International Math Olympiad (IMO) is the hardest math test for high schoolers. —USA beat the favorite China with a mostly Chinese-American team —India came 4th, 1pt short of Korea —Haojia Shi [China] got a perfect score 2x in a row, the 7th to ever do that 6 questions, NINE hrs!

Want more experts without the cost of increasing, under-utilized parameters? Check out our new paper: S’MoRE — Structural Mixture of Residual Experts for LLM fine-tuning, which extends the concept of routing flexibility into structural flexibility. How? 💡 Instead of selecting

Honored to serve as corresponding author on our #APA25 Poster of Distinction: “Longitudinal Trends of Pain Medication Prescriptions and Risks of Opioid-Related Disorder in Chronic Pancreatitis.” Grateful to work with amazing collaborators from PUMCH, Johns Hopkins, and Cedars.

Honored that our work, “Unequal Burdens: Racial, Ethnic, and Sex Disparities in Chronic Pancreatitis-Related Mortality”, has been selected for oral presentation at #APA2025 in San Diego! Nov 14 | 5:10 PM PT, Loews Coronado, San Diego. Session: Inclusivity in Training & Patient

Agent training can be viewed in many ways: learning from world-model interactions (or experience) is one; but in practice, especially for multi-agent, co-evolving the models + framework just works. That’s Mixture-of-Minds. With a few specialized small models and a simple

Instead of post-training each individual model for its own optimal outcome, mixture-of-minds optimizes the outcome for the whole agents system. It proves with great “agents” team collaboration, weak base models are able to beat frontier models GPT5/Gemini2.5 Pro in complex domain

Now you don’t even need to buy a GPU.

📢 Paper Alert LLM agent intelligence is different from human beings, so they don’t need to communicate in human’s way either. Together with Yujia Zheng Zhuokai Zhao prof Kun Zhang-in pursuit of Causality with ML and other partners, more than happy that ThoughtComm was accepted as a #neurips2025 spotlight

I am speaking at 2025 APA Annual Meeting. Please check out my talk if you're attending the event! - via #Whova event app

Too late to join the #openreview bug party, but honestly, probably too lazy to check who my reviewers were anyway😂. Good or bad, it is what it is... That said, I am genuinely grateful for my #NeurIPS2025 reviewers, and I can't wait to be in San Diego next week to meet more

Happening right now. Stop by our #NeurIPS2025 posters at 2706 and 3501!

#NeurIPS2025 Poster session continued, come to our poster at 1812 for some sweet S’MoRE 🤗

Also come check out this sick HARDMath2 benchmark at #NeurIPS2025 poster 1516 by Erik Wang !

Happening right now, Jason Weston presenting our DreamGym at #NeurIPS2025 Room 11AB

Meta × TBD Lab × CMU × UChicago × UMaryland In our latest work, we introduce Token-Level LLM Collaboration via FusionRoute 📝: arxiv.org/pdf/2601.05106 LLMs have come a long way, but we continue to face the same trade-off: – one huge model that kind of does everything, but is

DreamGym has been accepted to #ICLR2026 🚀

❓Is GRPO reaching its limits? While GRPO is powerful for reasoning, it has two major "kryptonites": high variance in small groups 🔊 and vanishing gradients when every answer in a group is wrong. 😢 Enter EBPO. 🧵👇 🔗: arxiv.org/pdf/2602.05165 #LLM #GRPO

We have just open-sourced the code. github.com/xiongny/Fusion…. Check it out!

Dr Yi Jiang to present and moderate at #DDW2026

Our commentary on the SIMBA trial is now out in Gut. A “negative” trial with positive lessons! Appreciate the SIMBA investigators for moving chronic pancreatitis research forward.