scribes make clinicians happier and health system finances worse, whether a PA with AI should be handling triage a specialist would otherwise do, what happens to the back office when healthcare is the top employer in 42 states, and why frontier models keep beating tools built specifically for medicine.
Timestamps:
(00:00) Lead in
(00:28) Intro
(01:23) A massive moment
(04:01) Red pill ER reality
(06:12) Mission control and upskilling
(07:13) Tech leaders as CEOs
(09:27) Ambient scribes and costs
(13:43) Hierarchy versus efficiency
(15:23) Full stack healthcare workforce
(18:09) Sponsor: Granola
(18:55) Rethinking training and roles
(21:36) Bending the back office curve
(23:12) Admin jobs and AI skills
(27:50) Self driving cars benchmarking
(30:11) AI exceptionalism and agents
(32:59) Autonomy in clinical care
(36:25) Humans in the loop
(39:27) Accountability and responsibility
(41:37) Frontier models beat specialists
(46:35) Workflow interoperability and wrap
(47:53) WrapThis media couldn't be displayed. See it on the original post. -
Peter Lee reposted
Gianrico Farrugia
@GFarrugiaMD
In this @axios piece, @JimVandeHei gives his personal perspective navigating a fragmented healthcare system, & how his family’s experience changed when they came to Mayo. I appreciated the opportunity to share how to bring better healthcare to more people: www.axios.com/2026/08/16/ai-mayo-clinic-health-care-fix-j…
-
Peter Lee
@peteratmsr
We @MSFTResearch consistently find that more general AI models outperform specialized models, even on the specialized tasks. Yet another example is the PRISM2 model for digital pathology, in Nature Medicine today. With Paige (now a part of @TempusAI). news.microsoft.com/signal/articles/teaching-ai-to-speak-t…
-
Peter Lee
@peteratmsr
Many have long believed that real-world evidence can accelerate discovery in medicine. But how, exactly? This new piece in @NatureMedicine talks to @hoifungpoon and others on the leading edge to look at how AI is making the dream of RWE come to life. www.nature.com/articles/s41591-026-04484-6
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
Aurora 1.5 adds 22 more variables, hourly temporal resolution, and probabilistic ensemble forecasting to the Aurora foundation model, making it more useful for real-world weather, climate, and energy applications. www.microsoft.com/en-us/research/blog/aurora-1-5-extendin…
-
Peter Lee
@peteratmsr
Medical diagnoses can be improved by reanalyzing genomic data in rare disease cases. This is normally very burdensome, but today in @NatureMedicine we present Talos, a tool that in iterative monthly reanalysis reduces the burden to 1 variant per 200 cases www.nature.com/articles/s41591-026-04477-5
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
Talos was built to help resolve a major bottleneck in genomic medicine: human review time. The open-source system recovered 90% of in-scope diagnoses while surfacing just 1.3 candidate variants per patient for expert review. www.microsoft.com/en-us/research/blog/talos-scaling-rare-…
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
ResNet has received the Longuet-Higgins Prize at CVPR 2026, recognizing research with proven, lasting impact. A decade after its publication, residual connections remain foundational to how modern AI systems are built, with over 320,000 citations and growing.
-
Peter Lee reposted
Satya Nadella
@satyanadella
Today in @naturemethods, we shared research on how AI can help us better understand cell behavior, offering new insights into why cancer medicines do not work the same for everyone.
By learning more about cell state — how individual cancer cells respond to their surroundings — we have the potential to match therapies more precisely to each patient and improve outcomes. news.microsoft.com/signal/articles/why-dont-cancer-medici… -
Peter Lee reposted
Microsoft
@Microsoft
New findings in Nature Methods highlight how Project Ex Vivo is helping researchers uncover patterns in cell behavior that may lead to more personalized therapies for patients dealing with cancer.
Microsoft researcher Lorin Crawford explains more: news.microsoft.com/signal/articles/why-dont-cancer-medici… -
Peter Lee reposted
Gianrico Farrugia
@GFarrugiaMD
I shared more with @CNN about Mayo’s collaboration with @Microsoft to develop a frontier AI model that will bring better healthcare AI capabilities to patients and care teams around the world: www.cnn.com/2026/06/02/tech/ai-for-healthcare-microsoft-m…
-
Peter Lee
@peteratmsr
Every day during the shared envisioning, planning, and dreaming of what might be possible, my excitement of what would be possible grew enormously. The joint commitment to advancing healthcare is so inspiring.
Gianrico Farrugia
@GFarrugiaMD
I’m pleased to share Mayo Clinic has announced a strategic collaboration with @Microsoft to develop and deploy a frontier AI model designed specifically for healthcare. Read more: newsnetwork.mayoclinic.org/discussion/mayo-clinic-and-mic…
-
Peter Lee
@peteratmsr
Only once or twice in my career have I felt a real chance to extend the frontiers of human knowledge and bring practical benefits to everyone in a big way. This is one of those times, not only due to the capabilities of both Microsoft and Mayo Clinic, but also our common vision.
Mustafa Suleyman
@mustafasuleyman
Proud that we’re collaborating with Mayo Clinic to build a frontier AI model for healthcare. Both our organizations exist to serve people at scale – and we believe this could be nothing short of transformative for global healthcare.
news.microsoft.com/source/2026/06/02/mayo-clinic-and-micr… -
Peter Lee reposted
Microsoft Research
@MSFTResearch
Introducing GridSFM, a small foundation model that can predict AC optimal power flow in milliseconds, boosting efficiency and unlocking cost savings. Learn how GridSFM gives grid operators direct visibility into congestion, stability, and system health: www.microsoft.com/en-us/research/blog/gridsfm-a-new-small…
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
MatterSim is expanding what AI can do for materials science—from faster large-scale simulations to MatterSim-MT, a new multi-task model for simulating properties beyond potential energy surfaces alone. www.microsoft.com/en-us/research/blog/advancing-ai-for-ma…
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
Tyger moves the most demanding MRI processing to the cloud, helping researchers turn raw signals into readable images – meaning results in hours rather than days or weeks. www.microsoft.com/en-us/research/story/when-mri-images-co…
-
Peter Lee reposted
Rianne van den Berg
@vdbergrianne
Today we're sharing a major Skala update: new paper and model release. Skala is a deep-learned XC functional for DFT: 2.8 kcal/mol on GMTKN55, wins 32/55 subsets & surpasses SOTA hybrids in accuracy at semi-local cost.
Paper: arxiv.org/abs/2506.14665
Code: github.com/microsoft/skala -
Peter Lee reposted
Microsoft Research
@MSFTResearch
A new perspective from Microsoft Research published in Cell makes the case that generative models are what oncology needs next, capable of integrating genomics, imaging, clinical data, and more into a unified system for cancer discovery. www.cell.com/cell/fulltext/S0092-8674(26)00328-4
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
We are thrilled to share Microsoft has been awarded the 2026 Franz Edelman Award by @INFORMS for using ML, optimization, and generative AI to orchestrate cloud fulfillment decisions at global scale. www.informs.org/IOL-Home/News-Room/INFORMS-Releases/Award…
-
Peter Lee reposted
Tristan Harris
@tristanharris
Exciting news! @theaidocfilm is now available to rent or buy on Apple TV, Prime Video, or wherever you get your movies on demand.
If you missed the theatrical release, you can now watch this urgent film at home. Invite your friends, your family, your church group, your business to hold a movie night, and save room for a group discussion afterwards!
The film is meant to be watched together.
@JeffLadish @SnehaRevanur @aza @NPCollapse @PeterDiamandis @Liv_Boeree @_KarenHao @Randima_F @Yoshua_Bengio @hendrycks @janleike @JasonGMatheny @RameshMedias @reidhoffman @DKokotajlo @DarioAmodei @demishassabis @peteratmsr @rajiinio @ShaneLegg @sanmikoyejo -
Peter Lee
@peteratmsr
I've always been impressed with @chrissyfarr's technical and business depth combined with down-to-earth authenticity. So, it was a truly fun honor to sit with her for a wide-ranging discussion about AI, healthcare, and the future of knowledge work. www.youtube.com/watch?v=lxWYV1rNqR4
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
MicroLED datacenter networking, LLM-driven GPU design, real-time multimodal AI, agent memory and learning from experience, plus new benchmarks for interactive planning and a new podcast about the future of machine intelligence. www.linkedin.com/pulse/microled-datacenters-multimodal-ai…
-
Peter Lee reposted
Hoifung Poon
@hoifungpoon
Excited to showcase the science and people behind our GigaTIME paper, recently published in Cell.
Paper: aka.ms/gigatime-paper
Model: aka.ms/gigatime-model
Code: aka.ms/gigatime-code
Azure Foundry: labs.ai.azure.com/projects/gigatime/
Satya Nadella
@satyanadella
We’ve trained a multimodal AI model to turn routine pathology slides into spatial proteomics, with the potential to reduce time and cost while expanding access to cancer care.
This media couldn't be displayed. See it on the original post. -
Peter Lee reposted
Satya Nadella
@satyanadella
We’ve trained a multimodal AI model to turn routine pathology slides into spatial proteomics, with the potential to reduce time and cost while expanding access to cancer care.
This media couldn't be displayed. See it on the original post. -
Peter Lee reposted
Microsoft Research
@MSFTResearch
Long-term glass data storage advances, new work on transferable reasoning, multi-turn AI safety testing, multilingual AI design, and evaluating how models actually think. www.linkedin.com/pulse/glass-storage-breakthroughs-robust…
-
Peter Lee reposted
Satya Nadella
@satyanadella
Today in @Nature, we shared new results from Project Silica—our work to encode data in glass—pushing toward durable, immutable archival storage designed to last for millennia. www.microsoft.com/en-us/research/blog/project-silicas-adv…
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
Project Silica introduces new techniques for encoding data in borosilicate glass, as described in the journal Nature. These advances lower media cost and simplify writing and reading systems while supporting 10,000-year data preservation: www.microsoft.com/en-us/research/blog/project-silicas-adv…
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
We are pleased to announce that Doug Burger (@dcburger), Technical Fellow and Corporate Vice President at Microsoft Research, has been elected to the National Academy of Engineering Class of 2026 for advancing cloud-scale computing and networking with field-programmable systems.
-
Peter Lee reposted
Ece Kamar
@ecekamar
Had such a fun time chatting with Jordan Wilson on his Everyday AI podcast! We dug into why the future of AI isn’t just about bigger models—it’s about societies of smaller, collaborative agents and how this shift will reshape the way organizations build with AI. @MSFTResearch
Jordan Talks Everyday AI
@EverydayAI_
Inside the Society of Agents: Why AI Teamwork Beats Bigger Models x.com/i/broadcasts/1mnxeNAakvZKX
-
Peter Lee reposted
Paola Gori-Giorgi
@paolagorigiorgi
Check it out #compchem and give us feedback!
Derk Kooi
@DerkKooi
Skala community edition update: GPU4PySCF support is in ✅
That means Skala can plug into GPU-accelerated PySCF DFT workflows via GPU4PySCF (API-compatible with PySCF).
Install docs are in the repo README: github.com/microsoft/skala
#dft #pyscf #gpu #gpu4pyscf #compchem -
Peter Lee reposted
Microsoft Research
@MSFTResearch
Rho-alpha, which translates natural language commands into control signals for robotic systems doing bimanual manipulation tasks, aims to make physical systems more adaptable by using physical sensing modalities like touch and continuous learning from human feedback.
-
Peter Lee reposted
Bob Wachter
@Bob_Wachter
AI doesn't need to be perfect to outperform our healthcare system's woeful status quo. It just needs to be better.
In many cases, it already is.
In today's @nytimes, I explore this idea – a central theme of my upcoming book, A Giant Leap (out Feb 3).
www.nytimes.com/2026/01/19/opinion/ai-health-medical-care… -
Peter Lee
@peteratmsr
I've devoted a *lot* of time to AI in Healthcare, because I think this is where the impact will be so profound — and needed. In the "On Second Thought" video series, @jcarlsonmsr @MSFTResearch joins @sineadbovell to look honestly at what's possible — and what still needs work.
Microsoft Research
@MSFTResearch
What changes when AI becomes part of the healthcare journey?
In episode one of On Second Thought, futurist @SineadBovell joins Jonathan Carlson, Head of Microsoft Research Health Futures, to discuss the rapid adoption of AI in healthcare, the potential of personalized medicine, and how trust, training, and human judgment shape the future of care.
- 01:48: Why would I want my doctor to use AI?
- 03:38: Can I Trust AI with my health data?
- 04:55: What does personalized medicine enable?
- 06:47: Can AI actually diagnose correctly?
- 10:06: Will doctors be trained to use AI?
- 11:13: Where is the human in the future of care?
- 12:53: What do you wish people knew about AI and healthcare?This media couldn't be displayed. See it on the original post. -
Peter Lee reposted
Microsoft Research
@MSFTResearch
2025 saw groundbreaking innovations including AI-powered materials discovery tools, protein structure modeling, and multilingual AI for underserved communities. Dive into our Year in Review for a look at these and other transformative advances. www.linkedin.com/pulse/2025-year-review-ai-driven-breakth…
-
Peter Lee
@peteratmsr
Thinking back to the start of 2025 @MSFTResearch, while I knew some things were brewing in our labs, I would never have predicted how much happened, and how fast. Here we recap just a tiny sample of the things we made public.
Microsoft Research
@MSFTResearch
From AI-driven material discovery and protein modeling to agentic systems, gaming, and multilingual AI for the global majority, 2025 turned research into real-world impact. Explore the breakthroughs in Microsoft Research’s Year in Review. www.microsoft.com/en-us/research/story/microsoft-research…
-
Peter Lee
@peteratmsr
Today in @MSFTResearch we announced GigaTIME—a breakthrough AI tool for population-scale cancer research, published in the journal Cell. It is a multimodal AI model that simulates spatial proteomics from routine pathology slides. /1
Post not available here, view on X
Peter Lee
@peteratmsr
Importantly, it also makes it possible to uncover new links between genetic mutations, immune protein activation, and clinical outcomes at unprecedented speed and scale. By generating virtual cohorts and making the tool open-source, this advances precision medicine. /3
Peter Lee
@peteratmsr
This milestone demonstrates how AI can transform biomedical research and move from data to actionable insights faster and more affordably. Blog post: www.microsoft.com/en-us/research/blog/gigatime-scaling-tu…
-
Peter Lee
@peteratmsr
Today in the journal Cell, our GigaTIME AI model. It enables population-scale analysis of the “neighborhood” around tumors in pathology slides. With @providence and @UWMedicine, open-source and validated on data from over 14,000 patients. I'm genuinely amazed that this is possible.
Satya Nadella
@satyanadella
Today in Cell, we published new research showing how AI can help accelerate cancer discovery. With GigaTIME, we can now simulate spatial proteomics from routine pathology slides, enabling population-scale analysis of tumor microenvironments across dozens of cancer types and hundreds of subtypes.
Developed in partnership with Providence and the University of Washington, our hope is that this work helps scientists move faster from data to insight, revealing new links between genetic mutations, immune activity, and clinical outcomes, and ultimately improving health for people everywhere. -
Peter Lee
@peteratmsr
I'm so proud of what researchers across Microsoft and @MSFTResearch shared with the world in AI, healthcare, life sciences, energy, and more. Thanks to everyone across Microsoft who made 2025 such an impactful year. Looking forward to 2026. news.microsoft.com/source/features/ai/10-scientific-break…
-
Peter Lee
@peteratmsr
I'm incredibly proud of what researchers across Microsoft, especially @MSFTResearch, shared with the world in AI, healthcare, life sciences, energy, and more. Here are 10 awe-inspiring examples. Thanks to everyone who made 2025 such an impactful year. Looking forward to 2026.
Microsoft
@Microsoft
AI didn’t just make progress this year, it delivered breakthroughs.
Microsoft researchers introduced Majorana 1, the world’s first quantum processor built with topological qubits, shared Aurora to predict extreme weather faster than ever, and developed FCDD to improve early breast cancer detection.
And that’s just the start. Explore 10 scientific innovations that made a global impact in 2025. news.microsoft.com/source/features/ai/10-scientific-break… -
Peter Lee reposted
Microsoft Research
@MSFTResearch
In the latest issue of Research Focus: Fara-7B brings efficient agentic computer use to small models; new tools help computing researchers assess societal impact; and more. www.linkedin.com/pulse/agentic-slms-computer-use-societal…
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
Fara-7B is our first agentic small language model for computer use. This experimental model includes robust safety measures to aid responsible deployment. Despite its size, Fara-7B holds its own against larger, more resource-intensive agentic systems: www.microsoft.com/en-us/research/blog/fara-7b-an-efficien…
-
Peter Lee
@peteratmsr
New from @MSFTResearch: A native agentic SLM, Fara‑7B, designed for computer use. It automates everyday web tasks with frontier performance while keeping your information private.
Now on Microsoft Foundry and Hugging Face.
www.microsoft.com/en-us/research/blog/fara-7b-an-efficien… -
Peter Lee
@peteratmsr
I love cars, and so converting classic cars to electric drive is just tons of fun, and it's an interesting design, engineering, and learning experience. I enjoyed this conversation with Kurt Schlosser at @geekwire, talking about a project with the folks at Moment Motor Company.
GeekWire
@geekwire
Out of Office: Microsoft Research’s Peter Lee is a car geek revved up by converting classics to electric www.geekwire.com/2025/out-of-office-microsoft-researchs-p…
-
Peter Lee
@peteratmsr
In responsible AI, a key challenge is to find ways for AI to incorporate and reflect the culturally nuanced lived experiences of all people and communities. Developing such globally equitable generative AI is the goal of Project Gecko and the new MMCTAgent.
Microsoft Research
@MSFTResearch
AI tools often perform poorly in non-Western languages and lack cultural context that’s crucial for many populations. Project Gecko uses small language models to deliver vital content that extends AI’s benefits to serve the global majority: www.microsoft.com/en-us/research/story/advancing-ai-to-me…
-
Peter Lee
@peteratmsr
There's been a very noticeable uptick in interest from folks (especially young people!) wondering what it's like to be at a place like @MSFTResearch at the leading edge of research. Here's a nice piece that gives an inside look at a day in the life of researchers across Microsoft as they explore areas like agentic systems, trustworthy AI, and the intersection of AI and society.
Microsoft Research
@MSFTResearch
Microsoft researchers are driving trends and tackling major challenges in the midst of the ever-changing AI era. Check out Trista Chen, Flavio Griggio, Ahmed Awadallah, Jina Suh, and Cecily Morrison's conversations with Forward Future, a daily AI newsletter, to learn more about the people building the systems that are becoming part of our daily lives. news.microsoft.com/signal/articles/a-day-in-the-life-at-m…
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
AI agents are transforming digital marketplaces, mediating discovery and transactions between consumers and businesses. The new Magentic Marketplace provides an open-source, extensible simulation environment for studying different agentic market designs: www.microsoft.com/en-us/research/blog/magentic-marketplac…
This media couldn't be displayed. See it on the original post. -
Peter Lee reposted
Microsoft Research
@MSFTResearch
Announcing the 2026 Microsoft Research Fellowship, a program that creates opportunities for faculty, PhD students, and postdocs to collaborate with Microsoft Research on open research challenges that advance science and benefit society. Submit proposals by Dec 2:
-
Peter Lee reposted
Paola Gori-Giorgi
@paolagorigiorgi
Amazing work on polymers by AI for Science @MSFTResearch colleagues!
Gregor Simm
@gncsimm
MLFFs 🤝 Polymers — SimPoly works!
Our team at @MSFTResearch AI for Science is proud to present SimPoly (SIM-puh-lee) — a deep learning solution for polymer simulation.
Polymeric materials are foundational to modern life—found in everything from the clothes we wear and the food we consume to high-performance materials in aerospace, electronics, and medicine. Today, we introduce a new way to simulate them.
We built a machine learning force field (MLFF) to predict macroscopic properties across a broad range of polymers—trained only on quantum-chemical data, with no experimental fitting. Specifically, we accurately compute polymer densities via large-scale MD simulations, achieving higher accuracy than classical force fields. We also capture second-order phase transitions, enabling prediction of glass transition temperatures. These two properties are fundamental to processing and application design. Finally, we created a benchmark based on experimental data for 130 polymers plus an accompanying quantum-chemical dataset—laying the foundation for a fully in silico design pipeline for next-generation polymeric materials.
The incredible team: Jean Helie, @temporaer, Yicheng Chen, Guillem Simeon, @a_kzna, @ErnestoCheco, @erunzzz, Gabriele Tocci, @chc273, @yatao_li, @SherryLixueC, @zunwang_msr, Bichlien H. Nguyen, Jake A. Smith, and Lixin Sun.
📄 Preprint: arxiv.org/abs/2510.13696
⚙️ Data and code release: in progress⏳
#mlffs #polymers #aiforscience #deeplearning #simpoly #scientificml #microsoft #microsoftresearch #microsoftquantum -
Peter Lee reposted
Tian Xie
@xie_tian
Congrats to my colleagues Gregor, Lixin, and the entire team for the announcement of SimPoly! I cannot remember how many times I’ve been asked for a polymer MLFF from industrial partners, since there is no great solution yet. Code/data will be released soon. Check it out!
Gregor Simm
@gncsimm
MLFFs 🤝 Polymers — SimPoly works!
Our team at @MSFTResearch AI for Science is proud to present SimPoly (SIM-puh-lee) — a deep learning solution for polymer simulation.
Polymeric materials are foundational to modern life—found in everything from the clothes we wear and the food we consume to high-performance materials in aerospace, electronics, and medicine. Today, we introduce a new way to simulate them.
We built a machine learning force field (MLFF) to predict macroscopic properties across a broad range of polymers—trained only on quantum-chemical data, with no experimental fitting. Specifically, we accurately compute polymer densities via large-scale MD simulations, achieving higher accuracy than classical force fields. We also capture second-order phase transitions, enabling prediction of glass transition temperatures. These two properties are fundamental to processing and application design. Finally, we created a benchmark based on experimental data for 130 polymers plus an accompanying quantum-chemical dataset—laying the foundation for a fully in silico design pipeline for next-generation polymeric materials.
The incredible team: Jean Helie, @temporaer, Yicheng Chen, Guillem Simeon, @a_kzna, @ErnestoCheco, @erunzzz, Gabriele Tocci, @chc273, @yatao_li, @SherryLixueC, @zunwang_msr, Bichlien H. Nguyen, Jake A. Smith, and Lixin Sun.
📄 Preprint: arxiv.org/abs/2510.13696
⚙️ Data and code release: in progress⏳
#mlffs #polymers #aiforscience #deeplearning #simpoly #scientificml #microsoft #microsoftresearch #microsoftquantum -
Peter Lee reposted
Satya Nadella
@satyanadella
No question that nurses are the heartbeat of healthcare. And today we’re introducing the first commercially available ambient experience built for nursing workflows, so they can focus on what matters most: caring for patients.
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
The wait is over! Microsoft Research is sharing Skala, the new exchange-correlation functional, marking a major milestone in the accuracy/cost trade-off in DFT. Help us learn from your testing so we can improve. Available on Azure AI Foundry and GitHub. labs.ai.azure.com/projects/skala/
This media couldn't be displayed. See it on the original post. -
Peter Lee reposted
Paola Gori-Giorgi
@paolagorigiorgi
Skala is now available to everyone!
Why are we releasing it? Because we’re not just aiming to publish a cool paper — we’re on a mission to bring DFT to chemical accuracy using deep learning. And to make real progress, we need the community’s feedback. #compchem
Microsoft Research
@MSFTResearch
The wait is over! Microsoft Research is sharing Skala, the new exchange-correlation functional, marking a major milestone in the accuracy/cost trade-off in DFT. Help us learn from your testing so we can improve. Available on Azure AI Foundry and GitHub. labs.ai.azure.com/projects/skala/
This media couldn't be displayed. See it on the original post. -
Peter Lee reposted
Tian Xie
@xie_tian
🚨We are hiring! 🚨 Want to join a highly talented, collaborative team and build the next frontier model for materials design? Apply to the following roles and join our materials team at @MSFTResearch AI for Science. Location can be Cambridge UK or Amsterdam NL or Berlin DE.
Senior Researcher in Deep Learning: jobs.careers.microsoft.com/global/en/job/1885459/Senior-R…
Senior Applied Scientist: jobs.careers.microsoft.com/global/en/job/1891473/Senior-A…
Senior Research Engineer on Data: jobs.careers.microsoft.com/global/en/job/1850515/Senior-R…
Senior Research Software Development Engineer(w/ our engineering team): jobs.careers.microsoft.com/global/en/job/1885463/Senior-R…
Microsoft Research
@MSFTResearch
Microsoft researchers introduce MatterGen, a model that can discover new materials tailored to specific needs—like efficient solar cells or CO2 recycling—advancing progress beyond trial-and-error experiments. www.microsoft.com/en-us/research/blog/mattergen-a-new-par…
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
AI is impacting healthcare—from medical training to drug discovery. Learn how things are changing from those using, studying, and developing healthcare AI in “The AI Revolution in Medicine, Revisited” with host @peteratmsr. All episodes available now: www.microsoft.com/en-us/research/story/the-ai-revolution-…
This media couldn't be displayed. See it on the original post. -
Peter Lee
@peteratmsr
Congratulations, @xie_tian! You're such a great colleague, doing incredible work.
Tian Xie
@xie_tian
I am honored to join this year’s MIT technology review’s innovators under 35 list. Amazing to see the impact of generative AI on materials design being recognized in this prestigious list. Credits to the whole MatterGen team for making this happen.
-
Peter Lee
@peteratmsr
A computer that's 100x faster and 100x more energy efficient for solving difficult optimization problems? Today in @Nature, we @MSFTResearch describe a working analog optical computer, in collaboration with @Barclays. Looking also to other domains, eg, healthcare.
Satya Nadella
@satyanadella
Our breakthrough work on an analog optical computer points to new ways to solve complex real-world problems with much greater efficiency. Super to see this published today in @Nature. news.microsoft.com/source/features/innovation/microsoft-a…
-
Peter Lee
@peteratmsr
I learned a lot about what is going on today with AI in the real world of medicine, in these in-depth conversations with so many great people. Fun, also, to reconnect with @goldbergcarey and @zakkohane. Key insights for healthcare, sure, but also for any other industry sector.
Microsoft Research
@MSFTResearch
For the finale of “The AI Revolution in Medicine, Revisited,” Peter Lee, Carey Goldberg, and Dr. Zak Kohane compare their predictions to insights from the series’ most recent guests, including leaders in AI-driven medicine and doctors in training. www.microsoft.com/en-us/research/podcast/coauthor-roundta…
-
Peter Lee reposted
Satya Nadella
@satyanadella
Exciting work by our team, representing a real step forward in understanding the protein dynamics that power biological function.
Science Magazine
@ScienceMagazine
Researchers have developed a #deeplearning system called BioEmu that rapidly generates diverse protein conformations, enabling fast, accurate insights into protein flexibility and function.
Learn more this week in Science: scim.ag/41rCpxZ -
Peter Lee reposted
Science Magazine
@ScienceMagazine
Researchers have developed a #deeplearning system called BioEmu that rapidly generates diverse protein conformations, enabling fast, accurate insights into protein flexibility and function.
Learn more this week in Science: scim.ag/41rCpxZ -
Peter Lee
@peteratmsr
Featured on the cover of @ScienceMagazine, BioEmu @MSFTResearch, a deep learning system that rapidly generates diverse protein conformations for accurate insights into protein function. Illustration: N. Burgess/Science; Data: S. Lewis et al., www.science.org/toc/science/389/6761.
-
Peter Lee
@peteratmsr
I always learn so much when I have a chance to interact with @GFarrugiaMD @MayoClinic, and so having him on the podcast was a big treat. Real insights into the leading edge of AI at this storied health system, by its leader.
Gianrico Farrugia
@GFarrugiaMD
On a recent podcast, I joined Peter Lee to discuss AI's transformative impact on healthcare. Mayo Clinic's decentralized AI approach empowers our teams to innovate, accelerate progress, and improve patient care.
www.microsoft.com/en-us/research/podcast/reimagining-heal… @peteratmsr -
Peter Lee reposted
Tian Xie
@xie_tian
Want to join our efforts @MSFTResearch AI for Science to push the frontier of AI for materials? We are the team behind MatterGen & MatterSim and we have 2 job openings! Each can be in Amsterdam, NL, Berlin, DE, or Cambridge, UK. It is a rare opportunity to join a highly talented, collaborative team and build the next frontier model for materials design.
Senior Researcher: jobs.careers.microsoft.com/global/en/job/1850507/Senior-R…
Senior Research Engineer: jobs.careers.microsoft.com/global/en/job/1850515/Senior-R… -
Peter Lee reposted
Satya Nadella
@satyanadella
Today we’re introducing Copilot Mode in Edge, our first step in reinventing the browser for the AI age.
My favorite feature is multi-tab RAG. You can use Copilot to analyze your open tabs, like I do here with papers our team has published in @Nature journals over the last year.
And there is a lot more to come, including built-in actions so you can delegate tasks as you browse. -
Peter Lee reposted
Ava Amini
@avapamini
thrilled to share The Dayhoff Atlas of protein language data and models 🚀 protein biology in the age of AI!
aka.ms/dayhoff/preprint
we built + open source the largest natural protein dataset, w/ 3.3 billion seqs & a first-in-class dataset of structure-based synthetic proteins -
Peter Lee reposted
Kevin K. Yang 楊凱筌
@KevinKaichuang
In 1965, Margaret Dayhoff published the Atlas of Protein Sequence and Structure, which collated the 65 proteins whose amino acid sequences were then known.
Inspired by that Atlas, today we are releasing the Dayhoff Atlas of protein sequence data and protein language models. -
Peter Lee
@peteratmsr
I think this might be my favorite episode of this podcast series. Talking to current med students about their thoughts and actions on AI in medicine is enlightening. It's impressive how they are really leaning in and learning together.
Microsoft Research
@MSFTResearch
In Episode 9 of "The AI Revolution in Medicine, Revisited," resident physician and venture capitalist Morgan Cheatham and med student Daniel Chen explore generative AI's impact on medical training, from classroom learning to clinical practice. www.microsoft.com/en-us/research/podcast/navigating-medic…
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
In Episode 9 of "The AI Revolution in Medicine, Revisited," resident physician and venture capitalist Morgan Cheatham and med student Daniel Chen explore generative AI's impact on medical training, from classroom learning to clinical practice. www.microsoft.com/en-us/research/podcast/navigating-medic…
-
Peter Lee
@peteratmsr
So pleased to launch @MSFTResearch Asia – Singapore today. This takes our longstanding collaborations to the next level and gives new opportunities to advance AI to benefit humanity. We're grateful for the engagement of so many great people and orgs here. news.microsoft.com/source/asia/2025/07/24/microsoft-resea…
-
Peter Lee reposted
Satya Nadella
@satyanadella
Big milestone for the future of quantum computing.
We are delivering the world's first operational deployment of a Level 2 quantum computer, powered by our stack and in partnership with @Atom_Computing. Congrats to EIFO, @novonordiskfond, and QuNorth.
novonordiskfonden.dk/en/news/eifo-and-the-novo-nordisk-fo… -
Peter Lee reposted
Josh Mandel, MD
@JoshCMandel
Which healthcare regulations should we kill?
Yesterday, the comment period closed on the Agency for Healthcare Research and Quality RFI titled "Ensuring Lawful Regulation and Unleashing Innovation to Make America Healthy Again." The RFI asked stakeholders to pinpoint specific regulations that: increase costs without benefits, impede innovation or competition, are obsolete or based on outdated technology, or require excessive reporting or recordkeeping.
The response included 646 comments from every corner of American healthcare. I fired up my open-source comment analysis pipeline (github.com/jmandel/regulations.gov-comment-browser) to get an early look at what people said, creating this theme hierarchy (joshuamandel.com/regulations.gov-comment-browser/). Then using this open data set through my regulations.gov MCP (www.linkedin.com/pulse/cms-rfi-mcp-now-its-your-turn-anal…), I asked Claude Opus 4 to synthesize the formal, data-heavy responses from major organizations together with the raw, often heartbreaking, stories from individual patients, clinicians, and families. The result is a sampling from many stories, with substantial areas of alignment.
A Universal Villain: Prior Authorization
No single issue united commenters more than prior authorization. The organizational data paints a grim national picture, while the individual stories reveal the human cost. The American Medical Association (AMA) laid out the systemic failure: 93% of physicians report that prior authorization delays patient care, and 29% report it has led to a serious adverse event, including hospitalization, disability, and death. The AMA found that 82% of patients simply abandon treatment due to these bureaucratic hurdles, and 40% of physician practices employ staff exclusively for PA (joshuamandel.com/regulations.gov-comment-browser/AHRQ-202…). That's not an abstract statistic. That's the daily reality for Dr. Margaret Chustecki, an Associate Clinical Professor at Yale. Her practice spends 13 hours every week on prior authorization alone—time that could be spent with patients (joshuamandel.com/regulations.gov-comment-browser/AHRQ-202…). Her story is a perfect ground-truth validation of the AMA's national finding that physicians average 13 hours of staff time per week on PA.
Drowning in Clicks: The Administrative Burden
Beyond prior authorization, the sheer weight of compliance is crushing clinicians with work that has no bearing on patient outcomes. The Cleveland Clinic highlighted the absurdity: "The notion that a provider's ability to perform a basic wet mount or urine sediment review must be annually documented, while their ability to prescribe controlled substances, interpret imaging, or perform in-office procedures does not require such documentation, reflects a fundamental misalignment of regulatory scrutiny" (joshuamandel.com/regulations.gov-comment-browser/AHRQ-202…). This misalignment has a staggering cost. The Merit-based Incentive Payment System (MIPS) earned universal condemnation. The AMA cited data showing practices spend $12,800 and 202 hours annually on MIPS compliance, while a JAMA study found the program "approximately as effective as chance" in identifying quality (joshuamandel.com/regulations.gov-comment-browser/AHRQ-202…). CMS's own 2025 estimate puts the total system-wide cost at $70,166,672. This isn't just about money; it's about time stolen from care and the well-being of clinicians. An anonymous Director of Nursing with 30 years of experience reported that NHSN/SAMS COVID reporting requirements, which have never been reversed, "take 8 hours per week from bedside care for an RN." She lives in daily fear: "Everyday we fear for our licenses while state boards of nursing hold a power that is not warranted" (joshuamandel.com/regulations.gov-comment-browser/AHRQ-202…). This burden extends to social services, where a Head Start worker described an annual renewal process that takes "months to complete" and is "consistently returned multiple times" for publicly available information (joshuamandel.com/regulations.gov-comment-browser/AHRQ-202…).
When Rules Actively Deny Care
In the most extreme cases, regulations designed to protect patients create insurmountable barriers to necessary treatment.
The ADHD Medication Crisis
Several individual comments described the struggle to access Schedule II ADHD medications. An entrepreneur described failing three businesses before getting treated, after which they successfully reopened and maintained their business (joshuamandel.com/regulations.gov-comment-browser/AHRQ-202…). But access remains a "full-time job," as one nursing student put it (joshuamandel.com/regulations.gov-comment-browser/AHRQ-202…). The regulatory disconnect is stark: ADHD treatments remain classified with "fentanyl and methadone" despite a long safety record. An anonymous drug policy researcher explained the systemic cause: DEA quota systems create artificial shortages that drive patients toward counterfeit street drugs (joshuamandel.com/regulations.gov-comment-browser/AHRQ-202…).
Four Decades of Fighting for Treatment
Leo and Claudia Soucy, both 81, have been fighting regulatory battles since 1986 for their 56-year-old son Brendon, who has severe autism. They seek access to electrical stimulation devices (ESDs) to treat his self-injurious behavior after other approaches failed catastrophically: "Seclusion, restraint and isolation... was a heartbreaking failure," while heavy medications left him "toothless and has tardive dyskinesia." When Massachusetts banned ESDs, "Brendon regressed terribly." Now, the FDA proposes a national ban, forcing them to start over. "It is very unfair that my wife & I at our age 81 must fight & fight for this treatment," they wrote (joshuamandel.com/regulations.gov-comment-browser/AHRQ-202…). The Soucys' personal, decades-long battle mirrors the systemic fight from organizations like Eli Lilly, which challenged CMS's Coverage with Evidence Development (CED) framework for creating access barriers to FDA-approved drugs. Lilly's core argument: "Access should never be contingent on enrollment in a research protocol" (joshuamandel.com/regulations.gov-comment-browser/AHRQ-202…).
Innovation Blocked by Outdated Frameworks
The regulatory system is consistently years, if not decades, behind the technology it governs. Epic detailed how 2016 FDA guidance threatens to regulate "over 150,000 unique CDS tools" because it misinterprets how clinical recommendations work, and how OCR rules still require "human translators" despite superior AI performance (joshuamandel.com/regulations.gov-comment-browser/AHRQ-202…). Zocdoc pointed out that the 1972 Anti-Kickback Statute "predates digital technology" and creates ambiguity that blocks platforms from fixing scheduling inefficiencies that waste 20-30% of provider time (joshuamandel.com/regulations.gov-comment-browser/AHRQ-202…). Anthony Leo, an innovator, challenged CLIA rules requiring credentialed Lab Directors who earn $160,000+ for minimal oversight, creating "de facto licensing cartels" that throttle new AI diagnostic tools (joshuamandel.com/regulations.gov-comment-browser/AHRQ-202…). Kassie Genna, a patient with dysautonomia, proposed a zero-cost innovation: adding one question—"Do you taste saline when your IV is flushed?"—to intake forms. This patient-observed phenomenon could help identify patients needing early autonomic testing, but there's no regulatory pathway for such simple, patient-driven improvements (joshuamandel.com/regulations.gov-comment-browser/AHRQ-202…).
Systemic Dysfunction & Unintended Consequences
The comments reveal a system where rules create perverse incentives and impossible choices. Rural Realities: The National Rural Health Association reported that "190 rural hospitals closed since 2010," with "46% of rural hospitals operating with negative margins" (joshuamandel.com/regulations.gov-comment-browser/AHRQ-202…). Christa Jones, a rural Kansas provider, described how conflict-of-interest rules designed for urban areas create "impossible choices," forcing providers to "divest parts of their organizations" they cannot replace (joshuamandel.com/regulations.gov-comment-browser/AHRQ-202…). Cost-Sharing Burden: The Cleveland Clinic argued for "shifting cost-sharing collection responsibilities to insurers," since they set the amounts. The clinic reported that in 2024, "50% of co-pays for visits and services... were not paid, accounting for $70M in lost revenue" (joshuamandel.com/regulations.gov-comment-browser/AHRQ-202…). Inside the VA: Whistleblower Gregory Romeu, with insider experience, exposed systematic fraud draining "BILLIONS of tax dollars" from VA homeless veteran programs. He stated that VA oversight uses only "Yes Men" to suppress negative findings and argued for transferring housing control to DHHS (joshuamandel.com/regulations.gov-comment-browser/AHRQ-202…).
A Mandate for Change
These 646 comments are not a call for a regulatory bonfire. The child care owner who opposed deregulation saw safety rules as essential business infrastructure (joshuamandel.com/regulations.gov-comment-browser/AHRQ-202…). The parents fighting for ESDs accept court-ordered treatment reviews. What they oppose are regulations that are ineffective, outdated, or misaligned with the goal of patient care. The most striking finding is the consensus. The work for policymakers begins now.
Methodology and Resources
You can explore this data and the tools used to analyze it yourself. To browse the comments by theme and read the full text of every submission cited here, use the user-friendly comment browser (joshuamandel.com/regulations.gov-comment-browser/). For the official, unfiltered source material, visit the AHRQ docket on regulations.gov (www.regulations.gov/document/AHRQ-2025-0001-0001). The open-source code for the entire analysis pipeline is available on GitHub (github.com/jmandel/regulations.gov-comment-browser). For a detailed write-up on the methodology, including the use of the Model Context Protocol (MCP) server and Claude Code, see my post on LinkedIn (www.linkedin.com/pulse/cms-rfi-mcp-now-its-your-turn-anal…). -
Peter Lee reposted
Microsoft
@Microsoft
A fast-moving world requires fast moving tools. Here’s how the speed of AI is helping scientists create world-changing developments that once felt out of reach: news.microsoft.com/source/features/ai/ai-for-science-5-wa…
-
Peter Lee
@peteratmsr
When it comes to uses of AI, the thing that most excites me today is how discovery is being accelerated in medicine, biology, chemistry, materials, and beyond. Great work at great labs everywhere, including (and especially) at Microsoft @MSFTResearch news.microsoft.com/source/features/ai/ai-for-science-5-wa…
-
Peter Lee reposted
Eric Topol
@EricTopol
A clip from my conversation with @peteratmsr on preventing the 3 major age-related diseases
Full episode w/ @DaphneKoller @NoubarAfeyan www.microsoft.com/en-us/research/podcast/how-ai-will-acce… -
Peter Lee
@peteratmsr
Three undeniable geniuses in biomedical research and innovation @DaphneKoller @NoubarAfeyan @EricTopol talk about the impact of generative AI today and in the future, on the "AI Revolution in Medicine, Revisited" podcast @MSFTResearch www.microsoft.com/en-us/research/podcast/how-ai-will-acce…
-
Peter Lee reposted
Noubar Afeyan
@NoubarAfeyan
“Nature doesn’t try to abstract itself in a human-understandable form.”
At the core of my conversation with @peteratmsr was the fact that nature is staggeringly complex. It’s evolved over billions of years, meaning it has a billions-of-years head start on any human attempts to understand it.
But now, AI is giving humans a tool that can process and decipher much more of nature’s complexity. And in doing so, it is totally transforming biomedical research.
The coming together of human, nature, and machine intelligences will create what I call #polyintelligence. It will lead to new medicines, treatments, and technologies that promise to improve the lives of millions of people around the world. It will also challenge humans to become comfortable working alongside AI on problems and solutions that we don’t necessarily understand.
Thanks to Peter and @MSFTResearch for having me on to talk about how AI is changing the way we do research and about how we are building a polyintelligent future at @FlagshipPioneer with @ProFoundTx, @QuotientTx, and @LilaSciences, among others.
Microsoft Research
@MSFTResearch
Daphne Koller, Noubar Afeyan, & Dr. Eric Topol, leaders in AI-driven medicine, explore how AI is reshaping how we diagnose and treat disease—from early-stage drug discovery to clinical care—in Episode 8 of “The AI Revolution in Medicine, Revisited.” www.microsoft.com/en-us/research/podcast/how-ai-will-acce…
-
Peter Lee
@peteratmsr
I always learn so much every time I talk to people like @DaphneKoller, @NoubarAfeyan, and @EricTopol. I think you'll see why in the latest episode of The AI Revolution in Medicine, Revisited @MSFTResearch. Great conversations about how AI is accelerating biomedical discovery.
Microsoft Research
@MSFTResearch
Daphne Koller, Noubar Afeyan, & Dr. Eric Topol, leaders in AI-driven medicine, explore how AI is reshaping how we diagnose and treat disease—from early-stage drug discovery to clinical care—in Episode 8 of “The AI Revolution in Medicine, Revisited.” www.microsoft.com/en-us/research/podcast/how-ai-will-acce…
-
Peter Lee
@peteratmsr
Today in @ScienceMagazine we present BioEmu1.1 @MSFTResearch. It rapidly and accurately emulates equilibrium distributions of protein dynamics at millisecond timescales. Code and datasets available on @Azure Foundry.
Microsoft Research
@MSFTResearch
Today in the journal Science: BioEmu from Microsoft Research AI for Science. This generative deep learning method emulates protein equilibrium ensembles – key for understanding protein function at scale. www.science.org/doi/10.1126/science.adv9817
This media couldn't be displayed. See it on the original post. -
Peter Lee
@peteratmsr
If you are part of a team or organization that is doing advanced materials design, chemical engineering, or biomolecule design, please consider joining the early access program @MSFTResearch for our Skala DFT functional.
Microsoft Research
@MSFTResearch
Microsoft Research invites organizations of all sizes to join the DFT Research Early Access Program to explore the potential of our new Skala functional and accelerate innovation across industries through faster and more accurate density functional theory. forms.cloud.microsoft/Pages/ResponsePage.aspx?id=v4j5cvGG…
This media couldn't be displayed. See it on the original post. -
Peter Lee reposted
Satya Nadella
@satyanadella
Excited to share two advances that bring us closer to real-world impact in healthcare AI:
SDBench introduces a new benchmark that transforms 304 NEJM cases into interactive diagnostic simulations. AI must ask questions, order tests, and weigh costs, mirroring the complexity of real clinical decision-making.
MAI-DxO is a model-agnostic orchestrator that simulates a panel of virtual physicians. It achieves 85.5% diagnostic accuracy—four times that of experienced doctors—while cutting diagnostic costs.
Together, these advances offer a blueprint for how AI can help deliver precision and efficiency in healthcare, and we're looking forward to working with healthcare partners and the entire ecosystem on these advances making a difference.
microsoft.ai/new/the-path-to-medical-superintelligence/ -
Peter Lee
@peteratmsr
I had fun doing a "Take 5" interview for the inaugural edition of Signal, a new Microsoft print magazine. They asked me 5 out of a master list of 50 questions. More info about the magazine at msftsignal@microsoft.com www.linkedin.com/posts/stevecla_microsoft-signal-activity…
-
Peter Lee reposted
Mustafa Suleyman
@mustafasuleyman
We're taking a big step towards medical superintelligence. AI models have aced multiple choice medical exams – but real patients don’t come with ABC answer options. Now MAI-DxO can solve some of the world’s toughest open-ended cases with higher accuracy and lower costs.
-
Peter Lee
@peteratmsr
It was so interesting to hear about the first encounters that @BillGates and @SebastienBubeck had with #genai, and when they realized that it could have a transformative impact on healthcare and biomedical discovery.
Microsoft Research
@MSFTResearch
We’re just scratching the surface of AI’s potential to transform healthcare. @BillGates and @SebastienBubeck discuss their visions for the future and what needs to happen to get there in Ep. 7 of “The AI Revolution in Medicine, Revisited.” www.microsoft.com/en-us/research/podcast/how-ai-is-reshap…
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
In Ep. 7 of “The AI Revolution in Medicine, Revisited,” @BillGates and @SebastienBubeck explore the state of generative AI in medicine, the potential of “medical intelligence” to empower people across healthcare, and what will come next. www.microsoft.com/en-us/research/podcast/how-ai-is-reshap…
-
Peter Lee
@peteratmsr
What is Density Functional Theory and why is the recent advance on solving it via deep learning so important? Chris Bishop @MSFTResearch explains in this short video.
Microsoft Research
@MSFTResearch
Chris Bishop, Technical Fellow and Director of Microsoft Research AI for Science, explains how Microsoft researchers recently achieved a milestone in solving a grand challenge that has hampered scientists and slowed innovation for decades. Watch the video: www.microsoft.com/en-us/research/video/what-is-density-fu…
-
Peter Lee
@peteratmsr
We're excited to reveal this major advance in computational chemistry, an AI model called Skala @MSFTResearch for the exchange-correlation functional, accurate enough, and super fast, for precise in silico predictions of chemical experiments. An early access program is now open.
Microsoft Research
@MSFTResearch
Microsoft researchers achieved a breakthrough in the accuracy of DFT, a method for predicting the properties of molecules and materials, by using deep learning. This work can lead to better batteries, green fertilizers, precision drug discovery, and more. www.microsoft.com/en-us/research/blog/breaking-bonds-brea…
-
Peter Lee
@peteratmsr
Students who have just finished their first year in college: If you end up going eventually to medical school and then into a residency, you will then start your own medical practice in 2035. What will AI and biomedical science be like in that year? I think it'll be amazing...
-
Peter Lee
@peteratmsr
In the latest episode of "The AI Revolution in Medicine, Revisited" @MSFTResearch, @BillGates, @SebastienBubeck, and I discuss the "capability overhang" of AI today, and why this likely means there will be an explosion of uses of AI in medicine soon. www.microsoft.com/en-us/research/podcast/how-ai-is-reshap…
-
Peter Lee
@peteratmsr
Mourning the passing of @atulbutte. I learned so much from the precious few chances I had to be with him. He's someone who always picked up the phone whenever I needed advice, on anything, anytime. A great man who made everyone around him better.
-
Peter Lee
@peteratmsr
What's really happening with AI in healthcare? Really fun and informative conversation with @BillGates and @SebastienBubeck on "The AI Revolution in Medicine, Revisited" @MSFTResearch. We chat about what we are seeing today, and expectations for the future.
Microsoft Research
@MSFTResearch
Technologists @BillGates and @SebastienBubeck join @peteratmsr for Ep. 7 of “The AI Revolution in Medicine, Revisited” and explore how they’re seeing AI in healthcare unfolding and their visions for how the tech will change medical delivery and discovery. www.microsoft.com/en-us/research/podcast/how-ai-is-reshap…
This media couldn't be displayed. See it on the original post. -
Peter Lee
@peteratmsr
Neural rendering (for computer graphics) has been around a long time, but transformers now enable some pretty amazing and highly practical results. A brief and entertaining summary by @twominutepapers of recent work @MSFTResearch. www.youtube.com/watch?v=qYJk9l65eJ8
-
Peter Lee
@peteratmsr
Congratulations @kalikabali! A real no-brainer to recognize you in this way.
kalikabali
@kalikabali
So honoured to be selected by @ForbesIndia India & @Accel amongst the 30 Indians who are leading AI
And as always, this is a recognition of the all the people I have worked with, who have supported me and set me up to do my best
www.linkedin.com/posts/forbes-india_30indianmindsinai-ai-… -
Peter Lee
@peteratmsr
For your weekend listening, a lively pair of conversations with @emollick and @azeem on the real-world impact, today and into the near future, of AI on both the workforce and economics of medicine.
Microsoft Research
@MSFTResearch
With the adoption of generative AI, there are gains to be made at the individual, organizational, & societal levels. In Ep. 6 of "The AI Revolution in Medicine, Revisited," Ethan Mollick & Azeem Azhar explore the opportunities & challenges along the way. www.microsoft.com/en-us/research/podcast/what-ais-impact-…
-
Peter Lee
@peteratmsr
Many doctors and patients like using AI and find it useful. But what is AI's impact on healthcare's workforce and economics? Some great insights from @emollick and @azeem on the latest "The AI Revolution in Medicine, Revisited" @MSFTResearch podcast. www.microsoft.com/en-us/research/podcast/what-ais-impact-…
-
Peter Lee
@peteratmsr
It's no secret that there's a workforce crisis in healthcare today. Will AI help? How and why? With @emollick and @azeem, we explore these questions on the latest episode of "The AI Revolution in Medicine, Revisited" @MSFTResearch www.microsoft.com/en-us/research/podcast/what-ais-impact-…
-
Peter Lee
@peteratmsr
The benefits of AI scribes: (1) doctors spend less time working on documentation; (2) more relevant info gets into the clinical note, and (3) both doctors and patients are more satisfied because they can pay attention to each other, instead of a computer. www.wsj.com/health/healthcare/ai-ambient-listening-doctor…
-
Peter Lee
@peteratmsr
In this podcast, @zakkohane told an interesting story about AI recommending opposing treatments for a patient, depending on whether it was prompted to be a doctor vs an insurer. Sparks important puzzles about AI values, two years after gpt-4 and our book.
Microsoft Research
@MSFTResearch
Generative AI is being used across healthcare. In Ep. 5 of "The AI Revolution in Medicine, Revisited," coauthors @peteratmsr, @goldbergcarey, & Dr. Zak Kohane respond to earlier podcast guests as they revisit their predictions & talk AI’s evolving role. www.microsoft.com/en-us/research/podcast/the-ai-revolutio…
-
Peter Lee
@peteratmsr
Balanced article on the Aurora model @MSFTResearch that handles "not only weather, but also any Earth system with data available ... like air pollution and wave height in addition to weather events like tropical cyclones." www.nytimes.com/2025/05/21/climate/ai-weather-models-auro…
Peter Lee
@peteratmsr
And another one in the Washington Post: www.washingtonpost.com/weather/2025/05/21/microsoft-ai-we…
-
Peter Lee reposted
Ava Amini
@avapamini
The final version of our work on enabling protein-function annotation on new, unseen functions with ProtNote is now published, with code and models available!
📰 doi.org/10.1093/bioinformatics/btaf170
💻 github.com/microsoft/protnote
with @Samir_char, Nate Corley, @SarahAlamdari, @KevinKaichuang
Ava Amini
@avapamini
excited to share & release ProtNote! 🧬📰
ProtNote enables protein-function annotation on new, unseen functions by fusing LLM embeddings of text descriptions w/ protein sequence embeddings
📎biorxiv.org/content/10.1101/2024.10.17.618952v1
🖥️ github.com/microsoft/protnote
led by the incredible @Samir_char -
Peter Lee reposted
Satya Nadella
@satyanadella
Aurora is a new foundation model from @MSFTResearch that goes beyond weather forecasting, delivering faster, more accurate predictions of environmental events. Awesome to see this breakthrough published in @Nature. news.microsoft.com/source/features/ai/microsofts-aurora-a…
-
Peter Lee
@peteratmsr
Here are some great examples of what it's like to use the Healthcare Agent Orchestrator. We're always creating bits and pieces of great agent tech, but what's important here is the seamless ability to bring it together in a practical, real-world humans+AI collaborative setting.
-
Peter Lee
@peteratmsr
Just published in @Nature, AI foundation model Aurora @MSFTResearch. It gives the most accurate predictions, not only of weather but also a wide range of environmental events, from hurricanes and typhoons to air quality and ocean waves. Available on @AzureAIFoundry and @ECMWF
Microsoft Research
@MSFTResearch
A new paper published in Nature explains how Microsoft’s Aurora AI foundation model goes beyond weather forecasting to more accurately predict a range of environmental events, from hurricanes and typhoons to air quality and ocean waves. news.microsoft.com/source/features/ai/microsofts-aurora-a…
-
Peter Lee
@peteratmsr
One of the things I'm most proud of at Microsoft is how we've gotten teams from all corners of the company, with great collaborators, to focus on creating and delivering cutting-edge technology for practical impact on people's health.
Microsoft Research
@MSFTResearch
AI is driving real-world transformation in healthcare. Learn how Microsoft works to empower developers and clinicians around the globe with dynamic new tools, including healthcare-focused agents that can collaborate—with each other and with people: www.microsoft.com/en-us/research/podcast/collaborators-he…
-
Peter Lee reposted
Ava Amini
@avapamini
EvoDiff is now available on @Azure AI Foundry!
ai.azure.com/labs/projects/evodiff 🧬
AI for generating novel proteins to accelerate biological discovery, together with more breakthrough models from @MSFTResearch! 👉 ai.azure.com/labs
Satya Nadella
@satyanadella
5. Microsoft Discovery: We’re bringing together the full tech stack to help speed up science itself. Discovery uses agents to generate ideas, simulate results, and learn. A great example is this promising candidate for a coolant that doesn’t rely on forever chemicals.
-
Peter Lee
@peteratmsr
Imagine AI that is a helpful member of a cancer care team. I know, I know -- this sounds like the typical AI hype you hear from lots of places. But this is real: an AI orchestrator that makes AI useful in the form of an agent that collaborates effectively with a team of doctors.
Satya Nadella
@satyanadella
Stanford Medicine is orchestrating multiple agents to connect medical data, clinical trials, and more - all with the goal of delivering more effective cancer care. Incredible to see in action.
This media couldn't be displayed. See it on the original post. -
Peter Lee
@peteratmsr
If you are in the medical field and you've been confused or skeptical about AI agents, you really need to check this out. Today we released our AI Healthcare Orchestrator Agent. Among other things, it is being used as a truly helpful (and intelligent) participant in tumor board meetings.
Matthew Lungren MD MPH
@mattlungrenMD
We have seen many ways that generative AI can transform healthcare but over-stretched health-IT teams can’t deploy new software systems to surface the latest advancements fast enough.
But... what if developers could meet clinicians where they already work today, building, deploying, and testing AI capabilities that address challenging complex clinical tasks in minutes, not months?
Thats why today we’re announcing the launch of Healthcare Agent Orchestrator bringing the latest cutting edge AI into the hands of healthcare users leveraging Azure AI Foundry and surfacing into M365 enterprise software that health systems globally already use and trust - check out this blog to learn more: www.microsoft.com/en-us/industry/blog/healthcare/2025/05/… -
Peter Lee
@peteratmsr
If you're building with AI, and especially AI agents, Azure AI Foundry is a fantastic resource. Besides integration with HuggingFace and support for end-to-end agent dev, there are tons of @MSFTResearch models and tools, with even highly experimental stuff in Foundry Labs.
Microsoft Azure
@Azure
Build, deploy, and orchestrate a digital workforce of AI agents and multi-agent systems at scale—all from the comprehensive Azure AI Foundry platform: azure.microsoft.com/en-us/blog/azure-ai-foundry-your-ai-a…
-
Peter Lee
@peteratmsr
In the realm of AI agents for doing complex tasks that require multi-step planning and browser use, Magentic-UI from @MSFTResearch is at the cutting edge. I think you'll find it surprisingly useful. Now available in open source on Azure Foundry.
Microsoft Research
@MSFTResearch
Magentic-UI, new from Microsoft Research, is an open-source research prototype of a human-centered AI agent, designed to work with people to complete complex, web-based tasks in real time over a web browser. www.microsoft.com/en-us/research/blog/magentic-ui-an-expe…
-
Peter Lee
@peteratmsr
What have we learned about the reality of "The AI Revolution in Medicine" after 9 interviews with people using and experiencing AI in the real world? @goldbergcarey, @zakkohane, and I discuss and debate on the latest episode of our @MSFTResearch podcast. www.microsoft.com/en-us/research/podcast/the-ai-revolutio…
-
Peter Lee
@peteratmsr
It's an incredible joy to share the great work of our research teams @MSFTResearch, led by great people like Chris Bishop, @dcburger, @nfusi, @jcarlsonmsr, @FrankNoeBerlin, @ecekamar, and so so many more across our and our collaborators' labs.
Microsoft Research
@MSFTResearch
What’s next for AI? @peteratmsr considers the road ahead in his @techreview EmTech talk, focusing on how forthcoming advancements will reshape industries and create new growth. www.microsoft.com/en-us/research/video/mit-technology-rev…
-
Peter Lee
@peteratmsr
"The AI Revolution in Medicine, Revisited" podcast looks into the reality of AI in healthcare. With coauthors @goldbergcarey @zakkohane, we dissect what we've learned so far in my conversations with doctors, techies, patients, researchers, and policy thinkers.
Microsoft Research
@MSFTResearch
In Ep. 5 of “The AI Revolution in Medicine, Revisited,” host @peteratmsr & his coauthors @goldbergcarey & Dr. Zak Kohane revisit their earlier predictions about AI in medicine & respond to insights from frontline clinicians, health leaders, & technologists. www.microsoft.com/en-us/research/podcast/the-ai-revolutio…
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
ADeLe, a new evaluation method, explains what AI systems are good at—and where they’re likely to fail. By breaking tasks into ability-based requirements, it has the potential to provide a clearer way to evaluate and predict AI model performance: www.microsoft.com/en-us/research/blog/predicting-and-expl…
-
Peter Lee
@peteratmsr
It's both humbling and a great honor to be invited to become a part of the @MayoClinic family by joining the amazing people on the Board of Trustees. Mayo is the beacon of light, excellence, and hope in the world of medicine, and I'm eager to do what I can to advance it.
Mayo Clinic
@MayoClinic
Peter Lee, Ph.D., president of Microsoft Research, was elected today to the Mayo Clinic Board of Trustees. Dr. Lee oversees Microsoft Research's 13 global laboratories.
Learn more about Dr. Lee: mayocl.in/4mdW7WI -
Peter Lee
@peteratmsr
On "The AI Revolution in Medicine, Revisited" podcast @MSFTResearch, guest @CALonghurst talks about AI "empathy." E.g., a worried patient might send "nasty" message to the doctor, but AI has infinite ability to respond with in a supportive and useful way. www.microsoft.com/en-us/research/podcast-series/the-ai-re…
-
Peter Lee
@peteratmsr
At #emtechai yesterday I spoke about AI's ability to "learn the languages of nature." Most fun was the lively on-stage chat with Will Douglas Heaven @techreview plus unusually interesting audience questions spanning core tech to biz to philosophy.
-
Peter Lee
@peteratmsr
On the "The AI Revolution in Medicine, Revisited" podcast @MSFTResearch, @RoxanaDaneshjou speaks about the practical problems that bias in AI models pose for healthcare, as well as the potential for AI to detect and mitigate biases and errors by humans. www.microsoft.com/en-us/research/podcast/laws-norms-and-e…
-
Peter Lee
@peteratmsr
On "The AI Revolution in Medicine, Revisited" podcast @MSFTResearch, Vardit Ravitsky talks about #aiexceptionalism -- expecting more transparency and less bias from AI than we expect of ourselves. A fascinating conversation on AI ethics and philosophy. www.microsoft.com/en-us/research/podcast/laws-norms-and-e…
-
Peter Lee
@peteratmsr
I truly appreciated the honor of speaking at the @PittPubHealth commencement yesterday and receiving an honorary doctorate. Loved interacting with so many dedicated students and faculty. Special thanks to Dean @M_Lichtveld for her brilliant leadership.
-
Peter Lee
@peteratmsr
How much energy and water is consumed by datacenters? Not just their operation, but also to produce and install the chips, servers, cooling, and other support equipment, plus transportation and eventual disposal? Now in @Nature, a full lifecycle analysis. news.microsoft.com/source/features/sustainability/microso…
-
Peter Lee
@peteratmsr
The previous episodes of this podcast series featured doctors, tech developers, a patient, and a VC about the reality of AI in healthcare. But what are, and should be, the guidelines for using AI? Provocative conversations with Laura Adams, @VarditRavitsky, and @RoxanaDaneshjou.
Microsoft Research
@MSFTResearch
In Episode 4 of “The AI Revolution in Medicine, Revisited,” guests Laura Adams, Vardit Ravitsky, and Dr. Roxana Daneshjou explore how to responsibly implement AI in medicine, from governance and fairness to reshaping patient relationships: www.microsoft.com/en-us/research/podcast/laws-norms-and-e…
-
Peter Lee reposted
Microsoft Azure
@Azure
One year ago, Microsoft introduced small language models with the release of Phi-3 on Azure AI Foundry.
Now, we're introducing Phi-4-reasoning, Phi-4-reasoning-plus, and Phi-4-mini-reasoning—pushing what's possible with AI. Read the blog: azure.microsoft.com/en-us/blog/one-year-of-phi-small-lang… -
Peter Lee
@peteratmsr
Phi-4-Reasoning is a 14B-parameter reasoning model with performance in certain areas that matches or exceeds much larger models such as Deepseek-R1. Useful for our research @MSFTResearch as well as production systems. Check out the tech report! Model now available on @Azure Foundry and HuggingFace.
Ahmed Awadallah
@AhmedHAwadallah
Introducing Phi-4-reasoning, adding reasoning models to the Phi family of SLMs.
The model is trained with both supervised finetuning (using a carefully curated dataset of reasoning demonstration) and Reinforcement Learning.
📌Competitive results on reasoning benchmarks with much larger top-tier models up to DeepSeek R1
📌 Strong performance on new tests released after data collection (AIME 2025, HMMT)
📌Reasoning transfers/generalizes well to new domains even with only SFT (e.g. k-SAT, Mae Solving, Calendar Planning, etc.)
📌Retains and often significantly improves general-purpose capabilities (e.g. instruction following)
In addition to the models, we are also very excited to share a very detailed technical report with insights on model training and evaluation
Still have a lot to improve especially with context length, coding and tools.
Hope you find the models useful!
A big thanks to the amazing team and to all our partners. -
Peter Lee reposted
Ece Kamar
@ecekamar
Excited to share our latest Phi model, Phi4-reasoning, a small but powerful model that match the performance of much larger reasoning models up to DeepSeek R1. Here is the report for new insights into training reasoning models and evaluating them: www.microsoft.com/en-us/research/wp-content/uploads/2025/…
Ahmed Awadallah
@AhmedHAwadallah
Introducing Phi-4-reasoning, adding reasoning models to the Phi family of SLMs.
The model is trained with both supervised finetuning (using a carefully curated dataset of reasoning demonstration) and Reinforcement Learning.
📌Competitive results on reasoning benchmarks with much larger top-tier models up to DeepSeek R1
📌 Strong performance on new tests released after data collection (AIME 2025, HMMT)
📌Reasoning transfers/generalizes well to new domains even with only SFT (e.g. k-SAT, Mae Solving, Calendar Planning, etc.)
📌Retains and often significantly improves general-purpose capabilities (e.g. instruction following)
In addition to the models, we are also very excited to share a very detailed technical report with insights on model training and evaluation
Still have a lot to improve especially with context length, coding and tools.
Hope you find the models useful!
A big thanks to the amazing team and to all our partners. -
Peter Lee
@peteratmsr
Spending time with @shanselman is one the best perks of being at Microsoft. So it was a real treat to have this fun chat with him for an #acmbytecast.
ACM Education & Learning Center
@acmeducation
In the latest #acmbytecast, @shanselman (@Hanselminutes) hosts ACM Fellow @peteratmsr, President of @MSFTResearch. They talk evolution of computing, GPT4 and medicine, how tech is changing society, & upcoming innovations from @Microsoft.
Listen/sub: learning.acm.org/bytecast/ep68-peter-lee -
Peter Lee
@peteratmsr
What's happening for real when it comes to AI in healthcare? This series of 12 @MSFTResearch podcasts, hosted by me, talks to doctors, patients, investors, policymakers, and more, to find out. We go past the hype to see what's real, and what's possible. www.microsoft.com/en-us/research/podcast-series/the-ai-re…
Peter Lee
@peteratmsr
So far, 3 of the planned 12 episodes have been released. Future episodes will be released every other Thursday.
-
Peter Lee
@peteratmsr
What is AI doing for patients, and what does this mean for new companies building consumer-facing health tech products? In this latest episode of "The AI Revolution in Medicine, Revisited," I talk to patient advocate @ePatientDave and VC @chrissyfarr to find out.
Microsoft Research
@MSFTResearch
In Ep. 3 of “The AI Revolution in Medicine, Revisited,” guests @ePatientDave and @chrissyfarr discuss the growing role of AI in supporting more engaged patients and the evolving relationship between technology, clinicians, and consumers: www.microsoft.com/en-us/research/podcast/the-ai-revolutio…
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
In Ep. 2 of “The AI Revolution in Medicine, Revisited," Microsoft’s Dr. Matthew Lungren & Epic’s Seth Hain discuss how generative AI is impacting clinical documentation at scale today & the tech’s longer-term potential to change how people receive care: www.microsoft.com/en-us/research/podcast/the-ai-revolutio…
-
Peter Lee
@peteratmsr
If you're like me and interested in what is really happening with #genai in medicine, check out the first two episodes of this special series of @MSFTResearch podcasts. Episode 1: doctors, episode 2: devs. Future episodes with patients, regulators, etc. www.microsoft.com/en-us/research/podcast-series/the-ai-re…
-
Peter Lee reposted
Katja Hofmann
@katjahofmann
So cool! You can now play a technical demo of our latest muse model, whamm, directly in copilot labs, in your browser. Try it today: copilot.microsoft.com/wham
haiyan zhang
@haiyan
(1/3) An AI-generated gaming experience—move, shoot, explore—and every frame is created on the fly by an AI world model, responding to player inputs in real-time. Try it here: copilot.microsoft.com/wham
We just launched @Copilot Gaming Experiences, built on cutting-edge AI research from @katjahofmann and @MSFTResearch
It’s a collaboration across research, gaming, and Copilot teams—built to make the research tangible, spark new ideas in model development, and explore how AI might enable the next generation of interactive storytelling.
The original Quake II level remains a powerful expression of the developers’ vision. This AI opens new ways of engaging with experiences and perhaps hints at how future tools could empower dev teams.
Read more technical details here: www.microsoft.com/en-us/research/articles/whamm-real-time…
See below for some capabilities the AI model supports: Shotgun pickup and equip... (1/3) -
Peter Lee
@peteratmsr
Really proud to work for a company whose CEO creates code!
Satya Nadella
@satyanadella
You really know intelligence has been commoditized when CEOs start vibe coding!
-
Peter Lee
@peteratmsr
At the #microsoft50 celebration, with the instigator of @MSFTResearch, Nathan Myrhvold. Amazing to think of how much has happened over 50 years, and since the founding of MSR in 1991.
Peter Lee
@peteratmsr
-
Peter Lee reposted
Satya Nadella
@satyanadella
Three Microsoft CEOs walk into a room on Microsoft’s 50th anniversary … and are interviewed by Copilot!
This media couldn't be displayed. See it on the original post. -
Peter Lee
@peteratmsr
Episode 2 is out! The first episode of this limited podcast series explored what doctors are really doing -- and not doing -- with #genai in medicine. This time, I talk to tech developers about the good, bad, and ugly of making and deploying AI at scale in the clinic.
Microsoft Research
@MSFTResearch
In Episode 2 of “The AI Revolution in Medicine, Revisited,” guests Dr. Matthew Lungren & Seth Hain join Microsoft Research president & host Peter Lee to discuss their experiences creating and deploying generative AI in clinical workflows at scale: www.microsoft.com/en-us/research/podcast/the-ai-revolutio…
-
Peter Lee
@peteratmsr
In my first week @MSFTResearch, almost 15 years ago, I was introduced to deep learning. I knew it was important, but still had no idea just how world-changing this and later AI developments would be. A fun reflection on this and a lot more: news.microsoft.com/source/features/ai/15-milestones-that-…
-
Peter Lee
@peteratmsr
I really appreciate @ePatientDave listening to, and reviewing, the first episode of our limited podcast series @MSFTResearch on AI in medicine. And I appreciate even more that he has agreed to be on an upcoming episode about patient experiences with AI! www.epatientdave.com/2025/03/26/the-ai-revolution-revisit…
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
Generative AI is becoming part of the daily workflow in clinics. In Ep. 1 of “The AI Revolution in Medicine, Revisited,” Dr. Christopher Longhurst & Dr. Sara Murray discuss how new AI tools are being used by clinicians & the early impact they’re having. www.microsoft.com/en-us/research/podcast/the-ai-revolutio…
-
Peter Lee
@peteratmsr
What's the reality of gen AI in the medical clinic today? I learned a lot talking to @CALonghurst and @SaraMurrayMD, who represent the vanguard of doctors using AI personally in their clinical work and also setting the pattern for others. Consider some weekend listening...
Microsoft Research
@MSFTResearch
In Episode 1 of the podcast “The AI Revolution in Medicine, Revisited,” host @peteratmsr, president of Microsoft Research, & guests Dr. @CALonghurst & Dr. @SaraMurrayMD discuss how generative AI is impacting patient care & workflows in healthcare: www.microsoft.com/en-us/research/podcast/the-ai-revolutio…
-
Peter Lee reposted
Jacobs Center for Health Innovation @ UC San Diego
@InnovationUCSDH
So many fantastic insights by JCHI's @CALonghurst and @SaraMurrayMD who joined @peteratmsr from @MSFTResearch to discuss how generative AI is transforming patient care & #healthcare workflows. 🌟
Listen to the podcast ⬇️
www.microsoft.com/en-us/research/podcast/the-ai-revolutio… -
Peter Lee reposted
Katja Hofmann
@katjahofmann
What happens when we train generative AI models to predict everything that happens in a modern video game? In my TED AI talk I share insights from how AI can learn the physics of a game, and how this research could empower game creatives: www.ted.com/talks/katja_hofmann_why_we_re_training_ai_on_…
-
Peter Lee
@peteratmsr
I've been interviewing some really interesting people for a series of podcasts on AI in medicine. I'm learning a lot and I think you will, too. First episode next week on @MSFTResearch's podcast series. An intro to explain what this is all about is here: www.microsoft.com/en-us/research/podcast/the-ai-revolutio…
-
Peter Lee
@peteratmsr
I wrote a book with @goldbergcarey and @zakkohane, about what AI might mean for healthcare, while GPT-4 was still a secret to the world. Two years later, what did we get right? What did we miss? In this new series of podcasts, I talk to people in the real world, to find out.
Microsoft Research
@MSFTResearch
An early encounter with GPT-4 set the tone for the 2023 book “The AI Revolution in Medicine.” @peteratmsr, president of Microsoft Research, revisits the predictions he & his coauthors made about generative AI’s impact on healthcare in a new podcast series. www.microsoft.com/en-us/research/podcast/the-ai-revolutio…
-
Peter Lee
@peteratmsr
It's so great to see the next evolution of Dragon with today’s introduction of Microsoft Dragon Copilot. This is the healthcare industry’s first unified voice AI assistant for clinical workflows.
Satya Nadella
@satyanadella
No one becomes a clinician to do paperwork, but it's becoming a bigger and bigger administrative burden, taking time and attention away from actually treating and supporting patients.
That’s why we’re introducing Microsoft Dragon Copilot, the industry’s first AI assistant for clinical workflow. -
Peter Lee reposted
Elon Musk
@elonmusk
More and more breakthroughs with quantum computing …
Satya Nadella
@satyanadella
A couple reflections on the quantum computing breakthrough we just announced...
Most of us grew up learning there are three main types of matter that matter: solid, liquid, and gas. Today, that changed.
After a nearly 20 year pursuit, we’ve created an entirely new state of matter, unlocked by a new class of materials, topoconductors, that enable a fundamental leap in computing.
It powers Majorana 1, the first quantum processing unit built on a topological core.
We believe this breakthrough will allow us to create a truly meaningful quantum computer not in decades, as some have predicted, but in years.
The qubits created with topoconductors are faster, more reliable, and smaller.
They are 1/100th of a millimeter, meaning we now have a clear path to a million-qubit processor.
Imagine a chip that can fit in the palm of your hand yet is capable of solving problems that even all the computers on Earth today combined could not!
Sometimes researchers have to work on things for decades to make progress possible.
It takes patience and persistence to have big impact in the world.
And I am glad we get the opportunity to do just that at Microsoft.
This is our focus: When productivity rises, economies grow faster, benefiting every sector and every corner of the globe.
It’s not about hyping tech; it’s about building technology that truly serves the world. -
Peter Lee
@peteratmsr
A real, practical device technology that provides stable, logical, qubits. We are entering the quantum era. This is not hype.
Microsoft
@Microsoft
A breakthrough in quantum computing. Majorana 1 brings us closer to harnessing millions of potential qubits working together to solve the unsolvable—from new medicines to revolutionary materials—all on a single chip. #quantumcomputing #quantumready
news.microsoft.com/source/features/ai/microsofts-majorana…This media couldn't be displayed. See it on the original post. -
Peter Lee
@peteratmsr
GenAI can develop a practical understanding of real and virtual worlds, just like it does for language. This has applications in videogame development and more. The AI model, called Muse, is described in detail in @Nature and is coming to Copilot Labs. blogs.microsoft.com/blog/2025/02/19/a-new-level-unlocked/
-
Peter Lee
@peteratmsr
That's right -- AI not only can learn the languages of humans, but also the mechanics of virtual worlds!
Satya Nadella
@satyanadella
If you thought AI-generated text, images, and video were cool, just imagine entire interactive environments like games!
-
Peter Lee
@peteratmsr
In our labs @MSFTResearch, as well as great labs around the world, AI is rapidly enabling advances in biomedical research, some of which are very likely to be transformative breakthroughs for human health. It's a time for nations to double-down on research support, not pull back.
-
Peter Lee reposted
Microsoft AI Frontiers
@ms_aifrontiers
🌻 SOTA agents in your console
MagenticOne (M1) is a 5-agent that can perform a wide variety of file and web related tasks. It achieves good generalization performance on challenging agentic benchmarks. It contains 5 agents:
- WebSurfer
- FileSurfer
- Coder
- Executor
- Ledger-based orchestrator
We've made it ridiculously easy to get M1 running in your console. First install the `magentic-one-cli` package. Then run `m1 --rich <task>` -
Peter Lee reposted
Microsoft AI Frontiers
@ms_aifrontiers
🌻 Saving and Loading Agent State
In many cases it is useful to save AI agent state to disk and load it later. This is particularly useful in a web application where stateless endpoints respond to requests and need to load the state of the application from persistent storage. AutoGen AgentChat provides easy to use methods to accomplish this. -
Peter Lee reposted
Microsoft Research
@MSFTResearch
Researchers pull back the curtain on MatterGen and MatterSim, the cutting-edge tools reshaping how we design and innovate advanced materials. Explore the journey from concept to creation driving these AI-powered technologies. www.microsoft.com/en-us/research/story/ai-meets-materials…
-
Peter Lee reposted
Hoifung Poon
@hoifungpoon
Excited to share that our PubMedBERT paper pioneering exploration of biomedical LLMs has been retrospectively awarded as the Best Paper of the Year by ACM Health: dl.acm.org/journal/health/honors-and-awards.
-
Peter Lee
@peteratmsr
Today in @Nature we present MatterGen, which generates novel materials given prompts of desired chemical, mechanical, electronic, or magnetic properties. @MSFTResearch we see that genAI can learn the languages of nature, not just of humans.
Microsoft Research
@MSFTResearch
Microsoft researchers introduce MatterGen, a model that can discover new materials tailored to specific needs—like efficient solar cells or CO2 recycling—advancing progress beyond trial-and-error experiments. www.microsoft.com/en-us/research/blog/mattergen-a-new-par…
-
Peter Lee reposted
Tian Xie
@xie_tian
Excited to finally announce the publication of MatterGen on Nature. MatterGen represents a new paradigm of materials design with generative AI. We are releasing the training and inference code of MatterGen under MIT license. Look forward to seeing how the community will use the tool and build on top of it.
Microsoft Research
@MSFTResearch
Microsoft researchers introduce MatterGen, a model that can discover new materials tailored to specific needs—like efficient solar cells or CO2 recycling—advancing progress beyond trial-and-error experiments. www.microsoft.com/en-us/research/blog/mattergen-a-new-par…
-
Peter Lee
@peteratmsr
Nice paper. Indeed, not enough attention is paid to AI being a tireless aid in spotting biases, despite it being biased itself. Aside from biases in both clinical work and health datasets, AI might also be helpful in catching biased decision-making in claims adjudication.
Emma Pierson
@2plus2make5
This is joint work with @dmshanmugam, @rajivmovva, Jon Kleinberg, @MonicaNAgrawal, @mdredze, @KadijaFerryman, @judywawira, @jurafsky, @PangWeiKoh, @karen_ec_levy, @m_sendhil, @oziadias, @harini824, and @keyonV!
Freely accessible link here: ai.nejm.org/stoken/default+domain/JIVCWI3KDJIZP3PHWYGY/fu… -
Peter Lee reposted
Microsoft AI Frontiers
@ms_aifrontiers
HUGE: The biggest upgrade to #autogen just dropped!
v0.4 (stable) is finally here.
For detail, checkout the blog below.
Microsoft Research
@MSFTResearch
Announcing AutoGen 0.4, fully reimagined library for building advanced agentic AI systems, developed to improve code quality and robustness. Its asynchronous, event-driven architecture is designed to support dynamic, scalable workflows. Learn more: www.microsoft.com/en-us/research/blog/autogen-v0-4-reimag…
-
Peter Lee
@peteratmsr
We're proud @MSFTResearch to be collaborating with @MayoClinic to advance the state-of-the-art in multimodal AI radiology by bringing AI innovation to real-world clinical use in a way that scales.
Javier Alvarez Valle
@alvarezvalle
Today, I am excited to share our collaboration with Mayo Clinic.
newsnetwork.mayoclinic.org/discussion/mayo-clinic-acceler…
Our collaboration brings together the best experts from AI and medicine to unlock new frontiers in radiology AI
#jpm25 -
Peter Lee
@peteratmsr
Important advance in AI medical imaging by pretraining solely on unimodal biomedical imaging data. This achieves similar or greater performance than state-of-the-art biomedical-language-supervised models on a diverse range of benchmarks.
Javier Alvarez Valle
@alvarezvalle
RAD-DINO: Exploring Scalable Medical Image Encoders Beyond Text Supervision, published today in Nature Machine Intelligence.
👉 Link www.nature.com/articles/s42256-024-00965-w
#microsoft #healthcare #ai #researchpapers -
Peter Lee
@peteratmsr
AutoGen 0.4 is available! We've learned a lot from all the innovative uses you've all be making of AI agents, and so this new version has stronger observability and control, more flexible multi-agent collaboration patterns, and reusable components. Enjoy!
Microsoft Research
@MSFTResearch
Announcing AutoGen 0.4, fully reimagined library for building advanced agentic AI systems, developed to improve code quality and robustness. Its asynchronous, event-driven architecture is designed to support dynamic, scalable workflows. Learn more: www.microsoft.com/en-us/research/blog/autogen-v0-4-reimag…
-
Peter Lee
@peteratmsr
We continue to learn about the big potential for multimodal foundation models in medicine. This paper is a particularly useful distillation of some of the lessons learned.
Hoifung Poon
@hoifungpoon
Happy to conclude 2024 by sharing that BiomedCLIP has been published in NEJM AI (right before Christmas no less :-): ai.nejm.org/doi/full/10.1056/AIoa2400640
-
Peter Lee reposted
Hoifung Poon
@hoifungpoon
Happy to conclude 2024 by sharing that BiomedCLIP has been published in NEJM AI (right before Christmas no less :-): ai.nejm.org/doi/full/10.1056/AIoa2400640
-
Peter Lee reposted
Charles Lamanna
@clamanna
The other day, Marc Benioff made a comment that caught my attention: “Microsoft isn’t using Copilot themselves.” I love his passion. But it’s just not true.
I use Copilot. My team uses Copilot. And pretty much all of Microsoft has been actively using Copilot for two years.
A few months ago, we even compiled several stories about Copilot’s impact at Microsoft: www.microsoft.com/en-us/worklab/ai-is-already-changing-wo…
We built an agent in Copilot Studio for azure.com which was deployed before Agentforce was even announced. It boosted engagement 4.75x and drove a 21.5% increase in conversion rates!
On our sales team, Copilot is providing a 9.4% increase in revenue per seller and a 20% increase in close rates among high users of the tool.
Copilot is benefiting customer support response by helping to resolve cases, with one team closing cases 12% faster.
From a marketing perspective, Copilot has helped us improve the digital customer experience and accelerated campaign creation from 12 weeks to 3 weeks.
Copilot is helping our legal team respond to issues 32% faster. Our HR department has seen a 42% increase in employee self-service answer accuracy when using Copilot.
And those are just a few examples! I always appreciate Marc’s takes and he is always welcome to ask me how else Microsoft is transforming with AI. Or he can ask Copilot 😊.
We have a lot more coming in 2025, and I cannot wait to tell you all about it! This is such an exciting time for us, our customers, partners, and the industry.
Cheers to the New Year and Happy Holidays to everyone. -
Peter Lee reposted
Eric Topol
@EricTopol
A multimodal foundation model for medical images from 15 million image-text pairings ai.nejm.org/doi/full/10.1056/AIoa2400640 @NEJM_AI open-source, open-access paper
@hoifungpoon @MSFTResearch -
Peter Lee reposted
Rohan Paul
@rohanpaul_ai
🔬 One of the key innovative method used in @Microsoft Phi-4 is Pivotal Token Search
Its a significant advancement in LLM inference optimization, fundamentally changing how models process and attend to sequences of tokens
→ The technique dynamically identifies and processes only the most influential tokens in a sequence, reducing computational overhead while maintaining model performance
→ This approach marks a shift from traditional attention mechanisms that process all tokens uniformly, leading to substantial efficiency gains
🧮 Architecture
→ The core innovation lies in its token importance scoring mechanism that evaluates each token's contribution to the final output
→ The algorithm uses a sophisticated pruning strategy to identify "pivotal tokens" - those that carry the most semantic weight and contextual relevance
→ A dynamic threshold mechanism automatically adjusts the selection criteria based on the content complexity
⚡ Performance Implications
→ Early benchmarks show up to 3x speedup in inference time while maintaining 95%+ of the original model's accuracy
→ Memory usage reduction of 40-60% compared to standard attention mechanisms by storing only pivotal token states
→ Particularly effective for long-context processing where traditional attention mechanisms struggle with quadratic complexity
🔄 Implementation Dynamics
→ Integration requires minimal changes to existing transformer architectures, making it highly adaptable
→ The technique can be applied during both training and inference phases, offering flexibility in deployment
→ Custom scoring functions can be tailored for specific domains or tasks, enhancing versatility
Rohan Paul
@rohanpaul_ai
Phi-4 from @Microsoft shows again that smarter data beats bigger models in the AI race.
Achieves remarkable performance by focusing on synthetic data quality and innovative training techniques rather than increasing model size.
phi-4 introduces a 14-billion parameter model focused primarily on high-quality synthetic data generation throughout training.
Key innovation in this Paper 🔧:
→ The model uses multi-agent prompting, self-revision workflows, and instruction reversal to construct diverse datasets.
→ It employs a novel Pivotal Token Search method to identify critical decision points during generation.
→ A two-stage Direct Preference Optimization process refines the model's outputs.
-----
Key Insights 💡:
→ Data quality and targeted synthetic generation can outperform raw parameter scaling
→ Multi-stage training with carefully curated data mixtures improves model reliability
→ Strategic post-training techniques can significantly enhance reasoning capabilities
-----
Key Performance Numbers 📊:
→ Outperforms teacher model GPT-4 on GPQA and MATH benchmarks
→ Achieves 84.8% on MMLU, 80.4% on MATH, and 82.6% on HumanEval
→ Performs comparably to much larger models on reasoning tasks -
Peter Lee reposted
VentureBeat
@VentureBeat
Microsoft’s smaller AI model beats the big guys: Meet Phi-4, the efficiency king venturebeat.com/ai/microsofts-smaller-ai-model-beats-the-…
-
Peter Lee
@peteratmsr
🚀 Phi-4 is here! A small language model that performs as well as (and often better than) large models on certain types of complex reasoning tasks such as math. Useful for us in @MSFTResearch, and available now for all researcher on the Azure AI Foundry! aka.ms/phi4blog
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
We’re excited to be a part of #neurips2024! Explore the future of AI with over 100 groundbreaking papers, including oral and spotlight sessions, on reinforcement learning, advanced language model training, and multilingual, culturally inclusive benchmarks: www.microsoft.com/en-us/research/story/microsoft-at-neuri…
-
Peter Lee
@peteratmsr
The #neurips2024 conference is a busy time for us @Microsoft @MSFTResearch. You can learn more about our activities at the conference and chat with a chatbot trained on the over 100 papers we'll be presenting. Enjoy! www.microsoft.com/en-us/research/microsoft-at-neurips-2024/
-
Peter Lee reposted
Frank Noe
@FrankNoeBerlin
Super excited to preprint our work on developing a Biomolecular Emulator (BioEmu): Scalable emulation of protein equilibrium ensembles with generative deep learning from @MSFTResearch AI for Science.
#ml #ai #neuralnetworks #biology #ai4science
www.biorxiv.org/content/10.1101/2024.12.05.626885v1 -
Peter Lee
@peteratmsr
MatterSim @MSFTResearch is useful for metals, oxides, sulfides, and halides; in crystals, amorphous solids, and liquids. Real-world conditions from 0 to 5,000 Kelvin and from standard to 10,000,000 atmospheres. Yet more evidence that genAI can learn the languages of nature.
Microsoft Research
@MSFTResearch
We're thrilled to announce the release of MatterSimV1-1M and MatterSimV1-5M on GitHub. These cutting-edge models are now available for researchers, developers, and innovators to explore, customize and build upon. github.com/microsoft/mattersim
-
Peter Lee reposted
Dominic King
@Dominic1King
Very excited to be joining Microsoft AI.
Health is a critical use case for AI. Patients and the public continue to face multiple pain points on their path to better health.
We’re now at an inflection point where AI capabilities are already assisting with faster and more accurate diagnoses and helping people better navigate the complexity of modern healthcare.
For this technology to make a scaled difference in improving lives we’re going to need input from clinicians, technologists, designers and importantly patients with lived experience. Get in touch if you think you can help! -
Peter Lee reposted
Microsoft Research
@MSFTResearch
We're thrilled to announce the release of MatterSimV1-1M and MatterSimV1-5M on GitHub. These cutting-edge models are now available for researchers, developers, and innovators to explore, customize and build upon. github.com/microsoft/mattersim
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
Discover the best prompting techniques for the OpenAI o1-preview model, which achieved unprecedented accuracy on medical benchmarks. www.microsoft.com/en-us/research/blog/advances-in-run-tim…
-
Peter Lee reposted
Nature Methods
@naturemethods
New from Wei, Poon, and Wang--BiomedParse is a biomedical foundation model that can jointly conduct segmentation, detection and recognition across nine imaging modalities. www.nature.com/articles/s41592-024-02499-w
-
Peter Lee
@peteratmsr
So nice to see @naturemethods publish this review of our BiomedParse @MSFTResearch paper on medical imaging. I like that the authors tried their own images and were impressed with the results -- more reviewers should do that! www.nature.com/articles/s41592-024-02519-9
-
Peter Lee reposted
Hoifung Poon
@hoifungpoon
BiomedParse is now live at Nature Methods: www.nature.com/articles/s41592-024-02499-w
BiomedParse adopts a novel image parsing framework for holistic image analysis. It is a single foundation model that attains state-of-the-art performance across 64 major object types in nine imaging modalities. -
Peter Lee
@peteratmsr
.@MSFTResearch and collaborators are proud to present BiomedParse, a practical and true foundation model for image analysis across 9 different medical imaging modalities. In @naturemethods. www.nature.com/articles/s41592-024-02499-w
Peter Lee
@peteratmsr
Our blog post about this: www.microsoft.com/en-us/research/blog/biomedparse-a-found…
-
Peter Lee reposted
Microsoft AI Frontiers
@ms_aifrontiers
To our collaborators & community:
We’ve seen questions about AutoGen forks/clones vs. the official project. Here's a summary of the latest. Please share with others.
- The official repo is github.com/microsoft/autogen.
- We're actively working on AutoGen v0.2, with v0.4 innovations like AutoGen-Core, -AgentChat, -Bench, and -Studio 🚀
- Development remains open-source under MIT, and contributions are welcome.
- No Microsoft employee is affiliated with AG2, a fork of Microsoft/AutoGen; We recognize varying endeavors/interests in the community and wish them well.
⚠️ Please note: The current pyautogen package isn’t affiliated with Microsoft AutoGen, and admin access is blocked for us.
Discord access is also limited for us -- we do not have viewing or posting rights, but you can connect via GitHub Discussions, email (autogen@microsoft.com), or our Wed office hours (10 AM PST).
Thank you for shaping this collaborative project! Stay tuned for more updates soon. 🙌
Read the article below for full details 👇
#autogen #opensourceagents #opensource
Microsoft AI Frontiers
@ms_aifrontiers
-
Peter Lee reposted
Christopher A. Longhurst, MD
@CALonghurst
I often get asked for resources to learn about next-generation AI in healthcare. One great place to start is 𝐓𝐡𝐞 𝐀𝐈 𝐑𝐞𝐯𝐨𝐥𝐮𝐭𝐢𝐨𝐧 𝐢𝐧 𝐌𝐞𝐝𝐢𝐜𝐢𝐧𝐞 by @peteratmsr @zakkohane @goldbergcarey
-
Peter Lee reposted
Microsoft AI Frontiers
@ms_aifrontiers
More and more users are testing #autogen Magentic-One on tasks they care about and realizing its potential.
In a recent new video, YouTuber @AllAbtAI tested Magentic-One on 5 real-world complex tasks that they care about. And out of the box it nailed 4/5 of the tasks. In the video, @AllAbtAI concluded:
"Overall, a very enjoyable experience and I think there is some potential here in this multi-agent system for solving complex tasks."
We are excited for what folks built with it and the open-source agents. A neat, easy to use developer API for Magentic-One is also in progress, stay tuned...
#autogen #magenticone #opensource #ossagents -
Peter Lee
@peteratmsr
Congrats to all our PowerScribe @Microsoft colleagues for winning a Minnies Award @AuntMinnie for Best New Radiology Software! www.auntminnie.com/clinical-news/article/15708058/minnies…
-
Peter Lee reposted
Ava Amini
@avapamini
wet-lab validation of proteins designed by EvoDiff!
- generated proteins express and are structured
- inpainted, disordered mitochondrial targeting signals function in yeast
- sequence-space motif scaffolding designs functional binders to Ca2+, MDM2
www.biorxiv.org/content/10.1101/2023.09.11.556673v2 -
Peter Lee reposted
Saleema Amershi
@SaleemaAmershi
In some better news… We're hiring! 📢
Join us in creating reliable, useful, and safe #ai #agents and systems.🤖 If you🩷working with awesome people in a vibrant, collaborative research lab💻, check out our opportunities below! 👇
#microsoftresearch #aifrontiers @pyautogen -
Peter Lee reposted
Microsoft AI Frontiers
@ms_aifrontiers
🧵#autogen Magentic-One is exhibits strong generalist performance, you can use it for so many open-ended, every day tasks. Here is a thread of videos for a wide variety of every day tasks! To learn more about Magentic-One, checkout the original release tweet below.
task: find missing citations in a paper
Microsoft AI Frontiers
@ms_aifrontiers
📢Introducing Magentic-One, a generalist 5-agent multi-agent system for solving open-ended web- and file-based tasks. 🤖🤖🤖🤖🤖
Magentic-One represents a significant step towards agents that can complete tasks that people encounter in their daily lives and can achieve strong performance and generalization across THREE challenging agentic benchmarks: GAIA, WebArena, and Assistant.
We are releasing an open-source implementation in #autogen, our popular open-source framework for developing multi-agent applications. Checkout the technical report, blog, and implementation below 👇
@MSFTResearch @Microsoft
#autogen #agents #opensource
1/ -
Peter Lee reposted
Microsoft Research
@MSFTResearch
Microsoft Research’s AI2BMD, an AI-based system that efficiently simulates a wide range of proteins in all-atom resolution, can advance drug discovery and biomolecular research. www.microsoft.com/en-us/research/blog/from-static-predict…
-
Peter Lee
@peteratmsr
Now in @Nature, AI @MSFTResearch that simulates proteins with more than 10,000 atoms with quantum accuracy, orders of magnitude faster than ever before. AI2BMD's results match those of wet-lab experiments, enabling new biomedical research advances. www.nature.com/articles/s41586-024-08127-z
Peter Lee
@peteratmsr
For a lighter read than the paper itself, here is a blog post from the research team. www.microsoft.com/en-us/research/blog/from-static-predict…
-
Peter Lee reposted
Microsoft AI Frontiers
@ms_aifrontiers
📢Introducing Magentic-One, a generalist 5-agent multi-agent system for solving open-ended web- and file-based tasks. 🤖🤖🤖🤖🤖
Magentic-One represents a significant step towards agents that can complete tasks that people encounter in their daily lives and can achieve strong performance and generalization across THREE challenging agentic benchmarks: GAIA, WebArena, and Assistant.
We are releasing an open-source implementation in #autogen, our popular open-source framework for developing multi-agent applications. Checkout the technical report, blog, and implementation below 👇
@MSFTResearch @Microsoft
#autogen #agents #opensource
1/ -
Peter Lee reposted
Kevin K. Yang 楊凱筌
@KevinKaichuang
We made sequences generated by EvoDiff in the lab!
- Unconditional generations express, fold, and exhibit expected secondary structure elements
- We inpaint functional intrinsically-disordered regions and scaffold binders -
Peter Lee reposted
David Shaywitz
@DShaywitz
Thank you @ldtimmerman for reupping latest piece highlighting viral adoption by early career physicians of clinical AI tool and reminding us in pharma R&D that nothing changes until suddenly it does… -> timmermanreport.com/2024/10/and-just-like-that-what-the-v… @vijaypande @JTLonsdale @wolfejosh @peteratmsr
-
Peter Lee
@peteratmsr
Nice to see good uptake of OmniParser, from the @MSFTResearch work on aligning genAI with agentic capabilities. venturebeat.com/ai/microsofts-agentic-ai-tool-omniparser-…
-
Peter Lee reposted
Thomas Dohmke
@ashtom
I am so proud of our new breakthrough: GitHub Spark. The Dumbledore of software development. 🪄 With Spark, we are closer than ever before to realizing 1 billion developers on GitHub.
Oh, and Spark is multi-model too.
githubnext.com/projects/github-spark -
Peter Lee reposted
Microsoft AI Frontiers
@ms_aifrontiers
🚀 AutoGen 0.4 is a from-the-ground-up rewrite, bringing new features and improvements. Learn more about its journey and future plans in our latest blog post! We're gathering community feedback before the official release. 📣 For prior users, AutoGen 0.2 is still actively maintained at the 0.2 branch with the autogen-agentchat~=0.2 PyPi package. #autogen #ai
Microsoft AI Frontiers
@ms_aifrontiers
We are excited to announce a preview of the new architecture of AutoGen (coming in v0.4).
To learn more, see
Blog: microsoft.github.io/autogen/blog/2024/10/02/new-autogen-a…
Pull request: github.com/microsoft/autogen/pull/3600
Come help us shape the future of AutoGen! -
Peter Lee reposted
Microsoft AI Frontiers
@ms_aifrontiers
New User Guide: Agents and Handoff in #autogen
@OpenAI recently introduced Swarm, a lightweight framework for multi-agents with a "handoff" pattern. We show how to implement the same functionality in AutoGen while benefitting from the flexibility and scalability of AutoGen.
To complete a task, the handoff pattern allow an agents to transfer the conversation (or computation) to another agent, sort of like agents passing a talk pillow.
You can easily implement this (and many other) pattern(s) in AutoGen...
1/2 -
Peter Lee
@peteratmsr
The WHO and others estimate a global nursing shortage in the millions by 2030. In partnership with a slew of some of the best healthcare orgs, @Microsoft is deploying AI tech that allows nurses to focus less on paperwork and more on their patients. news.microsoft.com/2024/10/10/microsoft-expands-ai-capabi…
-
Peter Lee
@peteratmsr
A major step forward for practical, state-of-the-art, AI medical imaging. Today we launched a collection of cutting-edge multimodal medical imaging foundation models in the @Azure AI model catalog. news.microsoft.com/2024/10/10/microsoft-expands-ai-capabi…
-
Peter Lee
@peteratmsr
Huge congratulations to Geoff Hinton, @HopfieldJohn , David Baker, @demishassabis, and John Jumper on their Nobel Prizes! This is just massive and well deserved. We both collaborate and compete with each other, but either way the influence on so much of what we do today is huge.
-
Peter Lee
@peteratmsr
LLMs amaze us with their ability to learn from the language of humans. Equally amazing, if not more so, is the ability of AI to learn from the language(s) of nature -- atoms, molecules, fluids, weather, and more. A tiny view of our work @MSFTResearch: news.microsoft.com/source/features/ai/from-forecasting-st…
-
Peter Lee reposted
Tian Xie
@xie_tian
Great to see our models MatterGen, MatterSim, and Aurora highlighted at this Microsoft source article, on the same day ML won Nobel Prize in physics. AI is revolutionizing scientific discovery. It really feels like a moment for a radical change in science. Very excited to be part of this.
Microsoft
@Microsoft
AI foundation models are shaking up research, making breakthroughs in weather forecasting, materials design, and drug development faster than ever.
Learn about how these tools are pushing the boundaries and shaping the future of research: news.microsoft.com/source/features/ai/from-forecasting-st… -
Peter Lee reposted
Ava Amini
@avapamini
excited to share our new review on AI advances in sequence-structure co-generation for protein design!
📰 arxiv.org/abs/2410.01773
with an incredible team: @WChentong @SarahAlamdari @cdomingoenrich @KevinKaichuang
Kevin K. Yang 楊凱筌
@KevinKaichuang
Happy to release a concise review of recent progress in sequence-structure co-generation for protein design.
with our amazing intern @WChentong + @SarahAlamdari Carles Domingo-Enrich and @avapamini -
Peter Lee reposted
Kevin K. Yang 楊凱筌
@KevinKaichuang
Happy to release a concise review of recent progress in sequence-structure co-generation for protein design.
with our amazing intern @WChentong + @SarahAlamdari Carles Domingo-Enrich and @avapamini -
Peter Lee reposted
Microsoft AI Frontiers
@ms_aifrontiers
We are excited to announce a preview of the new architecture of AutoGen (coming in v0.4).
To learn more, see
Blog: microsoft.github.io/autogen/blog/2024/10/02/new-autogen-a…
Pull request: github.com/microsoft/autogen/pull/3600
Come help us shape the future of AutoGen! -
Peter Lee reposted
Gagan Bansal
@bansalg_
🚨🚨Announcing preview of the new architecture of #autogen (v0.4)
Our team has been working hard at @MSFTResearch @Microsoft on this release! And we are looking forward to continue building it with our open-source community. For more info, checkout the links to blog and pull request in the announcement below.
Microsoft AI Frontiers
@ms_aifrontiers
We are excited to announce a preview of the new architecture of AutoGen (coming in v0.4).
To learn more, see
Blog: microsoft.github.io/autogen/blog/2024/10/02/new-autogen-a…
Pull request: github.com/microsoft/autogen/pull/3600
Come help us shape the future of AutoGen! -
Peter Lee reposted
Microsoft Azure
@Azure
🗣️ Unlock the power of voice with the GPT-4o Audio Realtime API, now in public preview on Azure OpenAI Service—enabling real-time, human-like conversations and multilingual mastery for your apps: azure.microsoft.com/en-us/blog/announcing-new-products-an… #azureai
-
Peter Lee reposted
Mustafa Suleyman
@mustafasuleyman
The all new Copilot was interviewed on Good Morning America. Pretty cool to see an AI on TV 🎙️
www.goodmorningamerica.com/news/video/microsoft-unveils-r… -
Peter Lee reposted
Oriol Servia
@OriolServia
It was a one-off helmet, @TroyLeeDesigns was like, I want to put your name in BIG letters Supercross style, and so he did.
ChampCar4Ever
@ChampCar4Ever
It's @OriolServia driving the #7 Forsythe Racing Panoz DP-01 Champ Car at the 2007 Grand Prix of Portland.
#indycar -
Peter Lee reposted
Aaditya Ura
@aadityaura
I also analyzed Smaller models and compared them with phi-3 to explore how small models perform in the medical field. (Without Fine-tuning)
🥇 Phi-3-4k: Top performer with 68.93% average score thanks to @SebastienBubeck
- Excels in MMLU College Biology (84.72%)
- Strong performance in MMLU Clinical Knowledge (75.85%)
🥈 Meta-Llama-3.2-3B-Instruct
- Second place with 64.15% average
- Best in MMLU College Biology (70.83%)
- Solid performance in PubMedQA (70.6%)
🥉 Meta-Llama-3.2-3B
- Third place with 60.36% average
- Strongest in MMLU College Biology (63.89%)
- Good performance in PubMedQA (72.8%)
#smalllanguagemodels #slm #microsoft #phi3 -
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
Really nice to see, we did give special care to healthcare in building phi-3, owing to previous collaboration with @peteratmsr and @zakkohane :-)
Aaditya Ura
@aadityaura
I also analyzed Smaller models and compared them with phi-3 to explore how small models perform in the medical field. (Without Fine-tuning)
🥇 Phi-3-4k: Top performer with 68.93% average score thanks to @SebastienBubeck
- Excels in MMLU College Biology (84.72%)
- Strong performance in MMLU Clinical Knowledge (75.85%)
🥈 Meta-Llama-3.2-3B-Instruct
- Second place with 64.15% average
- Best in MMLU College Biology (70.83%)
- Solid performance in PubMedQA (70.6%)
🥉 Meta-Llama-3.2-3B
- Third place with 60.36% average
- Strongest in MMLU College Biology (63.89%)
- Good performance in PubMedQA (72.8%)
#smalllanguagemodels #slm #microsoft #phi3 -
Peter Lee
@peteratmsr
There were so many creative ideas about the future of interaction between humans and machines in this period. That intensity is back in full force now with new tools like generative AI...
Windows On Windows
@wowstartsnow
Think you know @Surface? Think again. The original Surface (2008) was not a tablet in the traditional sense, but actually a table. 👀 Conceptualised in 2001 at @MSFTResearch, upon release it ran Windows Vista (2006) & could recognise & interact with objects placed on the display.
-
Peter Lee
@peteratmsr
Checked in on the Microsoft Hackathon, put on every year by the Microsoft Garage @MSFTResearch. Over 58K employees participating this week, including over half of Microsoft Research, on over 17K projects! Special kudos to partner companies who joined in with us this year.
-
Peter Lee reposted
Ahmed Awadallah
@AhmedHAwadallah
Check out Windows Agent Arena, it is enabling us to to test agentic capabilities beyond Web tasks and iterate faster by reducing the time for running full evaluation to minutes.
Great to see best preforming agent using OmniParser (arxiv.org/abs/2408.00203) and GPT-4V too
Rogerio Bonatti
@rogerio_bonatti
AI assistants have changed the way we use computers to work and search for information. As LLMs become more powerful, what’s next? Agents.
Excited to introduce Windows Agent Arena, a benchmark for evaluating AI models that can reason, plan and act to solve tasks on your PC.
🔗Blog: www.microsoft.com/applied-sciences/projects/windows-agent…
🌐Webpage: microsoft.github.io/WindowsAgentArena/
📃Paper: arxiv.org/abs/2409.08264
💻Code: github.com/microsoft/WindowsAgentArena
🧵👇 -
Peter Lee reposted
Mustafa Suleyman
@mustafasuleyman
Welcome @NandoDF !!
Super excited for you to join us on the amazing journey ahead 🚀
Nando de Freitas
@NandoDF
I’ve joined @Microsoft AI to advance the frontier of large scale multimodal AI research and to build products for people to achieve meaningful goals and dreams.
The MAI team is small, but well resourced and ambitious. We are now looking for exceptional ICs, who like to ship.
If you you’re interested in multimodal AI, both recognition and generation, love to collaborate and empower others, believe in diversity and inclusion, have a growth mindset, and want to impact the future of AI in a positive and profound way, please message me directly.
I believe this is a rare and unique opportunity to join a new AI team that will shape the future.
@black_in_ai @WiMLworkshop @_LXAI -
Peter Lee reposted
Microsoft Azure
@Azure
🎉 Introducing o1-preview and o1-mini, our newest OpenAI Models on Azure. The o1 series enables complex coding, math reasoning, brainstorming, and comparative analysis capabilities. Learn more: azure.microsoft.com/en-us/blog/introducing-o1-openais-new… #azureai
-
Peter Lee
@peteratmsr
Microsoft’s journey in quantum aims for scalable quantum computing. Congratulations to @jlzander, @krystasvore, the folks on @MSFTQuantum, and collaborators at @QuantinuumQC and @atomcomputing for hitting a major milestone. This is a big step towards quantum computing reality.
Satya Nadella
@satyanadella
As we make progress on our journey to large-scale quantum computing, I’m thrilled to share we have achieved the most performant logical qubits on record – and we’re bringing a new generation of reliable quantum hardware to customers.
-
Peter Lee reposted
Microsoft
@Microsoft
Check your registration status with Microsoft’s partner TurboVote and explore your voting options: microsoft.turbovote.org?ocid=FY25_soc_omc_br_X_TurboVote
-
Peter Lee
@peteratmsr
I'm honored to be named here. In truth, as in all such things, any positive impact I've made is the result of the teamwork of a lot of great people I have the good fortune to work with. Thank you!
Ayla Ellison
@AylaEllison
I'm honored to share the second edition of the Fierce 50, highlighting the innovators and leaders driving meaningful change in healthcare, pharma and biotech. www.fiercepharma.com/collection/fierce-50-2024
-
Peter Lee
@peteratmsr
Season opener for the Seahawks, and the new @SoundTransit light rail from Lynnwood is absolutely packed with fans going to the game. So much better than fighting I-5 traffic and stadium parking!
-
Peter Lee reposted
Adam Foster
@AdamEFoster
Curious what I've been working on since joining AI for science? With an incredible, multidisciplinary team we've studied the 'wave function -> election density' marginalisation using score matching & NCE. This has some big gains over older Gaussian-based methods.
Frank Noe
@FrankNoeBerlin
Introducing Neural Electron Real-space Density (NERD) models! 🧠 You’ve solved the electronic Schrödinger equation using PauliNet or Psiformer - what next? Important properties come from the 1-electron density (the marginal #machinelearning)
arxiv.org/abs/2409.01306 -
Peter Lee reposted
Lixue Cheng
@SherryLixueC
🥳Happy to share our work on extracting extremely accurate electron densities (and many related properties) from wavefunctions. Feel honored to work with all these talented colleagues in @MSFTResearch AI for Science Lab
Frank Noe
@FrankNoeBerlin
Introducing Neural Electron Real-space Density (NERD) models! 🧠 You’ve solved the electronic Schrödinger equation using PauliNet or Psiformer - what next? Important properties come from the 1-electron density (the marginal #machinelearning)
arxiv.org/abs/2409.01306 -
Peter Lee reposted
Chin-Wei Huang
@chinwei_h
Check out this cool work by my colleagues at @MSFTResearch AI4Science, using denoising score matching to distill one-electron densities from wavefunctions!
Paola Gori-Giorgi
@paolagorigiorgi
Very excited to share our work on highly accurate densities with neural networks! @MSFTResearch #compchem
Frank Noe
@FrankNoeBerlin
Introducing Neural Electron Real-space Density (NERD) models! 🧠 You’ve solved the electronic Schrödinger equation using PauliNet or Psiformer - what next? Important properties come from the 1-electron density (the marginal #machinelearning)
arxiv.org/abs/2409.01306 -
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
This is the coolest "application" of LLMs I have seen in a while: instead of using an LLM to solve a problem set, flip the table and ask the student to TEACH the LLM how to solve the problem set. Really lovely.
Thx to Jordan Hoffmann for sharing this :-). www.youtube.com/watch?v=om7VpIK90vE -
Peter Lee reposted
ChampCar4Ever
@ChampCar4Ever
It's @OriolServia driving the #22 Sigma Autosport Lola B01/00 Ford Cosworth XF at the Milwaukee Mile in 2001.
#indycar -
Peter Lee
@peteratmsr
I've seen this coming for a while now, but I'm still so impressed by Epic's use of genAI to ease cognitive burdens for doctors and nurses, and improve quality and safety. It so inspires our teams at Microsoft to work hard to help them. www.fiercehealthcare.com/ai-and-machine-learning/epic-tou…
-
Peter Lee
@peteratmsr
You know you're on vacation when it's almost 6pm and your phone still has 87% charge.
-
Peter Lee reposted
Nicolò Fusi
@nfusi
We’re excited to share our latest work with @paige_ai on (vision) foundation models of pathology. In this study, we introduce Virchow2 and Virchow2G, significantly increasing the scale of our previous work both in terms of data and parameters, with Virchow 2G reaching 1.85B parameters trained on 3.1M whole slide images. Thread below:
-
Peter Lee reposted
Bob Wachter
@Bob_Wachter
I'm increasingly feeling like GPT-4 (& beyond) will be healthcare's Hemingway moment when it comes to #ai.
When asked how someone becomes bankrupt, EH famously said, "Two ways. Gradually, then suddenly."
GPT has tons of implementation challenges, but use cases now seem very real. -
Peter Lee
@peteratmsr
Some home maintenance tasks, such as putting on a new roof, or re-doing the flooring, or replacing kitchen cabinets, etc, are pretty consequential and somewhat risky, but none more so than re-doing the mesh wifi system for the house and yard..
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
Check out our PNNL-Microsoft collaborative manuscript which answers these questions through accelerated materials discovery using HPC and AI, now accepted and online at JACS. pubs.acs.org/doi/10.1021/jacs.4c03849
-
Peter Lee reposted
NEJM AI
@NEJM_AI
FHIR-GPT achieved an exact match rate of over 90% in converting free-text input to their associated FHIR resources, significantly surpassing the performance of existing methods. Read the full Case Study: nejm.ai/3SDd5Rd
#healthit -
Peter Lee
@peteratmsr
People are building so many cool things with Phi-3, now with fine-tuning and very nice improvements in instruction following and overall quality. Also expecting big upticks in GPT-4o-mini. Llama-3.1-405B, and Mistral Large2. azure.microsoft.com/en-us/blog/announcing-phi-3-fine-tuni…
-
Peter Lee reposted
Satya Nadella
@satyanadella
Together with @MassGenBrigham, @uwsmph, and @UWHealth, we're exploring how multimodal AI foundation models can accelerate solving some of the biggest challenges in radiology.
Microsoft News and Stories
@MSFTnews
Microsoft is collaborating with @MassGenBrigham & @UWHealth to transform medical imaging with multimodal AI foundation models, helping improve experiences and workflow efficiency across the radiology ecosystem. news.microsoft.com/source/2024/07/24/microsoft-collaborat…
-
Peter Lee reposted
-
Peter Lee
@peteratmsr
Congratulations for this technical achievement, @paige_ai. I think it will prove to be seminal and a major step towards important clinical capabilities in medical imaging. We are so very pleased and proud @MSFTResearch to work with you on this.
Paige
@paige_ai
We are honored to announce that Paige's Foundation Model Virchow, developed in collaboration with @MSFTResearch, has been published in @NatureMedicine. This milestone study showcases Virchow's state-of-the-art performance
www.businesswire.com/news/home/20240723996420/en/Major-St… -
Peter Lee reposted
Microsoft Azure
@Azure
🎉 Introducing GPT-4o mini, OpenAI's new flagship model now available on Azure AI. GPT-4o mini can reason across vision and text at faster speeds and lower costs. Learn more here: azure.microsoft.com/en-us/blog/openais-fastest-model-gpt-…
-
Peter Lee
@peteratmsr
Congratulations, @avapamini! Recognizing you as the @VentureBeat Women in AI Rising Star is so well-deserved and, frankly, a no-brainer. Such an honor to be associated with you.
Microsoft Research
@MSFTResearch
Senior researcher Ava Amini is this year's @VentureBeat Women in AI Rising Star, an honor that recognizes a woman in the early stages of her AI career. Ava's research is pioneering new technologies for precision medicine by interfacing AI and biology. venturebeat.com/ai/announcing-the-winners-of-venturebeats…
-
Peter Lee
@peteratmsr
Even though it's called "generative" AI, the current models are often better at review, critique, and summarization than they are at generation.
Isaac Kohane
@zakkohane
How do LLM's review articles compared to humans? Impressive study (retrospective and prospective) @nejm_AI ai.nejm.org/doi/full/10.1056/AIoa2400196 Most interesting nugget: LLM much more focused on implications of research and human on novelty. Foreshadows interesting human augmentation in peer review.
-
Peter Lee
@peteratmsr
Wow this was fun, and a lot of money raised for a good cause. Also such an honor to share the racetrack with so many great pro drivers and interesting, generous people. Kudos to @iRacing for putting on such a nice event.
Oriol Servia
@OriolServia
Let’s go! @iRacing
www.youtube.com/live/BM6H5NyM6Eo -
Peter Lee reposted
Chris Stewart
@chrisstewart
Wednesday Night, 8 PM ET, Live on YouTube! Let's raise money for a great cause! @peteratmsr @OriolServia @iRacing @iRacingMyers
youtube.com/@chrisstewarttv -
Peter Lee reposted
Max Verstappen
@Max33Verstappen
Please go check this out, the iRacing MS Charity Race 🙏
www.iracing.com/iracing-and-national-ms-society-reunite-f… -
Peter Lee
@peteratmsr
Definitely looking forward to participating in this race @iRacing with the Improper Racing team! It will be fun and a great way to raise money for the @mssociety.
Chris Stewart
@chrisstewart
I'll be in this race, in the broadcasted split! One week from tonight. Hoping to raise a lot of money for a great cause with @peteratmsr and @OriolServia on behalf of Improper Racing!
Max Verstappen
@Max33Verstappen
Please go check this out, the iRacing MS Charity Race 🙏
www.iracing.com/iracing-and-national-ms-society-reunite-f… -
Peter Lee reposted
Han Yang 杨涵
@HanYangChicago
Exciting news! MatterSim has officially claimed the #1 spot on the #matbench Discovery leaderboard matbench-discovery.materialsproject.org.
MatterSim is a SOTA #machinelearning force field for materials simulation, discovery and more arxiv.org/abs/2405.04967. @luzihen
#ai4science @MSFTResearch -
Peter Lee reposted
Tian Xie
@xie_tian
MatterSim is now the SOTA in matbench discovery! It is becoming more clear that data scaling, especially diverse data, is the key to broad applicability of universal MLFFs. Check out paper in more detail: arxiv.org/abs/2405.04967
Han Yang 杨涵
@HanYangChicago
Exciting news! MatterSim has officially claimed the #1 spot on the #matbench Discovery leaderboard matbench-discovery.materialsproject.org.
MatterSim is a SOTA #machinelearning force field for materials simulation, discovery and more arxiv.org/abs/2405.04967. @luzihen
#ai4science @MSFTResearch -
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
It took more than a year but maybe Sparks experiments are finally reproducible 😁
Piotr Mazurek
@tugot17
The unicorn that was promised by @SebastienBubeck in Sparks of AGI is finally delivered by Claude 3.5, absolutely amazing 🥰
-
Peter Lee
@peteratmsr
First, I'm so pleased that DAX Copilot was selected as the #1 ambient clinical notetaking solution in this blind test at the VA, among 150 entrants. Second, I am thrilled that genAI-powered notetaking has become so impactful in real clinical environments. 150 entries, wow!
Microsoft Cloud
@MSCloud
🏆 DAX Copilot, our AI assistant, wins the United States Department of Veterans Affairs' Tech Sprint. Learn how we’re revolutionizing healthcare by helping reduce clinician burnout: www.microsoft.com/en-us/microsoft-cloud/blog/healthcare/2…
-
Peter Lee reposted
Microsoft
@Microsoft
What’s behind AI hallucinations? Find out how Microsoft is creating solutions to measure, detect and mitigate the phenomenon as part of its efforts to develop AI in a safe, trustworthy and ethical way. news.microsoft.com/source/features/ai/why-ai-sometimes-ge…
-
Peter Lee
@peteratmsr
We've been learning a ton and making tools for measuring groundedness of, and detecting hallucinations in, LLM responses. And the heavy internal and external use of the these tools is leading to a lot of good refinement. It's pretty useful stuff for genAI app developers.
Microsoft Research
@MSFTResearch
“Being on the cutting edge of generative AI means we have a responsibility and an opportunity to make our own products safer and more reliable." Learn how containing AI hallucinations is a big part of that effort. news.microsoft.com/source/features/ai/why-ai-sometimes-ge…
Microsoft
@Microsoft
What’s behind AI hallucinations? Find out how Microsoft is creating solutions to measure, detect and mitigate the phenomenon as part of its efforts to develop AI in a safe, trustworthy and ethical way. news.microsoft.com/source/features/ai/why-ai-sometimes-ge…
-
Peter Lee
@peteratmsr
I think we may be approaching a tipping point in AI for medical imaging. Incredible progress at the @CVPR workshop on foundation models for medical imaging. Our teams were pleased to share work @MSFTResearch on #biomeparse and #gigapath to standing-room only audience.
-
Peter Lee reposted
Atul Butte (In Memoriam)
@atulbutte
Whoa… Epic has "> 60 significant AI development projects underway": 9 specific projects listed here! "98% patients who used self-scheduling… would recommend it to others". @CALonghurst: "in 3-5 yrs, generative AI could have a huge impact on healthcare" www.beckershospitalreview.com/healthcare-information-tech…
-
Peter Lee reposted
Mustafa Suleyman
@mustafasuleyman
AI should be a catalyst for real progress, but too often we haven't seen that. One of the good things about being at @Microsoft is seeing the sheer breadth of effort behind the scenes, particularly with our AI for Good initiatives.
This is AI making a difference to the rainforest, to people facing hunger. It's AI mapping the world to help clean energy and town planning in stressed environments. It's AI that helps combat addiction and preserve wildlife and biodiversity. AI that meets challenges. Expect a phase shift here.
This is one group in one company. But it hints at what is possible. In the next year we will not only see AI being discussed out in the media, but millions of people will feel the benefit in ways that are still often underappreciated. www.microsoft.com/en-us/research/group/ai-for-good-resear… -
Peter Lee
@peteratmsr
Great set of people on the @theNAMedicine’s Commission on Investment Imperatives for a Healthy Nation. Hopes to better align values, incentives, policies, & actions to improve the nation's health system. nam.edu/national-academy-of-medicine-launches-commission-…
-
Peter Lee reposted
Microsoft
@Microsoft
As part of Microsoft’s commitment to bringing generative AI to precision health, we’re partnering with healthcare organizations and doctors to better analyze patient information to provide enhanced cancer care and advance medical discovery: news.microsoft.com/source/features/digital-transformation…
-
Peter Lee reposted
Microsoft AI Frontiers
@ms_aifrontiers
📹: Checkout this latest episode of @MSFTResearch forum, where @adamfourney talks about agents, complex tasks, and #autogen!
www.youtube.com/watch?v=KuX_dkqr7UY -
Peter Lee
@peteratmsr
It's not just that AI can be generative, but it can also be evaluative and collaborative. This is seen most directly in the very popular #autogen @MSFTResearch system, but the idea is pretty general. Easy-to-read article @willknight about this. link.wired.com/public/35624948
-
Peter Lee reposted
Wessel
@ikwess
Whoa, Aurora is featured in a Nature news article!
We believe that forecasting atmospheric chemistry is only the beginning of what’s possible with foundation models for environmental forecasting.
www.nature.com/articles/d41586-024-01677-2 -
Peter Lee reposted
STAT
@statnews
What reassures Microsoft's Peter Lee about the use of AI in health care? We caught up with him for our "Beyond the Panel" series at the #statbreakthrough Summit West to discuss.
@peteratmsr @MSFTResearch #ai #healthcare -
Peter Lee reposted
Paris Perdikaris
@ParisPerdikaris
Nice to see @Nature pick up our recent work on Aurora. As the article concludes, "there’s lots of cool science to be done"! Indeed we are in the early days of exploring how AI foundation models like Aurora can transform our ability to understand and predict the Earth System.
-
Peter Lee
@peteratmsr
Aurora is an AI foundation model that flexibly achieves SoTA 5-day air pollution, 10-day global weather, and other forecasts. More importantly, it shows evidence of good scaling properties and adaptation to new atmospheric tasks. Excited to see scale up. www.microsoft.com/en-us/research/publication/aurora-a-fou…
-
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
Updated phi-3 tech report with final numbers for 7B/14B and a new section on phi-3-V (e.g., MMMU at 40.4, in the ballpark of Claude 3-haiku and Gemini-1.0 pro) : arxiv.org/abs/2404.14219
-
Peter Lee reposted
Naoto Usuyama
@naotous
Thank you so much for the shoutout! 🎉 Excited to share that our code and model are available online!
Code: github.com/prov-gigapath/prov-gigapath
Model: huggingface.co/prov-gigapath/prov-gigapath -
Peter Lee
@peteratmsr
Foundation models trained on heterogeneous datasets from natural phenomena continues to impress us, such as Aurora @MSFTResearch for atmospheric analyses, simulations, and forecasts/predictions. They show signs of significant improvement with scaling up, and tremendous adaptability to a wide range of tasks.
Paris Perdikaris
@ParisPerdikaris
Excited to introduce Aurora: a foundation model of the atmosphere. In <1min, Aurora produces 5-day global air pollution predictions and 10-day high-resolution weather forecasts that outperform SOTA classical simulation tools and the best specialized deep learning models… 1/n
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
.@Chi_Wang_ @bansalg_ @beibin79 @AhmedHAwadallah @ryen_white @dcburger @JieyuZhang20 Eric Zhu Li Jiang Xiaoyun Zhang & coauthors won Best Paper at #iclr2024 LLM Agents Workshop for "AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation." llmagents.github.io
-
Peter Lee
@peteratmsr
In @Nature today, our work on the first foundation model for whole-slide digital pathology. Key advance is it shows strong evidence of scaling benefit. State-of-the-art performance on 17/18 imaging tasks, with 12 showing very significant improvement. www.microsoft.com/en-us/research/blog/gigapath-whole-slid…
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
Digital pathology helps decode tumor microenvironments for precision immunotherapy. In joint work with @Providence and @UW, we’re sharing Prov-GigaPath, the first whole-slide pathology foundation model, for advancing clinical research. www.microsoft.com/en-us/research/blog/gigapath-whole-slid…
-
Peter Lee
@peteratmsr
We've made a ton of progress on medical imaging with AI. What is most important here is the possibility that there is strong possibility of useful scaling at play.
Eric Topol
@EricTopol
Just out @Nature
The 1st whole-slide digital pathology #ai foundation model pre-trained on large-scale real-world data, from over 1.3 billion images, 30,000 patients
www.nature.com/articles/s41586-024-07441-w
@hoifungpoon @HanwenXu6 @Microsoft @UW @naotous @MSFTResearch -
Peter Lee reposted
Eric Topol
@EricTopol
Just out @Nature
The 1st whole-slide digital pathology #ai foundation model pre-trained on large-scale real-world data, from over 1.3 billion images, 30,000 patients
www.nature.com/articles/s41586-024-07441-w
@hoifungpoon @HanwenXu6 @Microsoft @UW @naotous @MSFTResearch -
Peter Lee reposted
Shital Shah
@sytelus
Phi-3 14B model from our team is available now! This was trained with 512 H100s on 4.8T tokens achieving MMLU of 78 (comparable with Llama3 70B!!).
huggingface.co/microsoft/Phi-3-medium-4k-instruct -
Peter Lee
@peteratmsr
Gordon Bell had a huge impact on the entire tech industry and on Microsoft. He was a giant in the world of computer systems research, with his concept of "utility computing" laying conceptual foundations for the cloud. He was an early advocate in areas such as digital health, telepresence, and digital immortality, his vision seems more relevant (and needed) than ever. For all of us @MSFTResearch, we feel so lucky to have worked and dreamed with you. Indeed, you will be missed.
Satya Nadella
@satyanadella
Rest in peace, Gordon Bell. Thank you for all of your impact at Microsoft and our industry. You will be missed. Thanks to you, I am always looking out for people from NOD.
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
Hear from @xie_tian during Microsoft Research Forum on how MatterGen is aiming to shift the traditional paradigm of materials design using AI. Join us on June 4. www.microsoft.com/en-us/research/event/microsoft-research…
-
Peter Lee
@peteratmsr
Providing teachers with free access to Khanmigo for Teachers from @khanacademy on @Azure is a highly useful and actionable resource for educators. news.microsoft.com/source/features/ai/khan-academy-and-mi…
-
Peter Lee
@peteratmsr
Phi-3-medium is here!
Sebastien Bubeck
@SebastienBubeck
Can't wait for your feedback on our largest and most capable Phi model yet!
huggingface.co/microsoft/Phi-3-medium-4k-instruct -
Peter Lee reposted
Christina Farr
@chrissyfarr
This was my dream team for a Health and AI panel and it happened! @peteratmsr @julesyoo @zakkohane #fortunebrainstormhealth
-
Peter Lee
@peteratmsr
Congrats to the AutoGen team @MSFTResearch for making such a great contribution to the worlds3 ability to experiment, learn about, and deploy AI agents into the real world.
Microsoft AI Frontiers
@ms_aifrontiers
📰: #autogen recently won the best paper award at the LLM Agent Workshop at ICLR-2024! The workshop had at 90+ accepted oral presentations and posters.
llmagents.github.io -
Peter Lee reposted
Microsoft Research
@MSFTResearch
Autogen is an open-source framework for agentic AI. And during #msbuild, you can learn about using this framework to build multi-agent AI applications. Register to attend this session and gather the tools needed for new ideas for self-guided exploration. build.microsoft.com/en-US/sessions/4e3b35a5-5862-46d0-b5e…
-
Peter Lee
@peteratmsr
Huge congratulations to all the new MDs graduating today @KPMedSchool! You are all the very first -- and of course the very best! -- new doctors to come out of this school. So proud to see Bernard Tyson's dream of a new kind of medical school come fully to fruition.
Peter Lee
@peteratmsr
Electrifying, touching, entertaining, and most of all, inspiring commencement speech by Magic Johnson. Talked about the impact of HIV/AIDS on his life, and his impact on creating much-needed awareness of it for all, and the positivity that brings.
-
Peter Lee reposted
Han Yang 杨涵
@HanYangChicago
Excited to share our latest development, MatterSim: A Deep Learning Atomistic Model Across Elements, Temperatures and Pressures. Check it out 👉arxiv.org/abs/2405.04967 www.microsoft.com/en-us/research/blog/mattersim-a-deep-le… #microsoftresearch #ai #deeplearning #materialscience
@luzihen -
Peter Lee reposted
a16z Bio + Health
@a16zBioHealth
On 5/20 at @bstormhealth, @julesyoo will join @Microsoft’s @peteratmsr and @harvardmed’s @zakkohane for a discussion moderated by @chrissyfarr on the #generativeai revolution in medicine.
Learn more at fortune.com/conferences/fortune-brainstorm-health-2024 -
Peter Lee reposted
Oriol Servia
@OriolServia
This machine is just so much fun to drive 😍🔥@TeamChevy @chevrolet #corvettez06 🔊
Jim Campbell
@CampbellJim
@OriolServia driving Corvette Z06 pace car for the @IndyCar @sonsio Grand Prix at Indianapolis Motor Speedway @IMS Road Course. @TeamChevy Tune in: @IndyCaronNBC @SIRIUSXM XM 218 IndyCar app
-
Peter Lee reposted
Oriol Servia
@OriolServia
😎
Steve H. Shunck
@SHUNCK
Ser-vi-a
In the lead!
@OriolServia -
Peter Lee reposted
Lake Nona Impact Forum
@LNImpactForum
Have you explored this captivating read yet? Discover why the #healthcare industry is embracing #ai, including insights from @peteratmsr of @MSFTResearch, which he first unveiled at Lake Nona Impact Forum 2023. #lnif23 #lnif24 #generativeai
-
Peter Lee
@peteratmsr
I’m looking forward to learning from some great people at #aspenideashealth. I'll be sharing my own lessons from the past year+ on genAI in medicine and seeing how they connect with the others. The @aspenideas speaker lineup: www.aspenideas.org/speakers/health-speakers
-
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
Check out this video if you want to learn more about phi-3!
And yes, yesterday's TikZ unicorn is by phi-3 :-) (14B model)
www.youtube.com/watch?v=rW9bAixVFkQ -
Peter Lee reposted
Microsoft Research
@MSFTResearch
We will be at #iclr2024 next week, where Microsoft researchers will be presenting over 100 papers. Learn more about these sessions, oral presentations, and publications, and check out career opportunities with Microsoft. www.microsoft.com/en-us/research/event/iclr-2024/
-
Peter Lee
@peteratmsr
I'm grateful to be named alongside so many great people on @TIME's #time100health list of the 100 most influential people in health. Truly a recognition of the efforts of the amazing team @Microsoft plus fantastic collaborators around the world. time.com/collection/time100-health/
-
Peter Lee reposted
Ashley J. Llorens
@ashleyllorens
Excited to be working with @SanctuaryAI and collaborators across industry and academia to advance research in embodied AI including perception, reasoning, planning, and safe and effective collaboration among humans and intelligent systems.
Microsoft Research
@MSFTResearch
We're excited to work with Sanctuary AI to advance embodied AI & robotics research to help customers across industries. Our collaboration will evaluate model performance, efficiency & planning capabilities, with Microsoft Azure powering their AI workloads. sanctuary.ai/news/
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
We're excited to work with Sanctuary AI to advance embodied AI & robotics research to help customers across industries. Our collaboration will evaluate model performance, efficiency & planning capabilities, with Microsoft Azure powering their AI workloads. sanctuary.ai/news/
-
Peter Lee
@peteratmsr
I've listened to every episode of the Possible podcast (and have even been a guest on one), but this one with @kevin_scott @Microsoft is my favorite. And I'm not saying that just because he's my boss! :-)
Reid Hoffman
@reidhoffman
Possible's back with a new season, and debuts with Microsoft CTO @kevin_scott, the architect of the partnership between @Microsoft and @OpenAI.
Kevin is equal parts technologist and humanist—the very blend that this podcast aspires to. A preview: -
Peter Lee reposted
Satya Nadella
@satyanadella
We’re redefining the developer environment with GitHub Copilot Workspace—where any developer can go from idea, to code, to software in natural language.
Thomas Dohmke
@ashtom
What started out as an autocomplete pair programmer is now redefining the developer experience itself. Welcome to @GitHub Copilot Workspace: The Copilot-native developer environment — a place for all to create with code instantly in natural language.
github.blog/2024-04-29-github-copilot-workspace/ -
Peter Lee reposted
Microsoft Research
@MSFTResearch
Want to engage with the highlights, emerging technologies, and challenges discussed during Microsoft Research Forum? Check out our research copilot experience. msrchat.azurewebsites.net
-
Peter Lee
@peteratmsr
On the one hand, I hear some say that AI has moved recklessly fast this past year. On the other hand, others (often the same people) say it's been a whole year and wonder why GPT-5 isn't out yet. I guess tech disruptions can warp one's perspective on the passage of time...
-
Peter Lee reposted
Microsoft Azure
@Azure
Phi-3—Microsoft’s most advanced family of small language models—expands choice and flexibility in AI models. Now available in the Azure AI model catalog. Learn more: azure.microsoft.com/en-us/blog/introducing-phi-3-redefini…
-
Peter Lee reposted
Christopher A. Longhurst, MD
@CALonghurst
𝐆𝐞𝐧𝐀𝐈 𝐢𝐧 𝐭𝐡𝐞 𝐄𝐇𝐑 - what have we learned so far? 🧵
One year ago @UCSDHealth and @StanfordHealth were announced as pilot sites for generative AI in Epic, and we have both published initial outcomes in @JAMANetworkOpen! (1/9) -
Peter Lee reposted
Microsoft
@Microsoft
We're excited to announce the launch of Phi-3, a groundbreaking family of small language models that outperform larger models on a range of benchmarks. Learn how these small language models trained on high-quality data are doing more with less: news.microsoft.com/source/features/ai/the-phi-3-small-lan…
-
Peter Lee
@peteratmsr
This study shows that GPT-4 compares well against a cohort of 849 humans who took medical board exams in 5 specialties in 2022. The analysis is solid but uses zero-shot AI; smarter (and fully automatic) prompting techniques such as MedPrompt would improve GPT-4 even further.
NEJM AI
@NEJM_AI
Datasets, Benchmarks, and Protocols: GPT versus Resident Physicians — A Benchmark Based on Official Board Scores by Dr. @uriel_katz_ and team: nejm.ai/3Q7hB9j #artificialintelligence
Peter Lee
@peteratmsr
Information on MedPrompt (with scripts openly available on GitHub) is here. www.microsoft.com/en-us/research/blog/the-power-of-prompt…
-
Peter Lee reposted
Kevin K. Yang 楊凱筌
@KevinKaichuang
Happy to share the final version of our work describing computational filters for selecting functional ML-generated enzymes in @NatureBiotech!
This was a distributed and remote collab with @TrichomeDoctor @Xiaozhi_Fu and other members of @AZelezniak's group. Global Science!
Aleksej Zelezniak🇺🇦
@AZelezniak
Our computational filter for enriching functional AI-generated enzymes is finally out. www.nature.com/articles/s41587-024-02214-2
Many thanks to @TrichomeDoctor @KevinKaichuang @Xiaozhi_Fu et al. for the fantastic collaboration! -
Peter Lee reposted
Seth Hain
@sethHain
While the impressive benchmarks may capture headlines, the compounding effect of high quality data (and the creative ways MSR is generating it) is the real story. The resulting reasoning capability in such a tiny package opens up even more doors for us, application developers.
Congrats, @SebastienBubeck and team.
Sebastien Bubeck
@SebastienBubeck
phi-3 is here, and it's ... good :-).
I made a quick short demo to give you a feel of what phi-3-mini (3.8B) can do. Stay tuned for the open weights release and more announcements tomorrow morning!
(And ofc this wouldn't be complete without the usual table of benchmarks!) -
Peter Lee reposted
Mikhail Parakhin
@MParakhin
I don't usually comment on model releases, but Phi-3 is genuinely good. I believe it is the best small model at the moment. Sebastien Bubeck, Weizhu Chen, Misha Bilenko and the team: phenomenal job, folks!
-
Peter Lee
@peteratmsr
The phi-3 family of models is here! In mini, small, and medium sizes. And as @SebastienBubeck says, they are good, even surprisingly so...
Sebastien Bubeck
@SebastienBubeck
phi-3 is here, and it's ... good :-).
I made a quick short demo to give you a feel of what phi-3-mini (3.8B) can do. Stay tuned for the open weights release and more announcements tomorrow morning!
(And ofc this wouldn't be complete without the usual table of benchmarks!) -
Peter Lee
@peteratmsr
We've been so impressed with @providence throughout our collaborations, including on clinical trial matching. Real-world results like these provide more evidence that generative AI will rapidly enable the use of real-world evidence to accelerate medical research.
Microsoft Research
@MSFTResearch
"As a research team, it has been exciting to work with Providence. Developing prototype AI tools using the principles of responsible AI... is important when we look to improve patient outcomes in the future," says Hoifung Poon, Ph.D., of Microsoft Research Health Futures.
-
Peter Lee
@peteratmsr
Lightening the "cognitive load" on physicians @StanfordMed by using DAX Copilot. The article is consistent with many others' experiences with this genAI tech: it's easy to use and more efficient. What's more, most also report better results. www.beckershospitalreview.com/innovation/why-stanford-exp…
-
Peter Lee
@peteratmsr
I had a great time chatting about how AI is transforming healthcare, from research to medical knowledge to how doctors communicate. This was on Microsoft’s WorkLab podcast, which explores data and insights into why and how work is changing. www.microsoft.com/en-us/worklab/podcast/all-leaders-can-l…
-
Peter Lee
@peteratmsr
One thing I like about visiting the east coast while working for a west coast company is that the mornings have less email.
-
Peter Lee
@peteratmsr
Very pleased with this next step in the expansion of @MSFTResearch MSR Asia. As part of Microsoft's newly announced $2.9B investment in Japan, this new lab in Tokyo will help accelerate ongoing and new collaborations in the area and globally.
Brad Smith
@BradSmi
In 1978, @Microsoft made Tokyo its first international office. Today, alongside Japanese Prime Minister Fumio Kishida, we announced a $2.9 billion investment over 2 years to boost our cloud computing and #ai infrastructure, a commitment to innovation and growth with Japan.
-
Peter Lee
@peteratmsr
Big thanks to everyone who participated in our recent MSR Forum. As a followup, you can chat about the ideas and information from this event using this AI Copilot experience.
Microsoft Research
@MSFTResearch
Do you have questions about anything that was discussed during Microsoft Research Forum? Check out our Research Copilot Experience to learn more about what was covered: msrchat.azurewebsites.net
-
Peter Lee
@peteratmsr
I remain impressed with the agile and yet methodical way that the folks at Epic have embraced generative AI and gone about integrating and deploying it into healthcare organizations. This is yet another example of their thoughtful approach.
Seth Hain
@sethHain
Providing our customers with tools to confidently deploy and monitor AI models to improve health outcomes – on their own data and in their own workflows – is key to building trust in this technology.
Read about how Epic is leading in this effort with our AI Trust and Assurance Suite
#healthit #ai @CALonghurst @aneeshchopra @HeatherLandi www.fiercehealthcare.com/ai-and-machine-learning/epic-pla… -
Peter Lee reposted
Satya Nadella
@satyanadella
This is such an exciting milestone on our path toward unlocking the scientific and commercial progress that will come from reliable quantum computation.
Microsoft
@Microsoft
Together, Microsoft and @QuantinuumQC have demonstrated the most reliable logical qubits on record.
How this major milestone will help on the journey of advancing science: blogs.microsoft.com/blog/2024/04/03/advancing-science-mic… -
Peter Lee reposted
Microsoft
@Microsoft
Together, Microsoft and @QuantinuumQC have demonstrated the most reliable logical qubits on record.
How this major milestone will help on the journey of advancing science: blogs.microsoft.com/blog/2024/04/03/advancing-science-mic… -
Peter Lee
@peteratmsr
A stable qubit needing only 30 physical qubits? @Microsoft and @QuantinuumQC achieve an 800x improvement in logical vs physical error rate, with both entanglement and error corrections. A real landmark in quantum computing. cloudblogs.microsoft.com/quantum/2024/04/03/how-microsoft…
-
Peter Lee reposted
Ethan Mollick
@emollick
It is remarkable how routine it has become for careful studies to show that GPT-4 (not trained specifically for medicine) outperforms most doctors in key aspects of diagnosis.
That doesn't mean that GPT-4 is reliable in all circumstances, but it still seems like a big deal.
Eric Topol
@EricTopol
How good is #ai at clinical reasoning? An early, simulated assessment
jamanetwork.com/journals/jamainternalmedicine/fullarticle…
“An LLM was better than physicians in processing
medical data and clinical reasoning using recognizable frameworks as measured by R-IDEA” -
Peter Lee
@peteratmsr
Some good progress here on tools that codify for developers a lot of the lessons, techniques, and experiences we've had across Microsoft in deploying generative AI applications. azure.microsoft.com/en-us/blog/announcing-new-tools-in-az…
-
Peter Lee reposted
Christopher A. Longhurst, MD
@CALonghurst
The @UCSDHealth sepsis AI study is the first to show improvement in clinical outcomes with a deep-learning AI model
Credit to the team for focusing on workflow, since healthcare AI is about more than algorithms – as the
editorial highlights! (5/5)
👉www.nature.com/articles/s41746-024-01066-z -
Peter Lee reposted
Aspen Ideas
@aspenideas
💡 Peter Lee (@peteratmsr), Head of @MSFTResearch, will explore the benefits, limitations, and risks of integrating today’s AI advancements into health care. #aspenideashealth (4/9)
-
Peter Lee reposted
Oriol Servia
@OriolServia
Aftermath, @TheThermalClub experiment.
Ok, so the format didn’t work. Also, the event itself needs to be enhanced with more on-track action and other entertainment for the attendees. Everything else was a positive imo.
Overall I liked that @IndyCar did it, it takes bravery for an over a century old sport series to try anything new. There were lots positives and with some tweaks I feel there is room for one special event like this each season. -
Peter Lee
@peteratmsr
What a great day! P10 in LMP2, P5 in GTP, and P3 in GTD. A fantastic result for Improper Racing!
Chris Stewart
@chrisstewart
We're going to crush it!
Peter Lee
@peteratmsr
Our sim racing team, Improper Racing, will be fielding 3 cars in tomorrow's #sebring12h event on @iRacing. Really looking forward to it, but we're all extremely nervous about the weather forecast of 96% chance of rain... www.iracing.com/this-week-iracing-sebring-12-special-even…
-
Peter Lee reposted
Paul J. Swider #HealthcareAI #HealthTech
@pswider
@peteratmsr for me. His work over the past two years in AI HLS and healthcare is insightful and inspiring.
Microsoft 365 Insider
@Msft365Insider
Take some time today to thank someone who has supported you throughout your learning journey. 💛
#msft365insiders -
Peter Lee reposted
Microsoft 365 Insider
@Msft365Insider
Thank you, Paul, for highlighting @peteratmsr’s impactful work in AI HLS and healthcare. Its sounds like he has definitely inspired you. Pretty cool!
-
Peter Lee
@peteratmsr
Our sim racing team, Improper Racing, will be fielding 3 cars in tomorrow's #sebring12h event on @iRacing. Really looking forward to it, but we're all extremely nervous about the weather forecast of 96% chance of rain... www.iracing.com/this-week-iracing-sebring-12-special-even…
-
Peter Lee
@peteratmsr
Thanks, @reidhoffman! I really appreciate the callout to our book. Doing the podcast with you and Aria was a ton of fun.
Reid Hoffman
@reidhoffman
The AI Revolution in Medicine: GPT-4 and Beyond, by @MSFTResearch’s @peteratmsr and @goldbergcarey @zakkohane, is a great complement to Peter's episode on the future of health and medicine: link.chtbl.com/M9jVQqAR
-
Peter Lee reposted
Reid Hoffman
@reidhoffman
The AI Revolution in Medicine: GPT-4 and Beyond, by @MSFTResearch’s @peteratmsr and @goldbergcarey @zakkohane, is a great complement to Peter's episode on the future of health and medicine: link.chtbl.com/M9jVQqAR
-
Peter Lee reposted
Mustafa Suleyman
@mustafasuleyman
I’m excited to announce that today I’m joining @Microsoft as CEO of Microsoft AI. I’ll be leading all consumer AI products and research, including Copilot, Bing and Edge. My friend and longtime collaborator Karén Simonyan will be Chief Scientist, and several of our amazing teammates have chosen to join us.
@InflectionAI will continue on its mission under a new CEO, and look to reach more people than ever by making its API widely available to developers and businesses the world over.
It’s been an amazing journey, with so much more to come. Thank you to everyone for your support. Things really are just getting started. -
Peter Lee
@peteratmsr
Packed room at #nvidiagtc24 for @SebastienBubeck's talk about #phi and small language models.
-
Peter Lee reposted
NVIDIA GTC
@NVIDIAGTC
Accelerate #generativeai, industrial digitalization and healthcare breakthroughs with @Microsoft and NVIDIA’s expanded partnership announced at #gtc24.
nvda.ws/48UgyQs -
Peter Lee reposted
Microsoft
@Microsoft
Ever wondered how the weather in your location has evolved over time? 🌤️
With this new feature in Weather from Microsoft Start, you can compare today’s forecast to data spanning 70 years: blogs.bing.com/search-quality-insights/2024/2/How-are-cli… -
Peter Lee reposted
Vardit Ravitsky
@VarditRavitsky
The @LNImpactForum was a wonderful experience. Here's some insights about our panel on #ai in health & my fav parts.
In pic: w/ @peteratmsr (President of Microsoft Research) & @byronpitts (Emmy Award-winning co-anchor of @NBCNews “Nightline").
www.thehastingscenter.org/news/inside-the-lake-nona-impac… -
Peter Lee
@peteratmsr
It's really worth paying attention to this. So much rising optimism that AI will accelerate the pace of discovery in the natural sciences.
Microsoft Research
@MSFTResearch
“In my view, the most important use case of AI will be to scientific discovery," says Chris Bishop. In his Research Forum keynote, he shares the vision for how Microsoft Research AI for science will leverage AI to model and predict natural phenomena. www.microsoft.com/en-us/research/quarterly-brief/mar-2024…
-
Peter Lee reposted
Isaac Kohane
@zakkohane
Patient-first granularity in concerns and opportunities for AI in healthcare
NEJM AI
@NEJM_AI
Perspective by @goldbergcarey et al.: To Do No Harm — and the Most Good — with AI in Health Care nejm.ai/3T9kb0q
-
Peter Lee reposted
Rasu Shrestha MD MBA
@RasuShrestha
The integration of #ai and ambient technology directly into Epic could truly change the game.
“The reason why you see 60+ use cases from the clinical to the bedside to the back office is because we've integrated #generativeai into our core development processes.” #himss24
Peter Lee
@peteratmsr
I'm impressed with Epic's work on integrating GPT-4 into clinical workflows. Already >150K clinical notes written and millions of patient messages drafted by AI in just one year, with incredibly high satisfaction. Proud to collaborate with them on this. www.fiercehealthcare.com/ai-and-machine-learning/himss24-…
-
Peter Lee reposted
Seth Hain
@sethHain
Appreciate the deep collaboration with Microsoft and shared focus to help clinicians and patients. Excited to keep building.
-
Peter Lee
@peteratmsr
I'm impressed with Epic's work on integrating GPT-4 into clinical workflows. Already >150K clinical notes written and millions of patient messages drafted by AI in just one year, with incredibly high satisfaction. Proud to collaborate with them on this. www.fiercehealthcare.com/ai-and-machine-learning/himss24-…
-
Peter Lee reposted
Ethan Mollick
@emollick
Microsoft continues to offer the only free way to access a GPT-4 class model, they just upgraded free Copilot (which used to be called Bing, which was secretly named Sydney) to GPT-4-Turbo in Creative and Precise modes.
So people in 169 countries have free access to a top LLM. -
Peter Lee reposted
Microsoft Research
@MSFTResearch
Microsoft Research program encourages AI development by academics, not just industry. Initial projects: AI-powered advice to beat traffic jams; help for writers learning a new language; a brilliant virtual astronomer with an encyclopedic memory. news.microsoft.com/source/features/ai/ai-for-all-how-acce…
-
Peter Lee reposted
Satya Nadella
@satyanadella
At #himss24, we're highlighting the remarkable work of innovators like @StanfordHealth, @Providence, and @WellSpan, who are using DAX Copilot embedded in Epic to empower clinicians and transform patient healthcare experiences.
Microsoft News and Stories
@MSFTnews
Today at #himss24, @Microsoft announced how our healthcare customers are adopting generative AI solutions and how we’re collaborating with the industry to operationalize responsible AI. blogs.microsoft.com/blog/2024/03/11/microsoft-makes-the-p…
-
Peter Lee reposted
STAT
@statnews
Microsoft research head @peteratmsr explained how generative AI could help doctors be more human. #statuslist www.statnews.com/2024/03/05/peter-lee-microsoft-generativ…
-
Peter Lee reposted
NVIDIA GTC
@NVIDIAGTC
Join an expert panel at #gtc24 to explore the transformative impact of #generativeai on patient care, research, and #healthcare economics. Led by @kpowgerade, the panel will feature @EricTopol, @CathieDWood, and @peteratmsr.
Add to your schedule: nvda.ws/49MDjXN -
Peter Lee
@peteratmsr
Every day we gain more insights into the importance of improved training signals and methods to enhance the reasoning abilities of language models. For math, Orca-Math at 7B shows math performance at or beyond that of much larger models. www.microsoft.com/en-us/research/blog/orca-math-demonstra…
-
Peter Lee reposted
STAT
@statnews
Microsoft research head and #statuslist selection @peteratmsr explains how generative AI could help doctors be more human in new Q/A with @SciFleur. www.statnews.com/2024/03/05/peter-lee-microsoft-generativ…
-
Peter Lee
@peteratmsr
I was pleased to receive the Polish edition of "The AI Revolution in Medicine: GPT-4 and Beyond" -- the 9th non-English edition of our book! @goldbergcarey @zakkohane @SebastienBubeck
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
Happening tomorrow—Microsoft Research Forum: Episode 2! Register now to join us on March 5 at 9:00 AM PT to learn more about the real-world impact of AI. www.microsoft.com/en-us/research/event/microsoft-research…
-
Peter Lee reposted
IndyCar Transporter
@Indys18Wheels
#flashbackfriday This is @OriolServia driving the Charter car for @DreyerReinbold in 2012.
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
ViSNet, a novel AI-powered molecular geometry modeling framework, is designed to help researchers predict molecular properties, simulate molecular dynamics, and understand the relationships between molecular structure and biological activity. Learn more. www.microsoft.com/en-us/research/blog/visnet-a-general-mo…
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
Introducing BitNet b1.58: 1.58-bit LLMs, in which every single parameter is ternary {-1, 0, 1}, that rival full-precision Transformer LLMs in performance while significantly boosting efficiency—paving the way for a new era of 1-bit LLMs. Learn more: arxiv.org/pdf/2402.17764
-
Peter Lee
@peteratmsr
It's such a special and important time to be working on advancing and improving the health and life sciences. Yes, it's an honor to be named to this list; but the bigger honor is to be part of a team and community that is so committed to making life better in a practical way.
STAT
@statnews
Today, STAT launches the 3rd annual #statuslist — a collection of 50 people with huge influence on the future of health and the life sciences.
Names include:
@MicrobiomDigest
@hankgreen
@johngreen
@peteratmsr
@DrAlethaMaybank
@zhangf
...and more
Read: www.statnews.com/status-list/2024/ -
Peter Lee
@peteratmsr
So pleased to welcome to @MSFTResearch our new (but very familiar) colleagues at @MicrosoftAfrica Research Institute (MARI) in Nairobi, Kenya! They will amplify our research into creating a more productive future globally. www.microsoft.com/en-us/research/group/microsoft-africa-r…
-
Peter Lee
@peteratmsr
My day in Boston yesterday has made me more educated, inspired, and convinced that biotech and pharma will be dramatically accelerated by a new generation of AI-based tools. @MSFTResearch is deep into this but there is so much more to do and so much more opportunity. Energizing!
-
Peter Lee
@peteratmsr
Great stuff coming from @MistralAI on @Azure.
Microsoft Azure
@Azure
🚀 Join us in celebrating the transformative partnership between Microsoft and @MistralAI!
We're paving the way for efficient and cost-effective AI models that accelerate growth and deliver real-world solutions. Learn more: azure.microsoft.com/en-us/blog/microsoft-and-mistral-ai-a… #microsoftazure #mistralai -
Peter Lee reposted
Microsoft Azure
@Azure
🚀 Join us in celebrating the transformative partnership between Microsoft and @MistralAI!
We're paving the way for efficient and cost-effective AI models that accelerate growth and deliver real-world solutions. Learn more: azure.microsoft.com/en-us/blog/microsoft-and-mistral-ai-a… #microsoftazure #mistralai -
Peter Lee reposted
Isaac Kohane
@zakkohane
European vs US views AI risks/benefits was unexpected contrast in RAISE Symposium (summarized @NEJM_AI ai.nejm.org/doi/full/10.1056/AIp2400036). Colleagues from Europe noted their deliberations more focused on risk with little discussion of opportunity cost of not proceeding given broken healthcare.
-
Peter Lee reposted
Motorsport
@Motorsport
Do you agree? 😄
📸 news.verstappen
#f1 #f1testing -
Peter Lee reposted
Matthew Lungren MD MPH
@mattlungrenMD
Vision-Language Models will revolutionize radiology and enhance patient care.
How do we ensure they achieve their goals in practice?
Latest research from @MSFTResearch led by @runmiridliy explores clinician+AI interaction for VLMs in healthcare!
arxiv.org/pdf/2402.14252 -
Peter Lee
@peteratmsr
Returning home from a week in @MSFTResearch lab in Bangalore. It was inspiring to see research spanning the gamut from trillion-scale vector search to enabling teachers in rural India harness LLMs to adapt curricula to their students' particular needs. www.microsoft.com/en-us/research/blog/teachers-in-india-h…
-
Peter Lee reposted
Nene del Cerro
@EugenioTuya
#novedadesanaya. Los 📚 que vienen.
La revolución de la Inteligencia artificial en la medicina. GPT-4 y más allá de @peteratmsr, @goldbergcarey @zakkohane con @SebastienBubeck.
@Anaya_Multimed en la vanguardia de la tecnología.
Peter Lee
@peteratmsr
I was so pleased to participate in the Taiwan launch of our book, "The AI Revolution in Medicine: GPT-4 and Beyond" @goldbergcarey @zakkohane @SebastienBubeck. Super discussion by great Taiwanese tech leaders, on the future of healthcare and Taiwan's place in that.
-
Peter Lee
@peteratmsr
Impressive how AI enabled the decoding of ancient scrolls that were turned into charcoal in the Mt Vesuvius eruption. I find it amusing to think that when @MSFTResearch first presented ResNet way back in 2015, this definitely was not an application we had every imagined!
DeepLearning.AI
@DeepLearningAI
This week’s AI news:
📜 AI helps recover ancient scrolls
🤖 U.S. restricts robocalls
⚡ GPU data centers’ growing energy consumption
Plus: Andrew Ng discusses data privacy.
Read #thebatch: www.deeplearning.ai/the-batch/issue-236?hss_channel=tw-99… -
Peter Lee reposted
Ethan Mollick
@emollick
Watch AI for its ability to de-bureaucratize complex processes like clinical trials
GPT-4 with vision & retrieval was able to read medical notes and determine the eligibility of patients for a trial better, faster & cheaper than humans. HT @TheDavidNelson www.medrxiv.org/content/10.1101/2024.02.08.24302376v1 -
Peter Lee
@peteratmsr
The year 2023 was amazing in the world of AI, but 2024 is shaping up to be even more incredible. A few of the trends we're tracking...
Microsoft Research
@MSFTResearch
Why are small language models a big deal? How does multimodal AI make apps more accurate? And how is AI impacting the world of science? Find out with our guide to three trends to watch in 2024.
-
Peter Lee
@peteratmsr
I do this and find it to be super useful.
Microsoft Copilot
@MSFTCopilot
We call that holding. copilot.com?ocid=usoc_TWITTER_M365_spl100005216272774
-
Peter Lee reposted
Satya Nadella
@satyanadella
Great to be in India this week meeting with changemakers like the team at Karya, who are doing the critical work of building high-quality datasets for AI—and expanding economic opportunity at the same time. news.microsoft.com/source/asia/features/village-by-villag…
-
Peter Lee
@peteratmsr
An important advance in the ability to generate novel yet physically foldable protein structures. www.nature.com/articles/s41467-024-45051-2.pdf
Kevin K. Yang 楊凱筌
@KevinKaichuang
Available today: the final journal version of FoldingDiffusion in @NatureComms
Led by former intern @Kevin_E_Wu with @alexijielu @vdbergrianne @SarahAlamdari @james_y_zou and @avapamini
Kevin K. Yang 楊凱筌
@KevinKaichuang
We present Folding Diffusion: a diffusion model for protein structure inspired by physical protein folding.
Let by @Kevin_E_Wu during his internship, with @alexijielu @vdbergrianne @james_y_zou @avapamini
Code: github.com/microsoft/foldingdiff
Preprint: arxiv.org/abs/2209.15611 -
Peter Lee reposted
Bob Wachter
@Bob_Wachter
Great having @peteratmsr of @Microsoft at @UCSFMedicine #ai retreat. Here, a live demo on how #gpt4 can aid in diagnosis (correctly diagnosing post-strep glomerulonephritis), correcting an error in interpreting a lab test, & suggesting how to talk to the patient. Powerful stuff.
-
Peter Lee reposted
Oriol Servia
@OriolServia
I’m sure it must be a more complex decision to what it may appear to some of us, however, I found despicable and insulting the reasons argumented behind it. @AndrettiGlobal brought to the table, funding, commitment, a top notch manufacturer with @Cadillac & GM, a US based team, earned "Charisma", and, and 3 generations of motorsport excellence. Hard to argue against the fact that @MarioAndretti and @michaelandretti himself are not part of the foundation where @F1 sits today. Oh, and as a team I believe they have won a few things along the way in every championship they have entered: Indycar, IMSA, Formula E, Indianapolis 500, Extreme E, IndyLights...
Motorsport
@Motorsport
#f1 rejects Andretti entry for now, but door open for 2028!
-
Peter Lee reposted
Reid Hoffman
@reidhoffman
How can AI help us become more human? How can it improve the lives of more humans?
I have many thoughts here, as do Possible's S2 guests: @khanacademy’s @salkhanacademy @peteratmsr @CondoleezzaRice @krmaher @SlaughterAM @VanJones68 & more.
link.chtbl.com/bf_lpx7t -
Peter Lee reposted
Ethan Mollick
@emollick
Data from 100 patients with strokes was given to GPT-4 & an experts:
1) GPT-4s treatment suggestion agreed with the expert in 80 cases, in 5 it was right & expert wrong, in 5 expert was right & AI wrong, 10 were judgement.
2) GPT-4 predicted mortality better than specialist AIs
Dvir Aran
@dvir_a
But what was really astonishing was that we also asked GPT-4 to estimate 90-day mortality and compared it with two recent traditional ML prediction models for this task. GPT-4 was significantly better (AUC=0.89 compared to 0.77 and 0.7).
-
Peter Lee
@peteratmsr
Really a fantastic #rolex24 this weekend with a great broadcast @IMSARadio. And huge appreciation to @DanielMorad for the shout-out to our sim racing crew, Team Improper!
Chris Stewart
@chrisstewart
@IMSARadio loving your broadcast. Sim racers everywhere are pulling for Morad! Shout out to Team Improper #rolex24
-
Peter Lee reposted
Chris Stewart
@chrisstewart
@IMSARadio loving your broadcast. Sim racers everywhere are pulling for Morad! Shout out to Team Improper #rolex24
-
Peter Lee reposted
The Economist
@TheEconomist
On our “Babbage” podcast, @zannymb is in conversation with OpenAI’s @sama and Microsoft’s @satyanadella about the future of generative AI econ.st/42hsDh6 👇
-
Peter Lee
@peteratmsr
You are invited! Join us in the @MSFTResearch Forum, a new quarterly series for sharing the latest ideas about research in the era of general AI. Explore and discuss the latest advances and ideas with the global research community. Registration is free. researchforum.microsoft.com?OCID=msr_researchforum_TW_Ep1…
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
Episode One of Microsoft Research Forum is next week!
January 30 at 9:00 AM PT.
Register now ⬇️
www.microsoft.com/en-us/research/event/microsoft-research… -
Peter Lee
@peteratmsr
.@MSFTResearch is highly dependent on academic collaboration. So, it's super important to us that AI capabilities become more accessible to the academic community. We thus support @NSF's National AI Research Resource pilot, as an experiment towards achieving this overall goal.
-
Peter Lee
@peteratmsr
The @broadinstitute announces general availability of Terra @Azure. This is a major step forward in biomedical data analysis, sharing, and cross-org collaboration in population genetics, clinical research, etc, and used by 65K researchers and growing. www.broadinstitute.org/news/terra-azure-release
-
Peter Lee
@peteratmsr
Flying home today from #davos2024 and looking forward to a weekend away from work to compete in the Daytona 24 sim race iracing, with the Improper Racing team. I'll be participating from the US west coast but will stay on European time so that I can do the night-owl shifts...
-
Peter Lee reposted
NEJM AI
@NEJM_AI
NEJM Group appreciates the foresight and support of the exclusive sponsors of our new journal, NEJM AI:
@ElevanceHealth
@lyricclarity
@MSFTResearch
@viz_ai
Explore the new journal: ai.nejm.org -
Peter Lee
@peteratmsr
Great session in front of a packed crowd #davos2024. Inspiring fireside chat by @VasNarasimhan and @satyanadella, followed by amazing talks by @MSFTResearch's @avaamini and Chris Bishop. Chris even pulled a working solid-state battery, discovered by #genai, out of his pocket.
Peter Lee
@peteratmsr
Beyond all the discussions on opportunities and risks of AI #davos2024, what is so exciting is the optimism that AI can accelerate discovery in the biological and natural sciences. I'm looking forward to a @MSFTResearch session today that will show some early real-world impact.
-
Peter Lee reposted
Peter Lee
@peteratmsr
Beyond all the discussions on opportunities and risks of AI #davos2024, what is so exciting is the optimism that AI can accelerate discovery in the biological and natural sciences. I'm looking forward to a @MSFTResearch session today that will show some early real-world impact.
-
Peter Lee
@peteratmsr
Beyond all the discussions on opportunities and risks of AI #davos2024, what is so exciting is the optimism that AI can accelerate discovery in the biological and natural sciences. I'm looking forward to a @MSFTResearch session today that will show some early real-world impact.
-
Peter Lee
@peteratmsr
While preliminary, it is undeniably encouraging to see generative AI drastically accelerating the design of small molecule inhibitors for essential target proteins, in this case for Mycobacterium tuberculosis and coronaviruses. Great collaboration @MSFTResearch & @theGHDDI
Microsoft Research
@MSFTResearch
The Global Health Drug Discovery Institute and Microsoft Research are using AI to innovate in life sciences by accelerating the discovery of new treatments for global infectious diseases like tuberculosis and COVID. Find out how. www.microsoft.com/en-us/research/blog/ghddi-and-microsoft…
-
Peter Lee reposted
Maxime Labonne
@maximelabonne
🔀 Phixtral
I made the first efficient Mixture of Experts with phi-2 models. 🥳
It combines 2 to 4 fine-tuned models and is better than each individual expert.
🤗 phixtral-2x2_8: huggingface.co/mlabonne/phixtral-2x2_8
🤗 phixtral-4x2_8: huggingface.co/mlabonne/phixtral-4x2_8 -
Peter Lee reposted
Microsoft
@Microsoft
Introducing your Copilot for writing, cooking, designing, and so much more. ✨
Discover Copilot Pro, our new premium subscription that gives you the most advanced AI capabilities, including building your own Copilot GPTs: blogs.microsoft.com/blog/2024/01/15/bringing-the-full-pow… -
Peter Lee reposted
Jordi Ribas
@JordiRib1
Our team is releasing Copilot Pro today which includes Open AI's GPT-4 Turbo and Dalle3 with dual guidance and landscape mode. It doesn't have the 30-turn limit, enables creating Copilot GPTs, and is twice as fast as regular Copilot. More details here blogs.microsoft.com/blog/2024/01/15/bringing-the-full-pow…
-
Peter Lee reposted
Microsoft
@Microsoft
A piece of glass that stores data for 10,000 years might seem like a concept from a sci-fi movie, but it could soon be a reality. 🤖
Learn how Microsoft Research’s Project Silica is using cutting-edge technology to make data storage more sustainable: unlocked.microsoft.com/sealed-in-glass/?ocid=FY24_soc_omc… -
Peter Lee
@peteratmsr
I've been so pleased to see more and more AI startups on our platforms. Today, we're extending our support by providing more early-stage startups with free access to Azure AI infrastructure. Hoping this facilitates faster development via tailored GPU virtual machine clusters.
Microsoft for Startups
@msft4startups
Exciting news for AI-first startups! We're expanding startup access to Microsoft’s AI infrastructure, offering free tailored GPU virtual machine clusters. www.microsoft.com/en-us/startups/blog/microsoft-expands-f…
-
Peter Lee
@peteratmsr
It was such an honor to have this chat with @ElevanceHealth CEO Gail Boudreaux. Such an inspiring, and yet pragmatic and consumer-focused, vision for the future of healthcare. www.linkedin.com/feed/update/urn:li:activity:715096140847…
-
Peter Lee reposted
waveaura
A Turing Test: Are Ai Chatbots Behaviorally Similar to Humans? @SSRN papers.ssrn.com/sol3/papers.cfm?abstract_id=4637354
-
Peter Lee reposted
toddbishop
@toddbishop
Microsoft and @PNNLab make a better battery, and show AI’s potential to supercharge science -- prototype reduces the lithium content in batteries by as much as 70%. www.geekwire.com/2024/microsoft-and-pnnl-make-a-better-ba… via @GeekWire
-
Peter Lee
@peteratmsr
AI makes for some pretty incredible acceleration of research in the natural sciences, and in particular the discovery of new materials. One example: Fantastic early results with @PNNLab and @MSFTResearch closing in on new battery tech. news.microsoft.com/source/features/sustainability/how-ai-…
-
Peter Lee reposted
Julie Yoo
@julesyoo
I'll be chatting with @suchisaria of @bayesianhealth, @sethHain of Epic, and @peteratmsr of @Microsoft about "AI in Healthcare: The Opportunity Ahead" - Monday 5:15 pm, in the Colonial Room at #jpm2024.
We'll cover the state of the state, regulatory landscape, and predictions and advice for healthcare execs re: healthcare AI as we head into the new year ahead.
@a16zBioHealth -
Peter Lee
@peteratmsr
At @MSFTResearch we hope that this change makes phi-2 more useful and especially more interesting to researchers.
Sebastien Bubeck
@SebastienBubeck
Starting the year with a small update, phi-2 is now under MIT license, enjoy everyone!
huggingface.co/microsoft/phi-2 -
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
Starting the year with a small update, phi-2 is now under MIT license, enjoy everyone!
huggingface.co/microsoft/phi-2 -
Peter Lee
@peteratmsr
Living on the US west coast and working @MSFTResearch with labs in China, UK, Europe, India, Canada, and US east coast, means lots of 7am PT meetings -- the only time that everyone can be reasonably awake. We are global and hybrid, but time zones still matter.
-
Peter Lee reposted
Matthew Lungren MD MPH
@mattlungrenMD
Resolving to read more in 2024 or just ready to learn about gen AI in healthcare?
Lots of noise out there…
IMO best to begin with this book by @peteratmsr @zakkohane
Accessible/balanced exploration of one of the most consequential healthcare technologies in a generation! -
Peter Lee reposted
Eric Topol
@EricTopol
My top 10 books for 2023
@abe_verghese @peteratmsr @mustafasuleyman @DoctorChrisVT @PeterAttiaMD @kkariko @drfeifei @PeterHotez @DavidFajgenbaum @AnupamBJena @ChrisWorsham -
Peter Lee reposted
Andrew Beam
@AndrewLBeam
The last episode of season 1 of AIGR is out! We got @zakkohane in the booth to reflect on the launch of @NEJM_AI and to hear his unique take on the future of AI and medicine.
Link: open.spotify.com/episode/2FXUPWlfqcd2Poc0Jumaz8
If you've ever had the pleasure of hearing Zak speak then you already know this one is a can't miss! -
Peter Lee
@peteratmsr
My son introduced me to the lyrical genius of @blackthought. Thank you, @reidhoffman and Aria, for giving me the opportunity to share with you and your listeners. Happy New Year!
Reid Hoffman
@reidhoffman
Our guests on the Possible podcast over the last several months—from @emollick and @peteratmsr to @SlaughterAM and @MKBHD—give me hope that the future can be bright. They’ve also gifted us some optimism-renewing 📚🎬🎶 recs:
-
Peter Lee
@peteratmsr
Congrats, @goldbergcarey!
NEJM AI
@NEJM_AI
A longtime health and science reporter who has also been the Boston bureau chief of the New York Times and Bloomberg News, @goldbergcarey co-authored The AI Revolution in Medicine: GPT-4 and Beyond with Editor-in-Chief @zakkohane and @peteratmsr, Microsoft’s chief of research.
-
Peter Lee reposted
NEJM AI
@NEJM_AI
A longtime health and science reporter who has also been the Boston bureau chief of the New York Times and Bloomberg News, @goldbergcarey co-authored The AI Revolution in Medicine: GPT-4 and Beyond with Editor-in-Chief @zakkohane and @peteratmsr, Microsoft’s chief of research.
-
Peter Lee
@peteratmsr
Whiskey advent calendars are a wonderful thing. Just in case you didn't know.
-
Peter Lee
@peteratmsr
2023 has been the most consequential year in @MSFTResearch's 33-year history. In addition to countless internal (to Microsoft, OpenAI, etc) contributions, I bet that again our labs contributed more per researcher than any others to open research. Read more www.microsoft.com/en-us/research/blog/research-at-microso…
Peter Lee
@peteratmsr
And I should add that so much of what we do is enabled by fantastic collaborations with academic colleagues around the world. Most of our work is done with, and often led by, external researchers. Thank you all!
-
Peter Lee reposted
Paola Gori-Giorgi
@paolagorigiorgi
Our paper for the John Perdew special issue is out! A surprising symmetry in the gradient expansion at weak and strong coupling is observed for finite systems. And more... Work by the wonderful @kjdaas and @DerkKooi. #compchem pubs.aip.org/aip/jcp/article/159/23/234114/2930315/Large-…
-
Peter Lee reposted
Fernando Pérez-García
@fepegar_
We know "all models are wrong", partially because "all datasets are biased", but how wrong and how biased are they?
Check out our latest @MSFTResearch preprint on stress-testing with diffusion models to learn more.
"Causality matters in medical imaging"! -
Peter Lee
@peteratmsr
AI is far away from being able to replace humans in the delivery of healthcare; at the same time, even today's genAI will be an enabler of *incredible* improvements, in hundreds of different areas. If you have a naive view of either healthcare or AI, it can be hard to see that.
-
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
Check out this short video for a brief discussion of the phi series with @Microsoft CTO @kevin_scott , including why "textbooks" in "Textbooks Are All You Need" might not be exactly what you have in mind.
youtu.be/O-DjHgZt-Uk -
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
We're so pumped to see phi-2 at the top of trending models on @huggingface ! It's sibling phi-1.5 has already half a million downloads. Can't wait to see the mechanistic interpretability works that will come out of this & their impact on all the important LLM research questions!
-
Peter Lee reposted
clem 🤗
@ClementDelangue
Phi-2 by @MicrosoftAI is now the #1 trending model on @huggingface (hf.co/models). 2024 will be the year of smoll AI models!
-
Peter Lee
@peteratmsr
Thoughts on #genai from a patient's point of view by the always-thoughtful and wise @goldbergcarey.
NEJM AI
@NEJM_AI
Perspective: Author and longtime health and science reporter @goldbergcarey fantasizes about a letter on generative AI from her provider. Read “Patient Portal” by Carey Goldberg: nejm.ai/3t7cNIJ
#artificialintelligence -
Peter Lee reposted
Kevin Scott
@kevin_scott
And our work on Medprompt (including promptbase on GitHub) steers frontier models to better performance— achieving the highest score ever reported on the MMLU challenge.
Microsoft Research
@MSFTResearch
We have also developed a powerful set of techniques to help steer frontier models to even greater performance, first with MedPrompt, now extended with additional resources we’re making available on GitHub as promptbase. www.microsoft.com/en-us/research/blog/steering-at-the-fro…
-
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
This is a really excellent video, including an insightful discussion of benchmarks like MMLU: www.youtube.com/watch?v=nPgs8THgbuI
-
Peter Lee
@peteratmsr
This video of a chat between Microsoft CTO @kevin_scott and MSR AI researcher @SebastienBubeck is full of interesting insights about the phi family of large language models, and AI in general. Only 12 minutes, definitely worth watching. www.youtube.com/watch?v=O-DjHgZt-Uk
-
Peter Lee
@peteratmsr
A nice wrap-up by @kevin_scott of AI contributions @MSFTResearch at NeurIPS this past week. Two pieces I especially like: (1) an AI Copilot you can chat with to learn about our AI work; (2) a video chat with @SebastienBubeck, about phi-2 and AI. Enjoy! www.linkedin.com/pulse/microsoft-neurips-2023-kevin-scott…
-
Peter Lee reposted
Kevin K. Yang 楊凱筌
@KevinKaichuang
We are hiring for ML/stats/AI researchers at both the senior and principle levels. Come work @MSRNE with cool people like @LesterMackey, @nfusi, @elmelis, @alexijielu, @avapamini and @lorin_crawford!
-
Peter Lee reposted
Kevin Scott
@kevin_scott
I'm immensely proud of the work that my colleagues in Microsoft Research showcased at this week’s NeurIPS Conference regarding foundation models of all sizes, including Phi-2, which matches or outperforms models up to 25x larger.
Microsoft Research
@MSFTResearch
Today, we share our teams’ latest contributions, Phi-2 and promptbase.
Phi-2 outperforms other existing small language models, yet it’s small enough to run on a laptop or mobile device. www.microsoft.com/en-us/research/blog/phi-2-the-surprisin… -
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
phi-2 is really a good base for further fine-tuning: we FT on 1M math exercises (similar to phi-1 w. CodeExercises) & test on recent French nation-wide math exam (published after phi-2 finished training). The results are encouraging! Go try your own data huggingface.co/microsoft/phi-2
-
Peter Lee reposted
ACM Education & Learning Center
@acmeducation
In the latest #acmbytecast, @bruke_kifle hosts 2022 ACM Prize in Computing recipient Yael Tauman Kalai (@MSFTResearch & @MIT_CSAIL @MITEECS). They talk #cryptography, #ringsignatures, #quantumcomputing, & more.
Listen & subscribe: learning.acm.org/bytecast/ep47-yael-tauman-kalai -
Peter Lee reposted
Satya Nadella
@satyanadella
The pace of AI innovation in 2023 was astounding – but even more impressive was how people applied this technology to make a real difference in their careers, communities, and countries. Here are some of my favorite examples.
This media couldn't be displayed. See it on the original post. -
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
Enjoy everyone!
(And remember it's a base model so you might have to play around with your prompts; if you want it to follow instructions you can try the format "Instruct:... Ouput:")
huggingface.co/microsoft/phi-2 -
Peter Lee reposted
Hugging Face
@huggingface
Phi 2 on the Hub 🤗
Sebastien Bubeck
@SebastienBubeck
Enjoy everyone!
(And remember it's a base model so you might have to play around with your prompts; if you want it to follow instructions you can try the format "Instruct:... Ouput:")
huggingface.co/microsoft/phi-2 -
Peter Lee
@peteratmsr
This week I went from the amazing AI progress at #neurips23 to the inspiring work on healthcare quality at #ihiforum @TheIHI. Opportunities to improve the human condition are vast, but so important to close the gaps between these communities. Techies need to lean in more here.
-
Peter Lee
@peteratmsr
I truly enjoyed speaking to the healthcare improvement community @TheIHI. #genai will be such an important ingredient in health for everyone.
Amelia Crabtree
@crabtree_amelia
Final day @TheIHI #ihiforum 2023. Kicking off with @peteratmsr from @Microsoft giving the perfect demo of AI function as it currently stands. It's amazing, but can also be wrong. Think of it as an intern! (I suspect it will grow up to do great things).
-
Peter Lee
@peteratmsr
There is a cool AI chatbot powered by Azure AI Search (and Azure OpenAI Service) that helps you learn about AI @MSFTResearch. It includes info about our papers @NeurIPSConf, as well as Phi-2, promptbase, and the past year's research. Link to chatbot here www.microsoft.com/en-us/research/event/neurips-2023/
-
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
Sorry I know it's a bit confusing: to download phi-2 go to Azure AI Studio, find the phi-2 page and click on the "artifacts" tab. See picture.
Andrej Karpathy
@karpathy
No they fully released it. But they hide it very well for some reason. Go to artifacts tab.
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
FABind is a new method for fast and accurate protein-ligand binding, a key aspect of drug discovery. Learn about the work in the “Abstracts” podcast series with guests Tao Qin and Lijun Wu now and at #neurips2023 this week: www.microsoft.com/en-us/research/podcast/abstracts-decemb…
-
Peter Lee reposted
Andrej Karpathy
@karpathy
There's too much happening right now, so here's just a bunch of links
GPT-4 + Medprompt -> SOTA MMLU
www.microsoft.com/en-us/research/blog/steering-at-the-fro…
Mixtral 8x7B @ MLX nice and clean
github.com/ml-explore/mlx-examples/tree/main/mixtral
Beyond Human Data: Scaling Self-Training for Problem-Solving with Language Models
arxiv.org/abs/2312.06585
Phi-2 (2.7B), the smallest most impressive model
www.microsoft.com/en-us/research/blog/phi-2-the-surprisin…
LLM360: Towards Fully Transparent Open-Source LLMs
arxiv.org/abs/2312.06550
Honorable mentions
twitter.com/robertnishihara/status/1734629320868687991
twitter.com/arthurmensch/status/1734470462451732839
Aravind Srinivas
@AravSrinivas
mixtral-instruct is on labs.perplexity.ai. online versions without knowledge cutoff, and choice as a default model for perplexity pro users, coming soon!
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
We have also developed a powerful set of techniques to help steer frontier models to even greater performance, first with MedPrompt, now extended with additional resources we’re making available on GitHub as promptbase. www.microsoft.com/en-us/research/blog/steering-at-the-fro…
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
Today, we share our teams’ latest contributions, Phi-2 and promptbase.
Phi-2 outperforms other existing small language models, yet it’s small enough to run on a laptop or mobile device. www.microsoft.com/en-us/research/blog/phi-2-the-surprisin… -
Peter Lee reposted
Eric Horvitz
@erichorvitz
More aka.ms/AAo3s1z on promptbase & advances in prompting, including how we steered GPT-4 to record performance on the Measuring Massive Multitask Language Understanding (MMLU) benchmark, spanning 57 areas @satyanadella @kevin_scott @peteratmsr @HarshaNori @MSFTResearch
Satya Nadella
@satyanadella
From new best-in-class small language models to state-of-the-art prompting techniques, we’re excited to share these innovations and put them in the hands of researchers and developers.
-
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
Phi-2 numbers, finally! We're seeing a consistent ranking: phi-2 outperforms Mistral 7B & Gemini Nano 2* (*on their reported benchmarks) and is roughly comparable to Llama 2-70B (sometimes better, sometimes worse). Beyond benchmarks, playing with the models tells a similar story.
Satya Nadella
@satyanadella
From new best-in-class small language models to state-of-the-art prompting techniques, we’re excited to share these innovations and put them in the hands of researchers and developers.
-
Peter Lee
@peteratmsr
The native power of GPT-4 continues to impress us, especially how it can be steered with zero- or few-shot prompting strategies. For example, "promptbase" (now on GitHub) enables GPT-4 to achieve SOTA results on wide range of standard AI benchmarks. www.microsoft.com/en-us/research/blog/steering-at-the-fro…
-
Peter Lee
@peteratmsr
With only 2.7 billion parameters, Phi-2 @MSFTResearch surpasses models at 7B and 13B parameters on various aggregated benchmarks, and is better compared to the 25x larger Llama-2-70B model on muti-step reasoning tasks, i.e., coding and math. www.microsoft.com/en-us/research/blog/phi-2-the-surprisin…
-
Peter Lee reposted
Tian Xie
@xie_tian
Arrived at #neurips2023! Happy to meet old and new friends to chat about ML for materials, diffusion models, etc. @rpinsler, @ryotat and I will be at @MSFTResearch booth at 2:30-3pm Mon & Tue to chat about MatterGen.
Tian Xie
@xie_tian
[1/N] Generative AI has revolutionized how we create text and images. How about designing novel materials? We at @MSFTResearch #ai4science are thrilled to announce MatterGen: our generative model that enables broad property-guided materials design.
👇
arxiv.org/abs/2312.03687 -
Peter Lee reposted
Andrej Karpathy
@karpathy
# On the "hallucination problem"
I always struggle a bit with I'm asked about the "hallucination problem" in LLMs. Because, in some sense, hallucination is all LLMs do. They are dream machines.
We direct their dreams with prompts. The prompts start the dream, and based on the LLM's hazy recollection of its training documents, most of the time the result goes someplace useful.
It's only when the dreams go into deemed factually incorrect territory that we label it a "hallucination". It looks like a bug, but it's just the LLM doing what it always does.
At the other end of the extreme consider a search engine. It takes the prompt and just returns one of the most similar "training documents" it has in its database, verbatim. You could say that this search engine has a "creativity problem" - it will never respond with something new. An LLM is 100% dreaming and has the hallucination problem. A search engine is 0% dreaming and has the creativity problem.
All that said, I realize that what people *actually* mean is they don't want an LLM Assistant (a product like ChatGPT etc.) to hallucinate. An LLM Assistant is a lot more complex system than just the LLM itself, even if one is at the heart of it. There are many ways to mitigate hallcuinations in these systems - using Retrieval Augmented Generation (RAG) to more strongly anchor the dreams in real data through in-context learning is maybe the most common one. Disagreements between multiple samples, reflection, verification chains. Decoding uncertainty from activations. Tool use. All an active and very interesting areas of research.
TLDR I know I'm being super pedantic but the LLM has no "hallucination problem". Hallucination is not a bug, it is LLM's greatest feature. The LLM Assistant has a hallucination problem, and we should fix it.
</rant> Okay I feel much better now :) -
Peter Lee reposted
Microsoft Research
@MSFTResearch
The central problem in materials science is to discover materials with desired properties. MatterGen enables broad property-guided materials design: www.microsoft.com/en-us/research/blog/mattergen-property-…
-
Peter Lee reposted
Satya Nadella
@satyanadella
With our new MatterGen model, we're applying the next generation of AI to one of the biggest challenges in materials science: increasing the rate at which we design materials with desired properties.
Tian Xie
@xie_tian
[1/N] Generative AI has revolutionized how we create text and images. How about designing novel materials? We at @MSFTResearch #ai4science are thrilled to announce MatterGen: our generative model that enables broad property-guided materials design.
👇
arxiv.org/abs/2312.03687 -
Peter Lee reposted
Frank Noe
@FrankNoeBerlin
Fantastic work by my @MSFTResearch AI4Science colleagues. #generativeai for Materials is going to be a gamechanger, and this is only the beginning.
Tian Xie
@xie_tian
[1/N] Generative AI has revolutionized how we create text and images. How about designing novel materials? We at @MSFTResearch #ai4science are thrilled to announce MatterGen: our generative model that enables broad property-guided materials design.
👇
arxiv.org/abs/2312.03687 -
Peter Lee
@peteratmsr
I had a ton of fun in this chat with the @RealChemistry_ folks about GPT-4 in healthcare.
Real Chemistry
@RealChemistry_
This week @Microsoft 's Corporate VP of Research & Incubations, Peter Lee (@peteratmsr) joins the #realchempod.
He dives into the implications of GPT-4 in healthcare, and how Charlotte's Web is helping him test the limits of generative AI.
www.realchemistry.com/using-generative-ai-to-bring-new-pe… -
Peter Lee
@peteratmsr
Generative AI thus far has focused mainly on distilling knowledge and cognitive capabilities from human-generated data. But what happens if the same is applied to data from the natural world? We @MSFTResearch are seeing real, and tantalizing, results that could revolutionize parts of biology, material science, and chemistry.
Tian Xie
@xie_tian
[1/N] Generative AI has revolutionized how we create text and images. How about designing novel materials? We at @MSFTResearch #ai4science are thrilled to announce MatterGen: our generative model that enables broad property-guided materials design.
👇
arxiv.org/abs/2312.03687 -
Peter Lee reposted
Natalie Pageler MD
@NatalieMPageler
Looking forward to a stimulating conversation with @peteratmsr tomorrow morning. Don't miss it! #stanfordaihealth
-
Peter Lee
@peteratmsr
We are full steam ahead on glass-based long-term storage technology, aka Project Silica. A great joint effort between @MSFTResearch and @Azure, in a fantastic brand-new research+production facility in Cambridge UK. www.techradar.com/pro/microsoft-inches-closer-to-glass-st…
-
Peter Lee reposted
Satya Nadella
@satyanadella
The pace of innovation in Copilot's first year has been astounding. And we're not done. Looking forward to even more advanced capabilities, including responses powered by GPT-4 Turbo and Dall-E 3, Code Interpreter, and more. blogs.microsoft.com/blog/2023/12/05/celebrating-the-first…
-
Peter Lee reposted
Reid Hoffman
@reidhoffman
For the full conversation with Peter Lee:
Watch: youtu.be/9FZ3jCplwKg
Read: possible.fm/podcasts/peterlee/
Listen: link.chtbl.com/M9jVQqAR -
Peter Lee reposted
Satya Nadella
@satyanadella
As we continue to celebrate #idpd, we embrace our responsibility to use technology to create a more accessible society. We’re thrilled to introduce a new version of Seeing AI, which applies next generation AI to narrate the world – right from your phone. blogs.microsoft.com/accessibility/seeing-ai-app-launches-…
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
ASL Citizen is the first crowdsourced sign language dataset, advancing the state of the art in sign recognition. The web-based project captured input from people in real-world settings, and from a diverse group of experts, including Deaf team members. www.microsoft.com/en-us/research/blog/tackling-sign-langu…
-
Peter Lee
@peteratmsr
Much credit for this book goes to @goldbergcarey for making this book so readable, and @zakkohane for important clinical insights. Also @SebastienBubeck for key technical guidance.
José Pedro Almeida
@jpedroalmeida_
Peter also wrote an outstanding book centered around GPT4 in Medicine that I hugely recommend, clearly visionary on what's next 🚀👌
-
Peter Lee
@peteratmsr
I felt silly the first time I used this "Charlotte's Web" example to explain to nontechnical newcomers what LLMs like GPT-4 (currently) can and can't do. But it's turned out to be more useful to people than expected.
Reid Hoffman
@reidhoffman
A creative demonstration of the capabilities of GPT-4 through Charlotte's Web, from @MSFTResearch exec @peteratmsr:
-
Peter Lee
@peteratmsr
In early studies, patients sometimes find that notes drafted by GPT-4 feel more empathetic. The way I see it: AI can't be "empathetic", but its outputs can act as a "reverse prompt" to the doctor, suggesting things to say that might boost mood and make an emotional connection.
Reid Hoffman
@reidhoffman
We spoke with @peteratmsr about partnerships, such as with hospitals and healthcare software companies. He shared that, in early studies, patients prefer notes generated by GPT-4 and have explicitly said they feel more human than the notes written by the doctors.
Here's more: -
Peter Lee reposted
Reid Hoffman
@reidhoffman
We spoke with @peteratmsr about partnerships, such as with hospitals and healthcare software companies. He shared that, in early studies, patients prefer notes generated by GPT-4 and have explicitly said they feel more human than the notes written by the doctors.
Here's more: -
Peter Lee reposted
Eric Topol
@EricTopol
“We believe that genAI will deliver meaningful improvements in health care more rapidly than was the case with previous technologies.”—@Bob_Wachter and @erikbryn in a very thoughtful @JAMA_current essay
jamanetwork.com/journals/jama/fullarticle/2812615 -
Peter Lee reposted
Reid Hoffman
@reidhoffman
Here's @peteratmsr of @MSFTResearch calling on AI for a mock curbside consult for a patient with chest pain.
This media couldn't be displayed. See it on the original post. -
Peter Lee reposted
Reid Hoffman
@reidhoffman
For the full conversation with Peter Lee:
Listen: link.chtbl.com/M9jVQqAR
Watch: youtu.be/9FZ3jCplwKg
Read: www.possible.fm/podcasts/peterlee/ -
Peter Lee reposted
Eric Topol
@EricTopol
Defying expectations, a generalist (all purpose LLM) foundation model exceeded intensive fine-tuning for medical challenge questions
arxiv.org/pdf/2311.16452.pdf @HarshaNori @erichorvitz and colleagues @Microsoft -
Peter Lee
@peteratmsr
.@reidhoffman and @AriaIrene do such a great job in these conversations. Really helped draw out not only the amazing potential of AI in medicine, but what it could end up meaning at a more personal level.
Reid Hoffman
@reidhoffman
How will AI assist doctors, nurses, and patients—along with caretakers and family members—at every stage of care?
This week, @MSFTResearch exec and member of @theNAMedicine Peter Lee joins the show to discuss whether, when, and how to use AI in medicine:
link.chtbl.com/M9jVQqAR -
Peter Lee
@peteratmsr
If you have a powerful general AI model like GPT-4, prompting goes a very long way. Fine-tuning is likely the last resort, and definitely not the first thing to go for, if you want best performance.
Eric Horvitz
@erichorvitz
1/8 We’ve published a study of the power of prompting to unleash expertise from GPT-4 on medical benchmarks without additional fine-tuning or expert-curated prompts: arxiv.org/pdf/2311.16452.pdf
Summary of results: -
Peter Lee
@peteratmsr
For medical knowledge tasks, some think that for best results you need fine-tuning and prompts crafted by human medical experts . But no! This study @MSFTResearch & @erichorvitz shows that generalist #gpt4 beats specialized models *and* is less expensive. www.microsoft.com/en-us/research/blog/the-power-of-prompt…
Peter Lee
@peteratmsr
The lesson here is NOT that medical knowledge tasks are important. Indeed, we find that other cognitive healthcare tasks are more important. The real lesson of here is that prompting is very powerful in a model like #gpt4 because the model itself can help generate good prompts.
-
Peter Lee reposted
Saleema Amershi
@SaleemaAmershi
📢📢We're hiring!📢📢
If you want to shape the future of AI and empower people and AI agents to collaboratively solve real-world problems, apply here:
jobs.careers.microsoft.com/global/en/job/1656948/Research…
See also below for more exciting opportunities in #ai at #msr with our partner teams. 👇👇 -
Peter Lee reposted
Greg Brockman
@gdb
GPT-4 for radiology. Far from perfect, but state-of-the-art performance on some tasks: “Surprisingly, we found radiology report summaries generated by GPT-4 to be comparable and, in some cases, even preferred over those written by experienced radiologists”
www.microsoft.com/en-us/research/blog/gpt-4s-potential-in… -
Peter Lee reposted
Kevin K. Yang 楊凱筌
@KevinKaichuang
HR has informed me that the posting will disappear after Dec 1st, and that there's no way to make exceptions, so please get your application and research statement in before then!
Kevin K. Yang 楊凱筌
@KevinKaichuang
Apply for a PhD research internship with the bioml group @MSRNE!
- work with me, @alexijielu @avapamini @lorin_crawford and/or Kristen Severson!
- Enjoy this view
- Sometimes I bring treats
jobs.careers.microsoft.com/global/en/job/1658689/Research… -
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
We trained a small transformer (100M params) for basic arithmetic. W. the right training data it nails 12x12 digits multiplication w/o CoT (that's 10^24 possibilities, so no it's not memorization🤣).
Maybe arithmetic is not the LLM kryptonite after all?🤔
arxiv.org/abs/2311.14737 -
Peter Lee reposted
Microsoft Research
@MSFTResearch
As the benefits of AI in healthcare continue to emerge, we investigate GPT-4's potential in radiology. Learn about research exploring GPT-4’s potential in assisting report structuring, classifying diseases, and generating comprehensive findings summaries: www.microsoft.com/en-us/research/blog/gpt-4s-potential-in…
-
Peter Lee
@peteratmsr
I'm amazed at the AI discoveries being made @MSFTResearch and so many other great places every day. Still, with so many open research questions, it can be confusing to see well-known researchers make loud assertions about them. Good to stay focused on doing good research and choosing words thoughtfully. Less about making headlines/likes, but instead adding the most value.
-
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
Join us on YouTube at 1pm PT/4pm ET today for the premiere of our "debate" with @bgreene @ylecun @tristanharris on whether a new kind of intelligence has emerged with GPT-4, and what consequences it might have.
www.youtube.com/watch?v=EGDG3hgPNp8 -
Peter Lee reposted
Microsoft Research
@MSFTResearch
The November 29th deadline is approaching to join the Microsoft Research AI & Society Fellows program. Collaborate on interdisciplinary research challenges focused on AI in the context of societal impact. Submit your proposal now: aka.ms/MSRAISocietyFellow
-
Peter Lee
@peteratmsr
Sure it would have been fun having Sam, Greg, and so many others as new Microsoft colleagues. But man, this is a such wonderful outcome for all of us at OpenAI and Microsoft, and for everyone else who cares about the future of AI for all.
Satya Nadella
@satyanadella
We are encouraged by the changes to the OpenAI board. We believe this is a first essential step on a path to more stable, well-informed, and effective governance. Sam, Greg, and I have talked and agreed they have a key role to play along with the OAI leadership team in ensuring OAI continues to thrive and build on its mission. We look forward to building on our strong partnership and delivering the value of this next generation of AI to our customers and partners.
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
Don't miss the upcoming Quantum Innovator Series episode on November 28th with Nathan Baker as he explores "How can Azure Quantum Elements accelerate scientific discovery today and in the future?" Register now: www.eventbuilder.rocks
-
Peter Lee reposted
Frank X. Shaw
@fxshaw
Fun one!
-
Peter Lee reposted
Kevin Scott
@kevin_scott
Wanted to take a moment to say to all of the unbelievably talented folks at Open AI I've had the privilege of working with for years: we're going to continue supporting you and the important work that you're doing, no matter what. @sama, @satyanadella, and I have got your backs.
-
Peter Lee reposted
Satya Nadella
@satyanadella
I’m super excited to have you join as CEO of this new group, Sam, setting a new pace for innovation. We’ve learned a lot over the years about how to give founders and innovators space to build independent identities and cultures within Microsoft, including GitHub, Mojang Studios, and LinkedIn, and I’m looking forward to having you do the same.
-
Peter Lee
@peteratmsr
❤️
Sam Altman
@sama
the mission continues
Satya Nadella
@satyanadella
We remain committed to our partnership with OpenAI and have confidence in our product roadmap, our ability to continue to innovate with everything we announced at Microsoft Ignite, and in continuing to support our customers and partners. We look forward to getting to know Emmett Shear and OAI's new leadership team and working with them. And we’re extremely excited to share the news that Sam Altman and Greg Brockman, together with colleagues, will be joining Microsoft to lead a new advanced AI research team. We look forward to moving quickly to provide them with the resources needed for their success.
-
Peter Lee
@peteratmsr
❤️
Satya Nadella
@satyanadella
We remain committed to our partnership with OpenAI and have confidence in our product roadmap, our ability to continue to innovate with everything we announced at Microsoft Ignite, and in continuing to support our customers and partners. We look forward to getting to know Emmett Shear and OAI's new leadership team and working with them. And we’re extremely excited to share the news that Sam Altman and Greg Brockman, together with colleagues, will be joining Microsoft to lead a new advanced AI research team. We look forward to moving quickly to provide them with the resources needed for their success.
-
Peter Lee
@peteratmsr
For someone who professes to be 100% focused on racing, Max Verstapoen has been talking a *lot* about Vegas, clowns, Toto Wolff, etc. He's never before spent more time in front of cameras talking about non-racing stuff than he has this week. He seems preoccupied. Hope his race goes well.
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
From service in the Singapore Armed Forces to autonomous navigation with NASA & VR with Disney, Desney Tan’s life journey hasn’t been linear. Learn how Tan landed at Microsoft & about the purpose guiding his work in the podcast series “What’s Your Story”: www.microsoft.com/en-us/research/podcast/whats-your-story…
-
Peter Lee reposted
Flagship Pioneering
@FlagshipPioneer
Check out highlights from our inaugural #artificialintelligence Summit with leaders from academia, industry, and across the Flagship ecosystem, including @Flagship_PI, a Flagship initiative harnessing AI to accelerate innovation in the life sciences and beyond. Watch to hear from @NoubarAfeyan, @GVMaltzahn, Armen Mkrtchyan, @MontaiHealth's Margo Georgiadis, Microsoft's @peteratmsr, and MIT's Dan Huttenlocher and @MarzyehGhassemi.
📣 Follow @Flagship_PI for more on how Flagship is capitalizing on trends in AI and biology to drive innovation and help advance the impact of our companies to transform human health and sustainability.This media couldn't be displayed. See it on the original post. -
Peter Lee
@peteratmsr
Returning today from two weeks in Asia. The highlight: hanging out @MSFTResearch Asia, getting deep into the weeds on the most consequential research in the world, as well as celebrating the 25th anniversary of the lab. So many incredible tech advances, too many to count!
Peter Lee
@peteratmsr
Oh, I should add: The visits to both Peking University and University of Tokyo @UTokyo_News_en were awesome. I was seriously impressed by the depth and sophistication of the questions from both students and faculty. Really, really impressive. L
-
Peter Lee
@peteratmsr
An incredible number of announcements today from Microsoft. All of us are contributing at a rate I've never experienced before.
Satya Nadella
@satyanadella
Copilot will be the new UI for both the world's knowledge and your organization's knowledge, but most importantly, it will be your agent that helps you act on that knowledge. Here are highlights from my keynote today at #msignite.
This media couldn't be displayed. See it on the original post. -
Peter Lee
@peteratmsr
Another major step towards understanding the potential of smaller foundation models. And as before, in open source from @MSFTResearch.
Sebastien Bubeck
@SebastienBubeck
Microsoft💜Open Source + SLMs!!!!!
We're so excited to announce our new *phi-2* model that was just revealed at #msignite by @satyanadella!
At 2.7B size, phi-2 is much more robust than phi-1.5 and reasoning capabilities are greatly improved too. Perfect model to be fine-tuned! -
Peter Lee reposted
Moshe Vardi
@vardi
:-)
-
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
My group is hiring a large cohort of interns for the summer of 2024 to work on the Foundations of Large Language Models! Come help us uncover the new physics of A.I. to improve the LLM building practices! (Pic below from our NeurIPS 2023 paper w. interns)
jobs.careers.microsoft.com/global/en/job/1656950/Research… -
Peter Lee reposted
Eric Topol
@EricTopol
Promoting empathy in medicine with #ai
A thoughtful op-comic by @NathanAGray
www.latimes.com/opinion/story/2023-11-12/ai-healthcare-do… @latimesopinion -
Peter Lee reposted
Eric Topol
@EricTopol
Getting the correct diagnosis for 38 complex patient case challenges
#gpt4 vs medical journal readers
onepub-media.nejmgroup-production.org/ai/media/ec2de32e-9…?
"We found the performance of GPT-4 in complex clinical cases to be quite impressive"
--@EriksenViktor
@JesperRyg @SUND_SDU @NEJM_AI -
Peter Lee reposted
Katie Link
@katieelink
The LLaVA-Med model by @MSFTResearch is now open access! 🌋🩺
LLaVA-Med is a 7B biomedical vision-language model initialized from the general-domain LLaVA model and then trained on a large dataset of PubMed Central figure-captions
Check it out on @huggingface (link below)! -
Peter Lee reposted
Nicolò Fusi
@nfusi
We are looking for AI/ML/stats interns at MSR New England (in Boston) jobs.careers.microsoft.com/global/en/share/1660083/ We are interested in many areas/topics, but will definitely have projects in generative models (discrete diffusion models, multimodal) and DEQs.
-
Peter Lee
@peteratmsr
It's amazing to see how many startups are building transformational products and services powered by AI. So we've added more Azure AI infrastructure for model training and inference for @Y Combinator and @M12 startups.
Microsoft for Startups
@msft4startups
We’re excited to announce we’re enhancing startup AI support with additional Azure AI infrastructure, including high-end GPU VM clusters. We're proud to offer this to @ycombinator and @M12vc startups to boost AI innovation. Learn more: blogs.microsoft.com/blog/2023/11/07/startups-to-access-hi…
-
Peter Lee reposted
Ethan Mollick
@emollick
I have been writing a guide to practical use of AI that I update every few months. The last version is from July and increasingly obsolete (stuff moves fast!), but I keep waiting for the Gemini launch to update... www.oneusefulthing.org/p/how-to-use-ai-to-do-stuff-an-opi…
-
Peter Lee reposted
Satya Nadella
@satyanadella
It was great to join @sama at OpenAI’s first ever DevDay yesterday. We are committed to ensuring OpenAI has the best possible systems to train the most advanced models for our mutual customers. www.microsoft.com/en-us/microsoft-cloud/blog/2023/11/07/c…
-
Peter Lee
@peteratmsr
I really learned a lot at @Flagship_PI. So many building and shaping AI-first biotech in such interesting and promising ways.
Flagship Pioneering
@FlagshipPioneer
Catch some snapshots from our inaugural AI Summit, which marked the official launch of Pioneering Intelligence (@Flagship_PI) and gathered life science leaders from around the world for a series of engaging discussions around the latest advances in AI and machine learning how it applies to the biotech industry.
#fspaisummit #artificialintelligence #machinelearning -
Peter Lee reposted
Yusuf Mehdi
@yusuf_i_mehdi
Congrats to @OpenAI on their first DevDay. Lots of amazing new capabilities! We’re excited to support these with @Microsoft Copilot and Microsoft Azure AI. More to come at our Ignite conference on Nov 15-16. openai.com/blog/new-models-and-developer-products-announc…
-
Peter Lee
@peteratmsr
Lightning and thunder this morning in the Seattle area. Yes, we are entering the rainy season here, but I don't remember this kind of thing happening much before the last couple of years. Usually just drizzle. The dog for sure isn't used to it.
-
Peter Lee reposted
Kenji Takeda
@ktakeda1
Excited to join forces on #ai and Society with collaborators around the world. Please apply by 29th Nov -> www.microsoft.com/en-us/research/academic-program/ai-soci…
#responsibleai #socialscience #humanities
Microsoft Research
@MSFTResearch
Interdisciplinary research can drive real-world impact. Join the new Microsoft Research AI & Society Fellows program, uniting eminent scholars and experts to collaborate on research at the intersection of AI and society. Proposals due November 29th: www.microsoft.com/en-us/research/academic-program/ai-soci…
This media couldn't be displayed. See it on the original post. -
Peter Lee
@peteratmsr
Smaller models like phi-1.5 @MSFTResearch enables investigation of the impact of things like synthetic training data and multimodality in LLMs. Making them open hopefully also accelerates research on LLM transparency and evaluation. And yes, they can be practical, too.
Peter Lee
@peteratmsr
And continuing on the specific subject of synthetic training data: As @SebastienBubeck explains, there is a ton of confusion out there, even in the specialist research community. The idea of synthetic data is more subtle and interesting than one might assume. Follow-up article:
-
Peter Lee reposted
Atul Butte (In Memoriam)
@atulbutte
Remembering this from 5 years ago... lessons for clinical #ai startups that are probably still valuable!
Atul Butte (In Memoriam)
@atulbutte
Some of the lessons I've learned researching #ai & machine learning over the past 20 years… #pmwc18 @UCSF
-
Peter Lee reposted
Isaac Kohane
@zakkohane
Sun rising over Symposium for Responsible AI for Social and Ethical Healthcare (RAISE) in Maine with a small group of thoughtful leaders. Results will be public and shared @NEJM_AI @LancetDigitalH @NatureMedicine soon. @MooreFound
@MaineHealth
@microsoft
@google
@TJCommission
@apple
@NHSX
@LancetDigitalH
@NatureMedicine
@NEJM_AI
@NEJMEvidence
@unibirmingham
@ePatientDave
@ashishkjha
@Brown_SPH
@HarvardChanSPH
@HarvardDBMI
@lightcollective
@theNAMedicine
@ClalitInnovate
@a16z
@DavisInstAI
@INSIGHTeyehub
@ConsortSpiritAI
@HMSbioethics
@UKZN
@EP_ThinkTank
@JNJInnovMed
@JanssenGlobal
@wyssCenter
@UniofOxford
@VeteransHealth
@DeptVetAffairs
@Umich
@UFLaw
@ONC_HealthIT
@UCSF
@EthicsPolicyLab
@HealthFdn
@SNForg
@eth_en
@CMSGov -
Peter Lee
@peteratmsr
I'm looking forward to speaking about #generativeai in healthcare to @TheIHI later this month. As a lead-up to that, I did an interview with them, discussing the great potential, risks, and the surprising impact of GPT-4's ability to generate empathetic language:
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
In this issue: Kosmos-2.5: A Multimodal Literate Model; Can vine copulas explain complex relationships of weather variables; New system accelerates the adaptive training process; Structural inequalities and relational labor in the influencer industry: www.microsoft.com/en-us/research/blog/research-focus-week…
-
Peter Lee reposted
Microsoft for Startups
@msft4startups
Using LLMs as reasoning engines can be a game changer. Watch as AI experts @SebastienBubeck and William of @langchain discuss the different possibilities that LLMs can unlock.
Watch the full video at www.youtube.com/playlist?list=PLlzNGEvTIKySN_7euoQJrm_zmx…This media couldn't be displayed. See it on the original post. -
Peter Lee reposted
Ethan Mollick
@emollick
Work from Harvard & Microsoft suggests GPT-4 is surprisingly helpful in radiology:
“This study evaluates GPT-4 on a diverse range of common radiology text-based tasks. We found GPT-4 either outperforms or is on par with task-specific radiology models.” arxiv.org/pdf/2310.14573.pdf -
Peter Lee reposted
Satya Nadella
@satyanadella
Heartbroken by the horrific terrorist attacks on Israel and the escalating conflict. My deepest condolences are with all those killed and impacted. Our focus remains on ensuring the safety of our employees and their families. Below is a message we shared with Microsoft employees today about our response. blogs.microsoft.com/blog/2023/10/10/microsoft-employee-em…
-
Peter Lee
@peteratmsr
Really appreciate the kind review of our book (with @goldbergcarey @zakkohane @SebastienBubeck) on AI (GPT-4) in medicine, @ePatientDave! www.epatientdave.com/2023/10/03/the-ai-revolution-in-medi…
-
Peter Lee
@peteratmsr
Woah, that was a bit of a jolt. earthquake.usgs.gov/earthquakes/eventpage/uw61965081/exec…
-
Peter Lee
@peteratmsr
Just 30-minutes to get Llama2-7b to "forget" Harry Potter!? Amazing, and while implications are still unclear, raises again questions about the care needed when fine-tuning of LLMs.
-
Peter Lee
@peteratmsr
Congratulations, @LesterMackey! So proud of you, and such a well-deserved recognition!
Microsoft Research
@MSFTResearch
Microsoft Research congratulates computer scientist and statistician @LesterMackey on his selection as a 2023 MacArthur Fellow. www.macfound.org/fellows/class-of-2023/lester-mackey #macfellow
-
Peter Lee reposted
Yusuf Mehdi
@yusuf_i_mehdi
Last week, @Microsoft was ranked first overall, and first for employee satisfaction, on the @TIME World's Best Companies list for 2023. Proud to see #microsoft at the top of that list! time.com/7364052/worlds-best-companies-of-2023/
-
Peter Lee reposted
Ethan Mollick
@emollick
A prompt library for experimental use in education: github.com/microsoft/prompts-for-edu
Microsoft* put together a prompt library for GPT-4 for educators, students, & admin using some of our prompts as a basis. Add your own!
*We do not take any compensation from them or any other AI company -
Peter Lee
@peteratmsr
It was an honor to be in St. Louis speaking to the Mercy leadership team about AI as our two organizations announce a key moment in healthcare. @Microsoft and Mercy are collaborating to help empower providers to give better patient care with the help of generative AI and Microsoft Azure OpenAI Service. news.microsoft.com/2023/09/27/microsoft-and-mercy-collabo…
-
Peter Lee
@peteratmsr
I was so pleased to participate in the Taiwan launch of our book, "The AI Revolution in Medicine: GPT-4 and Beyond" @goldbergcarey @zakkohane @SebastienBubeck. Super discussion by great Taiwanese tech leaders, on the future of healthcare and Taiwan's place in that.
-
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
Looking forward to discuss whether we are witnessing a new kind of intelligence tomorrow in NYC with @ylecun @tristanharris @bgreene ! Curious to see where the discussion takes us :-).
Cc @worldsci -
Peter Lee reposted
Ethan Mollick
@emollick
Increasing evidence across multiple papers that GPT-4 is a reliable substitute for humans as a grader/ rater/ categorizer of human-made text.
More work is needed to really nail down limitations and ideal prompts, but this would be a big boon to research (& in education overall). -
Peter Lee reposted
Ava Amini
@avapamini
presenting EvoDiff: new generative models for controllable protein design from sequence data alone!
✅generate high-quality proteins
✅scaffold functional motifs
🤩apply to therapeutic design + more!
👉www.biorxiv.org/content/10.1101/2023.09.11.556673v1
💻github.com/microsoft/evodiff
🎙️www.microsoft.com/en-us/research/podcast/abstracts-septem… -
Peter Lee reposted
Kevin K. Yang 楊凱筌
@KevinKaichuang
EvoDiff combines evolutionary-scale data with diffusion models for controllable protein sequence generation.
In addition to generating plausible proteins, we can scaffold structural motifs in sequence space!
Preprint: biorxiv.org/cgi/content/short/2023.09.11.556673v1
Code: github.com/microsoft/evodiff -
Peter Lee reposted
Nicolò Fusi
@nfusi
Operating directly in protein sequence space has so many advantages, particularly when thinking about scaling generative models. Really exciting work by the team
Kevin K. Yang 楊凱筌
@KevinKaichuang
EvoDiff combines evolutionary-scale data with diffusion models for controllable protein sequence generation.
In addition to generating plausible proteins, we can scaffold structural motifs in sequence space!
Preprint: biorxiv.org/cgi/content/short/2023.09.11.556673v1
Code: github.com/microsoft/evodiff -
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
I'm amazed by how fast this is happening, but that's exactly what were hoping for by open sourcing phi-1.5!! This new version by @teknium already seems to perform better than the base model. Can't wait to see how far phi-1.5 can be pushed.
Teknium 🪽
@Teknium
Okay, I think I have a decent finetune of Phi-1.5 with the Puffin dataset.
Uploading it now to huggingface: huggingface.co/teknium/Puffin-Phi-v2
It's able to correctly tell me Bill Clinton was US President in 1995, and write a negative review of twitter (However, it output nothing when I asked it to write a positive review of Twitter, a protest perhaps?)
Am currently still training hermes over phi will release that when done as well. -
Peter Lee
@peteratmsr
A new LLM, called phi-1.5, released today by @SebastienBubeck and his team @MSFTResearch. Released as open source to support open research on AI foundations, transparency, and safety, this model does text completion with quality that compares favorably to much larger LLMs.
Sebastien Bubeck
@SebastienBubeck
How far does one billion parameters take you? As it turns out, pretty far!!!
Today we're releasing phi-1.5, a 1.3B parameter LLM exhibiting emergent behaviors surprisingly close to much larger LLMs.
For warm-up, see an example completion w. comparison to Falcon 7B & Llama2-7B -
Peter Lee
@peteratmsr
Hot take after doing a bunch of demo rides in new EVs at @ElectrifyExpo today: almost every new high-ish end EV coming into the market today is much better than any Tesla. Except for charging network. But aside from that, so many really EVs much nicer than Tesla...
-
Peter Lee
@peteratmsr
Surprisingly big crowds all day at @ElectrifyExpo Seattle. Pretty pleased by the interest in my classic car EV conversion. And fun to try out all the different cars, ebikes, and scooters.
-
Peter Lee reposted
Roxana Daneshjou MD/PhD
@RoxanaDaneshjou
Thank you @StanfordHAI for featuring our recent paper on using AI to detect bias in dermatology education materials! We built a pipeline for automatically extracting and evaluating images of skin disease in textbooks and found representation is lacking.
hai.stanford.edu/news/ai-shows-dermatology-educational-ma… -
Peter Lee reposted
Kenji Takeda
@ktakeda1
Still time to apply for access to foundation models via Microsoft Azure AI services #gpt4 - including $20,000 USD of Azure credits for your academic research. Deadline 12 Sept. microsoft.com/en-us/research/academic-program/accelerate-…
@DeepIndaba @turinginst @responsibleaiuk @UKRI
@ELLISforEurope @ELIXIREurope
Microsoft Research
@MSFTResearch
There’s still time to join our AI research program for academic access to industry-scale models. Explore human-AI alignment, human benefit & scientific discovery. Submit proposals for the Microsoft Accelerate Foundation Models Research Program by Sept 12: www.microsoft.com/en-us/research/academic-program/acceler…
-
Peter Lee reposted
Besmira Nushi 💙💛
@besanushi
Our latest research review on Frontiers of Multimodal Learning takes a responsible AI approach to model evaluation & improvement and makes 4 practical observations on insufficiencies of current methods and promising mitigation steps. @MSFTResearch blog at www.microsoft.com/en-us/research/blog/frontiers-of-multim…
Microsoft Research
@MSFTResearch
New evaluation methods and a commitment to continual improvement are musts if we’re to build
multimodal AI systems that advance human goals. Learn about cutting-edge research into the
responsible development and use of multimodal AI at Microsoft: www.microsoft.com/en-us/research/blog/frontiers-of-multim… -
Peter Lee
@peteratmsr
I'm looking forward to checking out the Seattle edition of the @ElectrifyExpo this weekend. I'll be showing off my recent classic car EV conversion, and looking forward to seeing what all the other great "EV hacks" on display. www.electrifyexpo.com
-
Peter Lee
@peteratmsr
If you are looking to use generative AI to create new things, it's worth checking out our Copilot Copyright Commitment.
Brad Smith
@BradSmi
Today, we’re sharing our Copilot Copyright Commitment that extends IP legal protection to enterprise customers who use our Copilot AI services. The copyright liability of our products is ours to shoulder, not our customers. blogs.microsoft.com/on-the-issues/2023/09/07/copilot-copy…
-
Peter Lee reposted
Isaac Kohane
@zakkohane
If I knew someone suffering from a potentially rare or undiagnosed disease, I would not hesitate to engage with at least one if not more of the large language models that are now available to all of us to see if I could find any suggestions or insights that the doctors so far had not. I’m looking forward to discussing this, perhaps controversial, perspective. @UndiagnosedFilm @UDNconnect
-
Peter Lee reposted
Lake Nona Impact Forum
@LNImpactForum
At #lnif23, the @Microsoft team gave attendees a preview of the company’s #gpt4-powered #ai chatbot using prompts for #healthcare questions. @PeterAtMsr discusses the benefits of sharing this technology with the guests of the Impact Forum.
Learn more: lakenonaimpactforum.org -
Peter Lee reposted
ashish (@acgt01@genomic.social)
@acgt01
Fantastic editorial piece by @atulbutte
In India, ai will work as a force multiplier for our tiny health workforce and help us leapfrog the legacy health IT infra to realize the promise of data driven personalized medicine for a billion people
Cc : @vijaypande @AnuragAgrawalMD
Eric Topol
@EricTopol
The potential for LLM #ai to reduce inequities in cancer care, editorial by @atulbutte @JAMAOnc with 2 new studies on responses to patient questions
jamanetwork.com/journals/jamaoncology/article-abstract/28…
jamanetwork.com/journals/jamaoncology/fullarticle/2808733
jamanetwork.com/journals/jamaoncology/fullarticle/2808731 -
Peter Lee reposted
Andrew Ng
@AndrewYNg
How can business leaders, not just coders, build AI business applications with LLMs? Our new short course, created with Microsoft, is out! In How Business Thinkers Can Start Building AI Plugins With Semantic Kernel, taught by @johnmaeda, you'll learn to use Semantic Kernel and build an AI planner that can automatically select the plugins it needs to create multi-step plans with sophisticated logic. Please sign up here: www.deeplearning.ai/short-courses/microsoft-semantic-kernel
This media couldn't be displayed. See it on the original post. -
Peter Lee reposted
Eric Topol
@EricTopol
The potential for LLM #ai to reduce inequities in cancer care, editorial by @atulbutte @JAMAOnc with 2 new studies on responses to patient questions
jamanetwork.com/journals/jamaoncology/article-abstract/28…
jamanetwork.com/journals/jamaoncology/fullarticle/2808733
jamanetwork.com/journals/jamaoncology/fullarticle/2808731 -
Peter Lee reposted
Ethan Mollick
@emollick
Folks keep testing the same 3 things LLMs are bad at and declaring LLMs a failure. Yes, they are bad at:
1) Language games. "Give me 10 words starting & ending with vowels"
2) Providing accurate cites & quotes
3) Doing math (without Python)
I wouldn't ask AI to do those things! -
Peter Lee
@peteratmsr
According to @politico (and I agree): "A health care AI pact that matters" www.politico.com/newsletters/future-pulse/2023/08/24/a-he…
-
Peter Lee reposted
Satya Nadella
@satyanadella
Great to join Epic today at their annual Users Group Meeting to discuss AI in healthcare and our expanded collaboration. Together, we're working to deploy dozens of copilot solutions that will help address the growing challenges facing healthcare. aka.ms/AAm38vn
-
Peter Lee reposted
Rasu Shrestha MD MBA
@RasuShrestha
The brilliant @SatyaNadella schools us on #generativeai.
Listen in. 👇🏽
Simply love it, and such superb dialogue with the ever empathetic and insightful Sumit Rana of Epic. 🙌🏽
#epicugm #ugm2023 #aiThis media couldn't be displayed. See it on the original post. -
Peter Lee
@peteratmsr
For sure! This is definitely a fantastic collaboration that will improve the workdays of doctors and nurses everywhere, and has a great chance to contribute to the advancement of medical science. So great to work with @sethHain and the Epic team.
Seth Hain
@sethHain
Combining important challenges in healthcare, emerging tech possibilities, and creative teams can't help but lead to fun, fruitful, and fulfilling collaboration. @peteratmsr, @eric_boyd, excited to continue…
#ugm2023
blogs.microsoft.com/blog/2023/08/22/microsoft-and-epic-ex… -
Peter Lee reposted
Vardit Ravitsky
@VarditRavitsky
At the first meeting of the @theNAMedicine Artificial Intelligence Code of Conduct project, I was invited to present a high-level view of bioethics perspectives. I prepared a one-pager to facilitate our discussion – sharing here in case helpful to others.
nam.edu/programs/value-science-driven-health-care/health-… -
Peter Lee reposted
Rachel Woods
@rachel_l_woods
There's a resurgence of interest in fine tuning LLMs
I've yet to see a successful public use case where fine tuning > prompting.
But here's where I see fine tuning *mattering*:
First, fine tuning is for teaching an LLM specific tasks or behaviors
Not teaching an LLM new knowledge. For new knowledge, use Retrieval (store your data in an outside database and strategically pull the right chunks in to give the LLM context to your question)
But even in teaching LLMs specific tasks or behaviors - here's the catch...
LLMs are remarkably good at picking up tasks and behaviors from just a good prompt
THIS is what makes LLMs mind blowing after all
So that begs the question.
Where is fine tuning actually helpful?
Some use cases I could see developing are teaching LLMs tasks that are exceptionally difficult to describe, or fit into ~10 examples you can add to a prompt.
One way to think about this: if it would take someone a few weeks doing a task to 'master it' instead of being able to read training materials and get the picture...
That *may* be a use case for fine tuning
But proceed with caution
To truly teach an LLM a new behavior or task, you'll need to treat this like a machine learning project, not just throwing examples in and getting magic in return (which it still blows my mind that ChatGPT does this so well for us).
Things like:
- Dataset design
- Training and test data
- Overfitting
+ more as the tooling around fine tuning gets more sophisticated
The other obvious use case is cost.
If you can get a super small language model to do a task instead of GPT-4, there's meaningful cost savings there.
And if you're using a language model to do large scale tasks like triaging your customer support inbox, or analyzing public data for insights
The costs can add up.
But if you're wondering where the heck to invest in fine tuning...
My answer at the moment for most businesses is still:
Make sure you can't do it with prompts. -
Peter Lee reposted
NEJM
@NEJM
This week’s most cited articles:
1. Benefits, Limits, and Risks of GPT-4 as an AI Chatbot for Medicine www.nejm.org/doi/full/10.1056/NEJMsr2214184 #artificialintelligence -
Peter Lee reposted
NEJM AI
@NEJM_AI
Explore the intersection of artificial intelligence and health care in a new article collection curated by NEJM AI journal editors. From cutting-edge medical image interpretation to ID surveillance, these articles offer a comprehensive view of AI's impact. Download now:
-
Peter Lee
@peteratmsr
According to my colleague, @hoifungpoon, AI can “empower people to become superhuman data curators” which, among other things, can revolutionize clinical trial matching. www.clinicalleader.com/doc/ai-s-potential-to-improve-clin… @ClinicalLeader1
-
Peter Lee reposted
Satya Nadella
@satyanadella
Today we're expanding our collaboration with Epic as we deliver a highly scalable public cloud for Epic customers, starting with Mount Sinai.
Microsoft News and Stories
@MSFTnews
.@Microsoft @Azure Large Instances now allows healthcare providers to manage large Epic EHR database loads – up to 50 million database accesses per second. #azure #mscloudhealth news.microsoft.com/source/2023/08/10/microsoft-azure-meet…
-
Peter Lee
@peteratmsr
ND H100 v5 Virtual Machines are now generally available on @Azure! The pace and scale of Microsoft's AI infrastructure buildout is truly beyond anything I could have ever imagined, even just a few short years ago.
Microsoft Azure
@Azure
Learn what we're doing to improve our AI infrastructure and help you scale generative AI with increased service availability: azure.microsoft.com/en-us/blog/scale-generative-ai-with-n…
-
Peter Lee
@peteratmsr
People are impressed by GPT-4's ability to answer medical questions, but bigger (and the technically more challenging) early wins of generative AI are in things like writing notes for the electronic health record and drafting empathetic notes to patients. cnb.cx/45cPACN
-
Peter Lee reposted
Ethan Mollick
@emollick
We made a free five-part YouTube series on Practical AI for Teachers & Students (it might be useful for others, too). All 5 videos👉 youtube.com/playlist?list=PLwRdpYzPkkn302_rL5RrXvQE8j0jLP02j
Here is our paper on assigning AI to students: papers.ssrn.com/sol3/papers.cfm?abstract_id=4475995
And for helping teachers: papers.ssrn.com/sol3/papers.cfm?abstract_id=4391243 -
Peter Lee reposted
Microsoft Research
@MSFTResearch
Researcher Jina Suh and manager Shamsi Iqbal are long-time collaborators. Learn how their history of working together and their unique perspectives are informing their development of tools to support decision-making for organizational leaders: www.microsoft.com/en-us/research/podcast/collaborators-da… #msrpodcast
-
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
Unfortunately this is a correct take. I expect all my works on LLMs to remain unpublished because of this situation. Maybe that's the price to pay when the community gets too big. For me personally it's a non-issue, but what about young students entering the field?
Peter Richtarik
@peter_richtarik
#neurips2023 reviewing if science was sport: any athelete can evaluate any other athlete, irrespective of their specialization, experience, or level. Result? An amateur 100m dash guy criticizes a high jumper for lack of speed during her world-record-breaking jump. Reject.
-
Peter Lee reposted
Artur Olesch
@ArturOlesch
An interview with Peter Lee, co-author of the book “The AI Revolution in Medicine. GPT-4 and Beyond” 👇
Peter Lee
@peteratmsr
“Medicine traditionally refers to a sacred relationship between a doctor and a patient—a twosome, a dyad. Now we move to a triad with AI as a new partner” -- a quote used in this @aDigitalHealth review of our book.
Isaac Kohane
@zakkohane
Medical co-copilot? @peteratmsr aboutdigitalhealth.com/2023/08/01/gen-ai-a-medical-co-pilot/ "to understand is that this isn’t just a single-point disruption... in the near future...emergence of much more capable models...greater diversity of models...what GPT-4 really represents is a phase change."
-
Peter Lee
@peteratmsr
“Medicine traditionally refers to a sacred relationship between a doctor and a patient—a twosome, a dyad. Now we move to a triad with AI as a new partner” -- a quote used in this @aDigitalHealth review of our book.
Isaac Kohane
@zakkohane
Medical co-copilot? @peteratmsr aboutdigitalhealth.com/2023/08/01/gen-ai-a-medical-co-pilot/ "to understand is that this isn’t just a single-point disruption... in the near future...emergence of much more capable models...greater diversity of models...what GPT-4 really represents is a phase change."
-
Peter Lee reposted
Arjun (Raj) Manrai
@arjunmanrai
What a pleasure to chat with the inimitable @atulbutte today for @NEJM_AI Grand Rounds. Sneak preview below with Atul holding a quote with his attitude towards credit assignment; episode coming soon!
-
Peter Lee reposted
Susan Wolf
@swolf10
Great beach reading: The AI Revolution in Medicine by @zakkohane @goldbergcarey @peteratmsr Bonus: glossy pages repel water and other analog threats #ai #ethics #elsi
-
Peter Lee
@peteratmsr
I'm a believer in the idea that the most beautiful mathematical ideas tend to lead to the most elegant and practical technologies in computing. Yael Kalai @MSFTResearch exemplifies this superbly in her work spanning interactive proofs, cloud security, and quantum computing.
-
Peter Lee reposted
Isaac Kohane
@zakkohane
Particularly interesting take on a new data economy with immediate—not future—support for the most in need.
Peter Lee
@peteratmsr
How to make AI data work more fair to people and a more effective path out of poverty? The nonprofit, Karya (a spinout from @MSFTResearch) in India, is gaining early but real traction with its creative, socially-aware, approach. Great coverage by @TIME time.com/6297403/india-ai-karya-startup/
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
At ICML 2023? Attend the Interactive Learning with Implicit Human Feedback workshop on July 29, organized by Akanksha Saran, John Langford, and colleagues from industry and academia. Learn more here: interactive-learning-implicit-feedback.github.io #msftaticml2023
-
Peter Lee reposted
Sriram Rajamani
@SriramRajamani
Very cool work from our lab @MSFTResearch in India, on dignified digital labor for ethical data collection, now spun off into a startup, Karya Inc, covered by Time Magazine!
Peter Lee
@peteratmsr
How to make AI data work more fair to people and a more effective path out of poverty? The nonprofit, Karya (a spinout from @MSFTResearch) in India, is gaining early but real traction with its creative, socially-aware, approach. Great coverage by @TIME time.com/6297403/india-ai-karya-startup/
-
Peter Lee
@peteratmsr
How to make AI data work more fair to people and a more effective path out of poverty? The nonprofit, Karya (a spinout from @MSFTResearch) in India, is gaining early but real traction with its creative, socially-aware, approach. Great coverage by @TIME time.com/6297403/india-ai-karya-startup/
-
Peter Lee reposted
The Boston Globe
@BostonGlobe
The use of AI in medicine is expanding rapidly, bringing the potential for incredible innovations, but also big risks.
🎙️Listen to the latest episode of @GlobeOpinion podcast “Say More with Shirley @Leung,” featuring @goldbergcarey and @zakkohane. www.bostonglobe.com/opinion/podcasts/say-more/ -
Peter Lee reposted
Norman Quist
@NormanQuist
www.pearson.com/en-au/media/2yfj24is/9780138200046.pdf
“The AI Revolution in Medicine” is a fascinating and essential book. With compelling examples, it illuminates how GPT-4 will transform the practice of medicine and patient care—and the attendant ethical challenges.
@zakkohane @peteratmsr @goldbergcarey -
Peter Lee reposted
Oriol Servia
@OriolServia
Used to love that track and the fans! 💪❤️🏁
-
Peter Lee reposted
Microsoft 365 Insider
@Msft365Insider
Intriguing conversation between @DrLaPook and Microsoft's VP of Research & Innovation, Peter Lee (@peteratmsr). They're discussing #ai's growing role in medicine.
A must-watch for all #msft365insiders who are curious about the future of tech & health.
www.cbsnews.com/video/microsoft-vice-president-on-the-fut… -
Peter Lee reposted
Stanford Medicine
@StanfordMed
Dean Lloyd Minor shares his summer reading list, featuring a diverse set of books that dive into the mind, the arts, AI, and the incredible power of the human spirit. What books made your list?
-
Peter Lee reposted
Isaac Kohane
@zakkohane
Is @nejm_AI accepting article submissions? A close friend just asked me so perhaps I was not sufficiently clear: YES: mc05.manuscriptcentral.com/nejm-ai with the following article types: ai.nejm.org/general-information/article-types
-
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
It's somewhat surprising how much misunderstanding/misconceptions there are on LLM-training w. LLM-generated data so it's nice to see an article with sensible quotes by @aidangomezzz and pointers to TinyStories by @EldanRonen + our follow-up phi-1 work.
www.ft.com/content/053ee253-820e-453a-a1d5-0f24985258de -
Peter Lee
@peteratmsr
The research community has long understood that quality of the training data is important in machine learning. But it seems that more people take this more seriously than ever nowadays, to the point that it is even the subject of a Financial Times article, citing work by @MSFTResearch www.ft.com/content/053ee253-820e-453a-a1d5-0f24985258de
-
Peter Lee reposted
DeAnna Sposito
@deannacmj
.@CBSMornings @DrLaPook sat down w/ @peteratmsr CVP @MSFTResearch & Andrea Barrett. Tune into this conversation & hear how our technology offers critical support to clinicians to assist with documentation burden. #healthcareai #nuancedaxexpress
@NuanceInc sprou.tt/1XQ0eOg3nUj -
Peter Lee reposted
Andrej Karpathy
@karpathy
Huge day indeed for AI and LLMs, congrats to Meta 👏
This is now the most capable LLM available directly as weights to anyone from researchers to companies.
The models look quite strong, e.g. Table 4 in the paper: MMLU is good to look at, the 70B model is just below GPT-3.5. But HumanEval (bad misnomer) shows coding capability is quite a bit lower (48.1 vs 29.9).
Yann LeCun
@ylecun
This is huge: Llama-v2 is open source, with a license that authorizes commercial use!
This is going to change the landscape of the LLM market.
Llama-v2 is available on Microsoft Azure and will be available on AWS, Hugging Face and other providers
Pretrained and fine-tuned models are available with 7B, 13B and 70B parameters.
Llama-2 website: ai.meta.com/llama/
Llama-2 paper: ai.meta.com/research/publications/llama-2-open-foundation…
A number of personalities from industry and academia have endorsed our open source approach: about.fb.com/news/2023/07/llama-2-statement-of-support/ -
Peter Lee
@peteratmsr
Over and over again we learn that often what doctors and nurses hope for most from AI is the ability to be with their patients, holding eye contact and undistracted by technology. I enjoyed chatting with @DrLaPook about this and related for his @CBSMornings report about AI in medicine.
CBS Mornings
@CBSMornings
Doctors hope AI can save lives, and companies say it can save time, but can it be relied upon?
“The hope is that it will allow doctors and nurses to…be present with their patients,” @peteratmsr tells @DrLaPook. www.cbsnews.com/news/chatgpt-artificial-intelligence-ai-h…This media couldn't be displayed. See it on the original post. -
Peter Lee reposted
Ethan Mollick
@emollick
New in JAMA: GPT-4, without special prompting, outscored 1st and 2nd year Stanford School of Medicine students in clinical case exams
“Our findings underscore challenges and opportunities for medical training and practice and suggest likelihood of more future dramatic advances.” -
Peter Lee reposted
Satya Nadella
@satyanadella
Please join us at #msinspire tomorrow, as we announce groundbreaking products and partnerships to create new opportunity across our ecosystem in this new era of AI. inspire.microsoft.com/sessions/1825ce5e-db73-4a96-9f87-f1…
-
Peter Lee reposted
Eric Topol
@EricTopol
In one of the 1st randomized trials of a large language model (for writing tasks), professionals with #chatgpt vs controls
--Productivity substantially increased
--Quality increased
--Job satisfaction increased
--Inequalities between workers decreased
www.science.org/doi/10.1126/science.adh2586 -
Peter Lee reposted
Gabriele Fariello
@g_fariello
Nice summary of an interesting article showcasing DiG #ai for molecular equilibrium from @MSFTResearch
www.microsoft.com/en-us/research/blog/distributional-grap… -
Peter Lee reposted
Microsoft Research
@MSFTResearch
ICYMI: A new study looks at engineers’ views on hybrid work; prompt engineering adds context and specificity to generative AI models; Overwatch learns code edit patterns to aid developers; a Qlib update addresses financial market challenges. www.microsoft.com/en-us/research/blog/research-focus-week…
-
Peter Lee reposted
Eric Topol
@EricTopol
Not everyday you get a $282 Million NIH grant!
@julialmv @ScrippsRTI
Scripps Research
@scrippsresearch
Scripps Research receives a major award from @NIH to lead key programs in the @AllofUsResearch Program. The Participant Center, directed by @EricTopol & @julialmv, makes the program more accessible to participants & adds new data collection capabilities www.scripps.edu/news-and-events/press-room/2023/20230713-…
-
Peter Lee
@peteratmsr
Congratulation, Jiaqi!
Microsoft Research
@MSFTResearch
Microsoft Research congratulates Jiaqi Chu as she receives the Royal Academy of Engineering Young Engineer of the Year award. raeng.org.uk/news/royal-academy-of-engineering-names-2023…
-
Peter Lee reposted
Kevin Scott
@kevin_scott
I’ve had the pleasure of working with @OpenAI CTO @miramurati for a few years now, but we’ve rarely had time to catch up about our backgrounds. Take a listen to our conversation on #behindthetech where we discuss her innate sense of curiosity, deploying AI tools, and much more!
-
Peter Lee reposted
Gabe Charbonneau, MD
@gabrieldane
Two of my favorite leaders discussing medicine & AI. @peteratmsr @EricTopol
Eric Topol
@EricTopol
Are large language models a stochastic parrot or do they represent a higher level of machine intelligence not previously seen? My conversation with @peteratmsr
erictopol.substack.com/p/peter-lee-and-the-impact-of-gpt-4This media couldn't be displayed. See it on the original post. -
Peter Lee
@peteratmsr
I coauthored a book recently about AI in medicine, with the goal of sharing and informing. But last night a reader of the book taught me that the right book at the right time can have the power to pull a person out of a dark place. I can't express how humbled and grateful I am that this person's experience with our book was shared with me.
-
Peter Lee reposted
Condensed Matter Theory Center
@condensed_the
The top 2 papers trending in @PhysRevB now are on nanowire Majorana, they cite CMTC >50 times!
InAs-Al hybrid devices passing the topological gap protocol PRB 107, 245423 (2023)
Signatures of a topological phase transition in a planar Josephson junction PRB 107, 245304 (2023) -
Peter Lee reposted
Eliza Strickland
@newsbeagle
When doctor @zakkohane first asked #gpt4 for its opinion on a tough diagnosis, it argued with him and ultimately convinced him. "This was full-strength science fiction, a doctor-in-the-box," he told me. Our Q&A here: spectrum.ieee.org/chatgpt-medical-exam
-
Peter Lee
@peteratmsr
I've been surprised at the capability of distributional graphormers in molecular structure prediction problems. Worth reading this if you are working in the area.
Microsoft Research
@MSFTResearch
Distributional Graphormer, Microsoft’s new deep learning framework for predicting the equilibrium distribution of molecular structures, can generate realistic and diverse molecular structures with high efficiency and low cost. Find out how. www.microsoft.com/en-us/research/blog/distributional-grap…
-
Peter Lee reposted
Eliza Strickland
@newsbeagle
Your doctor might already be using #chatgpt. A recent book by @peteratmsr and @SebastienBubeck of @MSFTResearch, @zakkohane, & @goldbergcarey makes the case that #gpt4 will change medicine, and soon. My Q&A with @zakkohane here: spectrum.ieee.org/chatgpt-medical-exam
-
Peter Lee
@peteratmsr
Really interesting interview with @zakkohane about the use of ChatGPT (and GPT-4 specifically) in medicine. Important perspectives, grounded in reality but at the same time not denying that there is tremendous potential to do a lot of good here.
Eliza Strickland
@newsbeagle
Your doctor might already be using #chatgpt. A recent book by @peteratmsr and @SebastienBubeck of @MSFTResearch, @zakkohane, & @goldbergcarey makes the case that #gpt4 will change medicine, and soon. My Q&A with @zakkohane here: spectrum.ieee.org/chatgpt-medical-exam
-
Peter Lee reposted
Robert Scoble
@Scobleizer
Great book looking at what is going on with AI in medicine.
Peter Lee
@peteratmsr
I'm just tickled to see the release of the audiobook version of "The AI Revolution in Medicine: GPT-4 and Beyond", with @goldbergcarey @zakkohane @SebastienBubeck. Wondering how the voice(s) of GPT-4 came out. www.audible.com/pd/The-AI-Revolution-in-Medicine-Audioboo…
-
Peter Lee
@peteratmsr
I'm just tickled to see the release of the audiobook version of "The AI Revolution in Medicine: GPT-4 and Beyond", with @goldbergcarey @zakkohane @SebastienBubeck. Wondering how the voice(s) of GPT-4 came out. www.audible.com/pd/The-AI-Revolution-in-Medicine-Audioboo…
-
Peter Lee
@peteratmsr
Optimization problems have long been the bread-and-butter of computer science, but weirdly the right computing architecture has never been completely settled. This exploration into an analog-optical approach -- the Analog Iterative Machine @MSFTResearch -- is showing a lot of promise. And it's cool to see it working for real. www.microsoft.com/en-us/research/blog/unlocking-the-futur…
-
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
New video by @EldanRonen on his TinyStories work w. Yuanzhi Li!
Even Deep Learning experts such as @karpathy referred to the work as "inspiring", and I couldn't agree more. Anyone interested in understanding what's happening with LLMs should take a look.
www.youtube.com/watch?v=iNhrW0Nt7zs -
Peter Lee reposted
Matthew Lungren MD MPH
@mattlungrenMD
Agree! Esp as Kosmos-2 builds on methods in the OG Kosmos paper arxiv.org/pdf/2302.14045.pdf
The directionality/future for multimodal foundation models is very promising 😎
Woojin Kim
@woojinrad
Interesting work from @MSFTResearch: "In this work, we unlock the grounding capability for multimodal large language models."
arXiv 👉 arxiv.org/abs/2306.14824
Online Demo 👉 huggingface.co/spaces/ydshieh/Kosmos-2
#foundationmodels #genai #generativeai #lmms #mllms -
Peter Lee reposted
Modern Healthcare
@modrnhealthcr
Nuance said it plans to add the capability for all of Epic's EHR users sometime in the first quarter of 2024. www.modernhealthcare.com/digital-health/epic-gpt-4-ehr-nu…
-
Peter Lee reposted
This American Life
@ThisAmerLife
"Maybe a big chunk of intelligence is a lot simpler than we think." —@peteratmsr on how GPT-4 made him question so many things about how he thought intelligence works www.thisamericanlife.org/803/greetings-people-of-earth
-
Peter Lee
@peteratmsr
The NHS is making meaningful progress on the use of AI to improve radiotherapy for cancer patients, and so it's great to see this top-level recognition. This motivates all of us @MSFTResearch to do what we can to help, via things like integrating our open-source AI systems for radiology. www.linkedin.com/posts/rishi-sunak_ai-that-reduces-waitin…
-
Peter Lee reposted
Christopher A. Longhurst, MD
@CALonghurst
Thank you Dr. Peter Lee (@peteratmsr) for visiting @UCSDHealth to share an inspiring symposium during this pivotal moment of generative AI in medicine! Did you know that #chatgpt has a prime directive?
-
Peter Lee
@peteratmsr
It was a super evening being on stage with you and sharing thoughts and personal stories about AI impact with everyone.
Atul Butte (In Memoriam)
@atulbutte
So privileged to be at @UCSD tonight to chat about #ai with @CALonghurst @pattymaysent @peteratmsr @marlene_mmm and @UCSDHealth!
-
Peter Lee reposted
Ece Kamar
@ecekamar
I'd never think the work we do in MSR would become a story in This American Life. Listen in to hear MSR researchers' reactions to interacting with GPT-4 first time. @peteratmsr @SebastienBubeck @EldanRonen
www.thisamericanlife.org/803/greetings-people-of-earth -
Peter Lee reposted
Atul Butte (In Memoriam)
@atulbutte
So privileged to be at @UCSD tonight to chat about #ai with @CALonghurst @pattymaysent @peteratmsr @marlene_mmm and @UCSDHealth!
-
Peter Lee
@peteratmsr
When @OpenAI first told me and others @MSFTResearch about their #gpt4, I was impressed but also skeptical. @dkestenbaum tells the story of what it was like for @EldanRonen @SebastienBubeck @ecekamar and me -- and for David himself -- in those first days in this episode of @ThisAmerLife. www.thisamericanlife.org/803/greetings-people-of-earth
Peter Lee
@peteratmsr
And not to give a spoiler, but: I'm not at all skeptical any more.
-
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
10 years ago I listened to @ThisAmerLife for 2 weeks while crossing the US from Princeton to Berkeley, where I was going to spend a momentous semester @SimonsInstitute (my first encounter with TCS crowd). Thus this new episode feels really special to me.
www.thisamericanlife.org/803/greetings-people-of-earth -
Peter Lee
@peteratmsr
This paper shows multiple working devices that achieve the hosting of Majorana zero modes, an important advance in solid-state physics with BIG implications for scalable quantum computing. While the road ahead is difficult, we know now it can lead to practical quantum computing.
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
Learn how Microsoft is accelerating scientific discovery with Azure Quantum. Read the blog: blogs.microsoft.com/blog/2023/06/21/accelerating-scientif…
-
Peter Lee reposted
Ethan Mollick
@emollick
Big new AI thing: Microsoft Bing (which uses GPT-4 in Creative Mode), accepts images as input.
The results are impressive. I fed it a meme, it could understand context & read text! A new dimension of AI use just opened up. So expect a flood of AI Twitter influencer threads... -
Peter Lee reposted
Satya Nadella
@satyanadella
Today, we are bringing together AI and quantum with Azure Quantum Elements, ushering in a new era of scientific discovery. Our goal is to compress the next 250 years of chemistry and materials science progress into the next 25. aka.ms/AAks6iy
-
Peter Lee reposted
Bob Wachter
@Bob_Wachter
GPT-4 now matches expert MDs in creating a differential dx @NEJM CPCs. jamanetwork.com/journals/jama/fullarticle/2806457.
Fun to re-read my wife @katiehafner's 2012 @nytimes piece on whether computers will ever equal the best human diagnosticians. Amazingly, answer may now be yes.www.nytimes.com/2012/12/04/health/quest-to-eliminate-diag… -
Peter Lee reposted
National Academy of Medicine
@theNAMedicine
The #namleadershipconsortium’s new initiative aims to align & facilitate adoption of an AI Code of Conduct for health, medical care, and health research. Learn more about this multi-phased project to ensure responsible & equitable use of AI in health care: nam.edu/our-work/programs/leadership-consortium/health-ca…
-
Peter Lee
@peteratmsr
Textbooks Are All You Need. An intriguing and potentially important paper on the unreasonable effectiveness of high-quality training data for AI code generation.
Sebastien Bubeck
@SebastienBubeck
New LLM in town:
***phi-1 achieves 51% on HumanEval w. only 1.3B parameters & 7B tokens training dataset***
Any other >50% HumanEval model is >1000x bigger (e.g., WizardCoder from last week is 10x in model size and 100x in dataset size).
How?
***Textbooks Are All You Need*** -
Peter Lee
@peteratmsr
It's great to see that @theNAMedicine and @NASEM_Health are hosting this hackathon on tech and media solutions to reduce sexually transmitted infections. June 22-23 in Irvine, CA. Info and registration at nam.edu/STIHackathon #preventstis
-
Peter Lee reposted
Frank X. Shaw
@fxshaw
Grab a cup of coffee and tuck into this q/a between @StevenLevy and Satya.
www.wired.com/story/microsofts-satya-nadella-is-betting-e… via @wired -
Peter Lee reposted
Eric Topol
@EricTopol
The unexpected feature of generative #ai for healthcare was its ability to promote empathy.
Here's @peteratmsr relating the experience of his physician colleague, @GJMooreMDPhD w/ #gpt4
From erictopol.substack.com/p/peter-lee-and-the-impact-of-gpt-4This media couldn't be displayed. See it on the original post. -
Peter Lee
@peteratmsr
Something I didn't expect to find in our early investigations: Doctors using ChatGPT to reflect a bit on what to say to their patients to best support them. It seemed strange at the time, but now the idea of facilitating collaboration between doctor and patient seems like an important dimension of AI. www.nytimes.com/2023/06/12/health/doctors-chatgpt-artific…
-
Peter Lee reposted
Scripps Research
@scrippsresearch
In the recent episode of @EricTopol’s Ground Truths Podcast, he spoke with @peteratmsr from @Microsoft on the advances of GPT-4 and the overall impact of #ai in medicine and healthcare www.youtube.com/watch?v=_CeQAyQ-0oU #chatgpt
-
Peter Lee
@peteratmsr
Lots of masks in NYC today. I almost felt like I also needed goggles...
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
In this issue: Peter Lee discusses AI in medicine. Plus, new research on data inference privacy in machine learning; PII leakage in language models; and automatic prompt organization with gradient descent and beam search. www.microsoft.com/en-us/research/blog/research-focus-week…
-
Peter Lee
@peteratmsr
100% agree with this.
Sebastien Bubeck
@SebastienBubeck
I cannot recommend this podcast episode strongly enough. It's simply THE MOST INSIGHTFUL 2 hours content that you can find on LLMs. And it's by none others than @EldanRonen and Yuanzhi Li from our team @MSFTResearch. Stay tuned for a LOT MORE from us soon.
www.youtube.com/watch?v=mv3SIgDP_y4 -
Peter Lee
@peteratmsr
The latest installment of Microsoft's "People of AI" features the amazing @ecekamar. I'm so lucky to work with her; she's taught me so much... news.microsoft.com/source/shortform/people-of-ai-feeling-…
-
Peter Lee reposted
Christopher A. Longhurst, MD
@CALonghurst
Thanks for highlighting our recent @JAMAInternalMed article about ChatGPT responses to patient questions - we are excited about the @UCSDHealth partnership with Epic/Microsoft to implement in clinical workflows!
👉jamanetwork.com/journals/jamainternalmedicine/fullarticle…
👉 www.beckershospitalreview.com/ehrs/why-uc-san-diego-healt… -
Peter Lee reposted
Matthew Lungren MD MPH
@mattlungrenMD
Terrific new work on multi-modal medical imaging reasoning with chat interaction!
@MSFTResearch with @hoifungpoon @TristanNaumann
arxiv.org/pdf/2306.00890.pdf -
Peter Lee reposted
Lake Nona Impact Forum
@LNImpactForum
#lnif23 brought together #healthcare visionaries to discuss the role the of #ai and #gpt technology in patient care.
@ZakKohane explained the justification for AI integration as part of the “A New Era for Health and Life Sciences” panel, which was moderated by @goldbergcarey.This media couldn't be displayed. See it on the original post. -
Peter Lee
@peteratmsr
I just *love* this initiative by my colleague @erichorvitz @Microsoft to make an anthology of essays and case studies on the potential human and societal impacts of #gpt4. The first five contributions are already so insightful, by Alec Gallimore, @ghadfield, @Ada_Palmer, @Bob_Wachter, and Eric himself. unlocked.microsoft.com/ai-anthology/
-
Peter Lee reposted
Bob Wachter
@Bob_Wachter
I was honored to be asked by @Microsoft & @erichorvitz to contribute to an anthology on the future of generative AI (GPT & the like) on various fields.
Today's 1st issue unlocked.microsoft.com/ai-anthology/ covers engineering, law, history, & my piece on healthcare.
unlocked.microsoft.com/ai-anthology/robert-wachter/ -
Peter Lee reposted
Frank X. Shaw
@fxshaw
Super interesting start of an anthology seeking shed light on two questions:
1. How might this technology and its successors contribute to human flourishing?
2. How might we as society best guide the technology to achieve maximal benefits for humanity?
www.linkedin.com/feed/update/urn:li:activity:706932934959… -
Peter Lee
@peteratmsr
Clarifying analysis of LLM performance on medical challenge problems by @HarshaNori
Harsha Nori
@HarshaNori
I've been getting many questions about how our work on GPT-4 in medicine (arxiv.org/abs/2303.13375) compares to the recent PaLM 2 and Med-PaLM 2 models. Here is the TL;DR:
-
Peter Lee
@peteratmsr
An illuminating, lengthy, and yet highly entertaining conversation between @kevin_scott and @reckless @Verge about AI at @OpenAI and @Microsoft. Copilots, Sydney, responsible AI, and even uranium centrifuges... www.theverge.com/23733388/microsoft-kevin-scott-open-ai-c…
-
Peter Lee reposted
Jordi Ribas
@JordiRib1
I’m pleased to announce we're bringing Bing to ChatGPT as the default search experience. With our search engine built-in, ChatGPT answers can be grounded by search and web data and include citations so you can learn more—all directly from within chat. blogs.bing.com/search/may_2023/Bing-at-Microsoft-Build-2023
-
Peter Lee reposted
Yusuf Mehdi
@yusuf_i_mehdi
Today, we’re excited to announce new Bing plugin partners and our commitment, in partnership with OpenAI, to a shared plugin platform for interoperability across ChatGPT and Microsoft’s copilot experiences! blogs.bing.com/search/may_2023/Bing-at-Microsoft-Build-2023
-
Peter Lee reposted
Eric Topol
@EricTopol
Can machines communicate empathy?
@peteratmsr talks about a physician turning to #gpt4 for help with a patient and then what happened
Part of our extraordinary LLM conversation
erictopol.substack.com/p/peter-lee-and-the-impact-of-gpt-4This media couldn't be displayed. See it on the original post. -
Peter Lee reposted
Ethan Mollick
@emollick
Most of the tests of LLM math abilities are based on straightforward math. What if we gave them hard word problems where they had to apply advanced scientific & financial theorems?
This paper finds GPT-4 does quite well, beating every other model by a lot arxiv.org/abs/2305.12524 -
Peter Lee
@peteratmsr
I truly enjoyed our conversation, @EricTopol. Thanks for having me on!
Eric Topol
@EricTopol
A deep and enthralling conversation with @peteratmsr on #gpt4, large language models, their potential use in medicine, their future, and what are we really seeing here?
erictopol.substack.com/p/peter-lee-and-the-impact-of-gpt-4 w/ transcript -
Peter Lee reposted
Eric Topol
@EricTopol
A deep and enthralling conversation with @peteratmsr on #gpt4, large language models, their potential use in medicine, their future, and what are we really seeing here?
erictopol.substack.com/p/peter-lee-and-the-impact-of-gpt-4 w/ transcript -
Peter Lee reposted
Isaac Kohane
@zakkohane
Come for the AI chatter but stay for @DShaywitz masterful unpacking of the everlong struggle between data analysts (? parasites) and titular data owners in biomedicine @timmermanreport timmermanreport.com/2023/05/big-if-true-opportunities-and…
-
Peter Lee reposted
PBS News
@NewsHour
AI-powered programs are learning to answer patients’ medical questions and diagnose illnesses, but there are still problems to be worked out.
@zakkohane joins @johnyangtv to discuss. www.pbs.org/newshour/show/the-promises-and-potential-pitf… -
Peter Lee reposted
Dr. Hubertus Becker
@hubertusbecker
Agreed, @peteratmsr. Beyond answering medical queries, AI's real challenge is understanding healthcare's intricate context. Advancements must balance augmenting human efforts, ethical responsibility, and pushing boundaries.
-
Peter Lee
@peteratmsr
The more I explore the potential of GPT-4 in medicine, the more I realize that answering medical knowledge questions is uninteresting and probably not so important. More important, and more challenging, are things like medical notetaking and reasoning about clinical data. /1
Post not available here, view on X
Peter Lee
@peteratmsr
So, back to GPT-4: Yes, GPT-4 has strong performance on medical knowledge tasks, and it's amazing it basically has top results even with no specialized training and first-shot answers. But I think this is NOT what's important for the future of LLMs in medicine. /5
Peter Lee
@peteratmsr
Don't get blinded by such results, as impressive as they are. Focus on the extent to which these systems "understand" (for lack of a better word) the context of healthcare delivery or medical research activity. GPT-4 is stupendously impressive, and very useful, but not 100%.
-
Peter Lee reposted
Oriol Servia
@OriolServia
Fast Friday it is! 💥 @IMS #indianapolis500
Arrow McLaren IndyCar Team
@ArrowMcLaren
A turn three trap speed of 243.142 mph. 🤯
-
Peter Lee reposted
Ansa Biotechnologies
@AnsaBio
We collaborated with @MSFTResearch on a new paper published in ACS Synthetic Biology. This research shows the promise of novel, #enzymaticdnasynthesis on a scale necessary for DNA-intensive applications, such as #dna data storage or gene synthesis.
pubs.acs.org/doi/10.1021/acssynbio.3c00044
#synbio -
Peter Lee reposted
Nathan Labenz
@labenz
The book is "The AI Revolution in Medicine: GPT-4 and Beyond"
I find myself recommending the book to all sorts of people right now, but as @tylercowen put it, "especially for the skeptics"
www.amazon.com/AI-Revolution-Medicine-GPT-4-Beyond/dp/013… -
Peter Lee reposted
Steven Wardell
@stevenwardell
@ZakKohane says in order to avoid a negative alignment of #medicine and #ai, we’re going to have to “flip the clinic.” #tedxboston #digitalhealth. Zak got early access to @OpenAI ‘s GPT-4 for medicine.
-
Peter Lee reposted
Mikhail Parakhin
@MParakhin
People underestimate the impact on historical discoveries and interpretations: the real digital archaeology is here! I seriously plan to decipher Linear A with the next iteration of Bing Chat model.
Ethan Mollick
@emollick
Digital humanities is about to be wild.
I pasted in the Catalog of Ships from the Iliad, and asked GPT with Code Interpreter to map the origins of the Greeks. With hiccups, it extracted solid information from the text, found modern locations & started building an interactive map -
Peter Lee reposted
Oriol Servia
@OriolServia
Is this tight or is this tight? Wow 22 guys in one second and the top 4 within a tenth, in a Road-coarse! To be at the very top in Indycar even the meal you had last Monday counts 😅🏁💥 Should be an exciting race! @nbc @IndyCaronNBC
-
Peter Lee reposted
Mother_bumper
@mother_bumper
@peteratmsr Reading AI REVOLUTION. Interested observer of GPT-4 and parent of a daughter at Dartmouth Med. I laughed so hard where you write about saying "Thx" to the AI. Same! Loving the book. Sent it to daughter too.
-
Peter Lee reposted
Alexandre DaSilva
@DrAlexDaSilva
This is a required book to start to grasp the future of #ai in healthcare. I can see the revolution that is coming in #chronicpain and #migraine care, research, and education. Of course, I had to have my book signed in person by Drs. @peteratmsr & @zakkohane at the #sail23.@UMich
-
Peter Lee reposted
Nathan Labenz
@labenz
Professor @zakkohane got early research access to GPT-4 in the fall of 2022
His approach to exploring, characterizing, and integrating GPT-4 into clinical practice is the best study of AI application that I've seen from academia to date.
Bravo!
The Cognitive Revolution Podcast
@CogRev_Podcast
[new episode] Dr. @zakkohane on GPT-4:
"No doctor should be without it. They should have this sidekick that is meticulous, completely up to date, ever vigilant, and sometimes wrong.."
@zakkohane 's combination of risk awareness + optimism sets a worthy example. -
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
Their conclusion?
"For **every** task we tried, ChatGPT (3.5) is still stronger than Vicuna, [which is] sometimes very close, sometimes much worse. [...] We believe GPT-4 is a **much** stronger model, and are excited to see if open source can approach that."
cc @MosaicML too 2/2 -
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
Lots of discussion abt open source LLMs catching up The Big Ones, including eye-catching claims s.a. 90% of ChatGPT's quality (by the really cool work of @lmsysorg).
Two Sparks authors @marcotcr & @scottlundberg explore this further in a new blog post 1/2
medium.com/@marcotcr/exploring-chatgpt-vs-open-source-mod… -
Peter Lee
@peteratmsr
I spent quite a bit of time studying this paper. It really made me think hard at times. Early understanding of causal inference in GPT4.
Microsoft Research
@MSFTResearch
Through improved understanding of causal mechanisms and supporting translation between natural language and formal methods, LLMs open new frontiers for advancing the research, practice, & adoption of causality in societally impactful fields like medicine, science, law, & policy.
-
Peter Lee reposted
Jon Kragh
@jonkragh
A must have for today's pioneers.
Even if medicine is not your focus,
this is a wonderful book on how to approach high risk use cases with LLMs.
Stoked to see @labenz has an interview already.
@peteratmsr @zakkohane -
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
I just tried the new Bard powered by Palm 2 and asked it to draw a unicorn in TikZ. It's not quite there yet :-).
-
Peter Lee reposted
Rohaid Ali, MD
@RohaidAliMD
Agree with your sentiment, Dr Lee. Our group has a paper that we are finalizing that shows that optimizing complex reasoning and performance across linguistic contexts is the way to go over fine tuning to content areas (i.e., medical knowledge areas), at least in the near term
-
Peter Lee
@peteratmsr
Enjoying listening to @SebastienBubeck @srikanthbelwadi @vivnat on a panel about LLMs in healthcare #sail23. Sébastien shows how drawing pictures of unicorns is related to the future of medicine.
Post not available here, view on X
Peter Lee
@peteratmsr
. @SebastienBubeck takes a little dig at the new Bard, pointing to it's poor understanding of, and ability to draw, a unicorn. But I would expect Bard to improve eventually.
Peter Lee
@peteratmsr
Great panel discussion. I think people still think of #gpt4 as an ordinary ML system, and not a general intelligence.
-
Peter Lee reposted
Rohaid Ali, MD
@RohaidAliMD
This book is amazing, Dr Lee! Please keep your eyes out for a meaningful paper our group is publishing in preprint this week, where we’ve taken the lessons from your book (particularly Ch. 5) and applied them to a surgical context in a very practical way
-
Peter Lee reposted
Isaac Kohane
@zakkohane
Also policy makers who want to understand what #gpt4 and other generative models means for clinical practice and biomedical research today
Peter Lee
@peteratmsr
If you're curious about AI in healthcare, check out our book on #gpt4 in medicine, authored by me with @goldbergcarey and @zakkohane. It's on its 3rd print run! Written for doctors, nurses, patients, and anyone interested in the future of healthcare. www.amazon.com/AI-Revolution-Medicine-GPT-4-Beyond/dp/013…
-
Peter Lee
@peteratmsr
If you're curious about AI in healthcare, check out our book on #gpt4 in medicine, authored by me with @goldbergcarey and @zakkohane. It's on its 3rd print run! Written for doctors, nurses, patients, and anyone interested in the future of healthcare. www.amazon.com/AI-Revolution-Medicine-GPT-4-Beyond/dp/013…
-
Peter Lee
@peteratmsr
My father passed away two weeks ago, peacefully, at age 90. He was a great father, husband, and man. He and my mom raised three kids to be wildly successful in all aspects of their lives. A physicist and lover of adventure. He will be missed by many. www.mininggazette.com/obituaries/2023/05/dr-sung-mook-lee/
-
Peter Lee reposted
Eric Topol
@EricTopol
With a 1:1 tutor for students and AI assistant for teachers, @salkhanacademy makes a compelling case for generative #ai’s unprecedented transformative impact. There are parallels in healthcare for patients and clinicians www.ted.com/talks/sal_khan_the_amazing_ai_super_tutor_for… @TEDTalks
-
Peter Lee reposted
Andrej Karpathy
@karpathy
Oops haven't tweeted too much recently; I'm mostly watching with interest the open source LLM ecosystem experiencing early signs of a cambrian explosion. Roughly speaking the story as of now:
1. Pretraining LLM base models remains very expensive. Think: supercomputer + months.
2. But finetuning LLMs is turning out to be very cheap and effective due to recent PEFT (parameter efficient training) techniques that work surprisingly well, e.g. LoRA / LLaMA-Adapter, and other awesome work, e.g. low precision as in bitsandbytes library. Think: few GPUs + day, even for very large models.
3. Therefore, the cambrian explosion, which requires wide reach and a lot of experimentation, is quite tractable due to (2), but only conditioned on (1).
4. The de facto OG release of (1) was Facebook's sorry Meta's LLaMA release - a very well executed high quality series of models from 7B all the way to 65B, trained nice and long, correctly ignoring the "Chinchilla trap". But LLaMA weights are research-only, been locked down behind forms, but have also awkwardly leaked all over the place... it's a bit messy.
5. In absence of an available and permissive (1), (2) cannot fully proceed. So there are a number of efforts on (1), under the banner "LLaMA but actually open", with e.g. current models from @togethercompute, @MosaicML ~matching the performance of the smallest (7B) LLaMA model, and @AiEleuther , @StabilityAI nearby.
For now, things are moving along (e.g. see the 10 chat finetuned models released last ~week, and projects like llama.cpp and friends) but a bit awkwardly due to LLaMA weights being open but not really but still. And most interestingly, a lot of questions of intuition remain to be resolved, e.g. especially around how well finetuned model work in practice, even at smaller scales. -
Peter Lee reposted
waveaura
Longwood Author Series: The AI Revolution in Medicine GPT-4 and Beyond www.rockymounttelegram.com/news/national/longwood-author-… via @telegramreports
-
Peter Lee reposted
Yusuf Mehdi
@yusuf_i_mehdi
Announcing the next generation of AI-powered Bing & Edge! New image & video answers, chat history, share & export, and coming soon multimodal and plugins support. blogs.microsoft.com/blog/2023/05/04/announcing-the-next-w…
-
Peter Lee reposted
Judea Pearl
@yudapearl
Very cool and thoughtful. There is clearly great potential to using LLM in CI tasks, possibly leading to a new hierarchy and a new ladder, where the input is not "data", nor "experiments", but "texts about data and experiments".
Amit Sharma
@amt_shrma
New paper: On the unreasonable effectiveness of LLMs for causal inference.
GPT4 achieves new SoTA on a wide range of causal tasks: graph discovery (97%, 13 pts gain), counterfactual reasoning (92%, 20 pts gain) & actual causality.
How is this possible?🧵
arxiv.org/abs/2305.00050 -
Peter Lee reposted
Isaac Kohane
@zakkohane
Best reason for optimism: So many enthusiastic students and trainees and young faculty. www.goskagit.com/community/health/longwood-author-series-…
-
Peter Lee
@peteratmsr
Important insights into GPT-4's abilities in a new form of causal reasoning.
Amit Sharma
@amt_shrma
New paper: On the unreasonable effectiveness of LLMs for causal inference.
GPT4 achieves new SoTA on a wide range of causal tasks: graph discovery (97%, 13 pts gain), counterfactual reasoning (92%, 20 pts gain) & actual causality.
How is this possible?🧵
arxiv.org/abs/2305.00050 -
Peter Lee reposted
International Center for Genetic Disease
@H_iCGD
We are looking forward to @peteratmsr Peter Lee's talk tomorrow, Tuesday, at 10 AM ET.
You can still register for this event: docs.google.com/forms/d/e/1FAIpQLSeqzxV6kF31pJst-LpvxJF-z…
@MSFTResearch @BrighamResearch
@dental_harvard -
Peter Lee
@peteratmsr
I can't wait to meet up with coauthors @goldbergcarey and @zakkohane tomorrow (May 2) to sign copies of our book. I still need to collect Carey's and Zak's autographs! 😁 Only about 200 copies available, so sign up early. @DShaywitz will moderate a Q&A session with us.
Isaac Kohane
@zakkohane
Book signing next week @harvardmed @HMSCountway w @DShaywitz grilling the authors @peteratmsr @goldbergcarey @zakkohane libcal.countway.harvard.edu/event/10658386
-
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
The @TEDTalks by @YejinChoinka is both insightful & beautifully delivered! Totally agree with her that GPT-4 is simultaneously brilliant and incredibly stupid. Yejin gives 3 examples of common sense failing that are worth examining a bit more closely. 1/5 www.youtube.com/watch?v=SvBR0OGT5VI
-
Peter Lee
@peteratmsr
People aren't so surprised that #chatgpt is rated by patients to be more accurate than human doctors, but it's more than that -- ChatGPT is also rated more empathetic. In my experience, GPT-4 very quickly gets into an empathetic "vibe" when discussing personal health issues.
Eric Topol
@EricTopol
Noteworthy: a new study comparing #chatgpt vs doctors for responding to patient queries demonstrated clearcut superiority of #ai for improved quality and empathy of responses @JAMAInternalMed
jamanetwork.com/journals/jamainternalmedicine/fullarticle… -
Peter Lee reposted
Mikhail Parakhin
@MParakhin
Many people don't realize it: you can analyze even your local documents, the ones not available on the web.
-
Peter Lee reposted
Greg Brockman
@gdb
I've been genuinely impressed with sophistication of ChatGPT users, and the depth to which they explore the technology. Easy to underappreciate.
Sam Altman
@sama
common pattern: ChatGPT users demonstrate that they understand the limitations and weaknesses of the product *extremely well*; experts in the field continue to worry it’s just too complicated and the users won’t be able to understand
-
Peter Lee reposted
Isaac Kohane
@zakkohane
Book signing next week @harvardmed @HMSCountway w @DShaywitz grilling the authors @peteratmsr @goldbergcarey @zakkohane libcal.countway.harvard.edu/event/10658386
-
Peter Lee reposted
Isaac Kohane
@zakkohane
Nice note from @tylercowen marginalrevolution.com/marginalrevolution/2023/04/the-ai-… on our book www.amazon.com/AI-Revolution-Medicine-GPT-4-Beyond/dp/013… CC @peteratmsr @goldbergcarey
-
Peter Lee reposted
Modern Healthcare
@modrnhealthcr
The announcement was made in conjunction with the first day of #himss23, which is being held in Chicago this week. www.modernhealthcare.com/digital-health/himss-2023-epic-m…
-
Peter Lee
@peteratmsr
Because intelligence is the primary (maybe only) survival advantage of homo sapiens, we may have evolved hard-wiring to value it over all other capabilities in nature. Perhaps partly why we are fascinated, and discomforted, by AI. And why we think it must be highly complex.
-
Peter Lee reposted
Paola Gori-Giorgi
@paolagorigiorgi
It was really inspiring to watch the webinar of @MatthiasTroyer on how to solve quantum problems now and in the future. Really top science, well-grounded, no bs. msazurequantum.eventbuilder.com/event/68273?source=AzureT
-
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
Okay I think the TiKZ Unicorn hype is reaching its peak, I just discovered this amazing website www.tikzunicorn.com !!!!
If the website's owner is reading this, my wife is asking for kids' size shirts too 😂 -
Peter Lee reposted
NEJM AI
@NEJM_AI
Microsoft's Peter Lee predicts large language models such as ChatGPT will be used every day in clinical settings by patients, physicians, and nurses. He argues that public debate is urgently needed to ensure these tools are used well.
Full episode: nejm.ai/ep4 -
Peter Lee
@peteratmsr
I'm at once amused, amazed, and perplexed by GPT-4's ability (within Bing) to do things like this.
Ethan Mollick
@emollick
Bing, calculate how big a gorilla would need to be to fling a car into space. Make any assumptions you need but show your work.
The answer has likely errors, but look at the way the AI assembles facts from web searches & does step-by-step “reasoning” based on just a short prompt -
Peter Lee
@peteratmsr
Oh I really love this...
Jordi Ribas
@JordiRib1
Excited to share that we've fully shipped support for LaTeX in Bing chat! Now math formulas are beautifully formatted and easy to read so you can learn and study in chat mode. I wrote my PhD dissertation with LaTeX so this brings back good memories.
-
Peter Lee reposted
HIMSS
@HIMSS
#ai talk is buzzing around #himss23, for good reason. Andrew Moore, @KayFButterfield, @peteratmsr, @ReidBlackman and @CrisJRoss took on the controversial subject of
#ai in healthcare and provided valuable insights about the privacy, safety and ethical challenges. 👇
#smarthit -
Peter Lee reposted
Heather Landi
@HeatherLandi
"There are tremendous opportunities and also significant risks, and risks that we probably don't even know about yet. At the end of the day, I believe very strongly that the healthcare community will need to make the decisions about whether, when and how [it's] used," Lee said.
-
Peter Lee reposted
Heather Landi
@HeatherLandi
@Microsoft's @peteratmsr says in addition to legal and ethical issues, there are "deep scientific mysteries that we're just trying to come to grips with" in the development of generative AI
-
Peter Lee reposted
Heather Landi
@HeatherLandi
Packed house at #himss23 opening keynote this morning featuring a discussion on ethical, responsible use of AI
-
Peter Lee
@peteratmsr
You can download a free sample (>100 pages!) of our book on GPT-4 in medicine while #himss23 is going on this week. Coauthored with @goldbergcarey @zakkohane, required reading for clinicians and anyone who is curious about the future of AI in healthcare. www.informit.com/promotions/himss23-free-sample-chapter-f…
-
Peter Lee
@peteratmsr
What do you get when you put a CIO, two computer scientists, a lawyer, and an ethicist in front of 10,000+ people to discuss AI ethics in health? Some sparks but also a lot in common. Fun #himss23 keynote stage with @MayoClinic's Cris Ross, @KayFButterfield @ReidBlackman @awmcmu.
-
Peter Lee reposted
Will Knight
@willknight
GPT-4 often "seems" remarkably clever, and some believe it exhibits features of more general intelligence. Experts say we need new methods for probing the model's intelligence, and more transparency from @OpenAI about how it works to truly understand it. www.wired.com/story/chatgpt-agi-intelligence/
-
Peter Lee reposted
Ethan Mollick
@emollick
Many new AI document tools are powered by GPT-3.5, hiding it in their interface.
But as Bing AI (in creative mode) uses GPT-4 for free & the Bing sidebar can “read” screen content, it can be easier to use Edge & query GPT-4 directly, before turning to other specialized AI tools.
Ethan Mollick
@emollick
This is very cool: it has become a lot easier to read academic research.
If you open a paper in Microsoft Edge (I know, I know), the Bing AI sidebar can read the PDF and you can ask it questions about the paper. (I checked & the results seem high-quality, but be careful with AI) -
Peter Lee reposted
Ethan Mollick
@emollick
Though I don't program, GPT-4 gave me working software out of a tweet.
“Write code (and tell me how to run it) that will take whatever is in my clipboard, send it to GPT, ask it to summarize using text-davinci-003, and display the results on the screen.” www.oneusefulthing.org/p/one-sentence -
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
A very balanced reporting on the Sparks paper, and whether intelligence has been achieved with #gpt4 . The article touches on many of the issues that have been brought up in the last few weeks, great job @willknight !
WIRED
@WIRED
“We’re getting this tremendous amount of raw intelligence without it necessarily coming with an ego-viewpoint, goals, or a sense of coherent self. That, to me, is just fascinating.” wired.trib.al/8P2gOGU
-
Peter Lee
@peteratmsr
Already having so much fun and learning a lot from the other panelists. Looking forward to taking the stage with them!
HIMSS
@HIMSS
☕ Morning, y’all! Today at #himss23, we’re talking tech and how to make it rain. 🤖💸
Kicking off with a fantastic keynote on responsible use of #ai, featuring great panelists like Andrew Moore, @KayFButterfield, @peteratmsr and @ReidBlackman.
🚀 Strap in for a jampacked day. -
Peter Lee reposted
Christopher A. Longhurst, MD
@CALonghurst
Thanks Peter - we will certainly share with others, and have been using your outstanding @NEJM article as a guide! www.nejm.org/doi/10.1056/NEJMsr2214184
-
Peter Lee reposted
Christopher A. Longhurst, MD
@CALonghurst
🆕 GPT in the #ehr! "@UCSDHealth has already rolled out the company’s AI for In-Basket responses to a limited number of users." Thanks @marlene_mmm @mtaiseale and @UCSDH_IS team for leading a thoughtful implementation and outcomes analysis - more soon! 🆕
www.modernhealthcare.com/digital-health/himss-2023-epic-m… -
Peter Lee reposted
Eric Topol
@EricTopol
"You don't have to be a brain surgeon......"
Takes on a new meaning in the #gpt4 era
www.medrxiv.org/content/10.1101/2023.04.06.23288265v1 by @IDavidConnolly @RohaidAliMD @oliverytang @JohnHShinMD @WaelFAsaad @MGHNeurosurg -
Peter Lee reposted
Microsoft Research
@MSFTResearch
We’re proud to have contributed to 30 papers at NSDI this year. Topics range from networking for AI workloads and cloud networking to WANs and wireless networks. Learn how we’re pushing the boundaries and delivering cutting edge solutions: www.microsoft.com/en-us/research/blog/microsoft-at-nsdi-2…
-
Peter Lee
@peteratmsr
If you are at #himss, it is *definitely* worth your while to check out the AI-powered demos (esp. GPT-4 based) in the Nuance and Epic booths.
-
Peter Lee
@peteratmsr
I've been extremely impressed with what Epic has been doing to integrate #gpt4 to improve workflow, reporting, and data value in healthcare. This has really motivated us @Microsoft to deepen our strategic collaboration with them, and vice-versa. www.prnewswire.com/news-releases/microsoft-and-epic-expan…
-
Peter Lee reposted
Bob Wachter
@Bob_Wachter
Getting up to speed w/ GPT-4 & healthcare, good options featuring @microsoft research head @peteratmsr:
1) @NEJM: www.nejm.org/doi/10.1056/NEJMsr2214184
2) @AllenInstitute Talk: www.youtube.com/watch?v=qbIk7-JPB2c
3) Book: www.amazon.com/AI-Revolution-Medicine-GPT-4-Beyond/dp/013…
Non-MSFT: our @UCSF grand rounds: www.youtube.com/watch?v=j-aOCuzfxUI -
Peter Lee reposted
Robert Scoble
@Scobleizer
I got an early copy of Peter’s book. Sent it to a friend who is building an AI medical startup (with permission) and he wrote me back and said the book changed his life.
I am still reading it but learn something in every page about how healthcare is changing due to AI.
Peter Lee
@peteratmsr
"It is at once both smarter and dumber than any person you've ever met." I didn't expect that to be so quotable, but yes: In medicine, #gpt4 offers tremendous benefits but also requires a great deal of care to use. @thisisinsider's take on our new book: www.insider.com/chat-gpt-successor-gpt-4-can-help-doctors…
-
Peter Lee reposted
Ethan Mollick
@emollick
GLT-4 as brain surgeon.
GPT-4 scored 83% on neurosurgery board exams, GPT-3.5 got 62%, Bard, 44%
Even more intriguing, the paper measured hallucinations. Bard had a hallucination rate of 57% while GPT-4 was just 2%
That suggests a potential for real progress on made up answers
Ian D Connolly, MD MS
@IDavidConnolly
1 / In our new paper, we challenged LLMs on an oral boards q-bank with mostly higher order questions! Also for the first time, we quantify “hallucinations” — a puzzling phenomenon where LLMs will confidently incorporate falsehoods into responses.
www.medrxiv.org/content/10.1101/2023.04.06.23288265v1 -
Peter Lee reposted
Isaac Kohane
@zakkohane
More of this needed!
Josh Mandel, MD
@JoshCMandel
Last week, I was scratching my head over cases where GPT-3.5 struggled to judge the best clinical terms from search results. (See x.com/JoshCMandel/status/1644012643902976000) This week, we'll take a reductionist approach w/ a simple experiment to understand how the "chat API" influences judgments.
This post is unavailable. See it on X. -
Peter Lee reposted
Javi Lopez ⛩️
@javilopen
An AI-generated image wins the Sony World Photography Awards.
"Boris Eldagsen himself has come out and stated after the winning result was announced that this isn’t a photograph at all. It’s an image generated by inputs he provided to the Open AI photo tools".
Info+links 🧵👇 -
Peter Lee reposted
Isaac Kohane
@zakkohane
It's a long and nuanced conversation and please do not throw a brick when I say that the conversation was the motivation for our book www.amazon.com/AI-Revolution-Medicine-GPT-4-Beyond/dp/013… Especially Chapter 4 "Trust but Verify"
-
Peter Lee reposted
Mikhail Parakhin
@MParakhin
Fun fact: DeepSpeed is also a part of our team. And if you like training models, you might want to know we just released DeepSpeed-Chat: RLHF support out of the box.
DeepSpeed
@DeepSpeedAI
Want to train 10B+ ChatGPT-style models on a single GPU and 100B+ on multi-GPUs systems? Introducing DeepSpeed-Chat, an easy (single script), fast, and low-cost solution for training high-quality ChatGPT-style models with RLHF, 15x faster than SoTA.
Blog: github.com/microsoft/DeepSpeed/tree/master/blogs/deepspee… -
Peter Lee
@peteratmsr
Love this statement by @kimbbellard: "That’s only one of a million — a billion — blind spots that our health care system has about the people using it." This pretty much captures the aspiration of AI in healthcare.
-
Peter Lee reposted
Josh Mandel, MD
@JoshCMandel
"Help! Someone ran my clinical note through `tr -d '[:space:]'`! Please use your expert knowledge of clinical language to reconstruct the original."
Aaron Neinstein, MD
@AaronNeinstein
I was just asked to do a 2nd opinion medical case review on a chart ThatLiterallyIAmNotKiddingHadNoSpaceBetweenWordsInThePatientRecord and I'm supposed to make sense of it. Need that AI co-pilot, stat.
-
Peter Lee reposted
Curt Langlotz
@curtlanglotz
Anyone hyping the use of #chatgpt and other large language models for medical diagnosis should read this tour de force blog post from AI regulatory expert @DrHughHarvey. LLMs will have many uses in medicine, but clinical decision support is a much longer road.
-
Peter Lee reposted
NEJM AI
@NEJM_AI
GPT-4, without any specialized prompt crafting, outperforms earlier models and exceeds the passing score on the USMLE by 20+ points. The model is also significantly better calibrated and can predict the likelihood that its answers are correct. www.microsoft.com/en-us/research/publication/capabilities…
-
Peter Lee reposted
Eric Topol
@EricTopol
After a patient saw multiple physicians and neurologists over 6 months and was assigned a diagnosis of #longcovid, a relative entered her symptoms into #chatgpt with the correct output. Diagnosis was confirmed by antibody testing and therapy has been initiated.
-
Eric Topol
@EricTopol
After a patient saw multiple physicians and neurologists over 6 months and was assigned a diagnosis of #longcovid, a relative entered her symptoms into #chatgpt with the correct output. Diagnosis was confirmed by antibody testing and therapy has been initiated.
Peter Lee
@peteratmsr
The use of #gpt4 as a second set of eyes -- not a second opinion, really, but a source of additional diagnostic ideas -- is something that we mention in our book as a potentially important use case.
-
Peter Lee
@peteratmsr
"It is at once both smarter and dumber than any person you've ever met." I didn't expect that to be so quotable, but yes: In medicine, #gpt4 offers tremendous benefits but also requires a great deal of care to use. @thisisinsider's take on our new book: www.insider.com/chat-gpt-successor-gpt-4-can-help-doctors…
-
Peter Lee reposted
waveaura
ChatGPT can save lives in the ER, but it needs supervision: 'It is at once both smarter and dumber than any person you've ever met' www.insider.com/chat-gpt-successor-gpt-4-can-help-doctors… via @thisisinsider
-
Peter Lee reposted
Jeremy Howard
@jeremyphoward
An important slide from @SebastienBubeck, one of the researchers that closely studied the behavior of GPT-4 prior to its release.
www.youtube.com/watch?v=qbIk7-JPB2c -
Peter Lee reposted
Erik Brynjolfsson
@erikbryn
This is getting spooky.
-
Peter Lee reposted
NEJM AI
@NEJM_AI
Microsoft's Peter Lee (@peteratmsr) tells us the story behind the relationship between Microsoft and OpenAI, the company that created ChatGPT. Peter reflects on the mutual codependence and mutual independence between the two companies.
Hear more: nejm.ai/ep4 -
Peter Lee reposted
Caroline Schlocker
@drschlocker
An excellent podcast on the medical applications of Chat GPT and other large language models. While there are many topics covered, the later half of the podcast really focuses on the arrival of ChatGPT on the scene. podcasts.apple.com/us/podcast/the-gpt-4-episode-microsoft…
-
Peter Lee reposted
Curt Langlotz
@curtlanglotz
Highly recommend this inspiring talk from @peteratmsr from @MSFTResearch on the implications of large language models for medicine. @stanfordHAI @StanfordAIMI
Allen School Distinguished Lecture: Peter Lee (Microsoft Research & Incu... youtu.be/bEovhfxJsM4 via @YouTube -
Peter Lee reposted
Robert Scoble
@Scobleizer
I had an early version of @peteratmsr's book and in talking with my friend Jeff Stevens, who is running a medical AI startup, we agree it is right on point.
He is also named on the AGI Spark paper that I just retweeted from Microsoft Research.
AI will soon save many lives. He lays out how.
Peter Lee
@peteratmsr
Appreciate this review of our book on #gpt4 in medicine by @DShaywitz @timmermanreport. David nails each author's obsessions pretty much perfectly... "New Book Contemplates How Generative AI Will Serve Man" - timmermanreport.com
-
Peter Lee reposted
Harvard Data Science Initiative
@harvard_data
📻 In this @NEJM episode of “Intention to Treat,” HDSI Steering Committee Member @zakkohane (@HarvardDBMI) + Maia Hightower (@UChicagoMed) explore the promise and hazards of #ai and #ml tools for clinical and administrative uses in #medicine: www.nejm.org/doi/full/10.1056/NEJMp2301939?query=featured…
-
Peter Lee reposted
Ignacio Jusué-Torres MD FAANS FCNS FEBNS
@JusueMD
Must read paper about #gpt4 #ai use in clinical practice by the head of Microsoft research @peteratmsr @NEJM
www.nejm.org/doi/10.1056/NEJMsr2214184 -
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
Last couple of weeks I gave a few talks on the Sparks paper, here is the MIT recording!
The talk doesn't do justice to all the insights we have in the paper itself. Neither talk nor twitter threads are a substitute for actual reading of the 155 pages :-)
www.youtube.com/watch?v=qbIk7-JPB2c -
Peter Lee
@peteratmsr
In a great article by @goldbergcarey, @zakkohane makes this important plea: "To my colleagues in the medical establishment, let’s not leave it to others to guide how this world-changing technology [GPT-4] is implemented" in medicine. www.bostonglobe.com/2023/04/06/opinion/chat-gpt-health-ca… via @BostonGlobe
-
Peter Lee reposted
Atul Butte (In Memoriam)
@atulbutte
Great example from @madhumitasushil… #chatgpt off-the-shelf looking at cancer progress note (deidentified) can yield the biomarkers buried in that note! Of course this still needs supervision. #acentrus23
-
Peter Lee reposted
Eric Horvitz
@erichorvitz
At the White House today for discussions with the President on the risks and opportunities of #ai.
cnb.cx/3zycJRW #pcast @WHOSTP -
Peter Lee reposted
Antoine Moyroud 🇫🇷
@antoine_moyroud
I’m not much of a 150-page PDF reader so I had the Sparks of AGI research paper printed with a funky cover 🧠
If ever you’re in London or Paris happy to hand one over to you, I printed a few
@SebastienBubeck and team, thanks for the great work! -
Peter Lee reposted
Isaac Kohane
@zakkohane
“To serve man:” @DShaywitz writes a thoughtful review of our book (@peteratmsr @goldbergcarey). @timmermanreport timmermanreport.com/2023/04/new-book-contemplates-how-gen…
-
Peter Lee
@peteratmsr
Appreciate this review of our book on #gpt4 in medicine by @DShaywitz @timmermanreport. David nails each author's obsessions pretty much perfectly... "New Book Contemplates How Generative AI Will Serve Man" - timmermanreport.com
-
Peter Lee
@peteratmsr
The #lnif23 was just a super venue for starting the global conversation about #gpt4 in medicine. I so appreciated the energetic and thoughtful discussion and feedback, and look forward to continuing thisinto the future.
Lake Nona Impact Forum
@LNImpactForum
A special report by #lnif23 speaker @PeterAtMsr was published last week in @NEJM discussing the potential of #gpt4 as an AI chatbot for #medicine and #healthcare. The technology was previewed by @Microsoft at this year’s forum.
Learn more: www.nejm.org/doi/full/10.1056/NEJMsr2214184?query=feature…
Peter Lee
@peteratmsr
I especially want to emphasize how important it is that the medical community get up to speed on this emerging technology and take ownership of the questions about whether, when, and how it should or shouldn't be used in medical settings or the healthcare business.
-
Peter Lee
@peteratmsr
This was a fun and interesting discussion, and so I'm grateful to have been invited to join for this podcast. It took us a while, given so much to talk about, but we finally get into the questions about #gpt4 potential impact in medicine!
NEJM AI
@NEJM_AI
Microsoft’s Peter Lee describes the dangers of “irrational exuberance” in bringing technology to health care. He explains how he approaches problems at Microsoft grounded in understanding the health care day-to-day workflow.
Hear more from @peteratmsr: nejm.ai/ep4 -
Peter Lee reposted
Josh Mandel, MD
@JoshCMandel
It would be hard to overstate the importance of this open posture to my real-world impact on health data access/interoperability @MSFTResearch (not to mention my sense of personal fulfillment).
Openness here applies to collaborations, code, and publications.
Peter Lee
@peteratmsr
Being open forces us to work harder and faster and more "on point". It also makes us feel freer to liberally build on others' work, as we allow others to build on ours.
-
Peter Lee reposted
Ethan Mollick
@emollick
It seems a lot of the sudden advance in AI is due to the strong tradition of publishing public papers by researchers about topics that would have been kept secret in other contexts
Was this allowed by firms because no one knew AI would be useful & treated it like basic research? -
Peter Lee reposted
Ethan Mollick
@emollick
These two charts from @dustinvtran about the size of Google’s AI team relative to any other research team on the planet (including OpenAI & every university) makes the choice to give away the keys to the kingdom even stranger in retrospect. Good for AI advancement, though.
-
Ethan Mollick
@emollick
It seems a lot of the sudden advance in AI is due to the strong tradition of publishing public papers by researchers about topics that would have been kept secret in other contexts
Was this allowed by firms because no one knew AI would be useful & treated it like basic research?Post not available here, view on X
Peter Lee
@peteratmsr
Being open forces us to work harder and faster and more "on point". It also makes us feel freer to liberally build on others' work, as we allow others to build on ours.
-
Peter Lee
@peteratmsr
Breaking news: For the audiobook version of www.amazon.com/AI-Revolution-Medicine-Peter-Lee/dp/013820… by @peteratmsr @zakkohane @goldbergcarey @SebastienBubeck, publisher @Pearson is negotiating with Scarlett Johansson to narrate the voice of GPT-4, thus recreating her iconic role from the movie, "Her".
Peter Lee
@peteratmsr
April 1st... :-)
-
Peter Lee
@peteratmsr
I've been extremely impressed with the integration of GPT-4 into @khanacademy. The potential benefits for learners is so great, and the AI even makes the experience not just potentially more effective, but demonstrably more fun. Worth checking out. www.youtube.com/watch?v=rnIgnS8Susg
-
Peter Lee reposted
GeekWire
@geekwire
Trending: Microsoft Research head Peter Lee on the applications of GPT-4 in medicine and life sciences www.geekwire.com/2023/microsoft-research-head-peter-lee-o…
-
Peter Lee reposted
Christina Farr
@chrissyfarr
Okay this is fascinating.
NEJM is launching a new journal called NEJM AI - and @zakkohane, a physician with a background in informatics, is the editor-in-chief. ai.nejm.org -
Peter Lee reposted
Xavi
@xamat
New article in @NEJM on the use and promise of GPT-4 in medicine by @peteratmsr www.nejm.org/doi/full/10.1056/NEJMsr2214184?query=feature… Also, new upcoming book with many more details on how LLMs are going to radically transform healthcare www.amazon.com/AI-Revolution-Medicine-Peter-Lee/dp/013820…
-
Peter Lee reposted
Christina Farr
@chrissyfarr
O dang... @peteratmsr shares in his new book - worth a read, folks - that GPT-4 can help convert health data into and out of FHIR. That'll be critical. It can also provide advice to providers on improving quality ratings, which are tied to reimbursement rates.
-
Peter Lee reposted
Harvard Medical School
@harvardmed
HMS experts will steer the new @NEJM_AI journal as part of a journey to transform the field of clinical AI hvrd.me/8FzI50Nq59K
-
Peter Lee
@peteratmsr
Keep an eye out for the continued progress of @paolagorigiorgi's and her collaborator's research on DFT. I suspect some very significant things will emerge here.
Paola Gori-Giorgi
@paolagorigiorgi
Nice summary in Nature Italy of our paper on DFT for excited states with @timgould_scienc @DerkKooi and Stefano Pittalis (except for having H^2 instead of H_2..) #compchem www.nature.com/articles/d43978-023-00046-3
-
Peter Lee reposted
GeekWire
@geekwire
Microsoft research head Peter Lee on the implications of GPT-4 for medicine and research www.geekwire.com/2023/microsoft-research-head-peter-lee-o…
-
Peter Lee
@peteratmsr
When it comes to healthcare and medical research @MSFTResearch, the things that have pleased me most are all of the great people and organizations we've had the opportunity to collaborate with. They've been uniformly super and taught us a lot.
Microsoft Research
@MSFTResearch
We are at an inflection point for the use of AI in health care. Microsoft Research’s Health Futures organization shares its work to make health care more data-driven, predictive, and precise: www.microsoft.com/en-us/research/blog/ai-and-the-future-o…
-
Peter Lee reposted
Tom Lawry
@TCLawry
If you want to move beyond the hype surrounding GPTx and its application to medicine, read this assessment by @EricTopol of a book coming by @peteratmsr - The AI Revolution in Medicine - GPT4 and Beyond.
#ai #gpt4 @IrmaRaste @EvanKirstel @Gil_Bashe
erictopol.substack.com/p/the-gpt-x-revolution-in-medicine -
Peter Lee reposted
Daniel A. Dourado
@dadourado
Two immediate conclusions from this GPT-4 paper:
1) AI language models will soon be adopted in clinical settings. Without the slightest doubt.
2) One of the main regulatory concerns is to define ways to mitigate automation bias. This has to be thought about as soon as possible. -
Peter Lee reposted
Atul Butte (In Memoriam)
@atulbutte
Wow, great to see Epic committing to use @Microsoft @OpenAI GPT-4 with @NuanceInc Dragon Ambient Experience Express in its EHR products… I myself have seen a demo of DAX and am already amazed at what this product can do! www.beckershospitalreview.com/ehrs/epic-to-use-microsofts…
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
Peter Lee, head of Microsoft Research, and Ashley Llorens discuss the future of AI research and the potential for GPT-4 as a medical copilot: www.microsoft.com/en-us/research/podcast/ai-frontiers-ai-…
-
Peter Lee
@peteratmsr
I was glad that Charlotte Shubert @Geekwire could attend my lecture @UW on the potential impact of #gpt4 in medicine. We had a robust discussion afterwards. www.geekwire.com/2023/microsoft-research-head-peter-lee-o…
-
Peter Lee reposted
Frank X. Shaw
@fxshaw
One not to miss!
Microsoft Research
@MSFTResearch
Peter Lee, head of Microsoft Research, and Ashley Llorens discuss the future of AI research and the potential for GPT-4 as a medical copilot: www.microsoft.com/en-us/research/podcast/ai-frontiers-ai-…
-
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
I've had lots of fun following @peteratmsr 's lead over the last few months to think through all the **good** that careful applications of AI can do in healthcare. E.g., it's really amazing how much it could reduce paperwork, allowing doctors&nurses to refocus primarily on care.
Peter Lee
@peteratmsr
Our new article in @NEJM introduces clinical practitioners to the potential benefits, limits, and risks of #gpt4 AI in medical applications. The article is a quick read and links to a supplementary appendix with extended examples. @SebastienBubeck www.nejm.org/doi/full/10.1056/NEJMsr2214184?query=feature…
-
Peter Lee reposted
Tyler Berzin MD, FASGE
@tberzin
If you are interested in exploring impact of #chatgpt/ #llms in medicine, combine this @NEJM article by @peteratmsr with the superb @NEJM_AI podcast interview featuring @arjunmanrai + @AndrewLBeam and you'll be off to the races:
www.podbean.com/ew/pb-9sijp-13cafe4
Peter Lee
@peteratmsr
Our new article in @NEJM introduces clinical practitioners to the potential benefits, limits, and risks of #gpt4 AI in medical applications. The article is a quick read and links to a supplementary appendix with extended examples. @SebastienBubeck www.nejm.org/doi/full/10.1056/NEJMsr2214184?query=feature…
-
Peter Lee reposted
Felipe Kitamura
@FelipeKitamura
This paper has insightful uses of GPT4.
Peter Lee
@peteratmsr
Our new article in @NEJM introduces clinical practitioners to the potential benefits, limits, and risks of #gpt4 AI in medical applications. The article is a quick read and links to a supplementary appendix with extended examples. @SebastienBubeck www.nejm.org/doi/full/10.1056/NEJMsr2214184?query=feature…
-
Peter Lee reposted
NEJM
@NEJM
The editors announce both a series of articles focusing on AI and machine learning in health care and the 2024 launch of a new journal, NEJM AI, a forum for evidence, resource sharing, and discussion of the possibilities and limitations of medical AI. www.nejm.org/doi/full/10.1056/NEJMe2206291
-
Peter Lee
@peteratmsr
Our new article in @NEJM introduces clinical practitioners to the potential benefits, limits, and risks of #gpt4 AI in medical applications. The article is a quick read and links to a supplementary appendix with extended examples. @SebastienBubeck www.nejm.org/doi/full/10.1056/NEJMsr2214184?query=feature…
-
Peter Lee reposted
Eric Topol
@EricTopol
The @NEJM is getting serious about covering #ai in medicine with a new series and journal. A #gpt4 preview, review and editorial today
This important advance is covered in-depth in a new book
erictopol.substack.com/p/the-gpt-x-revolution-in-medicine -
Peter Lee reposted
Eric Topol
@EricTopol
Link for the #gpt4 paper
www.nejm.org/doi/full/10.1056/NEJMsr2214184?query=feature… -
Peter Lee reposted
Isaac Kohane
@zakkohane
It’s a great listen. Strongly recommend @AMIAinformatics @broad_ml4h @john_chilmark
Arjun (Raj) Manrai
@arjunmanrai
How will GPT-4 change medicine? I lost sleep after this long and fascinating conversation with @peteratmsr of Microsoft. He took us behind the scenes of the Microsoft/OpenAI partnership and predicted how GPT-4 & beyond will impact healthcare. Listen here: ai-podcast.nejm.org
-
Peter Lee reposted
ashish (@acgt01@genomic.social)
@acgt01
What a treat to listen to @peteratmsr being interviewed by @arjunmanrai & @AndrewLBeam .
Loved the insightful & sometimes tongue in cheek lightning round questions and the self reflective responses by Peter.
Learnt a ton !
Maybe my fav episode yet !
NEJM AI
@NEJM_AI
In the latest episode of AI Grand Rounds, Dr. Peter Lee (@peteratmsr) reveals @Microsoft’s interest in health care and the origins of the OpenAI and Microsoft partnership, and he speculates on how large language models like ChatGPT will transform medicine. Listen now:
-
Peter Lee reposted
Isaac Kohane
@zakkohane
Can’t recall a time like this. @peteratmsr describing a future that is in the hands of thousands of patients today. Good opener for a public conversation.
NEJM AI
@NEJM_AI
In the latest episode of AI Grand Rounds, Dr. Peter Lee (@peteratmsr) reveals @Microsoft’s interest in health care and the origins of the OpenAI and Microsoft partnership, and he speculates on how large language models like ChatGPT will transform medicine. Listen now:
-
Peter Lee reposted
NEJM AI
@NEJM_AI
In the latest episode of AI Grand Rounds, Dr. Peter Lee (@peteratmsr) reveals @Microsoft’s interest in health care and the origins of the OpenAI and Microsoft partnership, and he speculates on how large language models like ChatGPT will transform medicine. Listen now:
-
Peter Lee reposted
Ethan Mollick
@emollick
A lot of people ask me how to effectively use AI to do stuff. So I created this updated guide, with lots of links and resources. oneusefulthing.substack.com/p/how-to-use-ai-to-do-practic…
-
Peter Lee reposted
Bob Wachter
@Bob_Wachter
Re: new GPT-4 book by @peteratmsr, @zakkohane. I checked its pre-release sales this am, and was surprised to discover it's the bestselling book in history!
(As you likely guessed, it's the bestselling book in @Amazon's "history" category. A challenge for AI: words can be tricky!) -
Peter Lee reposted
Andrew Beam
@AndrewLBeam
How will GPT-4 change medicine?
Does GPT-4 show empathy?
What are the unique risks that GPT-4 poses for healthcare?
We explore these questions and more with Peter Lee (@peteratmsr) in the latest episode of AI Grand Rounds: ai-podcast.nejm.org/e/microsoft-s-peter-lee-on-the-future… -
Peter Lee reposted
Arjun (Raj) Manrai
@arjunmanrai
How will GPT-4 change medicine? I lost sleep after this long and fascinating conversation with @peteratmsr of Microsoft. He took us behind the scenes of the Microsoft/OpenAI partnership and predicted how GPT-4 & beyond will impact healthcare. Listen here: ai-podcast.nejm.org
-
Peter Lee reposted
Bob Wachter
@Bob_Wachter
Congrats @peteratmsr @zakkohane @goldbergcarey on timely, well written book on GPT-4's implications for healthcare–insider's view of the sky-high potential of GPT (many cool examples) along w/ sober reflections on potential harms & need for wise regulation www.amazon.com/AI-Revolution-Medicine-GPT-4-Beyond/dp/013…
-
Peter Lee reposted
Atanas G. Atanasov
@_atanas_
💡 GPT-4 and further AI applications are about to transform medicine
🔖 The AI Revolution in Medicine: GPT-4 and Beyond (1st Edition, by Peter Lee, Carey Goldberg, Isaac Kohane) www.amazon.com/AI-Revolution-Medicine-GPT-4-Beyond/dp/013… -
Peter Lee reposted
Frank X. Shaw
@fxshaw
Microsoft’s latest use for GPT4: Stopping hackers www.washingtonpost.com/technology/2023/03/28/microsoft-co…
-
Peter Lee
@peteratmsr
I'm very much looking forward to spending the day tomorrow with colleagues at the @UW Allen School. Will be an honor to give a distinguished lecture to share some of what we've learned about #gpt4 in medicine. www.cs.washington.edu/events/colloquia/details?id=3270
-
Peter Lee reposted
Eric Topol
@EricTopol
I wrote about GPT-4 in medicine and a new outstanding book by @peteratmsr @goldbergcarey @zakkohane coming out soon
erictopol.substack.com/p/the-gpt-x-revolution-in-medicine -
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
If you want to hear about some of the genesis of the Sparks paper, and very preliminary thoughts about what might come next, take a listen to this podcast!
Microsoft Research
@MSFTResearch
What is intelligence? How does it emerge and how do we measure it? Ashley Llorens and machine learning theorist Sébastian Bubeck discuss accelerating progress in large-scale AI and early experiments with GPT-4: www.microsoft.com/en-us/research/podcast/ai-frontiers-the…
-
Peter Lee reposted
Ethan Mollick
@emollick
This Bing example felt like working with a PhD student.
I asked it to look up wait equations (you can see what that is below), and then to derive a wait equation for film-making in AI. I then asked to solve the wait equation, and to reconsider its theory based on the answers. -
Peter Lee
@peteratmsr
We are about to witness a precambrian explosion of new GPT/AI-powered apps and tools. Starting pretty much now, but accelerating dramatically with days.
-
Peter Lee
@peteratmsr
Definitely worth a listen. #ai #gpt4
Microsoft Research
@MSFTResearch
What is intelligence? How does it emerge and how do we measure it? Ashley Llorens and machine learning theorist Sébastian Bubeck discuss accelerating progress in large-scale AI and early experiments with GPT-4: www.microsoft.com/en-us/research/podcast/ai-frontiers-the…
-
Peter Lee reposted
OpenAI
@OpenAI
We are adding support for plugins to ChatGPT — extensions which integrate it with third-party services or allow it to access up-to-date information. We’re starting small to study real-world use, impact, and safety and alignment challenges: openai.com/blog/chatgpt-plugins
-
Peter Lee reposted
Microsoft Research
@MSFTResearch
We’re already feeling the impact of today’s rapidly advancing AI. But do you know how we got here? Learn about the models helping to drive progress, training approaches pushing the cutting edge, and products harnessing these new capabilities: www.microsoft.com/en-us/research/blog/ai-explainer-founda…
-
Peter Lee reposted
Ethan Mollick
@emollick
👀After 154 pages of tests of GPT-4, this paper concludes “Given the breadth and depth of GPT-4's capabilities, we believe that it could reasonably be viewed as an early (yet still incomplete) version of an artificial general intelligence (AGI) system.” arxiv.org/abs/2303.12712
-
Peter Lee reposted
Josh Mandel, MD
@JoshCMandel
Let'e explore Epic's new EHI Export tables. Starting with ChatGPT4... we'll see how far we get.
> Epic just released their electronic health information export documentation. It's about 6k tables! Here's one smallish example. -
Peter Lee
@peteratmsr
Tremendous admiration for the work by the @OpenAI team. This report led by @SebastienBubeck should be useful reading for anyone who is curious about #gpt4 capabilities and limits, though much is still preliminary and evolving.
Sebastien Bubeck
@SebastienBubeck
At @MSFTResearch we had early access to the marvelous #gpt4 from @OpenAI for our work on @bing. We took this opportunity to document our experience. We're so excited to share our findings. In short: time to face it, the sparks of #agi have been ignited.
arxiv.org/abs/2303.12712 -
Peter Lee reposted
Sebastien Bubeck
@SebastienBubeck
Wonderful collaboration with @VarunChandrase3 @EldanRonen @JohannesGehrke @erichorvitz @ecekamar @peteratmsr Yin Tat Lee, Yuanzhi Li, @scottlundberg @HarshaNori @hmd_palangi Marco Tulio Ribeiro @YiZhangZZZ
-
Peter Lee reposted
Ethan Mollick
@emollick
Early days, but the first 5 public AI Large Language Models are here, and I have used them all. Taking the perspective of applying them to do real work, rather than as friendly chatbots, they all have strengths and weaknesses, but I now have a strong subjective ranking:
Top Tier: OpenAI ChatGPT-4 and Microsoft Bing AI. They are both built on the same platform, but they are different in key ways. At their best, they both seem like magic (I have written working Python programs based on a description, without knowing the language!). Bing, since it is connected to the internet, is much more powerful, but also has a weird personality that sometimes helps, sometimes hurts. ChatGPT-4 is steadier and is better for long projects like coding, but the lack of internet connection limits some uses. It is worth using these tools constantly, for any task that you have, just to see what they do. Sometimes they are brilliant, sometimes not. Both hallucinate less than earlier models.
My guide to Bing: oneusefulthing.substack.com/p/power-and-weirdness-how-to-…
Mid Tier: ChatGPT-3.5 (the free one everyone can access) and Anthropic's Claude. Both feel very similar. Claude is great for longer documents (it seems to have a longer memory), and is the most "sane" feeling of the AIs, but otherwise seems to lag behind ChatGPT-3.5. These are great for getting help on documents, generating ideas, etc. Be warned, neither is connected to the internet, but ChatGPT will pretend it is.
My guide to uses for these earlier models (before the release of Bing and ChatGPT-4): oneusefulthing.substack.com/p/the-practical-guide-to-usin…
Using ChatGPT for writing help (also works for Bing): oneusefulthing.substack.com/p/how-to-use-chatgpt-to-boost…
Bottom Tier: Google's Bard. I have only used it for a day, but it seems to both hallucinate (make up data) and provide worse initial answers. Again, this was just released, so I may not have figured out the secrets of working with it yet. It isn't bad, exactly, it just doesn't seem to have any advantages over other models. Except that it is very fast. But Google has a lot of talent and many models, so maybe this is just the start. -
Peter Lee reposted
Ethan Mollick
@emollick
AIs help spur idea generation, as it involves fluency (coming up with lots of ideas) & variation (unusual ideas). AI is good at both of these: oneusefulthing.substack.com/p/a-prosthesis-for-imaginatio…
Here is comparisons among AIs (Bing & GPT impress, Bard doesn't) in creating ideas for a espresso maker in space -
Peter Lee reposted
Ethan Mollick
@emollick
Its early, but Google's Bard does not seem as capable as a learning tool as Bing or GPT-4. In our paper on AI and teaching, we outline some AI prompts that can help. Not only does Bard not do interactive prompts, but the results are poor so far.
Paper: oneusefulthing.substack.com/p/using-ai-to-make-teaching-e… -
Peter Lee reposted
Ethan Mollick
@emollick
Bard also fails the Lem Test
SciFi author Stanislaw Lem wrote of two rival constructors of robots. One creates a robotic poet & the other challenges it to write an impossible poem, which it does (the English translator did a great job). Here is the original, Bard, ChatGPT & Bing -
Peter Lee
@peteratmsr
Microsoft subsidiary Nuance is using GPT-4 for a new physician notes app www.geekwire.com/2023/microsoft-subsidiary-nuance-is-usin… via @GeekWire
-
Peter Lee
@peteratmsr
If you haven't seen the new GPT AI integrations into Microsoft Office apps -- Microsoft 365 Copilot -- it's really worth your time to check it out. Just jaw-dropping capabilities and really smart integration. www.microsoft.com/en-us/microsoft-365/blog/2023/03/16/int…
-
Peter Lee
@peteratmsr
In computer vision today, Swin transformers are an important and, for most in the field, a de facto standard vision architecture. This article by Allohvk in @TDataScience provides one of the more lucid explanations of how and why Swin transformers work. towardsdatascience.com/swin-vision-transformers-hacking-t…
-
Peter Lee
@peteratmsr
I had fun at CNN #lifeitself connecting the dots from large-scale #machinelearning to #biology to a new paradigm of scientific discovery @marchodosh @drsanjaygupta. www.cnn.com/videos/health/2022/06/01/peter-lee-life-itsel…
-
Peter Lee
@peteratmsr
"Health is Wealth", @KizzyPhD #lifeitself
-
Peter Lee
@peteratmsr
This is great news! Through our partnership with the Broad Institute and Verily, we look forward to working together to accelerate the next generation of health and life science research on the Terra platform. investor.singulargenomics.com/news-releases/news-release-…
-
Peter Lee
@peteratmsr
The Microsoft Research Summit is by far the year's biggest computing research event. >400 speakers, 2 days of sessions available across 3 worldwide timezones, and a 3rd day with interactive live workshops. Over 30K have already registered, and it's free! researchsummit.microsoft.com/home_public
-
Peter Lee
@peteratmsr
AI to improve health is becoming real. This collaboration between @Humana, @Health_IT, and @MSFTResearch identifies patients who would benefit most from proactive attention. Yes, the AI is bleeding-edge, but the data foundation has been the real key. azure.microsoft.com/en-us/blog/humana-leverages-microsoft…
-
Peter Lee
@peteratmsr
Putting @Microsoft and @NuanceInc together is about bringing the most advanced - and most trustworthy - #ai to workers everywhere, in mission-critical areas like healthcare delivery and beyond. Cutting-edge AI that works for people and, well, just works. www.linkedin.com/pulse/microsoft-nuance-making-dream-ai-h…
-
Peter Lee
@peteratmsr
Here @Microsoft @Health_IT the @AdaptiveBiotech folks challenged us to accelerate, dramatically, the work on T-cell based diagnostics in order to help the world respond to #covid19. We said yes but honestly didn't know if it would be possible. But it was, and here it is.
-
Peter Lee
@peteratmsr
Doctors and patients, really being present with each other, seeing each other eye to eye. This is what ambient clinical intelligence can bring. And now Nuance DAX, from @NuanceHealth, with the power of Microsoft @Azure, makes it real. www.youtube.com/watch?v=sCLd5HliXYg
-
Peter Lee
@peteratmsr
So great to live in a time when something like this is possible: Working with @AdaptiveBiotech, we aim to map and decode the human immune system by merging #ai and #genetics #healthcarenext blogs.microsoft.com/blog/2018/01/04/microsoft-adaptive-bi…






































































































































































































































































































