What is with this AI Image Look?

[DEF]

Brian applies that aesthetic standard to women, not to men. By definition, sexist





Not objectification then? If you keep this up, I will run out of facepalms.

We are having this problem with AI, of course (sexism, not fruit-picking) consequent from biases in the selected corpus of training data and applied procedures. We've also seen that, absent effective training, chatbot's can quickly turn into Nazis. These are issues to resolve, not to take for granted or ignore.

Of course, by definition "In social philosophy, objectification is the act of treating a person as an object or a thing."
And Brian can be favoring men, women, whatever he wants.
If Brian was loving Men, he could of course have his idols there.
It's fantasy, and it's normal.
Please accept.

 

[Axo1989]

Please accept.

Objectification is tangential to sexual orientation, obviously.

Per my last paragraph, obviously no.

 

[DEF]

Per my last paragraph, obviously no.

OK. Have a nice week, though

 

[AdamG]

Objectification is tangential to sexual orientation, obviously.

Per my last paragraph, obviously no.

OK. Have a nice week, though

OK. You each get the last word above. Done, now move on please.

 

[DEF]

OK. You each get the last word above. Done, now move on please.

Ah, the mythical ‘last word’! Didn’t see it coming, but I’m here for the mystery and the fanfare!

 

[Tim Link]

Add some noise. I think helps a little.

 

[PierreV]

The concept breakthrough is that the NN is using tensor maths to emulate biological neurons. By throwing more hardware at it you are effectively emulating the brain power of more complex species. If we haven't already, we will soon reach model complexity that rivals the number of neurons in the human brain.

Well, once you have reached that level, multiply it by 7000, give or take a bit.
Oh, and run it on a 20W budget.

600 trillion synapses and Alzheimers disease

Late-onset Alzheimer’s disease is extremely difficult to model in the laboratory, one reason why no successful therapy for it exists. But scientists are using new capabilities to create more accurate, more useful research tools, bringing new hope for progress and better outcomes for patients.

www.jax.org

What is the difference between a complex NN that appears to reason and understand abstractions, and a human brain that you claim can?

Which type of NN? Assuming you refer to LLMs which are but one family of similar architecture, the main difference is that biological brains are able to update their world models on the fly.

LLMs in the current architecture can't do that. If they could GPT-4 would have self-improved by itself into GPT-5 and so on. That is ultimately what LLM developers hope will happen, but we aren't there just yet. In fact, extremely reputable people believe there are fundamental obstacles in the current LLM architecture (LeCunn, Chollet, Marcus and many others).

At this point, the deployed LLMs are mostly improved (GPT-4o, Claude, etc) by "hidden chain of thoughts" which is, very roughly, exploring the many possible branches at every step of a reasoning, evaluating them and advancing step-by-step. That leads, to some extent, to combinatorial explosion and hugely increased computing power demands.

The buzzwords in 2025 will be swarms of AI-Agents, some of them very simple, some of them full models implementing the above and that will lead to a further increase in computing/energy demands, which is why the leading players are jockeying for nuclear power plants right now.

Can you actually prove that your human brain is capable of reasoning and understanding abstractions, rather than simply providing the trained output for a given state and input parameters?

Yes, because at any moment, should you be willing to do so, you are able to _learn_ something new and readjust large parts of your wetware connexions.

One very simple example of this is the ARC challenge. There is a one million dollar prize for you to grab if you manage to show your model is equivalent to a human brain. Kids as young as 6 yo can do it once they have grasped the concept.

ARC Prize

ARC Prize is a $1,000,000+ nonprofit, public competition to beat and open source a solution to the ARC-AGI benchmark.

arcprize.org

 

[Karmacoma]

Seems like everywhere I turn, people are using AI generated images of people. How do I know? They all have this look:

Same bronzy, very warm look with that diffused background that doesn't look like a camera lens. Foreground is also aways sharp. An AI detection site says the above is from MidJourney. Maybe those always have the same look?

Regardless, it is getting way too obvious and tiring to look at. Above is from Totem website by the way. I routinely see thumbnails in youtube generated the same way.

Image AI is still lying in the ''uncanny valley''

 

[MattHooper]

It was interesting that when AI was starting out with videos and images, not that long ago, that we were told to “ look at the hands” because AI for some reason was having trouble replicating human hands.

I wouldn’t say this was “ predicted” by Michael Crichton’s Westworld, but it’s amusing that this was the tell for the Westworld robots:

 

[Tks]

Much of this stuff depends on how seriously you want to take it. Most of the stuff marketed by companies for regular consumers to use on whatever hardware they have, yields low res, low fidelity output. If you're running this stuff locally, then the requirements climb very quickly as does the per-requisite knowledge required to get things acceptable enough to pass blind tests between real and fake. Hands are the tallest tale sign (and up until recently, teeth/tongues and smiles that facial expressions of shock), but even that can be solved if you're willing to do some manual work so the AI understands what something should look like.

Most of that level of work requires either lots of time, or basically the best GPU hardware you can get your hands on (VRAM being the primary limiter in most of this stuff).

Most of the AI generated stuff falls apart in terms of convincing likeness, but with upscalers, you can get stuff looking really nice even without much (if any) manual work.

 

[gorb]

Some of it can be pretty weird lol

 

[Somafunk]

Most of the AI generated stuff falls apart in terms of convincing likeness, but with upscalers, you can get stuff looking really nice even without much (if any) manual work.

Apart from ears in this instance

 

[Tks]

Apart from ears in this instance

This was a case where manual inpainting work wasn't used, just basic upscaling. There's other errors that aren't as glaring, but the whole point is to show that the typical stuff you see (the painterly blur-fest "HDR"-like colorized style) isn't the only thing type of image generation possible.

Not my work btw, just a random image I picked up to show an example that deviates from the typical low res stuff. The problem when going higher res, it becomes easier to see the anatomical flaws and such (in the same way a high megapixel camera in photography shows how poor someone's makeup is), so you're fighting a double edged sword at the very least.

 

[007Shorty]

Um...