AI timelines forecaster slightly raises confidence, timelines slightly shorter

The forecasting group's update reflects a maturing approach to predicting AI milestone timing. The original model relied on METR's measurements of coding time horizon to anchor estimates for Automated Coder, but the team identified fundamental weaknesses in that approach—chiefly, it remains unclear what time horizon value actually corresponds to AC, and there is significant disagreement among researchers on this mapping. Rather than leaving those problems unresolved, the forecasters chose to rebuild their model to address these gaps. The slight shortening of timelines paired with higher confidence suggests the new model either incorporates new evidence pointing to earlier achievement or better captures the uncertainty around the original estimates. The team's stated intention to issue regular updates as new evidence arrives signals an iterative, evidence-driven forecasting discipline rather than a one-time prediction.