Conversation

Notices

1. Embed this notice
   Felicitas Pojtinger 🌊 (pojntfx@mastodon.social)'s status on Friday, 09-Feb-2024 23:51:27 JST Felicitas Pojtinger 🌊

   Whenever I hear someone talk about how "crazy" it is that LLMs have "the things you are about to see might be misleading" warnings and that their output might be nonsense, wait until you hear about search engines

   • HistoPol (#HP) 🏴 🇺🇸 🏴 repeated this.
   • Embed this notice
     Felicitas Pojtinger 🌊 (pojntfx@mastodon.social)'s status on Friday, 09-Feb-2024 23:51:25 JST Felicitas Pojtinger 🌊

     in reply to

     If you believe that LLMs have no value for getting to learn a new topic because of hallucinations, you must also think the same thing about any kind of search engine, Wikipedia, the internet or really any kind of platform with user-generated content, that is to say literally any form of media without review

   • Embed this notice
     HistoPol (#HP) 🏴 🇺🇸 🏴 (histopol@mastodon.social)'s status on Friday, 09-Feb-2024 23:51:25 JST HistoPol (#HP) 🏴 🇺🇸 🏴

     in reply to

     • Cory Doctorow

     @pojntfx

     #LLMs #AI #Search

     Hm, not quite IMO.
     Apart from legal issues (copyright infringements), in the GOOD old Internet days, there were so many ways to evaluevaluate your source, a method all academics learn at college, even in pre-Internet times.

     With an amalgamate of sources, a guesswork of probabilities, and a method that would get even a bachelor student disqualified--no, or fishy citations--we are at the ultimate endgame of #enshittification:

     https://libre.video/videos/watch/fb9554dc-1d02-4287-88ef-a897e488e80c

     @pluralistic

   • Embed this notice
     Felicitas Pojtinger 🌊 (pojntfx@mastodon.social)'s status on Saturday, 10-Feb-2024 00:14:14 JST Felicitas Pojtinger 🌊

     in reply to

     • Cory Doctorow
     • HistoPol (#HP) 🏴 🇺🇸 🏴

     @HistoPol @pluralistic IMHO there should be no copyright, the fact that LLMs don't care about it is beautiful and should be encouraged. I want LLMs to be trained on every single document that Springer keeps away from the public. Same with every paywalled article.

     As for checking sources - well yes, obviously. The same applies to search engines and everything else.