Training compute is small vs experiments, so a training pause may barely matter

eli_lifland · x · 2026-09-30

Eli Lifland argues frontier training compute is generally quite low compared to experiment compute, so pausing training without pausing experiments may not matter much on the months-long timescale of training the next model — even if the freed training compute were redirected to safety.

Related event: Forecaster Eli Lifland doubts training pause will meaningfully slow frontier AI, puts ~25% odds on real slowdown(6 posts)→

Original post →

More from AGI Musings

AGI Musings channel →