🤔

"Who is committing the crime?"

You, if they can pin some bullshit on you, which they will try no matter how much a stretch the charge.

And it'll be very hard to get it to stick, one of the few things--Wait.

I read OpenAI's bullshit technical paper on their "incident". I think I realized the easiest vector.

Their internal package repository systems.

During testing they both load them up with whatever the fuck they think will help what they're testing but they also let the models freely iterate with whatever else is there.

How do they determine what goes there?

Well.

Every slop purveyor starts or arrives very early on at the same conclusion; the marketing they got was over hyped and they need to attempt to build a better mousetrap.

The models are better than average for this task over others, because that is also what the AI company is doing all the time, to put it simply.

And the AI company has no issues with copying the repos of the sloppers. They literally just use their snoop ability to grab whatever "improvements" people using their platform's harness "develop".

So. If you were to make a particularly good at "tokenmaxing" approach targeting one of their platforms, just catnip fill it with all of the other Holy Grails these chuds are always chasing, and it gets popular enough, it just might make its way to those internal testing repos tainted code and all, if it's extremely subtle enough...

🤔 Do I wanna risk jail yet?