Large language models can say things that sound right but aren’t supported by the given document; this is called a faithfulness hallucination.
Skyra is a detective-style AI that spots tiny visual mistakes (artifacts) in videos to tell if they are real or AI-generated, and it explains its decision with times and places in the video.