Jiarui Yao

Astra developed the key ideas for two decade-old conjectures in just two hours. 😱 @OpenAI @sama @gdb On Sept 6, Xiangxin @NickZhou523786, Jiaxi, and I asked GPT-6-Astra to tackle Chen @wjmzbmr1 & Li’s 2016 best-arm identification conjectures. It produced the key algorithm and proof framework for resolving both in the independent, unit-variance Gaussian setting. Claude Fable 5.1 cross-checked the proofs and found no issues. After preliminary human review, we formalized the main results in Lean 4, also with GPT-6-Astra, passing the kernel check on Sept 8. On Sept 10, CodeBuddy + hy4-preview discovered that another team had posted a concurrent solution on arXiv (arxiv.org/abs/2609.10529v1) the day before (Sept 9). Our proof takes a different route, and we also provide Lean formalization code. Manuscript + code: github.com/zhouxiangxin1998/… The way we do research may be changing faster than we expected. Also thanks @thsottiaux for the reset :)

Thrilled to share our paper MiCRo: Mixture Modeling and Context-aware Routing for Personalized Preference Learning (arxiv.org/pdf/2505.24846) won an EMNLP 2025 Outstanding Paper Award! 🎉🎉 Huge congrats to the team @evangelinejy99 @RuiYang70669025 @YifanSun99 @FengLuo895614 @rui4research, and big thanks to our advisors Prof. Tong Zhang and @hanzhao_ml!

I am at EMNLP 2025 HPC-AI! #emnlp2025 #hpcai

We introduce Gradient Variance Minimization (GVM)-RAFT, a principled dynamic sampling strategy that minimizes gradient variance to improve the efficiency of chain-of-thought (CoT) training in LLMs. – Achieves 2–4× faster convergence than RAFT – Improves accuracy on math reasoning benchmarks – Generalizes to reinforcement learning methods such as GRPO – Comes with theoretical convergence guarantees 📄 Paper: arxiv.org/abs/2505.02391 🔗 Code: (expected in a few hours) github.com/RLHFlow/GVM #LLM #MachineLearning #ReinforcementLearning #ChainOfThought #AIResearch