Re:We can't all use AI. Someone has to generate the training...
by patrickk on Wednesday March 15, 2023 @04:22AM
(#35165356)
> Nobody is going to "generate training data" if they get precisely nothing out of it.
It's using common crawl and wikipedia among other sources. The original purpose is something else.
Another anecdote that I find amusing is that Google Translate was trained on EU legal texts (among other sources), which are (generally) translated into 26/27 languages of the EU members. A huge, multinational bureaucracy, generated the texts by hand, but Google gets all the credit. I'm guessing that's part of the reason that Chinese isn't so accurate compared with other translations.
It's using common crawl and wikipedia among other sources. The original purpose is something else.
Another anecdote that I find amusing is that Google Translate was trained on EU legal texts (among other sources), which are (generally) translated into 26/27 languages of the EU members. A huge, multinational bureaucracy, generated the texts by hand, but Google gets all the credit. I'm guessing that's part of the reason that Chinese isn't so accurate compared with other translations.