>[0] For starters, most YouTube videos aren't manually captioned, so you're feeding GPT the output of Google's autocaptioning model, so it's going to start lear

>[0] For starters, most YouTube videos aren't manually captioned, so you're feeding GPT the output of Google's autocaptioning model, so it's going to start learning artifacts of what that model can't process. Whisper models are better than anything google has. In fact the higher quality whisper models are better than humans when it comes to transcribing text with punctuation.