Software engineers, developers and academic researchers have serious concerns about OpenAI’s Whisper transcripts, according to a report from the Associated Press.
Although there is no shortage of discussion about the tendency of generative AI to hallucinate – basically, to invent things – it is a bit surprising that this is a problem of transcription, where we would expect the transcription to closely follow the audio being transcribed.
Instead, researchers told the AP, Whisper introduced everything from racial comments to imaginary medical treatments into the transcripts. And this could be particularly disastrous as Whisper is adopted in hospitals and other medical settings.
A University of Michigan researcher studying public meetings found hallucinations in eight out of ten audio transcripts. A machine learning engineer studied more than 100 hours of Whisper transcripts and found hallucinations in more than half of them. And one developer reported finding hallucinations in almost all of the 26,000 transcripts he created with Whisper.
An OpenAI spokesperson said the company is “continuously working to improve the accuracy of our models, including reducing hallucinations” and noted that its usage policies prohibit the use of Whisper “in certain decision-making contexts at high stakes.”
“We thank the researchers for sharing their findings,” they said.