I will present this work at the ICML Multi-Agent System (MAS) workshop during the poster sessions. If you are interested in this work or self-play LLMs in general, please feel free to come chat with me!

🚀 Call for Papers — NeurIPS Conference 2025 Workshop Multi-Turn Interactions in LLMs 📅 December 6/7 · 📍 San Diego Convention Center Join us to shape the future of interactive AI. Topics include but are not limited to: 🧠 Multi-Turn RL for Agentic Tasks (e.g., web & GUI agents,

If you are interested in becoming a reviewer for the workshop, feel free to sign up here (expected workload: 2~3): docs.google.com/forms/d/e/1FAI…

Excited to share that FlexOlmo💪 is accepted as spotlight at NeurIPS Conference. See you at San Diego

Dear NeurIPS Conference PCs, I don't understand why we still need reviewers and area chairs if PCs are finally going to take over and overturn the AC decision without providing any reason, whereby our weeks of effort spent on rebuttals (both authors and reviewers) have been ignored.

Seems to me the "findings of *ACL" decision is really wise and visionary (5 years ago), compared to the "had to reject because of venue limit" thing

Borderline accept is the new strong reject #NeurIPS2025

Hi everyone! We're looking forward to the first NLP Seminar of the year! For this week's seminar, we are excited to host Tong Chen (Tong Chen) from University of Washington! If you are interested in attending remotely, please fill out the form below: forms.gle/E1iL719njyG1Nf…

Thanks for the tweet AK! We designed a game that works with ANY image pairs - synthetic scenes, charts, real photos. Self-play on these arbitrary visual inputs improves reasoning across the board. Scalable visual reasoning improvement without manual curation :)

🚨What if solving a problem correctly isn't enough—cuz the WAY to reason about it based on your audience matters just as much⁉️ We introduce ✨personalized reasoning✨: proactively asking user preferences and adapting HOW models think Frontier models are not doing well at this!🧵

Forget modeling every belief and goal! What if we represented people as following simple scripts instead (i.e "cross the crosswalk")? Our new paper shows AI which models others’ minds as Python code 💻 can quickly and accurately predict human behavior! shorturl.at/siUYI🧵

6 months after our paper release, I still recall the debates on removing the length normalization term in DrGRPO. And people gradually think DrGRPO is just about removing the std, ignoring the most important and subtle (length) bias we tried to point out to the community. Even

Excited to share Prompt Curriculum Learning (PCL) from AI at Meta - we improve performance-efficiency tradeoffs for reasoning RL by predicting prompt difficulty with a value model updated on-policy, and selecting intermediate-difficulty prompts that yield high effective ratios.

I am at #COLM2025 🇨🇦 this week. Happy to chat about anything about RL fine-tuning of LLM, multiagent learning and multi-LLM training. I will be presenting a talk at the AI Agent workshop on Friday afternoon.

🥳🥳🥳Very happy to be selected as one of the three outstanding paper awards at the AIA workshop at Conference on Language Modeling !!! Congrats Mickel Liu

Yushi is awesome, please come work with him!

Great paper! Thanks for citing our Self-play work [arxiv.org/pdf/2506.07468]! Glad to see the continued endeavor of multiagent training for safety alignment.

🚀 Exciting opportunity! We are hiring research interns (current PhD students) at Meta FAIR to advance multi-agent, multimodal AI! Work on text, audio, images & more, collaborate with top mentors, and help shape the future of AI at scale. Apply: metacareers.com/jobs/182171308…

My team at FAIR at Meta is recruiting interns for next summer! If you're a PhD student interested in questions around theory of mind in language models for social, multi-agent settings, and have relevant background and/or experience: metacareers.com/jobs/182171308…

Our latest work proposes a new metric for measuring deception in LLMs, based on whether the LLM causes the listener's beliefs to become less accurate. This correlates more strongly with human judgements of deception than 4 existing methods. We find that LLMs frequently engage in