I'm interested in system-2 thinking, catastrophic forgetting, and fair evals.
I have contributed to the following open-source repositories:
- 🏋🏻♀️ Reasoning Gym – RL environments for reasoning models.
- 📚 RLHF Book – An introduction to RLHF and post-training.
- 🔬 Language Model Evaluation Harness – A framework for few-shot evaluation of LLMs.
- 📈 Policy Gradients – Minimal hackable implementation of policy gradient methods.
- 🌍 OpenEnv – An interface library for RL post training with environments.
- 🏒 Laser Hockey – Winning entry for an RL tournament in laser hockey.
- 👾 Word Game Bench – Evaluating LLMs on Wordle and Connections.
- 📖 ML Interview Q&A – Booklet with popular questions and answers for ML interviews.
My work is used by AI labs such as DeepMind, Meta, and NVIDIA:
- 🏋🏻 Reasoning Gym: Reasoning Environments for RL with Verifiable Rewards – NeurIPS (Spotlight)
- 🌊 Momentum-based Weight Interpolation for Continual Learning – Interpolate @ NeurIPS (Best Paper Award)