Conversation

Notices

1. Embed this notice
   Scott Jenson (scottjenson@social.coop)'s status on Wednesday, 20-Mar-2024 03:05:53 JST Scott Jenson

   Pardon me if I'm not freaking out about "AI" like everyone else. I worked in AI in the 1980s, at Symbian when the first rule-based expert systems were built. The exact same types predictions and lazy extrapolations were given. If you don't know your history, you really are doomed to repeat it.
   1/3

   • Tim Chambers repeated this.
   • Embed this notice
     Scott Jenson (scottjenson@social.coop)'s status on Wednesday, 20-Mar-2024 03:05:49 JST Scott Jenson

     in reply to

     Every single "design tool" I've seen with LLMs right now looks good at VERY SPECIFIC things. Push it just a tiny bit outside of it's bounds and it looses the plot. Every single time. This is still valuable, sure, but only at the low level tedious work.

     I'm not a luddite! These tools are going to be capable, but mostly at taking routine tasks (e.g. layout in a known space) and making it super fast.
     2/3

   • Embed this notice
     Tim Chambers (tchambers@indieweb.social)'s status on Wednesday, 20-Mar-2024 03:06:51 JST Tim Chambers

     in reply to

     @scottjenson I continue to be surprised at how high the hallucination rate is. Makes it useless for the core task of research as you have to double check every apparant fact and link. Amazed at how many broken or made up URL's it gives for instance as sources.

   • Embed this notice
     Kristoffer Lawson (setok@attractive.space)'s status on Wednesday, 20-Mar-2024 03:15:42 JST Kristoffer Lawson

     in reply to

     • Tim Chambers

     @tchambers @scottjenson one should trust #AI #LLM facts just as much as you’d trust a drunk guy in a bar. For the same reasons. It may raise some interesting tidbit, but you absolutely have to check it if you plan to use it for anything. I actually think the use of LLMs for knowledge sources is a misguided use of the tech.

   • Embed this notice
     Tim Chambers (tchambers@indieweb.social)'s status on Wednesday, 20-Mar-2024 03:16:54 JST Tim Chambers

     in reply to

     @scottjenson Good point: not ALL tech gets better…sometimes you just hit a wall.

   • Embed this notice
     Scott Jenson (scottjenson@social.coop)'s status on Wednesday, 20-Mar-2024 03:16:55 JST Scott Jenson

     in reply to

     • Tim Chambers

     @tchambers "But it'll get better!" just isn't an answer. It *may* get better but I'd like these people to think more like an engineer and not an SF writer. BUILD SOMETHING! It will most likely fail and that failure tells you a tremendous amount about how deep the tech actually goes.