Whisper, an AI transcription tool by OpenAI, has been flagged for inaccuracies, including fabricated text.

These errors raise concerns, particularly in medical settings where accurate transcription is vital.

More than twelve experts, including software engineers and researchers, found multiple instances of inaccuracy in Whisper, referred to as “hallucinations.”

They have expressed concerns about its use in hospitals for doctor-patient consultations. This is despite OpenAI’s caution against its use in high-risk scenarios.

Documented errors include fabricated violent actions, racial comments, and fictional medical treatments.

Computer scientists also found similar transcription issues in their studies.

Whisper is integrated into platforms like Microsoft’s cloud services and is also widely used on the open-source platform HuggingFace.