LessWrong (Curated & Popular) "AI Futures Timelines and Takeoff Model: Dec 2025 Update" by elifland, bhalstead, Alex Kastner, Daniel Kokotajlo
14 snips
Jan 6, 2026 Get ready for an insightful dive into AI futures! The hosts revamp their timelines and takeoff models, pushing back full coding automation predictions by about three years. They explore various forecasting methods, from revenue extrapolation to compute anchors, and discuss how these impact expectations for superintelligence. Unpacking the three-stage model, they analyze the transitions from automating coding to the intelligence explosion. Expect stimulating debates on technical limits, revised probability estimates, and what future evidence might reshape their outlook.
AI Snips
Chapters
Transcript
Episode notes
Modeling Improves Forecast Transparency
- Models force explicit aggregation of many arguments and improve calibration of forecasts.
- Elifland argues modeling helps update transparently and incorporate new data over time.
Use The Interactive Model To Test Priors
- Play with the model's parameters and Monte Carlo settings to reflect your priors.
- The authors recommend using the interactive web app at aifuturesmodel.com to explore uncertainties.
Modeling Changes Moved Timelines More Than New Data
- Improvements to modeling AI R&D automation shifted their median AC forecast later by ~2-4 years.
- This change mattered more than new empirical evidence since April 2025.
