Every big model jump of the past two years was surpassed within three months
Afinetheorem · x · 2026-09-04
AI researcher Afinetheorem notes that every model with a big capability jump over the past two years — o1 included — has been caught within three months. His core argument: capabilities will keep improving, and treating AI policy and strategy as "today plus epsilon" remains a very common mistake.
More from AGI Musings
- Writers love AI drafts, readers hate them: suspected AI content gets skipped and authors punished — birchlse · 2026-09-04
- AI won't make the best lawyers cheaper — it may make them worth more — jkubicki · 2026-09-04
- Andreessen: AI agents meeting crypto is the most important investment theme of the next year — learnedall · 2026-09-04
- Millière vs Mitchell: can the intentional stance make an AI bot a genuine believer? — raphaelmilliere · 2026-09-04
- Millière continues: Odysseus Bot could earn the intentional stance if it pursues persistent goals — raphaelmilliere · 2026-09-04
- Frontier model reasoning shifts from sparse motifs to reasoning-first cognition, observer argues — teortaxesTex · 2026-09-04