Learning Beyond Gradients

Learning Beyond Gradients

Learning Beyond Gradients Jiayi Weng Continual Learning has remained hard largely because of catastrophic forgetting in neural networks: learn something new, and old capabilities can get overwritten. But what if we do not put all of our attention on neural network weights? Is there another way to make progress? As LLM agents get stronger, coding gets faster and better. But the phenomenon I find more interesting is different: a coding agent can keep reading failures, editing code, adding tests, and watching replays, and a program system can improve without training a new network or updating wei

saved by

• Davis Brown
• Vihan Tiwari
• Christina Lee

related reading

• [2606.10346] Reasoning or Memorization? Direction-Aware Diversity Exploration in LLM Reinforcement Learningarxiv.org
• Lil'Loglilianweng.github.io
• Debugging Reinforcement Learning Systemsandyljones.com
• You can’t imitation-learn how to continual-learn — LessWronglesswrong.com
• Deep Reinforcement Learning: Pong from Pixelskarpathy.github.io
• Why We Need Continual Learning | Andreessen Horowitza16z.com
• Q-learning is not yet scalableseohong.me
• pdfopenreview.net
• Deep Reinforcement Learning: Pong from Pixelskarpathy.github.io
• The Continual Learning Problemjessylin.com
• [2602.19362] LLMs Can Learn to Reason Via Off-Policy RLarxiv.org
• State of RL for reasoning LLMs | A. Weersaweers.de