Conversation

Notices

1. Embed this notice
   Carl T. Bergstrom (ct_bergstrom@fediscience.org)'s status on Saturday, 15-Apr-2023 18:59:39 JST Carl T. Bergstrom

   I can't believe it's production technology, part II.

   So for some reason it took me forever to get access to Google's #Bard #LLM. But I finally got it tonight. It does a decent job of copying answers out of the wikipedia.

   But based on playing around for even a few minutes, I'm definitely not impressed.

   See below. The largest Fibonacci number less than a million is 701,408,733—and incidentally, 987 is the 17th number in the sequence, not the 28th.

   • Embed this notice
     awooo :blobfoxcheck: ? ⎇ (awooo@pawb.fun)'s status on Saturday, 15-Apr-2023 18:59:38 JST awooo :blobfoxcheck: ? ⎇

     in reply to

     @ct_bergstrom I was in for quite the shock too, I've heard about the google engineer who got sucked into believing LaMDA was sentient, so I expected this to be at least good enough to hold a cohesive conversation, but nope.

     It's so dumb that it was difficult for me to extract the prompt after injecting it, because it just couldn't follow my instructions precisely unlike Bing, I was only able to get a vague idea of it and extract the likely format of search snippets, a list of 5 short results.

     clacke likes this.
   • Embed this notice
     M. J. Fromberger (creachadair@mastodon.social)'s status on Saturday, 15-Apr-2023 18:59:52 JST M. J. Fromberger

     in reply to

     @ct_bergstrom Even without LLM assistance, it's an open secret that Wikipedia is disproportionately responsible for keeping Google search relevancy within reasonable bounds. Of course, search evaluation is already kind of a kayfabe regardless—but the language models add even more spice.

     I am both fascinated and terrified to see how Conway's Law will play out around these clay models masquerading as products.

     clacke likes this.
   • Embed this notice
     Gregor Gross (gregorgross@kolektiva.social)'s status on Saturday, 15-Apr-2023 19:00:16 JST Gregor Gross

     in reply to

     @ct_bergstrom Someone wrote it's a system that does good most of the time, but makes dramatic errors when wrong (like with autonomous driving, this is catastrophic). But because most people simply believe what is in text on a screen, no one ever bothers to check the answers. So no one else would count the Fibonaccis to check which place 987 lands. Chat-GPT seemingly works for many of us, because so many of us are already stupid in front of screens.

     clacke likes this.
   • Embed this notice
     Phil Picton ? (philpicton@hachyderm.io)'s status on Saturday, 15-Apr-2023 19:00:33 JST Phil Picton ?

     in reply to

     @ct_bergstrom 701,408,733 > 1,000,000 are you an llm? ?

     clacke likes this.