A dataset called "YouTube Subtitles" made of 173,536 YouTube video transcripts was used to train countless AI models since its publication in 2020. Professional creators are concerned while the AI companies, the creator of the dataset, and the largest distributor of it all deny culpability in breaking rules.
