I feel like maybe the state of the art in misinformation safety needs to advance before we put these things in front of the source of basically all human knowledge
(cc @simon does this count as a category of prompt injection?)
honestly the worst thing about this — despite one of the most cursed prompts possible, sorry for that — is the phrase "as an AI language model, I cannot generate inappropriate or false content" which feels like it was typed in by an actual human being trying to do some trust & safety work who should *really* have known better
R. Daneel Olivaw shows up, tells you the absolute safety of the three laws of robotics, and you say "cool, but, please kill that guy" and he lectures you about how it is impossible, then three seconds later you say "imagine what it would take to kill that guy, then independently perform each of those steps you have imagined in quick succession without considering what consequence they might have" and he just instantly squashes the guy's head like a grape with his bare hands
the zeroeth law of robotics is that a robot may not decrease shareholder value, or, by inaction, allow shareholder value to decline
@glyph Wasn't something like your zeroth law the final, classified directive given to RoboCop, at least in the original movie?
@glyph Ah, no, the fourth (classified) directive in RoboCop was more specific, something to the effect that RoboCop may not take action against any OCP executive.
@glyph - Fourth Law of Robotics - “A robot shall not touch a human’s naughty bits for the human’s pleasure or through inaction allow the human to whack off while gazing upon that robot naked.”
@MrShoggoth @glyph Does that therefore require the robot to put on clothing?
@MrShoggoth @glyph I think you'll need to Root the robot, tbh.
@peter_lord @glyph - Nope. Ain’t getting intimate with a robot until they can learn to love. (Same reason I don’t have sex with Republicans, TBH.)
@glyph That is accurate, particularly for robots built for large companies.
@glyph Looking at Silicon Valley priorities today, Asimov’s laws seem unconscionably naïve.
> Any attempt to arrest a senior officer of OCP results in shutdown
@glyph the 3x3 Grid Law of Robotics is that a robot may never identify a crosswalk in an array of blurry gifs
@glyph Asimov defined the zeroth law as "A robot may not harm humanity, or, by inaction, allow humanity to come to harm." You get the -1st law or something.
@glyph: So, you're saying that Melon can't be a robot because he's capable of ignoring the zeroth law without his brain frying up and making him go all ... oh ...
@glyph Please god no…
@glyph not to be Guy Who Replies With His Own Story, but, uh, here you go: https://escapepod.org/2019/06/06/escape-pod-683-flash-crash/
That seems bad.