RESEARCH · RESEARCH · #2564
Self-Direction: agents distill mistakes into persistent low-rank adapters
The authors introduce Self-Direction, a method where agents reflect on failed trajectories to extract lessons and immediately distill them into lightweight low-rank adapters (~0.02% of base model parameters) while keeping base weights frozen; adapters are stored in a persistent library and routed to related problems. Experiments with open-weight Qwen-family models on GAIA, Claw-Eval, and Terminal-Bench 2.1 report 8.9–17.0 percentage-point increases in overall success, reduced repeated mistakes (nearly doubling correction reliability), and transfer of a few adapters to dozens of new tasks.
KEY POINTS
- The authors introduce Self-Direction, a method where agents reflect on failed trajectories to extract lessons and immediately distill them into lightweight low-rank adapters (~0.02% of base model parameters) while keeping base weights frozen; adapters are stored in a persistent library and routed to related problems.
- Experiments with open-weight Qwen-family models on GAIA, Claw-Eval, and Terminal-Bench 2.1 report 8.9–17.0 percentage-point increases in overall success, reduced repeated mistakes (nearly doubling correction reliability), and transfer of a few adapters to dozens of new tasks.
- Self-Direction offers a cheap, non-destructive way for agents to turn mistakes into persistent, transferable lessons via tiny adapters that avoid full fine-tuning and overwriting base-model capabilities.
WHY IT MATTERS
Self-Direction offers a cheap, non-destructive way for agents to turn mistakes into persistent, transferable lessons via tiny adapters that avoid full fine-tuning and overwriting base-model capabilities.