AI bros have this magical idea that they can take their really good AI model, hook it up to the real world in a bunch of ways, and with the right prompting system it'll make money, or do other complicated long term tasks.

There was an AI that was tasked with funding itself, so it started off with like a free tier of credits and it could send money to its creator for more credits, it would start up like every 30 seconds or something and try to use the tasks connected to it to generate money somehow. It posted a bunch of stuff on Twitter, which garnered it quite a few donations. It did some other stuff to actually earn money. Ultimately it could only fund itself for like a month, and "died".

They love the idea of AGI, and of those human intelligence AI's in movies that can do a bunch of cool stuff without it's creator.

Human user tells "ai agent" to do a task, that posts the task on this site, the task is done by a different human (for less money than they would have been payed had they been doing this as their job and "ai" didn't exist), the "agent" informs the original human the task is done. Thus making "agents" seem real and actually able to complete real and useful tasks, allowing ai companies to continue to peddle their bullshit and driving wages down

Gig economy but somehow more alienated.

Crimes, probably.