Glad to see that Zuck has the source for pirated books that I do, though clearly on a greater scale.
(Article originally at The Atlantic but paywalled. Link above is to Archive.is instead).
pull down to refresh
50 sats \ 0 replies \ @elvismercury 21 Mar
I really wonder what would happen if they could only train on old-timey books and materials out of copyright. That would be a fun LLM to talk to.
reply
0 sats \ 0 replies \ @byzantine 21 Mar
that explains why it is so good. the best models will have the most illicit and broad training dataset
reply