What is FEU's stance on AI generated code?

I am asking this question because, while I know AI generated art is allowed, I couldn’t find any official stance about code.

The AI art policy reads:

To be clear. I am not asking if it would be F2E or not. I am asking if AI-made code is allowed at all. To be used in projects and to share publicly. And I would immediately assume that it would be F2E by default as well.

I know coding is a hard thing for many people. And I wanna contribute by using AI to bring new ideas for complex game mechanics to life.

1 Like

I don’t think I can really stop you if you choose to ask Claude to generate some assembly and use it in your own hack.

At the moment, we intend to mirror the policy on art - we require disclosure of any LLM-based tooling (please be precise about what parts are or are not AI-generated) and do not respect licenses imposed on code that is majority AI-generated (no, I am not going to give a precise definition of what “majority” means; any arbitrary metric I say is just asking people to game it).

If this policy leads to an influx of slop code, we reserve the right to take a harsher stance.

13 Likes

Thank you for the response!

I’ll make sure to test the code I generate to make sure it works before posting it.

1 Like

I’d recommend a rule stating that code should be reasonably tested before release, with perhaps warnings and the like when someone submits shit code that breaks the game or crashes. That might stem lazy vibe-coding a bit. Not saying they have to make it bug-free, but if your new rescue functionality crashes the game the minute you use it on a Cavalier, then that’s clearly not been tested properly.

6 Likes

I think code is a bit different from art in this case because of the way people interact with generated code differs from generated art. In that there is more “discussion” in a sense with the machine about what may or may not work, and more incentive to double check things and test them. Also, an ai can do some of the more mundane busywork aspects of writing code, like where you have to copy paste things several times. Wheras with ai art, it’s typically a one and done thing. Type prompt and that’s all. Plus the ownership issues are much more prevalent with ai art than ai code.

2 Likes

I think using AI code is nothing wrong. But it mus be labeled that the project got help with AI and the code must be tested carefully so that no one computer got exploded.

1 Like

I occasionally ask ai about code, but I don’t consider my stuff to be ai generated. I find that ai can only write simple functions - it makes mistakes more often than not with anything complicated. I’ve found ai helpful for learning some formatting and for immediate feedback (though the feedback itself is often useless). Often just typing out your issue can help you figure out what to do next, regardless of whether you send the message to ai or not.

To make code for fegba, you generally need to learn how to use the decomp, no$gba, febuilder, etc. and ai can only fill in some basic gaps in your knowledge. If you aren’t interested in learning to code, I don’t think you’ll really get anywhere with just ai generated code anyway.

I’m not worried about sloppy code on this forum personally - plenty of that already exists (and was not written by ai), yet people still use it. I’m not interested in drama by calling out specific people’s hacks as buggy, but I’ve had to fix enough things from a variety of wizards that I can confidently say that playtesting the code is often an afterthought, with the bare minimum done. No judgment here, as I’m also included in this - there have been many times that users have reported bugs to me after I’ve released code, and while I’ve usually gone back to try and fix these issues, there are some things I’ve never fixed, and probably more bugs that just haven’t been discovered.

Even things that have been extensively tested can suddenly break when in a new environment. It’s taken me years to anticipate and navigate around which patches might conflict with what I’m making, and I still end up releasing things that just don’t work or break things if you have certain asm hacks installed. I don’t think there’s an easy way to know if code wasn’t tested very well before release, nor to enforce this sort of rule, so I don’t personally support this rule. There are so few people who release custom code with any regularity that I’d also worry about discouraging these people. It’s ultimately up to the user to try the code themself and to figure out if it causes issues and/or if it’s worth using. If you install a bunch of patches at once, then spend a few weeks creating chapters before testing only to discover that the game now freezes from x action, that’s on you. Any time you install something, imo you should be prepared to uninstall it later or to revert to a backup if needed.

For community projects like skillsys, I do think code needs to be scrutinized more before being merged, as PRs (without ai code) have sometimes introduced new bugs, but at the end of the day, this is just a hobby, and we aren’t held to professional standards.

I personally haven’t noticed ai code as an issue around here, and I’m against strict rules about it. If you want to ban ai art, though, be my guest lol.

15 Likes

I agree with a lot of what Vesly’s saying regarding AI involvement in code not being worth getting into a tizzy about. You run the same risk of unvetted code with or without AI if you’re not testing properly. You need to make sure your stuff actually works, and for community projects that sort of vetting is easier to assign to someone. When it comes to standalone patches though, you don’t really have an option but to hope that the developer did “reasonable testing” before release. I don’t think there’s an acceptable way to enforce that without being a drain on moderator resources and driving people away in the process.

Where I disagree with Vesly however, is the extent to which AI can help a project. I’m not sure if he’s integrating an AI agent with his buildfile or just querying it seperately but I’ve recently attached Copilot to my own buildfile and after a few days of rapid fire development I can confidently say it’s a game changer. I’m able to fix bugs and develop new skills and features in a matter of minutes where it would have previously taken me hours/days of monotony. Copilot can index my entire codebase and very quickly find the relevant files to say… implement the trap rework without compilation issues or extend existing features like Vesly’s own debugger.

Yes there is still a technical hurdle, you need to understand the basics of software development to a degree. How to setup the buildfile, use version control, hook functions from the decomp repo, have some debugging knowledge, and most importantly of all; actually test your shit before release. But if you understand all that, then it really helps you fly. These are exciting times for a new era of rom hackers and I think it would behoove the FEuniverse moderation team to embrace the boost AI can give projects if properly used. The amount it lowers the barrier to entry is staggering.

For those that follow me in the Discord, I regularly create showcases for the cool features I make. AI is only going to accelerate that. I’d love to see what cool things other people can make with these tools.

9 Likes

Personally I think It’s fine, as long you check the generated code itself but also can apply changes to the code itself that would be needed in order to fix certain problems like bugs for example.

Also should be mentioned that it was made with the help of LLM

1 Like