Safety10h ago
Safety
Proposed Long Self-Correction Framework For AI Safety
A new proposal suggests replacing the "AI Pause" concept with Long Self-Correction. The author argues that humans are currently too unstable to serve as safe alignment targets for powerful systems. This shift moves the focus from mere reflection to active human improvement. Practitioners must address human volatility before building advanced AI.