Epic Games
Don’t worry, Epic will just sue Valve over the monopoly they were handed by Epic and their actions.
This is going to be a dumpster fire and I can’t wait to sit and watch it burn
ccording to a new report by Bloomberg, developers on the game were told to implement an “AI-first” production model using Anthropic’s Claude AI models. The team is not to use any more “hand-written code,” and even artists are being pushed to use the technology that has become a punching bag among actual players. The director of the unannounced game reportedly told developers that all art should be “generated with full AI workflows.”
No handwritten code allowed is laughably stupid. This decision was made by someone who doesn’t understand code or llms.
AI Art is even more stupid, getting a diffusion model to depict something generic close to what you want is easy, but when you want more precise control it takes a ton of effort, and it’s a whole new area of knowledge. Sure, full models on servers with terabytes of RAM will follow instructions better than what I run on my GPU, but it’s still absurd.
The team is not to use any more “hand-written code,”
Sweet, so if someone gets a copy of the source code then it is public domain as the output of AI is not copyrightable under current US case law.
And simehow it will be Valve’s fault, according to Tim-boy
Do you think so? Most big companies require developers to use Ai.
Edit: I don’t understand why my question is downvoted.
AI can for sure be a useful tool in development - but forcing everyone to use just the one tool and removing developer agency is going to be a shitshow. You can absloutely make slop code usable and stable - you could make dreamweaver code or VS’s autogenerated C# usable too - but it’s not a magically quick process. Given the goal here is speed, this ill wind doth blow due trainwreck…
Not only that. LLMs have serious flaws like are much more costly than hand written code, are much more prone to errors, cant be reviewed or audited, and rarely employs the most efficient architectures. On top of that, execs think the memory is absolute and dont realize the more context you feed it, the more it has to guess.
LLM-only code is going to be terrible.
They’re using AI as the main development tool by the sound of it, not a supplementary tool. If they’re anything like the AI ads I have seen lately it will look like shit. I have no trust that this will create a cohesive product.
In related news there is a rising industry of services to detect, fix and remove generative code and slop and substitute tested human-made products, for when your vibe-coded project turns out to be completely buggy.
A quick reminder: LLMs can’t logic or do math. They can only pathway through their autocomplete hyperminds for previous instances of parts of the solution.
Sounds like sloppy garbage.
