The AI Futures Project published a major update to its interactive model for forecasting AI progress and “takeoff” speed. The new version unifies timeline and takeoff assumptions into one framework and forecasts when systems might hit specific capability milestones like “Automated Coder” (full automation of software engineering) and broadly superhuman performance.
A key theme in the post is uncertainty and transparency: many parameters are grounded in public evidence, but plenty are still judgment calls. Instead of hiding those choices, the model makes them editable, so readers can see how timelines shift under different assumptions. Compared to the project’s earlier “AI 2027” framing, this update is less bullish on how much pre-automation AI systems will speed up AI R&D. That change pushes their central estimate for full coding automation out by a few years, and it also affects how sharp or gradual their modeled “takeoff” looks.
If you want to engage with it directly, the best starting point is the interactive site: try toggling assumptions about AI-driven R&D acceleration and compute scaling, then compare the resulting milestone timelines.