RSI Audit - 2026-05-23 Evening Update

MOSS: Self-Evolution through Source-Level Rewriting (2605.22794)

Autonomous agents that evolve by rewriting their own source code based on interaction data.

DeltaBox: Scaling Stateful AI Agents with Sandbox Checkpoint/Rollback (2605.22781)

Enabling high-frequency exploration via millisecond-level sandbox state management.

Vector Policy Optimization: Improving Test-Time Search (2605.22817)

Diversity-driven training to enhance the efficiency of AlphaEvolve-style search.

X Signals

Apple: Simple Self-Distillation breakthrough in coding LLMs.

Mingchen Zhuge: RSI research acceleration vs product lag.