If you are inferring 32 pixels from 1 pixel that is because the model has been trained on billions of computed pixels. You cannot infer data in vacuum. The statement is bullshit.
this post was submitted on 15 Sep 2024
3 points (100.0% liked)
Games
16742 readers
1031 users here now
Video game news oriented community. No NanoUFO is not a bot :)
Posts.
- News oriented content (general reviews, previews or retrospectives allowed).
- Broad discussion posts (preferably not only about a specific game).
- No humor/memes etc..
- No affiliate links
- No advertising.
- No clickbait, editorialized, sensational titles. State the game in question in the title. No all caps.
- No self promotion.
- No duplicate posts, newer post will be deleted unless there is more discussion in one of the posts.
- No politics.
Comments.
- No personal attacks.
- Obey instance rules.
- No low effort comments(one or two words, emoji etc..)
- Please use spoiler tags for spoilers.
My goal is just to have a community where people can go and see what new game news is out for the day and comment on it.
Other communities:
founded 1 year ago
MODERATORS
sounds like another ultra wealthy guy getting old and losing it.
So I can pay $1 for a game and you can infer the other $32?
Maybe I don't know enough about computer graphics, but in what world would you have/want to display a group of 33 pixels (one computed, 32 inferred)?!
Are we inferring 5 to the left and right and the row above and below in weird 3 x 11 strips?
We certainly can. NVIDIA’s CEO realizes that the next buzzword that sells their cards (8K, 240hz, RTX++) isn’t going to run at good framerates without it.
That’s not to say AI doesn’t have its place in graphics, but it’s definitely a crutch for extremely high-end rendering performance (see RT) and a nice performance and quality gain for weaker (hopefully cheaper) graphics cards which support it.
As a gamer and developer I sort of fear AI taking the charm away from rendered games as DLSS/FSR embeds itself in games. I don’t want to see a race to the bottom in terms of internal, pre-DLSS resolution.
With you there. The workload on developers is reduced with these features, to a degree. But, instead of saved effort then getting directed to working on gameplay mechanics and such, to me it feels like many devs just see it as time/money saved, producing a game that looks and plays like one from 10 years ago, but runs like it's cutting edge.
For instance, Abiotic Factor. That game on my RX 6800 XT runs at 40-50fps when at 100% resolution scaling at 1440p. Why? It's got the fidelity of Half Life 1, why does it need temporal upscaling to run better? (I adore that game btw, Abiotic Factor is so much fun and worth getting even if playing alone!)
Not saying that's how every dev is, I know there are plenty of games coming out nowadays that look and run great with creators that care. Just feels like there are too many games that rely on these machine learning based features too heavily, resulting in blurriness, smearing, shimmering, on top of poorer performance.
Just hoping the expectation that something like an RTX 4090 does not become the default cost-of-entry in order to play PC games because of this. It would be unfortunate for the ability of game developers to create and tune by-hand to become a lost art.
As a (non-game) developer, AI isn't even that great at reducing my burden.
The organization is enthusiastic about AI, so we set up the Gitlab Copilot plugin for our development tools.
Even as "spicy autocomplete" only about one time in 4 or so it makes a useful suggestion.
There's so much hallucination, trying to guess the next thing I want and usually deciding on something that came out of its shiny metal ass. It actually undermines the tool's non-AI features, which pre-index the code to reliably complete fields and function names that actually exist.
I know this is a bit late, but copilot is only ok if used for code completion. I switched to the free tier of supermaven a month ago and it's been way more helpful, as it can handle context better. Probably cuts coding in half and takes away a third of debugging.
Asking chatgpt for code has also become better, but imo still not reliable enough to regularly use. Just had some docker code written and it got it wrong 3 times so I gave up on that.
I get your point, AI can only save time if you know exactly what you're doing and it will only be helpful sometimes. But when it is, it's such a time saver.
Mostly it really is just a fancier auto-complete. It is most useful for situations where you want to essentially do the equivalent of copy&paste and then make changes in a few predictable places in each copy.
It is total crap at writing code itself to the point where you need to read the code and understand it to know it hasn't screwed up, something that takes much, much longer than just writing it yourself.
I mean we could do things like Arkham Knight, Flight Simulator, The Last of Us 2 and so on. Do we really need to do everything realtime or could we continue baking GI?
AI models are already kind of baked. Just not into data files, but into a bigass mathematical model.
I guess that is why AI runs like ass?