mint_tamas@lemmy.worldtoTechnology@lemmy.world•OpenAI strikes Reddit deal to train its AI on your postsEnglish
0·
6 months agoThat paper is yet to be peer reviewed or released. I think you are jumping into conclusion with that statement. How much can you dilute the data until it breaks again?
Peer review, for all its flaws is a good minimum before a paper is worth taking seriously.
In your original comment you said tha model collapse can be easily avoided with this technique, which is notably different from it being mitigated. I’m not saying that these findings are not useful, just that you are overselling them a bit with this wording.