David Manheim on AI superforecasting and broken evals

davidmanheim · x · 2026-07-20

David Manheim discusses AI superforecasting and the state of AI evaluations on the FLI podcast.

He says forecasting ability has been getting much better, and highlights work by people like @dschwarz26, @BTurtel, and @BenTereick. The framing suggests that evaluation and forecasting are becoming important lenses for understanding current model capabilities.

The post itself is brief, but it points to a substantive discussion about how to measure model forecasting skill and why many AI evaluations may be failing to capture what matters.

Related event: Expert Discusses AI Superforecasting and Evaluation Challenges(3 posts)→

Original post →

More from AGI Musings

AGI Musings channel →