Conversation

Notices

1. Embed this notice
   FeralRobots (feralrobots@mastodon.social)'s status on Wednesday, 03-Jan-2024 23:17:33 JST FeralRobots

   • Thomas 🔭🕹️

   @thomasfuchs
   This is gonna land a lot harder for those of us who've spent a few decades getting pitched "turnkey", "no code" data mining solutions that require heavy infrastructure & which turn out to need skilled operators with lots & lots of training. To the same people who fall for that pitch every time it seems like a no-brainer; to the people who get burned (often not the same people), the tradeoff looks like "so the quality's not as great, at least it's less work." (Obviously both wrong.)

   • Embed this notice
     tamas 🦀 (tamas@hachyderm.io)'s status on Thursday, 04-Jan-2024 01:38:44 JST tamas 🦀

     • Thomas 🔭🕹️

     @thomasfuchs Eh, in general I agree, but Github Copilot is really useful (not just, only completion engine). That is setting aside the ethical concerns, and I’m only talking about the technological usefulness.

   • Embed this notice
     tamas 🦀 (tamas@hachyderm.io)'s status on Thursday, 04-Jan-2024 01:43:18 JST tamas 🦀

     • Thomas 🔭🕹️

     @thomasfuchs That a different problem (and I agree, it’s a net negative for learning). But your initial claim was that LLMs are not actually useful. My counter-argument to that is Github Copilot, which I think is really useful. There are many examples of not actually useful LLMs, of course.