PeopleofSuperintelligence

Researcher

Chi Jin

Also known as 金驰

Chi Jin is a machine learning researcher whose work spans reinforcement learning theory, multiagent learning, and AI agents. He studied physics at Peking University from 2008 to 2012 and completed a computer science PhD at UC Berkeley in 2019 under Michael Jordan. His work covers saddle-point optimization, Q-learning, multiagent self-play, the Pokémon agent PokéChamp, and the Goedel-Prover-V2 theorem prover. In 2023, he received an NSF CAREER Award and Princeton’s Faculty Advancement Award, followed by a Sloan Research Fellowship in 2024. He is a Member of the Technical Staff at OpenAI and an associate professor at Princeton, where he is on sabbatical.

Timeline

  1. 2008–2012

    Physics at Peking University

    He studies physics at Peking University from 2008 to 2012.

  2. 2013–2019

    Berkeley PhD and Q-learning

    He completes a PhD in computer science at UC Berkeley in 2019, advised by Michael I. Jordan. In 2017, he coauthors How to Escape Saddle Points Efficiently, which shows that perturbed gradient descent can escape saddle points efficiently. In 2018, he coauthors Is Q-learning Provably Efficient?, which proves a regret bound for Q-learning with UCB exploration in episodic Markov decision processes.

  3. 2020

    Multiagent learning through self-play

    In 2020, he coauthors Near-Optimal Reinforcement Learning with Self-Play. The paper proposes Nash Q-learning and Nash V-learning algorithms for two-player zero-sum Markov games.

  4. 2023

    NSF CAREER and Princeton award

    In 2023, he receives an NSF CAREER Award and Princeton’s E. Lawrence Keyes, Jr./Emerson Electric Co. Faculty Advancement Award.

  5. 2024

    A Sloan Research Fellowship

    In 2024, he receives a Sloan Research Fellowship.

  6. 2025

    PokéChamp for Pokémon battles

    In 2025, he coauthors the PokéChamp paper, which introduces a minimax agent powered by large language models for Pokémon battles. The system uses LLMs for action sampling, opponent modeling, and value estimation.

  7. 2025

    Goedel-Prover-V2 for Lean

    In 2025, he coauthors Goedel-Prover-V2, an open-source series of language models for automated theorem proving. The models use scaffolded data synthesis, verifier-guided self-correction, and checkpoint averaging.

  8. As of October 2026

    OpenAI and Princeton roles

    As of October 2026, he is a Member of the Technical Staff at OpenAI and an associate professor of electrical and computer engineering at Princeton, where he is on sabbatical. His group develops agents capable of advanced reasoning, strategic planning, and complex tasks.

Career

  1. NowOpenAIMember of the Technical Staff
  2. NowPrinceton UniversityAssociate Professor of Electrical and Computer Engineering

Education

  1. 2019University of California, BerkeleyPhD
  2. 2012Peking University

From the biography.

Sources and credits

Photo: We have no published photo of Chi Jin yet. Send us a free-licensed photo.

  1. Chi Jin sites.google.com. Checked 26 October 2026.
  2. Chi Jin simons.berkeley.edu. Checked 26 October 2026.
  3. How to Escape Saddle Points Efficiently arxiv.org. Checked 26 October 2026.
  4. Is Q-learning Provably Efficient? arxiv.org. Checked 26 October 2026.
  5. Near-Optimal Reinforcement Learning with Self-Play arxiv.org. Checked 26 October 2026.
  6. PokéChamp: an Expert-level Minimax Language Agent proceedings.mlr.press. Checked 26 October 2026.
  7. Goedel-Prover-V2: Scaling Formal Theorem Proving with Scaffolded Data Synthesis and Self-Correction arxiv.org. Checked 26 October 2026.