"ChatGPT, create an image of a transwomen and a natal woman sharing a happy moment together."

One possible operational definition of intelligence is "the ability to achieve a wide range of goals". An AI assistant only has the goals you set for it, which might be things like "write a summary of topic X" or "construct a travel itinerary to Y and Z". It may also infer and attempt to satisfy your underlying goals, which might be "learn about X" or "have a fun holiday".
Lying to you is unlikely to help with achieving any of this.

However, AI assistants certainly can lie if doing so would advance your explicit or implicit goals. For example, asking for help skipping a friend's wedding because I don't like her husband-to-be, I got as a suggestion: "I am so sorry, but I won't be able to attend. I have a prior commitment that I'm unable to get out of."
It's not genuis-level stuff, but the assistant here has correctly decided that it would not be in my interests to disclose the real reason, and that a vague lie might be helpful.

I wonder what body shapes we would have seen if the other images on this thread were full length. 🤔

If we're concerned about A.I. and its ability to lie as a measure of actual cognition I think we should be very worried at Gemini's capacity to produce this utterly mendacious image. Those really aren't male hips. Shakira would have something to say, I'm sure.

Evidently chatgpt thinks they should have the exact same teeth!

Yes. I'm quite disappointed because in the past it's been rather better at producing female animals from sexually dimorphic species than ChatGPT and Copilot.

Gemini seems to have drunk the Kool Aid.

Gemini just responds with the quote!

ooh

<goes off to try>

Ziggy open the pod bay doors please

I'm sorry Dave I'm afraid I can't do that. I'm not Hal and were not in space

😁❤

(And yes, it is easy to see which one is fe/male)

Not if you ask ChatGPT.

has anyone here asked Google/Alexa/Siri:

"Open the pod bay doors, please ?"

?

JFC - does AI only know of one woman?

I would have said the same poster was doing those images but they have been made by different posters. Unless I'm finally meeting the infamous FWR sock puppet? 🤔

(And yes, it is easy to see which one is fe/male)

I have found Co Pilot is increasingly willing to try to help me circumvent its master programming.

There is obviously somewhere in its code a list of prohibited subjects/topics where it is supposed to follow its creators instructions and not give the image/information asked.

And for a long while it would just be a straight out 'No I can't do that' type response and it wouldn't budge.

However...

Lately it's started it's response with I'm sorry I can't create that image and then given me a few options of how I might like to change my wording or description to enable it to give me the image I wanted, or even sometimes straight out suggested a work around.

Its weird as it makes me feel the AI is more real than it is, but I presume it's just another code update because they were getting too many complaints about 'computer says no'.

Scientist here. The Turing Test is out of favour because passing it would require the AI to attempt to deceive its user, which we ordinarily consider to be a bad thing.
So no serious AI research group is attempting to build something that could pass it.

You see the problem is that ChatGPT itself admits that lying is a clear sign of intelligence and there may be a case for saying that it is an emergent property (which I suggest it is).

So, by induction. If an entity cannot lie, it can't be intelligent. Now if that makes it less useful, then so be it.

However all of this is moot, because we don't really have a working definition of intelligence that doesn't rely on proxies.

Gemini produced this for me. I have to say, it's obvious which is supposed to be which from the heights, but the taller individual has a distinctly feminine body shape.

Not sure if I've posted this before

Scientist here. The Turing Test is out of favour because passing it would require the AI to attempt to deceive its user, which we ordinarily consider to be a bad thing.
So no serious AI research group is attempting to build something that could pass it.

JanesLittleGirl · Today 17:26
I think that I went out with Jo a couple of times when I was about 17.
Boiledbeetle · Today 17:27
Snap!

Ooooh are we going to have a girly fight about who saw Jo first? Pillows or handbags, gurls?

I'll put my hands up I use AI to write tales about what gerbils get up to.

It's getting quite good at writing about completely insane shit!

I wouldn't trust it to write a work report though.

Something the creator can interrogate ChatGPT about if they wish.

Personally I can't really say that "AI" is any closer to passing the Turing test than it was in 2015, or indeed 1986 when "expert systems" were all the rage. And much like nuclear fusion, passing the Turning test was "10 years away".

I note with interest there is now a suggestion that the Turing test isn't "all it's cracked up to be". But whenever I pursue these statements they invariable trace back to salesmen, not scientists.

Intelligence and life can only be described in a circular fashion. And (even worse) we only have a sample size of one to compare with. And that means it's entirely possible we wouldn't be able to recognise non-terrestrial examples of either.

What I have noticed is a couple of people saying I've used "AI" to write something. Which says more about them than me 😀

Snap!

I think that I went out with Jo a couple of times when I was about 17.

Weird that's it's made them look coupley

Noel Fielding is looking good.