Conversation

Notices

1. Embed this notice
   Scott Jenson (scottjenson@social.coop)'s status on Friday, 30-Aug-2024 08:29:08 JST Scott Jenson

   This "how many Rs in Strawberry" mistake with chatGPT is going around. I just tried it here:
   https://chatgpt.com/c/0b62bdd5-b33c-45d0-b0ed-f1c780d3533d
   There appear to be a fews camps trying to spin this:
   * a trivial 'gotcha'
   * a warning on the limits of LLMs

   There is likely a bit of truth in both. It clearly shows LLMs mimic language but not thought. It was incapable of responding to my very clear rebuttals.

   There are going to be profound uses of this tech, but we must see it, and it's limitations, clearly.

   • Embed this notice
     Eaton (eaton@phire.place)'s status on Friday, 30-Aug-2024 08:29:07 JST Eaton

     in reply to

     @scottjenson I think it’s obvious that any of these *individual* issues can be accounted for at the product pevel, so in that sense “number of R’s in strawberry” is a gotcha. but it is valuable as an example of the underlying process by whoch an LLM works, when not augmented by other types of rules/systems/etc. the “don’t be a hater, it will get better” folks seem to consistently miss that illustrative part…

   • Embed this notice
     Eaton (eaton@phire.place)'s status on Friday, 30-Aug-2024 08:31:36 JST Eaton

     in reply to

     @scottjenson i remember being a baby nerd around the time when expert systems were the absolute hotness, and everyone was convinced the march of innovation would blow past the “current limits.” Some problems yielded to that, others remained intractable or inefficient, and the “this technique will solve everything” fervor died out, leaving useful but less magical use cases…