Conversation

Notices

1. Embed this notice
   Paul Cantrell (inthehands@hachyderm.io)'s status on Wednesday, 29-Nov-2023 10:43:00 JST Paul Cantrell

   • malevolent dictator for life

   This from @astrid is a brilliant example of how GPT works, and how it doesn’t:
   https://fedi.astrid.tech/objects/2d4410a1-b827-4dd8-8422-c415ecb3929b

   The usual pattern of the Trolley Problem is, “Should you take action to kill a smaller number of people?” In that usual pattern, a utilitarian diverts the trolley.

   GPT has correctly identified the relevant pattern — and, having no comprehension, no model of truth, no sense of semantics other than word relationships, it repeats the pattern.

   Context? Spot on.

   Answer? Spot wrong.

   • Embed this notice
     Paul Cantrell (inthehands@hachyderm.io)'s status on Wednesday, 29-Nov-2023 10:51:54 JST Paul Cantrell

     in reply to

     • malevolent dictator for life
     • J Miller

     @JMMaok @astrid
     Yup. Or so •unimpressed• by it that we pose it trolley problems. Nothing entertains us techies quite like watching things malfunction.

     I’m more worried about gullible C-level taking it seriously and giving it actual trolley problems. Because that sort of thing is •already happening•.

     https://arstechnica.com/health/2023/11/ai-with-90-error-rate-forces-elderly-out-of-rehab-nursing-homes-suit-claims/

   • Embed this notice
     J Miller (jmmaok@mastodon.online)'s status on Wednesday, 29-Nov-2023 10:51:55 JST J Miller

     in reply to

     • malevolent dictator for life

     @inthehands @astrid

     This is a great example of what stresses me out about LLMs: techies being so impressed by « Grammarly for Python » that they start posing it trolley problems.