I am currently a researcher at Thinking Machines.

I was previously a researcher at OpenAI, where I worked on reinforcement learning, small models, and synthetic data. I led the release of 4o-mini, and contributed to other models such as o*-mini and o3 (see About).

I graduated from UC Berkeley, where I worked on reinforcement learning and offline sequence modeling. I was fortunate to be advised by Pieter Abbeel and Igor Mordatch.

Email: matches my arxiv papers

Blog

• Jul 2025 — The Only Important Technology Is The Internet
  Why you should care about product-research co-design, and what is the dual of RL?
  
• Jun 2025 — AI Models for Pokemon Games
  What can Pokemon teach us about designing interactive agents?
  
• Mar 2024 — Spending Inference Time
  How should we structure inference compute to maximize performance?
  
• Feb 2024 — LoRAs as Composable Programs
  How can we design LLMs to be future-proof operating systems?
  
• Jan 2024 — Unifying RLHF Objectives
  What are different RL algorithms actually doing?
  

Papers

• Summary — Towards a Universal Decision Making Paradigm
  How can we design a universal learning method for sequential decision making?
  
• Jun 2021 — Decision Transformer
  How can we perform reinforcement learning with autoregressive sequence models?
  
• Mar 2021 — Pretrained Transformers as Universal Computation Engines
  What are the limits of transfer of large pretrained language models?