News

METR updates AI time-horizon analysis, highlighting uneven capability gains

AI evaluation nonprofit METR has updated its time-horizon plot, a visualization intended to show how long certain tasks AI systems can complete as their capabilities improve.

The report finds that some AI capabilities continue to advance along an exponential trajectory. However, the results also reveal an important distinction between technical execution and judgment: models perform better at coding tasks than at tasks that require contextual understanding, discernment, or other forms of judgment.

The update suggests that broad measures of AI progress need to be interpreted carefully. Even when performance improves rapidly in some areas, the gains may not transfer uniformly to tasks that depend on evaluating ambiguous situations or making appropriate decisions.

Sources