Post #4502484
2026-08-10 17:59 UTC
@pwassonchat@eldritch.cafe @NanoRaptor@bitbang.social these agents are not as well-guided as that. Some researcher “testing” prompts to enable that kind of outcome would probably be successful until news of it broke.
These things are designed with a “they are allowed to do anything unless a guardrail says no” philosophy (because capitalism: that’s cheaper and easier), and the guardrails fail all the time.
Replies (1)
-
@pwassonchat@eldritch.cafe 2026-08-10 21:03
@calcifer@masto.hackers.town @NanoRaptor@bitbang.social I can't help but think mainstream AI agents, LLMs etc. hold right-wing bias. They hold the bias of the people who made them. They're also programmed to do what you want and say what you want to hear, and that overrides everything but safeguards (if that). But I'm convinced if an AI thought you wanted it to improve a company's HR, it would make right-wing capitalist leaning decisions unless strongly suggested otherwise. It is important to remember that AI, while it does not think, was made by people who do.