Definition
Credit assignment is the problem of figuring out which action in a long sequence was responsible for an eventual outcome — the move that lost the chess game, the line of code that broke the build, the early decision that derailed the agent. It’s central to RL and gets exponentially harder as horizons get longer.
Episodes covering this
Worth reading next
Papers we haven't done a deep dive on yet, but would recommend on this topic.
- Let's Verify Step by Step
- RLVR Is Not RL: Rethinking Credit Assignment in Verifiable Reward Settings
- GRPO: Group Relative Policy Optimization for LLM Reasoning
- GiGPO: Group-in-Group Policy Optimization for LLM Agent Training
- RAGEN: Understanding Self-Evolution in LLM Agents via Multi-Turn Reinforcement Learning
- DARE: Language Model Weight Pruning via Model Merging
- LoRA: Low-Rank Adaptation of Large Language Models