Opinion AI design discussion

[klier]

Math says so -

A RTX 4090 GPU sustained load is ~400 watts. So one of them running SDXL continuously for a day would consume .4kwh * 24 = 9.6 kwh/day.

An avg human adult including lifestyle consumption in the developed world consumes ~90 kwh/day. Source

So a GPU consumes ~10x < energy than the avg adult human.

And that is not even counting the non-productive years of a human life. Yes, humans are terribly inefficient machines if you reduce them to their mere physical output. But fortunately, we are so much much more than that.

Exactly. But these LLMs clearly do not use just one GPU. And the datacenters also don't work with off the shelf consumer graphics cards. Those are the H100 etc cards.

 

[Design FIRST]

The bad side is, now even mediocre/bad ideas can now become reality. So now everyone and their gold fish is going to design a new Ferrari! And given normal/Gaussian distribution of any skill (designing Ferrari in this case) in general population - there will be an overwhelming amount of mediocre crap (aka AI slop) drowning out any good ones!

LOL-a one sentence long prompt generated this:

 

[Matski]

The ability to produce something that looks like it was done by a professional does not on it's own make something art. Similarly, just because you can sketch, it doesn't make you a designer. AI slop is often just slop because of the minimal thought that goes into it by the user, and the minimal effort taken to refine it (or the limitations of whatever system is being used).

None of that makes the use of generative AI bad, but in my view, it does justify a rethink in how such output is critiqued.

 

[Design FIRST]

The ability to produce something that looks like it was done by a professional does not on it's own make something art.

Yes, I would be inclined to agree. Perhaps the only genuine "skill" involved regarding AI is the ability to envision a concept in one's head based on something already existing and clearly articulate what the result should be. Usually-but not always, the more precise the articulation, the more satisfying the result.

 

[martinbo]

I'll hold off on the debate around the drawbacks and merits of AI in the creative process for now.

What I will say is this: as a staff member who's watched this forum evolve over the years, I can't help but feel that certain types of contribution are detrimental to discussion and debate. This detriment is perceived by many - but not all - as a reduction of "quality" in this forum.

For me, it's all about context and the forum is very nicely laid out to provide members with logical placeholders (read - for context) in which to create threads and posts within threads. Personally, it's irritating to be alerted to a new post in a "Car Maker XYZ - what's next" thread only to see that it's nothing more than AI-generated imaginings with no basis for what is actually next for said manufacturer. This AI-generated slop is a focus robber away from the topic at hand i.e. what is genuinely next for that manufacturer.

The same sentiment stands for a model-specific thread and some pointless, dreamed-up image of what maybe, if-you-imagine-hard-enough, could be gets posted with tenuous relevance.

If you're so hell-bent on posting imaginary AI-generated images of what could be then simply create an AI-titled thread in the appropriate section and do it there.

 

[Matski]

Yes, I would be inclined to agree. Perhaps the only genuine "skill" involved regarding AI is the ability to envision a concept in one's head based on something already existing and clearly articulate what the result should be. Usually-but not always, the more precise the articulation, the more satisfying the result.

I think prompt engineering can genuinely be regarded as a skill in itself, just a very different one that hasn't traditionally been considered part of the process. I suppose in some respects it's more analogous to being a film director. A good director doesn't have to be a good writer, cinematographer or composer, but they do need to be good at communicating the vision in their head to the people that do have the skills to execute it.

.. but again, a lot of AI isn't being well-directed by the user, and it often lacks any solid vision to begin with. Most of what's spread on social media is a lazily directed effort based on a weak vision, and it can't gain any respect for technical proficiency in the medium it portrays, because the computer did that bit.

 

[Rolf]

Exactly. But these LLMs clearly do not use just one GPU. And the datacenters also don't work with off the shelf consumer graphics cards. Those are the H100 etc cards.

AI Data Centers are massive and use an enormous amount of electricity for their GPU banks and water for cooling them.
This is a highly interesting and informative video.

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies

 

[Sunny]

Exactly. But these LLMs clearly do not use just one GPU. And the datacenters also don't work with off the shelf consumer graphics cards. Those are the H100 etc cards.

Using H100 tilts the balance even more in favor of AI. They consume less (~350W) and can do 3x TFlops (330 tflops vs 1000 tflops) for FP16 operations (what SDLX/diffusion uses).

 

[Matski]

This is a highly interesting and informative video.

Colossus, named after the super computer that enslaved humanity in the Forbin Project. The Sci-Fi genre's "voice of world control" as presented in the real world by the world's richest fascist f#ckstain.

 

[klier]

Using H100 tilts the balance even more in favor of AI. They consume less (~350W) and can do 3x TFlops (330 tflops vs 1000 tflops) for FP16 operations (what SDLX/diffusion uses).

Yeah whatever the tech behind it may be, we agree.
And in not too long all these AI data centers will be powered by their own nuclear reactor. Energy and emissions solved in one sweep.

 

[Matski]

And in not too long all these AI data centers will be powered by their own nuclear reactor. Energy and emissions solved in one sweep.

I'm glad the data centres will have guaranteed cheap energy, this will really help those struggling Billionaires maximise their profits.

 

[Rolf]

The other major problem with data centers is the large amount of water they have to use for their cooling systems. This has become a major issue in some areas where the rivers are the main source of drinking water for communities and towns. In my opinion the best places to build data centers are near coastal areas. They can build distillation plants inside the data center site location and use sea water as the source. The distillation plant will create fresh water from the sea water to use for the cooling. They can use wave energy converters and install them under the water and use wave energy to generate the electricity.

 

[KiwiRob]

AI Data Centers are massive and use an enormous amount of electricity for their GPU banks and water for cooling them.
This is a highly interesting and informative video.

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies

You need to watch some of the youtube videos from the people in the community beside this facility, it's ruined there lives, the noise is 24/7, power bills have increased dramatically to subsidise the electricity used by this facility, drinking water has been contaminated and property values tanked.

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies

 

[Design FIRST]

The application of AI generated design may be "more acceptable" when utilized for mass-volume "bread and butter" sorts of vehicles.