返回资讯中心
外部精选
软件工程
#开发者实时流

Trying and Failing to Draw a Line Between Code and Art

There are hundreds of examples I’ve omitted. All of it is code. Every little animation, every decision about where to put sounds and music, every Easter egg, every particle effect, every game mechanic, all of it. Code.

Hacker News 最新em-tg2026/11/5 · 今天10 分钟阅读

以下正文同步自 Hacker News 最新,版权归原站所有,已转换为易读排版。

There are hundreds of examples I’ve omitted. All of it is code. Every little animation, every decision about where to put sounds and music, every Easter egg, every particle effect, every game mechanic, all of it. Code.

Websites and Undertale were the easiest examples I could reach for, but really there are so many examples I could have chosen. Ever heard of the demoscene? That’s a community where people make entire animated spectacles using nothing but code. There’s also Shadertoy, a website where people have taken a single tiny piece of the modern game development graphics pipeline and made art using only that. Minecraft’s rolling hills and winding caves aren’t drawn by an artist’s hand but indirectly by the programmer: specialized code creates unique terrain with each new game you start.

The reason it initially made sense to make a distinction between LLM-generated code and AI-generated images or sounds is that the latter can largely be created in one prompt while the former still required the involvement of a human programmer. If an LLM generates all the code but a programmer makes all the decisions, then surely we can still consider the result human-authored?

The problem with this view is that it’s based on an outdated understanding of LLM limitations. Unity recently announced Unity Spark, an LLM-based game authoring tool that can generate an entire game from a handful of prompts. Notably, it doesn’t appear to rely on “AI art” in the code-exclusionary sense, but instead composes games from collections of pre-made assets presumably made by human authors. Should games made with this tool be labeled as AI-generated on platforms that require it? I think it’d be difficult to argue that the answer is “no.”

I hope I’ve made it clear why excluding code from an AI content labeling system doesn’t make a whole lot of sense. Why then might Steam choose to do so? The short answer is they have to, otherwise every single game developer using a modern version of Unreal Engine would need to mark their games with the AI label.

Game engines are often thought of as tools, but in reality a game engine is both a tool and a set of components that eventually end up in the final game. Modern games load polygon data, draw 3d shapes on screen, render various lighting effects, animate models, simulate basic physics, and more. All of these things are driven by code written by engine developers or by their LLM-powered “agents.” Even games that don’t use engines will often still rely on a substantial amount of code written by others, much of it “poisoned” by LLM authorship.

It’s probably not fair to pick on Unreal Engine like this, but it makes for a good example thanks to the public statements by Tim Sweeney, CEO of the company responsible for developing it. Sweeney has publicly confirmed that Unreal Engine is developed using LLM-generated code, while publicly criticizing Valve for Steam’s AI-generated content disclosure policy. It’s easy to guess that Unreal Engine’s authors might be worried about what an exhaustive AI-generated content disclosure policy might mean for games developed with Unreal.

Even if the major engines and game libraries all promised to never use AI-generated code, or if Steam carved out an exception for a game’s dependencies, the poorly-kept secret in gamedev right now is that there has already been a widespread switch to using LLMs. The view that an LLM-coded game doesn’t “count” as AI-generated art is common among programmers, and both industry surveys and my industry connections confirm that LLM usage when writing game code is already ubiquitous.

Clearly there is a mismatch between the expectations of players, the expectations of game developers/distributors, and the reality of what it means to generate code with AI. As LLM-based coding tools improve, it will become increasingly difficult to rely on the code/asset distinction when marking and filtering AI-generated content. Soon the gaming community will be forced to decide what it wants the future of game development to look like.

正文由 探How 从来源站点 RSS 同步,内容未经改写;遇到排版缺失或需要图片、视频时请以原文为准。

打开原文
去社区讨论 →