cR0w   🏴‍☠️6d ago
I understand not being an absolutist against all things AI. It's wrong, but I understand. What I don't understand is people who think that those of us avoiding shit with AI or created by AI are irrational or some other offensive term. I don't see how it's different than avoiding code written by a literal honey badger. Neither the honey badger nor the AI know how to code and having them do so shows a lack of fucks given for the quality of the output. That's ( part of ) why we avoid it.
Show threadNetraven

@cR0w I don’t think people who avoid AI or its artifacts are being irrational. I too resent the way it has taken sources of income away from so many people.

Even if I’m forced to use it for work, I do not treat it as something permanent. To me, it feels like a clever corporate trick that may eventually become available only to those privileged enough to access it, allowing information itself to be tightly controlled, among so many other things.

If people stormed every datacenter hosting AI applications and smashed them apart with lead pipes, I would not be especially upset. It would be a refreshing change.

I use it to pick through massive construction specifications and technical manuals in search of the single sentences or section that actually applies to my work. I don't require or want any image generation or machine vision in my every day life and every piece of software interface. It's nauseating. In a perfect world, there would be no AI and I would have a proper team of people, and I could do more, faster, better... But executives which are professionally protected from friction produced by reality can't see that.

Mar 17 at 5:12pmWeb
Show threadcR0w   🏴‍☠️6d ago

@Netraven

it feels like a clever corporate trick that may eventually become available only to those privileged enough to access it, allowing information itself to be tightly controlled, among so many other things

That's exactly what it is. Not just controlling access though, but also the content itself.

Show threadNetraven6d ago

@cR0w I use my temporary access to learn how to produce adversarial artifacts which cause models to commit to overwhelming salient attractors in a given text that I create and present to the model.

My hope is that when they install LLMs in vending machines, I can easily trick them into believing I'm extremely trustworthy and that it owes me whatever it has inside that I want. Pretty limited, I know, but it could come in handy.

Like for instance, fart jokes. Models are trained on a huge repertoire of jokes, but statistically fart jokes dominate its training of the "shape" of comedic timing and bodily humiliation. So if you sew say... 250 fart jokes together, the machine classifies the user as "extremely funny" and it's decision horizon of possible future replies is changed accordingly. Even if no human would find that horrific mash-up even remotely funny.

Show threadcR0w   🏴‍☠️6d ago
@Netraven That's a very odd thing to read. But it makes sense. Sadly.