Calibrating in AI 2027 units: his ~2.5x uplift estimate tops AI Futures Model's ~1.3x
austinc3301 · x · 2026-11-10
austinc3301 calibrates competing AI-uplift estimates in AI 2027's units (how much faster AI makes algorithmic progress than without it):
- Ryan's reading of METR data implies 4x
- His own model gives 2.5x (80% CI: 1.3–4.6x)
- That lands roughly on AI 2027's schedule (2x by late 2026, 3x by Jan 2027) and well above the AI Futures Model's 1.3x for today
A useful side-by-side reconciliation of the main AI-speedup estimates.
More from AGI Musings
- Safety advocates launch pledge rejecting AI labs' 'knowingly reckless' 2027 acceleration plans — AndyMasley · 2026-11-10
- AutoDNA: AI agents now run multi-step wet-lab molecular biology experiments autonomously — james_y_zou · 2026-11-10
- Sentinel forecasters raise ASI alert to highest level after Anthropic's acceleration warning — teortaxesTex · 2026-11-10
- The Bitter Lesson Was Misread: More Compute Makes Your Algorithm Matter More — tw_killian · 2026-11-10
- Data provenance, not model quality, is why enterprise AI fails, argues defense consultant — Graham_dePenros · 2026-11-10
- Pitch your idea to Claude instead of VCs — it pays you a salary with unlimited tokens — LeviTurk · 2026-11-10