Combining compute caps with use lags and audits could delay frontier AI by ~1.5 years, analysis says
ronbodkin · x · 2026-11-19
ronbodkin discusses AI pacing governance: his model, where AI is slow to develop taste for slowly-validated research, yields 7 months delay from a 20% compute cap — comparable to a 9-month use lag or comprehension audits, each delaying TED-AI by about a year. Combining use lags with audits delays 1.5 years and slows pace throughout. He argues caps alone shift timelines without slowing takeoff, so they should be paired with speed limits or embedded-auditor comprehension audits.
More from AGI Musings
- Luiza Jarovsky's newsletter #346: five new AI governance papers on whether agents follow rules — LuizaJarovsky · 2026-11-19
- At MLOpsWorld, only 3-5 people in the room still read code—the 'post-literate engineer' era — annetgriffin · 2026-11-19
- AI-written text this year already exceeds the entire internet, and it's going private — bennash · 2026-11-19
- Gary Marcus slams Musk for planning to lease GPUs to Anthropic despite safety concerns — GaryMarcus · 2026-11-19
- Epoch's "automate Epoch" write-ups: math preprints acknowledging AI use jumped 4% to 25% in months — Jsevillamol · 2026-11-19
- Memo Akten & Katie Hofstadter podcast: why do we rush to call machines conscious? — memoakten · 2026-11-19