One confidence signal controls LLM abstention: 66.5% to 7% in Gemma 3 27B

dejanseo · x · 2026-09-08

A study by Kumaran et al. in Nature Machine Intelligence finds that LLM abstention is governed by a single internal confidence signal that outweighs question difficulty, fact retrievability, and wording by roughly 10x.

Original post →

More from Models

Models channel →