AI Capability Acceleration Looks Linear on Many Reasonable Scales, Researcher Argues
davidmanheim · x · 2026-11-17
In a thread with @TomDavidsonX and @robbensinger, the author argues the problem runs deeper than Labenz suggests: there is no absolute scale, and on many reasonable scales the acceleration is actually a linear trend, citing Greg Lewis' writeup. He adds that this does not mean accelerating capabilities won't kill us all — rather, people are updating to realize what should have been obvious: exponentials usually grow exponentially.
More from AGI Musings
- OpenAI staffer: science is where AI can do the most for society — shyamalanadkat · 2026-11-17
- Cambridge safety professor: AI labs knowingly risking all our lives commit historic crimes — DavidSKrueger · 2026-11-17
- Outlook outage celebration called 'best ad' for AI replacing white-collar work — RaphaelDabadie · 2026-11-17
- Is algorithmic efficiency really accelerating? EA Forum debate says AI capability has no good y-axis — davidmanheim · 2026-11-17
- Mathematician Steven Strogatz on a 46-min podcast: the numbers that run your life — stevenstrogatz · 2026-11-17
- In defense of the research blog: closer to arXiv rigor than critics admit — kohjingyu · 2026-11-17