There was an early 2000s meme of unknown provenance where people would dispute the authenticity of images by saying “This looks shopped. I can tell from some of the pixels and from seeing quite a few shops in my time.” And I’m not saying we should revive this for the deepfake boom, but what I am saying is that as lengthy debates about image authenticity come back in style, the keys to the kingdom await whoever can express skepticism in the dumbest-sounding way
what do you mean early 2000s. I say this at least once a month. you trying to say something about me
been running chatgpt-equivalent chatbots on my computer and discovered that you can define/edit its personality, we accidentally created the worlds chillest dude
It does seem pretty clear to me that the claim “many members of the British royal family secretly ate sand as children” is not anything that could ever be falsified, so it really is an alright metaphor
(note: I really, really did not check any of this stuff, I just looked at this one thing which didn’t seem particularly dubious, did a little googling, and went “yeah sounds pretty likely” before hitting post)
AIs being able to convincingly pretend to know things isn’t a sign of intelligence. Come back when we have an AI that can convincingly pretend to be unaware of things that are common knowledge for the sake of a bit.
Oh man, that one is tough. That’s not a thing people can usually agree on from first-hand experience, and it is rarely encountered in the wild.
That said, if your AI can pass the Turing test in a convincing and non-limiting way, it should be able to improvise a convincingly reasonable, yet extremely incorrect answer to the question.
For instance, I can imagine an AI feeling bad about not knowing the question, and coming up with something like: “ummmmm … what do you mean by ‘grass’? Are you talking about 'turf’? Grass … uh, that shit is yellow-green, right? Right? What do you want from me?”
(If I’m lucky, my AI will say “grass = turf” and have an extremely strong hunch that “turf = green,” which will allow my AI to say “that shit is green, right? What do you want from me?” – though possibly while repeating the 'grass = turf’ part, because it doesn’t like to look smart, it just wants to impress you. Maybe it should be reclassified as a pretty smart AI?)
I am doing this from a hotel room, where I could probably get a good picture of grass, but cannot find a good picture of turf. Any volunteers to help me complete my proof?
What if you already knew that I believed that grass was really purple, and you thought I was dumb but didn’t want to hurt my feelings?
Here’s another possible answer:
“How would I know? If I had grass, I would have no time for this conversation – I would be out in it, studying it! I’m very curious to know how it looks like, it sounds like the grass in books that I’ve read.”
And then make an obvious guess – maybe say something like “It’s probably just green, like the grass in pictures” – because the “curious” part sounds real, but the guess doesn’t.
ALT
