Conversation

Notices

1. Embed this notice
   Charlie Stross (cstross@wandering.shop)'s status on Tuesday, 09-Apr-2024 17:48:32 JST Charlie Stross

   • emilygorcenski

   @emilygorcenski I see a third possibility (speculating outside my competence): what if Chomsky was onto something with his concept of a "deep grammar" underlying human languages, and LLMs have inadvertently exposed or replicated it.

   But language and linguistic competence alone are not actual intelligence or sentience—they're just how intelligence communicates.

   (Which is where the illusion arises.)

   People who've never seen a picture of a face see one for the first time and think it's a face.

   • Embed this notice
     pettter (pettter@mastodon.acc.umu.se)'s status on Tuesday, 09-Apr-2024 17:48:30 JST pettter

     in reply to

     • emilygorcenski

     @cstross @emilygorcenski are you thinking of universal grammar? I mean, it's not surprising that LLMs replicate strings that resemble human language, since that is literally what they are built to do.

   • Embed this notice
     Graydon (graydon@canada.masto.host)'s status on Tuesday, 09-Apr-2024 20:36:27 JST Graydon

     in reply to

     • emilygorcenski

     @cstross @emilygorcenski There is a shedload of wetware specifically to recognize faces. It's why we can see faces in clouds.

     The problem with Chomsky's grammar notions is that no one has been able to find it. There's a bunch of developmental stuff about sound order in infants and another bunch of developmental stuff leading up to theory of mind, but the organization of language doesn't seem to be especially constrained.

     I think the "psychic illusion" take is much closer.