• 4 Posts
  • 98 Comments
Joined 2 years ago
cake
Cake day: April 27th, 2024

help-circle


  • Sorry for being late to the party, but I had to give my two cents on AI detectors. (specifically GPTZero & co., watermarking is different) I think it’s a bad idea to recommend them. These companies are overpromising their accuracy and essentially selling false certainty. This will also not fade over time since the inaccuracy is baked into the tech.

    Currently the performance may be good for basic use cases, but recent papers seem to indicate you can still dazzle them by burning more tokens or explicitly prompting against detectors. But from what I understand about GenAI training* this is an eternal cat & mouse game, since the AI vendors don’t want their output to be detectable.

    what I understand about GenAI training

    A basic technique of training a generative model is training a discriminative model alongside it. The gen model tries to synthesize output and the discriminative model tries to distinguish it from actual training data. The models are then trained alongside each other, building better and better discriminator models to sus out better and better generative ones.

    If these detector companies had some kind of breakthrough (better techniques, better datasets, whatever) to sus out generated output, they would be bought up to improve GenAI training. More likely, they don’t. Maybe they can dial into some typical usecases for specific models and perform highly there, but that will fail to generalize for the edge cases.





  • This is bullshit and not how pro-Israel influence in US politics works. Kirk was broadly accepted by the republican mainstream and didn’t stray too far from the party line. In fact he was for a while harrassed by Groypers in part because he refused to adopt bald-faced anti-semitism into his rethoric. This theory might well have originated in those circles.





  • This rational animation channel seems to focus pretty non-stop on AI-risk crank talking points. Can’t tell by skimming if they’re engaging critically, but it seems unlikely.

    And now you’re apparently inspired to make a cutsiefied rendering of a slop machine running prompts, which is likened to a child being encouraged to explore.

    I don’t wanna shit on the art or anything, but it seems like the general AI psyop is getting to you a little.



  • I work with GitLab a lot. I think it’s better for organizations than using GitHub, but my SW Engineer Spidey sense gives me death spiral vibes.

    They’ve swallowed the AI Kool-Aid and are redesigning the UI to push their Chatbot on you (disabled by my org fortunately, but I still see it). They are feature creeping the fuck out of their already bloated product and are engineering features specifically to keep lots of usefull stuff behind the paid tiers, which are getting more expensive.

    It’s fine for institutional use if you have sane admins, but long-term I 100% bet on Forgejo.