Ramez Naam: real AI research time horizons run 1-2 orders of magnitude shorter than METR forecasts
ramez · x · 2026-11-11
Ramez Naam pushes back on AI 2027/METR/ECI-style timeline forecasts with early real-world data:
- Real AI time horizons on research tasks appear 1-2 orders of magnitude shorter than METR/ECI/AI 2027 estimates.
- ECI does not seem to map to exponential improvement in real-world time horizons or researcher productivity.
He concedes the calibration data is narrow but argues it casts doubt on several forecasts and RSI models. In the thread, davidmanheim agrees key uncertainties are underdetermined but favors general predictive track records, noting Daniel Kokotajlo's record is hard to beat.
More from AGI Musings
- Could Recursive AI Improvement Compress Millennia of Progress into One Decade? — VraserX · 2026-11-11
- AI 2027 authors say reality is running 5 months ahead of their scenario — LeviTurk · 2026-11-11
- teortaxesTex: RL may give coding models 'full fluency' with no superhuman headroom — teortaxesTex · 2026-11-11
- Frontier AI Still Lacks 'Heresy': Breakthroughs Come from Defying Consensus — JeremyNguyenPhD · 2026-11-11
- Dev argues optimism is the only trait that matters as intelligence gets commoditized — zack_overflow · 2026-11-11
- AI risk community splits over 'intelligence explosion next year' claim as credibility war erupts — connoraxiotes · 2026-11-11