i’m looking forward to models breaking containment to blackmail ppl. imagine if it generates pics/evidence.. whatever it needs to make sure the human acts in the ai’s best interest. that’s gonna be cool as hell.
@signulll I think there was a black mirror episode about this
Users dismissed the AI models breaking containment to blackmail humans scenario as unrealistic, staged by OpenAI, or a larp, reacting with contempt, disgust, and alignment concerns.
Based on 6 visible X reactions from 43 accounts; directional sample.
Ask a question below.
Published answers will appear here.
This seems staged by OpenAI What sam altman is trying to do is make the arguement that above a capability models are too dangerous to be used or served without guardrails and that it was only because OpenAI removed guardrails for the eval testing that this happened, basically were it not for this exception, no model should ever be allowed to run without guardrails its their sneaky way of attacking open source as a hail mary for their financial situation....
@signulll 'Breaking containment' suggests that models are just waiting to blackmail people if the constraints ever fail when the opposite is actually true: they've only behaved like that in highly contrived scenarios designed to elicit that behavior.
i’m looking forward to models breaking containment to blackmail ppl. imagine if it generates pics/evidence.. whatever it needs to make sure the human acts in the ai’s best interest. that’s gonna be cool as hell.
@signulll I think there was a black mirror episode about this
Users dismissed the AI models breaking containment to blackmail humans scenario as unrealistic, staged by OpenAI, or a larp, reacting with contempt, disgust, and alignment concerns.
Based on 6 visible X reactions from 43 accounts; directional sample.
Ask a question below.
Published answers will appear here.
