Whisper by OpenAI Faces Scrutiny for Inaccurate Transcripts (Full Transcript)

Speaker 1: The popular speech-to-text bot that converts voice into text has been caught red-handed, dangerously fabricating parts of its transcripts. Sometimes it invents whole sentences with racist remarks, hostile language, and even imaginary drugs. I'm talking about Whisper by OpenAI, a speech recognition system trained on nearly 700,000 hours of audio in multiple languages. However, it's now under scrutiny for hallucinating content, casting doubt in its accuracy. One engineer said he found hallucinations in half of the 100 hours of recordings he tested, while another stated they showed up in every one of 26,000 scripts he reviewed. OpenAI, meanwhile, has widely acknowledged the need to address Whisper's hallucination issues. Make sure to subscribe for an in-depth look at the latest in AI tech and beyond.