Stable confidence beats high confidence

A new method improves LLM reasoning by checking whether a model's confidence holds steady under small changes, not just whether it is high.

Also today

Today's top five AI stories, one line each, at 7am Eastern. Free. One-click unsubscribe.