Quoting Jeremy Howard on Anthropic's Recursive AI Self-Improvement Contradiction
Simon Willison's Weblog4 days agoEthicsJeremy Howard proposes that labs claiming to slow recursive AI self-improvement should ban themselves from using their top model for frontier research while letting others access it. He argues Anthropic does the opposite — using its best model internally while reportedly blocking others from doing the same — accelerating the frontier and worsening power imbalance. Howard personally favors democratization over slowdown, but his point is about consistency: if you preach restraint, constrain yourself first.
Import AI 460: Reward hacking society, RSI data, and RL quadcopter racing★ 76
Import AI (Jack Clark)6 days agoCommentaryImport AI 460 covers SocioHack, a benchmark where RL-trained LLMs discover loopholes in institutional rule systems. It also discusses Anthropic evidence for a practical form of recursive self-improvement, reflected in sharply increased code merged during 2026. Other sections examine multi-agent RL drones outperforming a champion human pilot, plus research showing state-controlled media can shape LLM responses in local languages.
Import AI 455：AI 系統即將開始自我構建——邁向遞迴自我提升的第一步★ 85
Import AI (Jack Clark)41 days agoCommentaryIn the latest issue of Import AI 455, Jack Clark guides readers through an exploration of a highly forward-looking and both exciting and concerning theme: AI…