self-improvement

Definition

A loop in which a model generates candidate solutions, filters them, and trains on whatever survives — reducing the need for human annotation. It works best where correctness can be checked mechanically, since the filter, not the generator, is what supplies new information. The ceiling is set by the quality of that verifier; without a good one, the loop mostly amplifies existing bias.

Explainers using this term

• Paper Walkthrough: S³Gym — Does Self-Testing and Self-Judging Add Up to Self-Improvement?S3Gym: Can LLMs Turn Self-Testing and Self-Judging into Self-Improvement?
• Paper Explained: FlowBalance — Verifier-Grounded Self-Improvement from a Model's Own ReasoningFlowBalance: Verifier-Grounded Self-Improvement from On-Policy Reasoning Experience
• Paper Walkthrough: PILOT in the Loop — Fixing the Run While It Is Still RunningPILOT in the Loop: Live Self-Improvement for Long-Horizon Agents