Researchers now only forecast AI three months out, as Noam Brown says even insiders are surprised
victor_explore · x · 2026-09-20
Quoting a researcher on the Navier-Stokes effort who no longer forecasts AI capability more than three months ahead — twelve months used to be comfortable — and OpenAI's Noam Brown's remark that the people surprised by the pace are inside the labs, the author argues builders have compressed planning horizons to a quarter: if your roadmap assumes the same model behavior a year from now, you're forecasting further than the researchers will.
More from AGI Musings
- "We should accelerate harder": arguing doomers shouldn't get a veto on AI's future — VraserX · 2026-09-20
- Hassabis tells King Charles at AI safety meeting: confident we can address these risks — borowcy · 2026-09-20
- Open Source Isn't Dying of Code — It's Dying of a Generation's Engineering Culture, Accelerated 100x by AI — thedealdirector · 2026-09-20
- Tech hype fatigue: from Google Glass to 'AI swarms will kill us all' — BdR76 · 2026-09-20
- The compute-intelligence-discovery loop is closing, says AI commentator — Dr_Singularity · 2026-09-20
- US daily AI usage doubled in six months, up from 8% to 19% — The Decoder · 2026-09-20