Chi Jin

Life update🙂: I’m on sabbatical from Princeton and have started at OpenAI, working on building AGI. Happy to be back in the Bay Area after 6 years! Bay Area friends—DMs open for food & hikes.

Really excited about the new GPT-5.5 release — a new level of intelligence, and a new level of usefulness for everyday life. Also, try Codex to accelerate your daily work!

@ysu_nlp @NeoCognition Big congrats Yu!

Introducing @NeoCognition, the agent lab for specialized intelligence. Everyone needs experts, but human expertise does not scale. Backed by $40M seed funding, we build self-learning agents that specialize across domains to make expertise abundant.

Thrilled to share our Goedel-series project on code! We can now automatically generate not only verifiable mathematical proofs, but also verifiable coding/programs. Thanks for Ziran (@__zrrr__), Zenan, and all collaborators for the amazing work!

Introducing Goedel-Code-Prover 🌲 LLMs write code, but can they prove it correct? Not just pass tests, but construct machine-checkable proofs that a program works for ALL possible inputs. We built a system that does exactly this. Given aprogram and its specification in Lean 4, Goedel-Code-Prover automatically synthesizes formal proofs ofcorrectness. Our 8B model achieves 62% overall success rate across three benchmarks (Verina, Clever &AlgoVeri), a 2.6x improvement over the strongest baseline, surpassing both frontier LLMs (GPT/Gemini/Claude)and open-source theorem provers up to 84x larger (DeepSeek-Prover/Goedel-Prover/Kimina-Prover/BFS-Prover).

Check out the full manuscript about the largest AI Pokémon tournament we ran at NeurIPS 2025!

Honored to have contributed to substantial token efficiency gains powering the GPT Codex release!

Hard to believe it’s been 6+ years since we coauthored “Is Q-learning Provably Efficient?”—my first RL foundations paper, which also turned out to be super impactful. Really excited to collaborate again!

@shaneguML lol, would love to catch up in person!

@chijinML Love to catch up sometime! Are you going to work on Pokemon (along with other projects)? ;)

Life update🙂: I’m on sabbatical from Princeton and have started at OpenAI, working on building AGI. Happy to be back in the Bay Area after 6 years! Bay Area friends—DMs open for food & hikes.

This is a truly remarkable math theorem prover! — well ahead of competitors, near-saturating PutnamBench, and achieving much higher solve rates on the recent concluded Putnam 2025 with a suprisingly short amount of time.

How do we close the gap between specialist RL and generalist LLM agents? The PokeAgent Challenge @ NeurIPS 2025 was designed to map the Pareto Frontier with a state space complexity that makes RL envs cry We had 100 active teams with over 650 people (and growing) in our community leading to 15k+ battles and dozens of hours of AI speedrunning gameplay The result: RL specialists dominate in partially observable 2p0s games while hybrid RL-LLM solutions are best for high-stakes long-context tasks Winners Track 1 (Battling): PA-Agent (RL) & Foul Play (MCTS). Judges' Choice: Porygon2 (League), August (LLM) Track 2 (Speedrunning): Heatz (LLM+RL hybrid), Hamburg Pokerunners (Dreamer-based). Judges' Choice: Deepest (Least # actions LLM) We have a few interesting things coming in the future, including... -Our retrospective paper, new battling ladder for VGC, and Pokemon AI dataset dropping in late January 2026 -Our full LLM agent run of Pokemon Emerald playing live on youtube .com/@ PokeAgentChallenge -PokeAgent Challenge v2 next year... with a twist (potential sponsors---send me a DM!) -Join our community for more discussion discord. gg/ZU9BQGnuts On a personal note, this was an incredible feat to put together and I have to give a huge shoutout to @__jakegrigsby__ for countless late nights calls discussing infrastructure and potential sponsorship ---this wouldn't have happened without him. Huge thanks to Aaron Traylor and Minmin Chen for speaking at our workshop, @steph_milani @kiranvodrahalli @drfeifei @yukez @chijinML for organizing support, and @GoogleDeepMind for sponsoring

It was really nice talking to you all at such an interesting event. Special thanks to Seth and Jake for the tremendous effort in organizing!

Join us at the 5th MATH-AI Workshop at @NeurIPSConf now, our biggest year yet with a record 249 submissions!! 🎉 ➡️ mathai2025.github.io We’re honored to host an incredible lineup of speakers: @swarat @WeizhuChen @j_dekoninck @Leonard41111588 @HannaHajishirzi @chijinML @tkalil2050 @aviral_kumar2 @Besteuler @tengyuma @patrickshafto @tegmark @jonathan_thomm @dawnsongtweets ☕️🥐 Light breakfast provided (for anyone who didn’t get a chance to grab one!). Huge thanks to our sponsors for making this possible: @DARPA @HarmonicMath @awscloud @axiommathai #NeurIPS2025 #MATHAI #Math #AI

At #NeurIPS2025? Join us for a Social on Wednesday at 7 PM, featuring a fireside chat with Jon Kleinberg and mentoring tables. Ft. mentors @canondetortugas @SurbhiGoel_ @HamedSHassani @tatsu_hashimoto @andrew_ilyas @chijinML @thegautamkamath @MountainOfMoon + more!

How do we close the gap between specialist RL and generalist LLM agents? We're benchmarking it in Pokémon. Join us at the PokeAgent Challenge competition workshop @ NeurIPS 2025. 📍 Dec 7, 8AM in San Diego 🎮 Track 1: Competitive Pokémon (game-theoretic reasoning) 🗺️ Track 2: Speedrunning (long-horizon planning) Speakers from Google DeepMind, NYU, CMU, UT Austin, Princeton.

Super proud of my fantastic postdocs and graduate students taking their next steps at frontier labs 🎉 • Yong Lin (@Yong18850571) → Thinking Machine • Zihan Ding (@Hanry65960814) → Bytedance • Ahmed Khaled → Google It’s always bittersweet to say goodbye😢 but I couldn’t be more excited to see what you achieve next!

I am going to present two papers at #COLM2025 tomorrow from 4:30-6:30pm, as none of our leading authors can attend due to visa issues. Haven't done poster presentations for years 🤣🤣 .... so I will do my best! #76: LongProc #80: Goedel-Prover v1

Our Goedel-Prover V1 will be presented at COLM 2025 in Montreal this Wednesday afternoon! I won’t be there in person, but my amazing and renowned colleague @danqi_chen will be around to help with the poster — feel free to stop by!

Excited to share that I’ve been promoted to Associate Professor with tenure at Princeton!🎉 6 years may not be long, but AI research has evolved significantly during this period. Grateful to all my students, collaborators, colleagues for being with me on this remarkable journey!

🚀With early access to Tinker, we matched full-parameter SFT performance as in Goedel-Prover V2 (32B) (on the same 20% data) using LoRA + 20% of the data. 📊MiniF2F Pass@32 ≈ 81 (20% SFT). Next: full-scale training + RL. This is something that previously took a lot more effort on our Princeton cluster. Really impressed by how Tinker takes care of the heavy multi-GPU infra while still letting us control the algorithms. It feels great to focus on research instead of plumbing. This also echoes Thinking Machine Lab’s LoRA blog thinkingmachines.ai/blog/lora/ — makes me excited about what scalable and efficient systems like Tinker can bring to the research community. Thanks to the Thinking Machines team and Goedel team at Princeton @__zrrr__ @Yong18850571 @johnschulman2 @Tianyi_Zh @danqi_chen 🙌 Also check out our paper on Goedel prover: arxiv.org/abs/2508.03613