Maintaining Alignment during RSI as a Feedback Control Problem
beren — 2025-03-02 — LessWrong
Summary
Argues that maintaining alignment during recursive self-improvement (RSI) requires shifting from static alignment approaches to dynamic feedback control systems, identifying this as an overlooked framework that could address classic alignment theory’s pessimism about hitting precise value targets.
Source
- Link: https://lesswrong.com/posts/PhgEKkB4cwYjwpGxb/maintaining-alignment-during-rsi-as-a-feedback-control
- Listed in the Shallow Review of Technical AI Safety 2025 under 2 agenda(s):
- supervising-ais-improving-ais — Make AI solve it
- tiling-agents — Theory