Conversation

Notices

1. Csepp 🌢 (csepp@merveilles.town)'s status on Wednesday, 15-Mar-2023 08:24:03 JST Csepp 🌢

   @suricrasia Re: Coding Machines: I'm kind of looking forward to having AI that can write jailbreaks for all the old devices lying around.

   As for jobs, AI might be able to spit code out, but it will still need handholding for QA. But yeah, lots of menial programming jobs will likely be lost.

   But on the bright(?) side, we might have a huge energy crisis that makes all these power hungry model trainings cost prohibitive.

   • Csepp 🌢 (csepp@merveilles.town)'s status on Wednesday, 15-Mar-2023 08:24:00 JST Csepp 🌢

     in reply to

     @suricrasia A somewhat more esoteric application that I'm really curious how it'll develop is automatic theorem proving. This is kind of a subset of the QA thing I guess, but with different tools.
     There isn't nearly as much training data for developing Coq tactics and proof scripts as there is for writing JavaScript or Python, but unlike those languages, you'd be certain that the generated code (however unreadable) fits the spec. (barring the AI finding holes in the underlying logic)

     Adrian Cochrane repeated this.
   • Csepp 🌢 (csepp@merveilles.town)'s status on Wednesday, 15-Mar-2023 08:24:01 JST Csepp 🌢

     in reply to

     @suricrasia ...now that I write it down, a Red Queen's Race co-evolution scenario seems most likely.

     There was an interesting episode of uuh some podcast I'm too lazy to look up right now where the guest talked about how US defense would benefit from closing all security holes, because that would deny other states access to the cyber battleground and since the US has major advantages in industry and meatspace military, they would come out on top.

     Adrian Cochrane repeated this.
   • Csepp 🌢 (csepp@merveilles.town)'s status on Wednesday, 15-Mar-2023 08:24:02 JST Csepp 🌢

     in reply to

     @suricrasia The dark side of the Coding Machines scenario is that whichever nation state puts the most resources behind AI viruses will be able to fuck things up big time.
     New viruses might not even have to bother with precise targeting or prolonged avoidance of detection when a new virus can be generated in a few days.

     Buuut then again maybe this will lead to security holes being patched quicker??? Who knows.

   • David Mankins (lain_7@tldr.nettime.org)'s status on Wednesday, 15-Mar-2023 09:00:28 JST David Mankins

     in reply to

     • Adrian Cochrane

     @csepp @suricrasia @alcinnz I’ve heard that the Common Lisp these LLMs produce is terrible (no doubt due to a poverty of examples), so I expect Coq to be incorrect both syntactically, and logically. Maybe if one has a substantial body of Coq to train it on?