Jeremy Howard proposes rule to slow AI self-improvement

Simon Willison's Weblog

Howard suggests the top-ranked AI lab must not use its own model for frontier research, while others have access. He argues this would halt frontier advancement and reduce power imbalance, criticizing Anthropic's current approach.

AnthropicAI safety

Simon Willison's Weblog·Simon Willison

Jun 10, 3:23 PM