Yiping Wang

We only need ONE example for RLVR on LLMs to achieve significant improvement on math tasks! 📍RLVR with one training example can boost: - Qwen2.5-Math-1.5B: 36.0% → 73.6% - Qwen2.5-Math-7B: 51.0% → 79.2% on MATH500. 📄 Paper: arxiv.org/abs/2504.20571…

Code release! 🚀 Following up on our IMO 2025 results with the public LLM Gemini 2.5 Pro — here’s the full pipeline & general (non-problem-specific) prompts. 👉 [github.com/lyang36/IMO25] Have fun exploring! #AI #Math #LLMs #IMO2025

>>> Qwen3-Coder is here! ✅ We’re releasing Qwen3-Coder-480B-A35B-Instruct, our most powerful open agentic code model to date. This 480B-parameter Mixture-of-Experts model (35B active) natively supports 256K context and scales to 1M context with extrapolation. It achieves…

WHY do you prefer something over another? Reward models treat preference as a black-box😶‍🌫️but human brains🧠decompose decisions into hidden attributes We built the first system to mirror how people really make decisions in our #COLM2025 paper🎨PrefPalette✨ Why it matters👉🏻🧵

🚨 Olympiad math + AI: We ran Google’s Gemini 2.5 Pro on the fresh IMO 2025 problems. With careful prompting and pipeline design, it solved 5 out of 6 — remarkable for tasks demanding deep insight and creativity. The model could win gold! 🥇 #AI #Math #LLMs #IMO2025

New SOTA for 26-circle packing: @ypwang61 achieved 2.635977 sum of radii using OpenEvolve (evolutionary optimization framework). Progress: AlphaEvolve paper reported 2.635, OpenEvolve made improvements, now new record at 2.635977. #CirclePacking #SOTA #Optimization #OpenEvolve

New blog post about asymmetry of verification and "verifier's law": jasonwei.net/blog/asymmetry… Asymmetry of verification–the idea that some tasks are much easier to verify than to solve–is becoming an important idea as we have RL that finally works generally. Great examples of…

Remember the drama about RLVR techniques that got improved accuracy with Qwen models basically no matter what you did – incorrect rewards, one example, entropy minimization, whatever? Then there was a refutation (pic). Now there are objections to the refutation too…

Becoming an RL diehard in the past year and thinking about RL for most of my waking hours inadvertently taught me an important lesson about how to live my own life. One of the big concepts in RL is that you always want to be “on-policy”: instead of mimicking other people’s…

I gave an invited lecture at New York University for @hhexiy's class! I covered three ideas driving the LLM revolution: scaling, emergence, and reasoning. I tried to frame them in a way that reveals why large LMs are special in the history of AI. Slides: docs.google.com/presentation/d…

Recent work has seemed somewhat magical: how can RL with *random* rewards make LLMs reason? We pull back the curtain on these claims and find out this unexpected behavior hinges on the inclusion of certain *heuristics* in the RL algorithm. Our blog post: tinyurl.com/heuristics-con…

(1/4)🚨 Introducing Goedel-Prover V2 🚨 🔥🔥🔥 The strongest open-source theorem prover to date. 🥇 #1 on PutnamBench: Solves 64 problems—with far less compute. 🧠 New SOTA on MiniF2F: * 32B model hits 90.4% at Pass@32, beating DeepSeek-Prover-V2-671B’s 82.4%. * 8B > 671B: Our 8B…

Happy to introduce Kimina-Prover-72B ! Reaching 92.2% on miniF2F using Test time RL. It can solve IMO problems using more than 500 lines of Lean 4 code ! Check our blog post here: huggingface.co/blog/AI-MO/kim… And play with our demo ! demo.projectnumina.ai

Why you should stop working on RL research and instead work on product // The technology that unlocked the big scaling shift in AI is the internet, not transformers I think it's well known that data is the most important thing in AI, and also that researchers choose not to work…

We built 200k-GPU clusters; We scaled up & curated higher-quality data; We scaled compute by 100x; We developed training & test-time recipes; We made everything RL native; We stabilized infrastructure and speeded up; That's how you turn RL into the pre-training scale. Yet I am…

Can data owners & LM developers collaborate to build a strong shared model while each retaining data control? Introducing FlexOlmo💪, a mixture-of-experts LM enabling: • Flexible training on your local data without sharing it • Flexible inference to opt in/out your data…

Reasoning can be made much, much faster—with fundamental changes in neural architecture. 😮 Introducing Phi4-mini-Flash-Reasoning: a 3.8B model that surpasses Phi4-mini-Reasoning on major reasoning tasks (AIME24/25, MATH500, GPQA-D), while delivering up-to 10× higher throughput…

🤔 How do we train AI models that surpass their teachers? 🚨 In #COLM2025: ✨Delta learning ✨makes LLM post-training cheap and easy – with only weak data, we beat open 8B SOTA 🤯 The secret? Learn from the *differences* in weak data pairs! 📜 arxiv.org/abs/2507.06187 🧵 below

Personalization methods for LLMs often rely on extensive user history. We introduce Curiosity-driven User-modeling Reward as Intrinsic Objective (CURIO) to encourage actively learning about the user within multi-turn dialogs. 📜 arxiv.org/abs/2504.03206 🌎 sites.google.com/cs.washington.…

Happy to share that ReasonIR is accepted by @COLM_conf! Synthetic data & test-time scaling are powerful tools to enable new capabilities for challenging tasks. I’m impressed by how quickly smaller retrievers and better rerankers have been developed with ReasonIR data! #COLM2025

EvalTree accepted to @COLM_conf 2025 - my first PhD work and first COLM paper 🙌! What would you like to see next—extensions, applications, or other directions? Always open to ideas! 🧐