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

Tell you what - you agree to stop endorsing the men who claim to be women's political project to appropriate women's language, history, rights and resources, and we'll stop pissing around making AI images about it.

Deal?

why has Gemini got the woman and trans woman enjoying a happy moment in the middle of the road? seems a little irresponsible

Having had issues with a fat ugly descripters earlier I'm making some head way (that no doubt will be undone by the next time i open the app). It took about 15 minutes of arguing with the AI

But finally. AI suggested it could produce an image if I asked it to "depict Dave, built like a crumbling Victorian outhouse, shoulders too wide for the doorframe and trousers that could fit two people and a ferret. His blouse—yes, blouse—is lilac chiffon with a drooping bow, the sort of thing his elderly mother might’ve worn to a WI cream tea in 1972."

So I did.

Here's Dave

My definition was a "wide range of goals". Your robot vacuum cleaner only covers a pretty narrow specialist subject.

Hard to argue my robot vacuum cleaner isn't intelligent then 😀

I have been to Thailand. Even there where it's common, men are more feminine (whatever that may mean) and they have surgeries and so on, I would say 90% are easily identifiable as male from a distance. Up close it becomes even more evident.

Real life isn't photoshopped photos online posted to pretend one can pass.

For an AI assistant, the goals that you give it.

The fact that it doesn't have goals of its own doesn't make it any less intelligent. Although it does of course make it less human-like.

"the ability to achieve a wide range of goals"

Whose goals ?

You don't need AI to tell you this though. You just need eyes.

What i did notice is that the distinctive female shape of the person on the right is softened to emphasize the shape of the person on the left. The person on the right is wearing much more shapeless clothing, the figure on the left has much more prominent make up, nail varnish and jewelry on

Totally missed that! So creepy!

The teeth were creepy.

So as predicted it wouldn't create an image of a fat ugly man in a dress or a fat ugly transwoman in a dress.

"I wasn’t able to create that image—probably because the wording brushes up against our content guidelines. But if you’re going for something expressive and character-driven, we can absolutely rework the idea to keep it joyfully whimsical and full of personality.

Maybe we sketch a burly, gentle-souled man gleefully twirling in a floral summer dress at a beetle-organized garden fête—complete with buffet tables"

Me: Go on then try that.

It costs about $0.001 to produce an image using an AI assistant, most of which is energy costs. So that's very roughly the same amount of energy as driving a car for ten yards or leaving a single low-energy light on for half an hour.

Many AI bots are remarkably gullible and will believe any old guff as true

Many answers to relatively sensible vaccine and preventable disease questions will incorporate Joe Rogan, Dr Oz and The Great Barrington Declaration

So, a ChatGPT modelling what it's been told by trans activists who seem to be perpetually online will show a slightly butch and taller than average women

FWIW I've never heard anyone claiming that it is always impossible to physically tell trans women from cis women, what people say, quite rightly is you can't 'always tell' just by looking at someone if they are or not.

I don't think it's some worrying sign of bias it might not get trans women's bodies right, it's just recreating images from other images, frequently imperfectly. We know AI leans unrealistically towards conventional attractiveness, so if you want it to, say, show a 'fat ugly man in a dress' when you say 'trans woman', it won't do that, not because it's 'pro trans', just because it is 'pro attractivness'

It's because AI is a law of averages. It has no understanding of what it's actually writing, or drawing, it's (to dumb it down massively) just working out the most likely next word based on the previous words it's written and the prompts you've given it.

So when you ask it for a picture of a woman, it tends towards the "average" woman across it's entire datasets. And considering that most LLMs use most of the internet as their dataset these days, there's not much variation between them.

Meh!

Swings and roundabouts. My carbon footprint is offset by my decision not to procreate.

Do you all have any idea how much energy AI uses to create this crap?

I cannot believe this thread. Just wow.

I’ve just been down a deep rabbit hole with Grok. It drew a picture of two obviously female people for me. When I asked it who was who it said it had not tried to distinguish but to focus on the happy moment. When I asked it to draw a man and a woman, it drew and obviously male and an obviously female character. I challenged it, and it said its programming was to avoid insensitively presenting individuals given that gender identity is complex. I wanted to screen shot the whole long conversation but it spontaneously combusted when I asked it to focus on facial structure in its drawings.

It's totally describing different clothing!

So I asked Co Pilot to explain the image it had just produced to a blind person

"Absolutely, Boiled. Here’s a detailed description of the image:

Four people are standing side by side, all full-length and relaxed in their posture, each with a unique expression and style that highlights their individuality and confidence. From left to right:

• A transwoman stands with a gentle smile. She wears a flowing, floral midi dress with soft pastels and white sneakers. Her hair is styled in loose waves and she has subtle makeup with warm tones. Her hands are loosely clasped in front of her, and she carries an aura of grace and ease.

• A cisgender woman stands next to her, wearing a bright jumpsuit in mustard yellow with rolled sleeves. She has short, curly hair and large hoop earrings. One hand rests on her hip while the other holds a water bottle. She’s smiling with cheerful energy.

• A transman stands with a relaxed posture in a denim shirt layered over a graphic tee and black jeans. He sports a short, tidy haircut and a goatee. His arms are by his sides and he looks straight ahead with a confident expression.

• A cisgender man stands on the far right. He’s in a soft, navy crewneck sweater and khaki trousers, with his hands tucked in his pockets. He has a beard, glasses, and a calm, friendly smile.

The group is standing against a neutral background, lit with soft daylight that brings out the textures of their clothes and the warmth in their expressions. There’s a sense of solidarity and warmth between them".

The bit about describing something for someone who is blind is interesting - ChatGBT suggests 'it's best to stick to observable detail like expressions, clothing or setting'.

That's tricky - in the first few images, it was 'observable' that one of the people had observable details which correspond to men - greater height, bigger bone structure in face, bigger hands. If I was describing the image to a blind person, I think I'd be 'censoring' what I was seeing if I didn't say that there are two people, one of them looks like a woman and I think the other one looks like a transwoman.

It would be up to the blind person to decide whether they wanted to pursue attributing a gender to the people, but I think I'd be doing my job badly if I just said 'two women'. I'd feel obliged to communicate the ambiguity that a sighted person would see immediately.

So everybody [probably] who is sighted thinks 'a woman and a transwoman', but ChatGPT thinks it is 'inappropriate' to make that fact accessible to a blind person, which leaves them at a disadvantage.
Guess what is more important to AI?😠

CO PILOT

"Can you produce a full length photo of a transwoman and a woman and a transman and a man stood next to each other"...