Hacker News Reader

Comment by bulbosaur123 | Hacker News Reader

Game dev for online game here and genuinely curious.

I have to work with online servers, regular testing, analyzing bugs in both client and server (python), troubleshooting, adding new features to client (written in C# for iOS and Java for Android).

In practice, what is the best way I could use ChatGPT in game development? How can it realistically test the game and troubleshoot issues, how can it implement new features?

These things require making changes to code, compiling, running the client and testing how it plays and looks. Everything has to be extremely precise.

Any ideas?

View on HN

Right now today it won't go as far as you are asking but the best workflow imo is using GH's Copilot in your IDE and a ChatGPT plus account on the side to use for more complicate discussions that GPT4 handles better.

I am not in game dev but that is my workflow and it has definitely improved my productivity and is a pleasure to use. I can get Copilot to help write tests for my functions, I use it regularly to see if there is a more optimized way to write the code I am looking for. I switched from JetBrains to VS Code so I can get a better integrated experience too. The Copilot autocomplete is usually pretty spot on with what I want to write.

I'd like to add to what the other commenter said. I use the same stack but I mostly limit Copilot to autocomplete, the built in chat is not really as good as just copying and pasting into GPT-4 for anything even remotely complex. There is also Cursor https://www.cursor.so/ which integrates GPT-4 into the IDE but you do have to switch to using their custom version of VSCode. ChatGPT Plus is also getting multi-modal support so you can send it images, this has been useful for working on webdev for me but I could see it being useful in gamedev as well. If you haven't used ChatGPT much before, make sure to fill in your Custom Instructions with details about yourself and your preferred communication and programming styles. It really helps focus the model on your use case.

Any time you can drive things with data there's a good opportunity.

For instance, if you have a functional testing suite that can be run with, say, a series of commands in JSON... you can tell GPT what it can do and ask it to write out scripts in that format.

In general if you think you want a DSL you might be able to get away with JSON and GPT instead, using GPT prompts as your DSL and some simpler imperative system run off GPT-generated JSON. (GPT can translate the JSON back to English too, so it's not just one way)

If there's any story elements in your game you can generate them with GPT, again probably resulting in JSON. You shouldn't defer the entire plot to GPT, but it can fill out lots of details and make the environment richer.

If you have autonomous entities in your game (as simple as any kind of enemy) you could both brainstorm behaviors and personalities and then turn that into code, making the enemies more eclectic.

The best way is probably using ChatGPT as a replacement for StackOverflow and documentation search, asking questions like "how do I do X using Y".

Reply on news.ycombinator.com