Conversation

Notices

1. Embed this notice
   Timnit Gebru (she/her). (timnitgebru@dair-community.social)'s status on Monday, 01-May-2023 18:36:07 JST Timnit Gebru (she/her).

   "Meta’s misstep—and its hubris—show once again that Big Tech has a blind spot about the severe limitations of large language models. There is a large body of research that highlights the flaws of this technology, including its tendencies to reproduce prejudice and assert falsehoods as facts."
   https://www.technologyreview.com/2022/11/18/1063487/meta-large-language-model-ai-only-survived-three-days-gpt-3-science/

   • clacke likes this.
   • Embed this notice
     CartyBoston (cartyboston@mastodon.roundpond.net)'s status on Monday, 01-May-2023 18:36:05 JST CartyBoston

     in reply to

     @timnitGebru Stephanie Arnett yay thanks

     clacke likes this.
   • Embed this notice
     CartyBoston (cartyboston@mastodon.roundpond.net)'s status on Monday, 01-May-2023 18:36:06 JST CartyBoston

     in reply to

     @timnitGebru shoutout please to the graphic artist at Technology Review who did the black bear cub in a spacesuit.

     clacke likes this.
   • Embed this notice
     dealyllama (dealyllama@masto.ai)'s status on Monday, 01-May-2023 18:37:49 JST dealyllama

     in reply to

     • Prof. Emily M. Bender(she/her)
     • mmitchell_ai

     @timnitGebru @mmitchell_ai @emilymbender Holy crap - thanks for the info and your work in this area. Tons to digest here if I can manage it :)

     clacke likes this.
   • Embed this notice
     Timnit Gebru (she/her). (timnitgebru@dair-community.social)'s status on Monday, 01-May-2023 18:37:50 JST Timnit Gebru (she/her).

     in reply to

     • Prof. Emily M. Bender(she/her)
     • mmitchell_ai
     • dealyllama

     @dealyllama our stochastic parrots paper which @mmitchell_ai and I got fired for writing with @emilymbender, her student Angelina and our coauthors who had to take their names of the paper Ben, Vinod and Mark. https://dl.acm.org/doi/10.1145/3442188.3445922

     clacke likes this.
   • Embed this notice
     dealyllama (dealyllama@masto.ai)'s status on Monday, 01-May-2023 18:37:51 JST dealyllama

     in reply to

     @timnitGebru any good resources / papers on how societal bias is reflected in trained models? Seems like a semi-frightening mirror to look in to

     clacke repeated this.
   • Embed this notice
     Joel Webber (j15r@mastodon.j15r.com)'s status on Monday, 01-May-2023 18:39:44 JST Joel Webber

     in reply to

     @timnitGebru It seems so unfortunate that, rather than using the opportunity provided by this work to explore the *limitations* of LLMs, the team chose to promote its use well beyond what it was capable of doing safely or reliably.

     It's "interesting" is no substitute for "useful and safe", and choosing to promote it for the latter undermines the former.

     clacke likes this.
   • Embed this notice
     Dave Wilburn :donor: (davemwilburn@infosec.exchange)'s status on Monday, 01-May-2023 18:40:13 JST Dave Wilburn :donor:

     in reply to

     @timnitGebru didn't we go through this before with SCIgen? Did they not realize that SCIgen was supposed to be a joke, not an instruction manual?

     clacke likes this.