Hospitals use a transcription tool powered by a hallucination-prone OpenAI model

OpenAI's Whisper, an AI transcription tool used in healthcare, occasionally creates false information in medical transcripts. A study found that Whisper hallucinated in about 1% of transcriptions, inventing sentences during silent periods. This is especially problematic for patients with language disorders. The tool is used by over 30,000 clinicians and 40 health systems, potentially affecting millions of medical conversations. OpenAI is working to reduce hallucinations and advises against using Whisper in high-risk situations.