Task-horizon doubling now 4.3 months; forecast professional AGI acceptance by late 2027

ChipHaseCoolGuy · reddit · 2026-09-12

A Reddit user applies the METR-style task horizon framework — the length of human tasks a frontier model completes at 50% reliability — to forecast AGI timing: minutes for GPT-4 in 2023, hours for GPT-5-class in 2025, workday-plus by late 2026.

Core claims

Evidence cited

Point estimates: early RSI now–2027; strong AI R&D automation 2026–2028; professional AGI acceptance late 2027; broad consensus 2028; full RSI 2029 (2027–2032); ASI 2030–31, using the OpenAI-charter definition of AGI.

Original post →

More from AGI Musings

AGI Musings channel →