Google/Penn State paper: Self-Improvement Memory lifts harness optimization gains by 2.36 points across models

rohanpaul_ai · x · 2026-11-18

A new Google and Penn State paper (arXiv:2611.18856) studies Continual Recursive Self-Improvement (CRSI): how harness optimizers can carry gains across benchmark or policy-model switches. They propose Self-Improvement Memory (SIM), which groups edits by recurring harness problem and records rationale, predicted/observed outcomes, and evaluation context. Across four domains SIM beats harness-only and root-cause-only continuation; in coding, paired continuations with SIM improve final scores by an average of 2.36 points over harness-only (p=0.003) across ten independent source runs.

Related event: Google Proposes Self-Improvement Memory for Cross-Model Agent Harnesses(3 posts)→

Original post →

More from coding & agent

coding & agent channel →