Game Development

For someone like me, who was a bit lost on the meaning of framework, framework here is what you use to build an engine (I thought it's on top of the engine instead)

Also, the beginning of the article is a bit messy and the author jumps around thoughts, but it gets an interesting read, and they even talk about how to actually use AI for benefit instead of for multiplying bugs:

"It's hard for me to talk about it without sounding like a cult member," Hall said sheepishly, when describing how he uses ChatGPT while working in Brutal. But he and Falanghe agreed—using LLMs has made language-based coding an easier task.

Not that much easier, to be clear. They both said that when querying an LLM, they rarely copy and paste whatever code it generates. Instead they ask questions about C# libraries or Vulkan documentation, and the software is able to return high-quality answers. Answers that normally require programmers to sit down for hours to pore over documentation or scour ancient forums to find that one post with the solution (which was probably written in 2014).

"An LLM is essentially tokenizing language, then putting masses of vectors around that to build linkages between those tokens," said Hall. "What could be better than a highly-structured, in fact brutally structured language?" Vulkan and the latest version of C# are very "highly structured, with very clear syntax."

Developers critical of ChatGPT maker OpenAI should be able to replicate this process on open source models like DeepSeek, Falanghe said (though he hasn't tried this himself).

This process doesn't work as well with Unity and Unreal because they're both "highly spatial" as a result of their visual scripting tools. A solution for one game's problem may not work with another because of the different scripted elements. LLMs scouring the web can't produce consistent answers.

It is also the opposite of vibe coding, a method where programmers tell an LLM what they want a system to do and it generates code—and it isn't code completion, where AI tools "predict" what someone is typing and finish the string for them to speed up their workflow. The only thing the LLM does for Brutal developers is speed up access to information, letting them research without watching a 40 minute YouTube video.

Maybe we will finally see no-vibe solutions, like we saw no-code solutions 🌚