▲ | everforward 2 days ago | |
No, I don’t think it applies here. The semantics and speech patterns were generated by a human, Whisper just transcribed them. There is some risk that Whisper transcribed inaccurately, but that’s less model collapse and more “the dataset is bad”. |