Revised Forecasts: The AI Futures Model and the Shift in Takeoff Timelines
Coverage of lessw-blog
In a significant update to forecasting literature, lessw-blog has released the "AI Futures Timelines and Takeoff Model: Dec 2025 Update." This analysis revisits the projected arrival dates for transformative AI capabilities, specifically focusing on the "Automated Coder" (AC) and subsequent superintelligence (ASI).
The debate surrounding AI "takeoff" speeds-how quickly artificial intelligence transitions from human-level competence to superintelligence-is central to global safety strategies and regulatory frameworks. Aggressive timelines often drive urgency in safety research and capital allocation, while longer horizons might allow for more robust governance structures. Understanding the specific variables that accelerate or decelerate this process is essential for realistic strategic planning.
The core of this update is a notable shift in expectation regarding the "Automated Coder" milestone-the point at which AI can fully automate software engineering. Unlike previous "AI 2027" models that predicted a rapid sprint to full automation, this revised model extends the timeline by approximately three years. The primary driver for this adjustment is a more conservative view on "pre-full-automation AI R&D speedups."
Essentially, the analysis argues that while AI will eventually automate its own improvement, the intermediate phase where AI merely assists human researchers may not yield the explosive productivity gains previously assumed. This suggests a potentially longer runway before the recursive self-improvement loop becomes fully autonomous. The post also introduces an interactive tool, aifuturesmodel.com, allowing users to adjust parameters and test how different assumptions regarding compute, algorithm efficiency, and investment impact the timeline.
For stakeholders in AI safety, policy, and investment, this post provides a nuanced counter-narrative to hyper-accelerated timelines. While the author candidly notes that the model relies on intuitive estimates for certain parameters, the resulting shift in the projected arrival of ASI highlights the sensitivity of these forecasts to assumptions about R&D efficiency.
We recommend reading the full post to understand the methodology behind these shifted timelines and to experiment with the interactive model.
Read the full post on LessWrong
Key Takeaways
- Timeline Extension: The updated model predicts a delay of approximately three years for the arrival of the "Automated Coder" compared to previous AI 2027 models.
- R&D Efficiency: The primary factor slowing the forecast is a reduced expectation for how much AI will accelerate R&D before it reaches full automation.
- Interactive Modeling: The release includes a web-based tool allowing users to toggle parameters and visualize different takeoff scenarios.
- Uncertainty Remains: The author emphasizes that the model relies on intuitive guesses for several parameters, underscoring the difficulty of empirical prediction in this domain.