Reframing LLM Hallucination Detection as OOD Detection

A new arXiv paper (2602.07253) proposes treating hallucination detection in large language models as an out-of-distribution (OOD) detection problem. The authors argue that next-token prediction can be viewed as a classification task, allowing OOD techniques from computer vision to be applied with modifications for language model structures. Their approach yields training-free, single-sample-based detectors that achieve strong accuracy on reasoning tasks, where existing methods often struggle. The work suggests that reframing hallucination detection as OOD detection offers a promising and scalable path forward.