I'm an AI researcher.

Fable achieved a significant breakthrough in one of our open problems. This is a problem where ChatGPT 5.5 could not even begin anything useful. The breakthrough seems legit (although not 100% checked yet), and Fable even claims to have a full solution. >10 hours total runtime so far. A 30 page document with the proofs of some lemmas not yet spelled out. We can not yet know whether Fable indeed has solved it, but even if it is just a partial solution, we are absolutely amazed. More details will follow, and once we are at the end of the story, I will also write a full substack post. Collaboration with István Vona, a postdoc in my group.

FrontierMath got updated to remove erroneous problems and the consequent score change is kinda wild

Co-Mathematician has retained its position relative to other models, save Fable which has taken the lead. I don't think Fable will stay there for long.

In two hours I'm hosting a panel discussion with Kevin Buzzard, Johan Commelin, and @AlexKontorovich about math Lean AI. There's still time to join us (see link below) or submit questions in the thread.

Sorry. I got to say this publicly. I really agree with @karpathy point here. My wife @mioana is a leading economist and we discuss this all the time. The singularity think of the AI community is rather misguided.

Sobering take-away from 1stproof (round 2) 1stproof.org/. OpenAI's vanilla prompt to 5.5pro tinyurl.com/yc8ymuna solves research math 10-40 x cheaper than custom prompts from academic teams. We used Gemini pro. Switching to 5.5pro improves results a lot but costs rise to the level of other academic pipelines :(

Yet another Co-Mathematician-enabled paper. Nice to see mathematicians working with the system to make progress. Follow Daniel Zheng for more AI-for-math announcement from Google.

If you were a company and you were going to benchmax once, when would you do it?

Vibe on Mythos seems to be that people don't like being told what they can't do with a model.

What if diffusion models could think ahead instead of being greedy at every step?🤔 We introduce: Learned Relay Representations for Forward-Thinking Discrete Diffusion Models

Queens club, London.

Warning: once you learn category theory, you'll never be able or willing to talk with people who don't know category theory.

Just heard Dimitri Bertsekas passed away. We would often chat after his Convex Analysis class. He was always very kind and encouraging of my theoretical pursuits. The guy wrote books like most people write papers. He was a true educator. Sorry that I didn't get to cross paths recently.

Google has provided high quality AI referee services to the most recent NeurIPS, ICML, and STOC conferences.

over the weekend i had another obvious thing to check, namely whether claude autonomously resolves the famed sum-product conjecture over the reals. answer: yes

This was awesome! Don't despair, even if you missed it, the talk will appear on the youtube channel youtube.com/@RLtheory soon!

*​When you think of AI, think of humanity too* ​This was the main message of the GStar Summit on AI & Humanity that I organized two days ago (as a breather after Google I/O). For a long time, I was deeply focused on advancing model capabilities toward superhuman reasoning, and I found immense joy when we accomplished our goal, e.g., our IMO-gold milestone in July 2025. I always assumed that someone, somewhere, would just take care of the social impact side for me. ​It really hit me in Feb 2026 when our math research agent, Aletheia, solved FirstProof problem #7 by flawlessly utilizing heavy mathematical machinery, as noted by our surprised mathematicians. I started to wonder: what's left for humanity? ​Discussing human values with my wife, Wendy Nguyen, and our dear friend Jean DeSombre, I realized I hadn't been thinking much about humanity while actively charting the course of frontier AI. But as our conversations continued over the past few months, I noticed myself becoming more "humanity-aware." ​For example, during a team lunch at Google DeepMind, we were chatting about what we would do after we "solved" AI for math. I heard a number of suggestions, e.g., robotics or world models, but nothing explicitly centered on humanity. I told the team that we need to think about empowering humans: one way is to build a model capable of asking questions, from basic paraphrasing all the way to forming conjectures, which would be incredibly useful for both education and model training! ​At GStar Summit 2026, aside from the deep dives into agentic AI with @edchi, @YiTayML, @preslav_nakov, Phong Nguyen, Myungsub Choi, Noriyuki Kojima, and Hung Bui, we dedicated half of the summit to talking about humanity! ​@PoShenLoh, awesome as always, proposed the "Thought Full" philosophy and reminded us how people find joy in helping others. Jay Kim shared with us new perspectives about human health in space, showing that space is not that scary to be a part of. Together with other speakers and panelist, Wendy Nguyen, Jean DeSombre, Marc Woo, Laurent El Ghaoui, Tuong Nguyen, Tuoc Huynh, Tuan Cao, and @CurtisSChin, we discussed all aspects of humanity in the age of AI. ​Our hope is that everyone who walked out of the summit can now find joy in talking and working with each other on human values (either before or after AI)! ​Thanks to everyone for coming from all over Asia Pacific and Silicon Valley! Many people told me that it's very rare in Vietnam for people to stay from 8am - 6pm with a fully-packed hall of 1,000 people! See you again soon!

your novel idea, when you ask an llm to fill in the details

Sold out before early bird deadline. I wish there had been a running progress bar because that was not on my bingo card. Hoping they set up a waiting list or some in person registrations.

Second math paper citing Google DeepMind's AI co-mathematician. Follow Daniel for future announcements.