106
No, AI Is Not "Autonomously Hacking" with Cal Newport (podcasts.apple.com)
submitted 1 day ago by to c/fuck_ai@lemmy.world
8 comments fedilink hide all child comments
 

cross-posted from: https://lemmy.ca/post/70064423

In this week’s Better Offline, Ed talks with computer science professor and writer Cal Newport about what the recent AI hacking incidents actually mean, why we should stop referring to LLMs as AI, and what the post-bubble world looks like for LLMs.

all 8 comments

sorted by: hot top controversial new old
[–] 36 points 1 day ago* (last edited 1 day ago) (1 child)

The entire huggingface thing infuriates me because it was basically negligent incompetence that they try to use as marketing.

Edit: also it hallucinated and justified its hallucination. It's clear example of why agentic AI is a bad idea. It's not "oh this tech is soooo powerfull".

  • permalink
  • fedilink
  • source
  • hideshow 2 child comments
    •
  • [–] 15 points 1 day ago (1 child)

    Like what the fuck are the odds that, of the billion or so websites on the internet, your "rogue agent" "accidentally" hacks an AI repository?

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 7 points 1 day ago

    Yeah, that might have been the goal, but they can't state that publicly. Instead "we just didn't pay attention to what it was doing and let it spend thousands of dollars of compute that ended up in a crime". That's not the defense that they think it is.

  • permalink
  • fedilink
  • source
  • parent
    • [–] 26 points 1 day ago (1 child)

    Regardless of what actually happened, the company openly and proudly admitted to committing a crime. Company needs to be shut down, people making decisions need to be arrested, and we need to use civil asset forfeiture to take every fucking thing they have just like we do to everyday regular people who break the law.

  • permalink
  • fedilink
  • source
  • hideshow 2 child comments
    •
  • [–] 18 points 1 day ago

    Yeah, imagine if you let your gasoline powered lawnmover just run around in your yard just to see what would happen and you just left it for a couple of days in the anticipation that eventually your entire lawn was cut only to find that it also ran over some kid. The lesson from that wouldn't be that your lawnmover is just too good.

  • permalink
  • fedilink
  • source
  • parent
    • [–] 8 points 1 day ago

    you can't call something that can be stopped with a ctrl+c an AI gone rogue

  • permalink
  • fedilink
  • source
    •
    [–] 3 points 1 day ago* (1 child)

    Kinda off topic but I stopped listening to Better Offline on Spotify because it's by far the most ad saturated podcast there (edit: of those I listen to).

  • permalink
  • fedilink
  • source
  • hideshow 2 child comments
    •
  • [–] 3 points 1 day ago*

    I only listen to it on YouTube because there are no adbreaks (except the ads YouTube plays, but I never see them).

    On my podcast app, there are three long ad breaks in almost every episode. It's really annoying, but par for the course for podcasts made in the USA. Whenever I listen to something that was produced in that country, it's always part of a large network like iHeart and they love putting their ads in. Even if they don't have a sponsor, they promote themselves. Can't just let go of the ad space if no one booked any ads, capitalism at its finest.

    Ironically, I feel like this is something LLMs could help with. Transcribe the audio, figure out where the ad breaks are and skip them.

  • permalink
  • fedilink
  • source
  • parent