Limit Autonomous Compute, Not Training Compute, for AI Safety
jachiam0 · x · 2026-08-11
AI researcher Jacob Hilton argues that instead of making laws to limit the compute (FLOPs) used in training runs, we should focus on limiting the amount of compute models can use autonomously without human review.
More from AGI Musings
- Study: AI Boosts Fossil Fuel Production More Than Green Energy, Raising Emissions — nordicinst · 2026-08-11
- AI Text Watermarking Called Ineffective: Rewriting with Another AI Easily Bypasses It — miniapeur · 2026-08-11
- Andrew Ng: Worrying About Evil AI Superintelligence Is Like Worrying About Mars Overpopulation — PMinervini · 2026-08-11
- Indifferent to AI? Creator Shares Backlash When Showing Off AI Art to Friends — Sufficient-Shock-641 · 2026-08-11
- The Real Threat of AI Agents: Incompetence Over Malice — bendee983 · 2026-08-11
- Personal Search Usage Drops 95% Amid AI Rise, Threatening Google — firstadopter · 2026-08-11