Paper summary
Paper summary: RRSI: Regularized Recursive Self-Improvement of Agent Harnesses
Automatically evolving an agent's harness (prompts, control flow, tools, memory) tends to overfit the tasks it evolves on; RRSI regularises the proposal and selection steps and keeps gains on out-of-distribution benchmarks.
- Recursive self-improvement of agent harnesses can memorise its training tasks, much like an overfit model.
- Limits on edit size, a critic against benchmark-specific tricks and a pruner for useless changes favour reusable mechanisms.