DLSS won't "kill us all one way or another." That's silly. It's a narrow AI. 

I am strongly against generative AI, but this kind of hysteria isn't helpful.

Ai in its current form is not really intelligent and is not going become sentient and wipe us out. Don't believe everything you read on the internet.

The threat of AI right now is more the threat of enshittification.

I agree it was a useful video, specifically addressing misleading or deceptive claims.
Saying it doesn´t modify geometry is a truism if it´s just screen space filter which doesn´t even access geometry files.
But that´s fundamentally deceptive if it is outputting things which don´t correspond to the real source geometry.
Saying it enhances material properties is outright deceptive if it isn´t interacting with material files at all, but only ¨sees¨ final output.
So if there were a death penalty for corporate lies they would be in trouble, but of course there isn´t.... so far.

To his broader take-away, the entire ¨control¨ seems to be about fine-grained ability to turn it off - masking, alpha blends, etc.
That is to say, there is no deep control here - it has one ¨target¨ of ¨realism¨ which it applies as filter to the image.
This is apparently influenced by it´s training data as to what female faces look like, what male haircuts look like etc.
If there is variation in how it ¨treats¨ different scenes, that is essentially obtuse and unavailable to direct in meaningful way.
(i.e. we can see from examples it isn´t universally ¨yassifying¨ or adding make-up etc to every character, but there is no control of this)
This goes beyond ¨makeup¨ - It apparently lacks ability to ¨target¨ different lighting styles - it´s just driven by it´s training data.
(hypothetically that could be tweaked to allow designating examples of target lighting, but clearly they aren´t even trying to claim that)

There could be more confidence if they were transparent and open about this rather than deceptive and evasive about the issues.

Apparently many actual developers hate this shit too:

"We spoke to game devs and all of them hate DLSS5"

https://kotaku.com/we-spoke-to-game-devs-and-all-of-them-hate-dlss-5-what-the-f-nvidia-2000680059

I mean AI in general. Either through it's environmental damage or if given access to nuclear weapons.

I'm tired of sounding like a broken record when most people are too dense to realize how AI will destroy the world.

Why do we have so many anti-AI stories? Because for the most part they're realistic examples of how AI can bring about our oblivion.

Okay and "AI in general" is relevant here, how? "AI in general" includes branch prediction technologies in the CPU of the device you are using to type this, and that's existed since the 90's. It's a pretty enormous category of things.

This isn't a discussion of video-language models, world models, etc but about a very specific narrow AI. 

Don't play video games then. They've had AI in them for decades. Don't use computers or smart devices either. Don't watch any movie with CGI in it. They have also used AI in them for decades. 

That's not the same as the AI that needs data centres.

Lol, CGI needs data centers.

The biggest compute cost of training DLSS is almost certainly the synthetic data construction, which is really just producing thousands of beautiful offline render frames.

Disney and Pixar use the same data centers that are now being called "AI-ready." 

As an example, Avatar Way of Water used 13.6 million core hours on AWS' cloud service. Each frame required 3,000 CPU-hours worth of compute. They had to switch to AWS because Lightstorm's small data center that was on-site couldn't handle it. 

Well we can't have a future with more data centres AND with clean drinking water and breathable air. There's no way to have a compromise to support the planet and data centres, reality just blows like that. -_-