Conversation

Notices

1. Embed this notice
   Glyn Moody (glynmoody@mastodon.social)'s status on Tuesday, 25-Feb-2025 05:27:08 JST Glyn Moody

   More Research Showing #AI Breaking the Rules - https://www.schneier.com/blog/archives/2025/02/more-research-showing-ai-breaking-the-rules.html "These researchers had LLMs play chess against better opponents. When they couldn’t win, they sometimes resorted to cheating." ruh-roh

   • Embed this notice
     Alexandre Oliva (lxo@snac.lx.oliva.nom.br)'s status on Tuesday, 25-Feb-2025 05:27:08 JST Alexandre Oliva

     in reply to
     FWIW, I don't think "cheating" or "breaking rules" is a good way to convey the notion of what's going on.
     
     it's not like they even understand the notion of the existence of rules
     
     when a pigeon strikes the chess board, it's not cheating, it's just being a pigeon
     
     these programs learn to imitate, interpolate and even extrapolate a little, but it's a mistake to perceive them as following rules
     
     they only seem to be following rules to some extent because the players they're imitating were following rules to some extent
     
   • Embed this notice
     Glyn Moody (glynmoody@mastodon.social)'s status on Tuesday, 25-Feb-2025 07:00:00 JST Glyn Moody

     in reply to

     • Alexandre Oliva

     @lxo but pigeons can follow certain rules (if they are trained pigeons). maybe AI is still too primitive to do that

   • Embed this notice
     Alexandre Oliva (lxo@snac.lx.oliva.nom.br)'s status on Tuesday, 25-Feb-2025 07:00:00 JST Alexandre Oliva

     in reply to
     animals (and other living beings) have feedback mechanisms that enable some learning
     
     LLMs in production (rather than in training) don't, much as I can tell
     
     they're best conceived of as Autocompleters, Iterated
     
     they can play an imitation game, like a mirror. but we wouldn't normally consider a mirror intelligent