Why is action chunking crucial for robot dexterity? 🤖 - We identify a natural tradeoff between temporal consistency and reactivity - New policy decoding technique that is *both* temporally consistent & fully reactive ICLR 2025 paper: arxiv.org/abs/2408.17355 A short thread 🧵

Heading to ICML Conference #ICML2025 this week! DM me if you’d like to chat ☕️ Come by our poster sessions on: 🧠 Optimizing Test-Time Compute via Meta Reinforcement Fine-Tuning (arxiv.org/abs/2503.07572) 🔍 Learning to Discover Abstractions for LLM Reasoning (drive.google.com/file/d/1Sfafrk…)

🚨 NEW PAPER: "RLAD: Training LLMs to Discover Abstractions for Reasoning"! We introduce reasoning abstractions: concise insights that help LLMs solve hard reasoning problems by guiding structured exploration. 📄 arxiv.org/abs/2510.02263 🌐 cohenqu.github.io/rlad.github.io/ 🧵[1/N]

🚨🚨New Paper: Training LLMs to Discover Abstractions for Solving Reasoning Problems Introducing RLAD, a two-player RL framework for LLMs to discover 'reasoning abstractions'—natural language hints that encode procedural knowledge for structured exploration in reasoning.🧵⬇️

Hierarchical RL for LLM reasoning. Paper: arxiv.org/abs/2510.02263

Excited to share that I’ll be presenting two papers at CoLM 2025! Cognitive Behaviors that Enable Self-Improving Reasoners (Session 1, Poster 26) Training LLMs to Discover Abstractions for Solving Reasoning Problems (Ram 2 Workshop, Talk 4 — Oct 10, 4:00 PM) If you’d like to

Check out our new paper on improving exploration in CoT for LLMs by generating abstractions! 👇 Rather than letting the LLM think longer and longer to explore, we can let it first produce concise insights that help guide structured exploration later. This works really well!