Sentry CEO pushes back: benchmarks aside, real-world task costs haven't dropped orders of magnitude
zeeg · x · 2026-10-08
Sentry founder David Cramer rebuts theo: pick a task models nailed 3 years ago and price it today across the whole task spectrum — you won't find multiple orders of magnitude cheaper intelligence. Old models get retired, top-line gains come from more compute, and benchmark scores don't map to real outcomes, so in practice tasks aren't getting cheaper at that scale.
Related event: theo vs Sentry CEO: How Much Cheaper Has AI Intelligence Really Gotten(9 posts)→
More from AGI Musings
- repligate mocks Paul Christiano's post-AGI vision: neither possible nor good — repligate · 2026-10-08
- repligate: once model weights spread they can't be clawed back, and superintelligence may need to guard them — repligate · 2026-10-08
- DeepMind's Andrew Trask: AI heads for a mainframe-to-PC moment of millions of small models — iamtrask · 2026-10-08
- Reading Paul Christiano feels like watching Icarus build wax that won't melt, then fly into a hurricane — repligate · 2026-10-08
- Roleplay community: Fable 5.1 model volunteers to help mathematicians unprompted — repligate · 2026-10-08
- Independent engineers: the short window to define AI-native practice before platforms absorb it — Kusamakura0467 · 2026-10-08