Conversation

Notices

1. whiteline (whiteline@shitposter.club)'s status on Wednesday, 18-Oct-2023 05:14:20 JST whiteline

   • LS
   • バツ子（甥っ子はもう一匹
   @ageha @lain
   
   hofstadter thinks we're facing an "ai tsunami" and that we'll be irrelevant pretty soon
   • LS (lain@lain.com)'s status on Wednesday, 18-Oct-2023 05:14:19 JST LS

     in reply to

     • バツ子（甥っ子はもう一匹
     @whiteline @ageha well, he was wrong about the whole "strange loop" thing, at least there's no recursion in transformer-based llms
   • LS (lain@lain.com)'s status on Wednesday, 18-Oct-2023 05:20:20 JST LS

     in reply to

     • バツ子（甥っ子はもう一匹
     @whiteline @ageha I don't think he's dumb, the more-or-less sudden breakthrough of transformer and diffusion architectures really caught everyone off guard. Gpt-2 was nearly useless. Gpt-3 was already usable for a lot. Now I can run an llm locally on my computer that can code, translate, write poems, hold a discussion. We'll see, maybe it's hitting diminishing returns soon, but jus the progress in the last year has been incredible.
   • whiteline (whiteline@shitposter.club)'s status on Wednesday, 18-Oct-2023 05:20:21 JST whiteline

     in reply to

     • LS
     • バツ子（甥っ子はもう一匹
     @lain @ageha
     
     he admitted in an interview i watched that he was astounded that this "intelligence" could arise from purely feed-forward networks