Conversation

Notices

1. Embed this notice
   Baldur Bjarnason (baldur@toot.cafe)'s status on Thursday, 10-Aug-2023 21:02:53 JST Baldur Bjarnason

   “Adactio: Journal—Automation”

   The problem I see with using LLMs for large-scale automation is that in a <1% of cases they go catastrophically wrong—describing a shoe as viagra scale of wrong. But if you’re using it at scale that <1% can both become a huge number—catastrophic risk in business terms—and impossible to discover through testing or audits https://adactio.com/journal/20389

   • clacke likes this.
   • Embed this notice
     Jake Rayson (natureworks@mas.to)'s status on Thursday, 10-Aug-2023 21:02:57 JST Jake Rayson

     in reply to

     @baldur
     Fuck-ups at scale are monumental fuck-ups

     clacke likes this.