Lmao, no. LLMs don't really help with game engine code. In every other dev space there are thousands upon thousands of open source projects and github repos to train on. For Unity and Unreal, they don't have enough training data and hallucinate all day long. I've tried with the notable recent models and get nothing but bs.
this post was submitted on 18 Aug 2025
1 points (100.0% liked)
Game Development
6221 readers
1 users here now
Welcome to the game development community! This is a place to talk about and post anything related to the field of game development.
founded 3 years ago
MODERATORS
Can't say I have experienced the same. The models I have tried out don't always get everything 100% correct, but if I ask it to provide an example of code for X or Y game mechanic in C# compatible with Unity for example, it usually is able to provide code that actually does what I asked for. Even for Unity compatible HLSL shader code, shockingly. This is certainly a more recent improvement, maybe within the last 2 years.
It's not always the most optimized or the most readable, and sometimes I have to ask it to regenerate the code a few times, but for the most part it provides code that works on the first try.
Does this compare to the code that I have previously written that I simply reuse in my ~~unending sea of abandoned~~ projects? Not usually. But for code I don't even have an idea of where to start, it is a pretty useful tool.
Study by Google who benefits from AI hype. Big doubt. But also I would be interested to know what "counts as using AI". Doing one query once per month just to try solve some specific issue vs daily use/vibe coding.
I tried using AI to create my games, but most of the time it just got in the way. And now that I am more conscious about harm and issues of AI, I refuse to use it unless in really specific scenarios.