Models went from grade-school math to hard problems in 2 years — and faster gains are coming
felixrieseberg · x · 2026-09-11
Felix Rieseberg argues people haven't internalized how fast models improve: two years ago they barely did grade-school math, today we debate which unsolved math problems they'll crack, and the pace is accelerating. He pairs this with a threat report his team published the same day, saying combining expected capability gains with threat analysis explains how frontier engineers really feel.
More from AGI Musings
- Bryan Johnson Applies 'Mechanistic Interpretability' to Women's Health, Says Hormonal Swings Fell 30% — menhguin · 2026-09-11
- Ex-AI-risk community urged to drop p(doom) claims that stall public engagement — agstrait · 2026-09-11
- 130B output tokens = 35,616 PhD-years: the scale argument for AI research output — YouJiacheng · 2026-09-11
- tszzl: AI extinction risk is tiny but orders of magnitude above anything else — and better models will help alignment — deanwball · 2026-09-11
- Frontier models like GPT-6 Astra excel at one thing: relentlessly pursuing verifiable objectives — daniel_mac8 · 2026-09-11
- Researcher quits Anthropic after 3 years of pretraining work: 'I haven't finished a thought in four years' — inductionheads · 2026-09-11