Excited to see this. 15 years ago, I spent most of my days doing CFD calculations and developing numerical solutions to Navier-Stokes equations. What a time to live!
We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics.
The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra.
The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier-Stokes equations can break down. It has remained unresolved for roughly 90 years.
Please deliver room temperature, ambient pressure superconductors 🤞
I spent much of the weekend talking with the team who did this work. Seb--and everyone else--acted with integrity and generosity throughout.
Initially we believed the other team had also solved the problem. We wanted to collaborate and do a joint release.
When we learned that they had Euler but not Navier-Stokes, we offered to let them go first, to suggest that they should be the ones to get the prize, and optionally for Tristan to be the lead author on a rewrite of the OpenAI proof. We felt it was challenging to offer the same to Levent (an Anthropic employee), who was not willing to talk or coordinate with us anyway. We were open to other solutions.
We would have greatly preferred coordination. We did not rush to publish even though the other team wasn't communicating with us. The team threatened us with unfounded accusations of plagarism.
Now that we can see their work, the approaches appear to be different. It is also worth noting that our latest model can solve many, many other math problems.
It is true that we tried this because there were rumors on the internet last week that Anthropic's models had solved a millennium problem and we were curious if ours could do it too.
Great move! We need a $100 openclaw subscription tier with unlimited gpt-5.3-codex-spark access
Peter Steinberger is joining OpenAI to drive the next generation of personal agents. He is a genius with a lot of amazing ideas about the future of very smart agents interacting with each other to do very useful things for people. We expect this will quickly become core to our product offerings.
OpenClaw will live in a foundation as an open source project that OpenAI will continue to support. The future is going to be extremely multi-agent and it's important to us to support open source as part of that.
Re-config this to accelerate scientific discovery would be amazing. Imagine explore vast concepts in science for weeks and find new theory for superconductivity that leads to new superconductors
Fun weekend playaround: browser-native M3GNet in JavaScript cchen.me/fun/
✅ Predict crystal properties: band gap, formation energy, refractive index, shear/bulk modulus
✅ Structure optimization + molecular dynamics
No backend, just client-side compute.
#AI4Science #AI4Materials
In reflecting on 2025, I keep coming back to how quickly AI moved from “interesting” to genuinely transformative for science and engineering.
Over the last few months in particular, I’ve felt a real shift. For the first time, AI agents can reliably take on meaningful research engineering tasks and produce results that are actually useful fairly autonomously after receiving a specific ask.
That shift has changed how I spend my time. My workflow has flipped: I now spend ~90% on problem definition, system design/architecture, validation, and review, and perhaps less than 10% on implementation. Not long ago, it was the other way around. It’s been both exciting and humbling. The leverage is real, but the responsibility to verify and steer the work is higher than ever.
I’ve also been energized by the momentum in AI for materials science and physical AI, especially seeing so many strong teams and startups forming around real technical problems and the Genesis Mission executive order. It feels like an inflection point, one where the frontier is less about whether these tools can work, and more about how we build trustworthy, repeatable workflows around them.
To close the year, I’m sharing a 148-page review paper generated by zero-shot an AI agent I built for fun. This is something I used to daydream about—and now it’s surprisingly accessible. It’s not perfect, but the trajectory is very promising. Moments like this give me a lot of hope that AI will significantly accelerate the pace of science. 2026 will be another big year for AI for Science.
Prompt: "Impacts and potentials of quantum computing for artificial intelligence"
Paper link: cchen.me/files/quantum_ai_re…
Blog: cchen.me/posts/2025-12-23-ai…
#AI4Science #AIAgent #materialsscience #physicalai #quantum
Wonderful idea
claude.ai/gift
It's that time of the year! Keep your loved ones thinking!
I know it is a hot area, still didn’t expect this 🤔
Jeff Bezos will be co-CEO of AI startup Project Prometheus theverge.com/news/821943/jef…
Very impressive! Strong open-source model for agentic tasks helps accelerate science.
🚀 Hello, Kimi K2 Thinking!
The Open-Source Thinking Agent Model is here.
🔹 SOTA on HLE (44.9%) and BrowseComp (60.2%)
🔹 Executes up to 200 – 300 sequential tool calls without human interference
🔹 Excels in reasoning, agentic search, and coding
🔹 256K context window
Built as a thinking agent, K2 Thinking marks our latest efforts in test-time scaling — scaling both thinking tokens and tool-calling turns.
K2 Thinking is now live on kimi.com in chat mode, with full agentic mode coming soon. It is also accessible via API.
🔌 API is live: platform.moonshot.ai
🔗 Tech blog: moonshotai.github.io/Kimi-K2…
🔗 Weights & code: huggingface.co/moonshotai
Congrats!
Polymers are central to materials science, yet AI for materials is still crystal-first. We hope this work helps change that.
Proud to support the team these last few years. Also shoutout to Piero @erunzzz and Guillem @guillemsimeon, both on X and quietly impactful.
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
Pleasantly surprised to receive an actual medal for our work on #Niobium.
Nb never ceases to fascinate me as a versatile element for advanced materials.
This project explored disordered rock-salt Nb oxides as a negative electrode for Li-ion batteries.
Original paper: nature.com/articles/s41563-0…
A lot of data to play with, anyone working on downsampling to create various sizes?
🔥 Today we announce the Meta OMol25 Electronic Structures Dataset - 500 TB of molecular data in collaboration with @mshuaibii and team at @AIatMeta. We envision a future where researchers can rapidly design molecules and peptides to treat diseases, discover catalysts to revolutionize synthesis and manufacturing, identify the next electrolyte to store and transport energy to protect the grid, and more. But these breakthrough discoveries require data.
Data to train next-generation AI models and interatomic potentials. Data to push the boundaries of what's computationally possible in molecular chemistry and lead the world in AI for science. Data that captures the full complexity of chemical systems, from small organic molecules to massive biomolecular complexes.
The OMol25 Electronic Structures dataset includes the raw DFT outputs, electronic densities, wavefunctions, and molecular orbital information for over 4M million high-accuracy quantum chemical calculations. We see this as a transformative opportunity to develop higher quality partial charges, partial spins, and advanced electronic features to unlock the next generation of physics-informed ML models.
The Materials Data Facility is proud to make these data available via the Eagle cluster at ALCF through a high-performance Globus endpoint. Given the dataset's unprecedented scale, we're first releasing all output data for a 4M random OMol25 split, with the full multi-petabyte dataset following based on community engagement.
For this first release, the data are quite raw, and as-created by the Meta team. There's a significant opportunity for the community to build tools that simplify access to these data, allow data query and browsing, create databases of calculated properties and descriptors, and much more. We intend to work on these topics with all of you.
We can't wait to see what you can do with these data!
Access Details: github.com/facebookresearch/…
Eagle was pioneered as the Petrel project, a new way to provide researchers access to high-quality, high-volume data by Ian Foster, Rachana Ananthakrishnan, Kyle Chard, Michael Papka, Rick Stevens, and others. Globus.org provides core platform capabilities (auth, data transfer, workflow automation, and compute) to over 600k researchers.
Thanks to support from NIST and James Warren for making the MDF vision of vast troves of open data to fuel discovery possible.
@mshuaibii, @zackulissi , @argonne, @argonne_lcf
What @xprize does is truly transformational. Honored to have joined the Deep Tech Brain Trust since May, contributing to AI & materials science innovation in service of XPRIZE’s mission to inspire and empower humanity toward an abundant, equitable future. Exciting times ahead!
We’re thrilled to share the release of our 2025 XPRIZE Impact Report. 🚀From carbon removal to space exploration, our prizes prove philanthropy can deliver 60x ROI in global impact. Learn how we're building the #BusinessOfBreakthroughs: xprize.org/2025-impact-repor…
Just reviewed a “review article” full of irrelevant and false concepts. When AI papers train the next wave of AI, we risk normalizing mediocrity and mistakes instead of pushing science forward.
Same prompt on one H100 NVL runs with 550 tok/s input and 65 tok/s output. Not bad. The model does seem to hallucinate more than closed top models, but still great progress. Will be exciting to see how the science community will leverage the model
