As I said before, I've never up to this point read anything bad about AI regarding Horizon Zero Dawn Remastered, and actually have seen it was a really impressive remaster of an already great looking game.

Hence why at least up until now, I'm not really worried about it at all.

The other studios mentioned, Polyphony Digital and Naughty Dog, with GT7 the AI use has been praised and is quite honestly impressive with Sophy, and if I would doubt Naughty Dog when it comes to delivering quality, well, let's say there wouldn't be a single studio in the world I wouldn't doubt then, as IMO it is by far the peak of gaming dev alongside Rockstar North, their attention to details, for both studios, is so above anything else in the industry it's not even funny. Many games on PS5 still lack considerably in various areas compared to Uncharted 4, which is a PS4 game that is 10 years old this year.

Feel like I dragged things on too much of a tangent and in too much of a negative direction so I'm just gonna sit this one out from here on.
Apologies for the derail.

Yes but 57% of people don't regularly stab people with kitchen knives... That's why I say when does the bad outweigh the convenience. 

And AI is a shortcut, tricks to do things without understanding. I dropped out of university when they were just learning us 'tricks' to pass exams rather than explaining the why. What do we lose when people let AI do the work for them.

As for HZD, I agree with you the color changes change the mood of the game. The new shots do not look inviting to me. Vibrant colors with lots of green is what I like about Horizon. 

As for GT7 AI, it sucks tbh. Sophy 3.0 is a mess in the Power Pack, drives too awkwardly and still takes itself out on the straight at Le Mans. It has lots of odd behavior and still can't take corners fast. PD would have been better off to apply that focus on the penalty system.

I think you meant someone else about Horizon as I never mentioned color changes. Actually one of the two things that prevented me from giving Horizon Zero Dawn a 10 was the extremely weird random lip sync it original had, which would be called AI slop nowdays but vastly predates it. Forbidden West didn't suffer from it but if AI can help we avoid what we had with Zero Dawn and that process took far too long on Forbidden West (thus why they tried avoiding it originally with Zero Dawn), then it's actually a pretty good thing.

Also, I don't have the power pack for GT7. But a quick search suggests you are right on that. Definitely didn't hear many complains before. Which doesn't mean it can't be improved.

Ah I see I got it double wrong, it was sc mentioning the changes and they actually made it more colorful not less. Which matches my screenshots from the original, lot of browns. The remaster looks more like FW.

Finish coffee first then reply haha.

Sophy was always iffy, cheating abusing track limits, 'accidentally' hitting you. PD try to solve it by making it less aggressive and stay within the lines. But now it's too easy and slow in the corners. One step forward, 2 steps back. Same as PD handled the penalty system before giving up on that. The regular pro AI now gives me more of a challenge than Sophy :/

I largely agree with this take. Comparing up-scaling tech to actual replacement of people in a creative process is day and night. I use DLSS in almost every modern game I play, both due to the looks of it having improved dramatically and the fact that I have to if I'm to get decent frame-rates. 

The above sentiment is also the main reason for the backlash against DLSS5; it supplants actual creative work by actual people rather than simply enhance and upscale what was there. The showing also served as an indicator as to where the industry might be going, or at the very least where hardware manufacturers would like it to go. 

AI can have its place, but shouldn't dictate the ebb and flow of entire industries (as per you last sentence). Right now, the amount of AI bullshit in internet browsers (the recent Gemini scandal is likely only the tip of the iceberg) and on modern phones bothers me much more than that applied in video games in most instances.  

The amount of content is mindblowing. The quality of the content overall is crap. 

I didn't mean to imply a no true Scotsman fallacy. Any book that gets published whether it is physical or digital is real. In the pre-internet days you did have a lot of incomplete and outdated data. But most authors knew how to use information literacy skills to vet the majority of their information. Sources were at the very least quality and in context at sometime. These days a tiktoker can make a 30 second video with tons of lies and half-truths, based on sources that were always garbage. It goes viral and spreads around the world. Same goes for tons of youtube videos. And it's overwhelmingly that sort of poorly researched, lowest common denominator slop content that dominates public discourse and the internet as a whole. 

What I'm saying here is that while the internet does have a ton of information, the vast majority of it is useless or hard to find. Meanwhile a public library has mostly information that is useful and easy to find. 

On topic: I'm so confused as to how AI even works when it comes to drawing over assets to make a higher poly model. And I don't get how a higher poly model can make something look better. Doesn't it bog down the GPU? What is mega-geometry? What are all the lame buzzwords that nvidia throws around? Someone explain what they do to me like i'm a 6 year old. I don't get it.

On the topic of whether NS2 Editions “need to exist” or “are cash-grabs”: Frankly, regardless of how much of a “cash grab” it might be, I genuinely do not care lol im gonna pick up my copy of Kirby Forgotten Lands NS2 Edition and have some time playing the heck out of 4k 60fps. If Nintendo wants $20USD, then they’ve got it. Thank you for the product, I’m gonna enjoy myself now. Same goes with PS5/PS4 upgrades to previous gen software. I eat those up!

Mega geometry is actually not a neural rendering technique, even though it is part of their "RTX" SDK along with neural rendering techniques. It's a non-ML algorithm & group of data-structures that depends on clustering sub-meshes into groups of close together triangles. 

For a regular BVH pipeline you have: geometry output/triangles -> BLAS (BVH for mesh level) -> TLAS (BVH for scene level) 

Mega Geometry adds an additional step between the geometry output and the traditional BVH passes: geometry output/triangles -> CLAS (Cluster level BVH) -> Cluster BLAS (BVH for mesh level, but now clustered using the CLAS) -> TLAS (BVH for scene level.) This allows Nvidia to accelerate and/or improve ray tracing. 

In the RTX suite, the more deep-learning oriented features are neural shaders, neural materials, and RTX texture compression. 

Neural shaders are exactly what their name implies. They're very small neural networks that take in a few dozen inputs about an asset and the scene and provides output values which affect the properties of a scene, material, or group of materials. Traditional shaders are programs that do this through intentional code. Neural shaders do this by learning what traditionally is done, and replacing expensive (compute-wise) procedural parts of the shader. The value is that they can more efficiently do this for a similar quality output than a traditional shader by taking learned associative short-cuts. 

Neural materials are a specific type of neural shader that focuses on making material representations more efficient. Neural shaders can affect much more than material properties, but when applied to materials in a very organized fashion, you can get a more efficient representation of the material and therefore improve quality by utilizing the extra saved compute. 

Neural texture compression is Nvidia's improved compression method that uses neural networks to reduce texture sizes without as large a loss in quality as non-NN methods. 

All of these happen on the actual assets, or in other-words they're part of the pipeline. DLSS 5 on the other-hand, is a screen-space effect that happens after the image is rendered (but using some buffer data to assist.)