✳flâneur — a map of the web's best reading
I underestimated AI capabilities (again) - by Ajeya Cotra
planned-obsolescence.org · 2,560 words · saved by 1 readers
Revisiting a prediction ten months early
I underestimated AI capabilities (again) Revisiting a prediction ten months early Ajeya Cotra Mar 05, 2026 109 32 9 Share On Jan 14th, I made predictions about AI progress in 2026. My forecasts for software engineering capabilities already feel much too conservative. In my view, METR (where I now work) has some of the hardest and highest-quality software engineering and ML engineering benchmarks out there, and the most useful framework for making benchmark performance intuitive: we measure a task’s difficulty by the amount of time a human expert would take to complete it (called the “time hori
Explore this link on the map →related reading
- Measuring AI Ability to Complete Long Tasks - METRmetr.org
- We spent 2 hours working in the future - METRmetr.org
- Clarifying and predicting AGI — LessWronglesswrong.com
- A (Slightly) Mechanistic Theory for Exponentially Increasing AI Time Horizons? — LessWronglesswrong.com
- AIs can now often do massive easy-to-verify SWE tasks and I've updated towards shorter timelines — LessWronglesswrong.com
- My picture of the present in AI — LessWronglesswrong.com
- AI 2027ai-2027.com
- When AI builds itself \ Anthropicanthropic.com
- How Does Time Horizon Vary Across Domains? - METRmetr.org
- My picture of the present in AI - by Ryan Greenblattblog.redwoodresearch.org
- Is there a Half-Life for the Success Rates of AI Agents? - Toby Ordtobyord.com
- Estimating the Productivity of an Autonomous AI Software Engineer | Cognitioncognition.ai