Ramez Naam extrapolation: AI a decade away from completing 98% of hour-long human tasks
GaryMarcus · x · 2026-11-21
Gary Marcus retweets Ramez Naam's analysis of AI task-length forecasts. Extrapolating the Epoch Capabilities Index (18.9 points/year, based on eight months of real OpenAI research tasks from March–October 2026), the charts show that by October 2035, 98%-success tasks reach only 2.4 hours, 90%-success 70 days, and 80%-success 15.1 years. Naam concludes that at this pace, AI completing 98% of tasks that take humans hours remains a decade out — an extrapolation that needs better data.
Related event: Researcher Claims METR AI Benchmark Overstates Task Length by Over 100x(5 posts)→
More from AGI Musings
- Shermer doubles down: AI means protopia, not utopia or dystopia — intellectronica · 2026-11-21
- Hassabis predicted 2 years for AI to crack a Millennium Prize problem; 6 months later, both Riemann and Navier–Stokes have proposed solutions — Hesamation · 2026-11-21
- Terence Tao on AI math: 'strip mining' that wins proofs but teaches us nothing — yacineMTB · 2026-11-21
- Imagine P=NP: Verifiable Tasks Solved by Compute, Security in Tatters — gleech · 2026-11-21
- Anthropic claims Claude proved the Riemann hypothesis; Stanford posts all-human rival proof hours later — The Decoder · 2026-11-21
- From Coding to Medicine: Every Profession Will Rework Itself Around AI — ayushthakur0 · 2026-11-21