[hattmall]

I mean there's no reason that the AI would need to act maliciously, thought it would be possible. I was thinking more along the lines of pre-pay for a year of server and tell the AI it has to start a business or something to pay its own bills.

[Nevermark]

If those prompt limits were reliable borders it would be very cool.

But that would mean the alignment problem was solved. When survival depends on watering down ethics, and there is a long slippery slope of ethical wells available to plum, depending on prompts for safety sounds risky.

Is that malicious? I don't think that would be considered malicious. We don't consider it immoral for starving people to steal food.

But, I am fascinated by the idea too! I just think it is a terrible idea (despite being almost certain to happen).