That's actually what the paper is about. I don't know why they didn't use that in the title. > Here we develop new methods grounded in statistics, proposing en

That's actually what the paper is about. I don't know why they didn't use that in the title. > Here we develop new methods grounded in statistics, proposing entropy-based uncertainty estimators for LLMs to detect a subset of hallucinations—confabulations—which are arbitrary and incorrect generations.