- cross-posted to:
- pcgaming@lemmy.ca
- cross-posted to:
- pcgaming@lemmy.ca
This video is a short documentary talking about how idTech developers and how they came together to make their own employee-owned indie studio. Good stuff!
This video is a short documentary talking about how idTech developers and how they came together to make their own employee-owned indie studio. Good stuff!
Companies firing employees for AI, only for that AI to help crack their games, leak their exclusives, and those fired employees now create their own studios making the market more competitive
It wasnt for AI in this case. IdSoftware had arguably the best engineers in the industry. Doom 2016 and Doom Eternal look beautiful and run on junk GPUs to this day.
Something happened after Doom Eternal and next game came up… Just mid. Amazing cutscenes, very mid gameplay, garbage performance due to mandatory ray tracing, while looking same as previous games.
Game didnt sell well, parent company fired people EA style.
Also they fucked Mick Gordon their music producer who did music for all doom games.
Mick Gordon was eight years old when first ‘Doom’ came out, and the first video game for which he did sound design came out in 2006.
Fuck, right. Dont know how did that slip through
Hang on… AI… is… Good?
Their first project:
“For our first project, Hidden Horses is developing a high-octane single-player FPS. Influenced by games like Smash TV and movies like 1987’s The Running Man, the game will take place on a deadly game show run by a sinister but charming AI host.”
Nah it’s unaligned and chaotic. It’s an industry disrupter because everyone wants to pretend it’s good at everything, wildly unsustainable, and a huge time bomb.
My point is it’s good because in this instance, it seems like we get something good out of it.
In the same way as Trump is good by making the gas price so high people would rather buy an EV than pay for gas.
Worked for me 🤣. Car payment is a wash because of the gas savings.
Always has been.
It’s the AI companies which are bad. AI itself is one of the dependencies for an actually free society where people only work because they actually want to. Still a long way to go for that. But the first steps have been made. Let’s hope that MAD keeps working until we are there.
I agree that the system we have now were everyone has to be shackled to a job in order to be considered valuable is terrible. And that a system were work is entirely at will is both desirable and achievable. But I’m afraid gen ai isn’t the way to that goal.
Gen AI is impressive but you have to understand how it works; it is just a statistical model, every word or character is decided by probablity. It will have a list of words (or changes to an image) that might come next and will have different probabilities for each one. One will be randomly picked according to their assigned odds and then the algorithm calculates the words and odds for the next word till it’s finished. The fact that this system produces such consistent output is an achievement. But there is an issue: the ai needs a constant flow of new training data or it’s outputs start to become very similar, it’s facts start to get out of date and it’s use of language starts to fall behind natural linguistic drift.
Leaving aside the rather weighty issue of where and how these ai companies have typically gotten the enormous amount of data they require; they need a constant supply of human generated data to train on. Training on only machine generated data threatens model collapse so there must be a majority of human generated data in the training data to prevent that.
To put it another way AI isn’t the solution to needing human effort its just laundering where the human effort is coming from.
In software development, there’s a concept of layers of abstraction, where each layer up implements functionality under increasingly approachable language. It’s “possible” to program using strictly binary, and early computers did exactly that - punch cards were instructions written in binary. Assembly is often the next layer which provides “readable” aliases to these binary instructions. Next up is compiled languages like C/C++ that takes even more readable code and the compiler then turns that into the binary instructions. Languages like Python or Javascript are translated by a runtime interpreter that takes the code being run and applies those instructions to pre-compiled binaries.
To me in the context of writing code, an LLM is basically the next abstraction layer on the stack.
Sure, LLMs and Stable Diffusion will not be the final forms of AI even though they are really impressive now.
But that they are just maths isn’t the limiting factor. Natural neural networks are just maths too.
The solution for removing the need of human labor is fully automating everything. But something needs to be the ghost in the shell. Surprisingly few tasks don’t need human-level intelligence. We know, neural networks can do it because humans can do it…
We just can’t make good enough artificial neural networks yet.
It’s like trump: pushing for electric cars, uniting the EU, and more!
/s but actually it’s true.
Just don’t tell him. And don’t discuss it with Muricans because it’s way less funny for those who are in that failed state.