Ishir Garg

AI Researcher · UC Berkeley

I'm an undergraduate at UC Berkeley studying Electrical Engineering & Computer Science and Engineering Mathematics & Statistics. I do research at Berkeley AI Research (BAIR), advised by Sergey Levine and Kurt Keutzer. Previously, I also worked with Dawn Song.

My current research interests focus on reinforcement learning and post-training for LLM and robotics agents: how models can learn from interaction, explore effectively, and improve themselves over time. Previously, I’ve worked on LLM memory and methods for synthesizing training data.

My messages are always open — feel free to reach out if you're interested in talking about reinforcement learning, agentic systems, robotics, or anything in between!

Ishir Garg

News

Publications

Agent at the center of a maze shaded by empowerment, with arrows in all four directions and a single arrow out of a dead end

Measuring Empowerment from Offline Data for Intrinsically Motivated ExplorationUnder Review

Ishir Garg, Vivek Myers, Qiyang Li, Catherine Ji, Benjamin Eysenbach, Sergey Levine

Preprint, 2026

Conversation flowing into a memory store and retrieval, with failure points marked

MemFail: Stress-Testing Failure Modes of LLM Memory Systems

Ishir Garg, Neel Kolhe, Dawn Song, Xuandong Zhao

Findings of EMNLP 2026

Seed problems expanding into novel, diverse clusters of generated problems

InfoSynth: Information-Guided Benchmark Synthesis for LLMs

Ishir Garg, Neel Kolhe, Xuandong Zhao, Dawn Song

Re-Data @ COLM 2026

Gradient projected orthogonally to previous-task directions over Fisher contours

Fisher-Orthogonal Projected Natural Gradient for Continual Learning

Ishir Garg, Neel Kolhe, Andy Peng, Rohan Gopalam

Preprint, 2026

Research

Industry

Awards