Perhaps the source training data will shift to transcripts or subtitles of podcasts, cable television, TV shows, and other sources that haven't (yet) been pollu

Perhaps the source training data will shift to transcripts or subtitles of podcasts, cable television, TV shows, and other sources that haven't (yet) been polluted. Combined with pre-LLM data sets.