AI capability gains per billion dollars falling 8x in 2 years — and ASI may never be general, argues ramez
burny_tech · x · 2026-11-12
ramez cites data showing Anthropic is gaining exponentially less capability per billion dollars spent — boosting AI capabilities is now 8x as expensive as two years ago, worsening on a log scale (including METR report and internal models data). He also challenges the assumption that superintelligence will be general:
- AI surpasses humans in chess and math, but those are highly verifiable domains; that doesn't mean it can master messy real-world problems
- "I'm not saying RSI is impossible. The feedback loop is just much weaker than you think"
- AlphaGo Zero played itself 29 million times; Waymo needed billions of simulated miles to match what a human teenager does in a thousand
- Once you need to touch the real world or wait for experimental results, general superintelligence is far from guaranteed
More from AGI Musings
- The new Turing test: can a human with AI pass as intelligent? — cocktailpeanut · 2026-11-12
- CS faculty job market 2025: AI keeps driving demand, but only at schools that can afford to grow — fortnow · 2026-11-12
- Heresy is still a human advantage: AI won't voice what society knows is wrong — JeremyNguyenPhD · 2026-11-12
- Gary Marcus Slams Dario Amodei's 'Age of Miracles' Rhetoric as Cult-Like — GaryMarcus · 2026-11-12
- Tyler Cowen Interviews Terence Tao — burny_tech · 2026-11-12
- OpenAI chief economist: AI agents still early for most firms, job impact overstated — soumitrashukla9 · 2026-11-12