OmniConfess: Eliciting Token Confessions to Mitigate Omni-Modal Hallucination

Researchers introduced OmniConfess, a training-free inference-time method that elicits token-level confessions from omni-modal large language models (OmniLLMs) to reveal which evidence sustains a generated commitment and to mitigate hallucination. The work targets OmniLLMs that unify text, images, audio, and video but hallucinate when generation relies on the wrong evidence, a gap existing inference-time methods rarely address.

Omni-modal large language models OmniLLMs unify text, images, audio, and video, yet hallucinate when generation relies on the wrong evidence. Existing inference-time methods can reduce hallucinations, but rarely reveal which evidence sustains a generated commitment. We introduce OmniConfess, a tra