Godot Foundation just posted an official statement about changes to the Godot Contribution Policies, specifically focused on applying stricter rules against AI contributions.
It was met with really positive feedback across the different platforms it was posted today.
What do you think about the changes?
This makes me very happy. The barrier of entry seems nice and low.
I am curious if the team has considered a similar policy to Asset Store Contributions? Since 4.7 released, there seem to be an inordinately high number of LLM-created and documented addons. Though that may just be my impression rather than fact.
I did see some pushback on bluesky, but I think that’s mostly people who never had to manage a massive project, who think the foundation should completely ban AI and enforce that rule.
But unfortunately that’s impossible to do, how would you even tell if something is or isn’t AI?
I know some people will say “Well just use AI checkers” but those are also just… an AI trying to figure out if a code is AI, and they are incredibly inaccurate.
They will say a code is AI generated if it’s clean and well organized. Proof? “Well it looks clean and organized”.
So in my opinion this is as best as you’ll get for the contribution policy.
Finally! This was long overdue.
Imo the policy should spill over to the asset store, and then to forum’s resources section. The slop poisons those “secondary” channels too. Not as critical as the source code, but it still slowly rots the whole ecosystem.
This is a very, very well needed addition to the contribution policies,
best it could be at the least.
I know it might not be appropriate to ask in this post (or should I start a new thread?).
Since Godot has imposed stricter restrictions on AI PRs, I’m curious whether there is already a fork of Godot that is more tolerant of AI-generated code, or even primarily focused on AI-driven development.
I’d really like to compare the differences between Godot with AI-generated code and without.
I think a fair few people miss the point, and it is the first point in the article
- Encouraging new contributors to become future maintainers, that involves teaching and growing the understanding of new contributors.
LLM code quality is not a major reason, though it does suck to wade through slop. Detecting LLM generation over human made code doesn’t matter, it’s about building up people and you just can’t do that with slop machines.
Zig is a systems language improving on C is near and dear to my heart, as with Godot I read every release notice and eagerly make use of new features and functions. They have similarly made a recent LLM pull request ban, for similar reasoning too; they want to develop relationships with core developers, not talk to a wall that doesn’t learn anything. Here’s their community manager’s article on the subject
I’m sure someone will make one and it’ll be a spectacular failure. LLMs are notoriously bad at making high performance code, which is what you need in a game engine. As a professional programmer who has used LLMs to code, I can tell you that they just slow one down in performance coding, and they burn you out if you don’t need performance because you’re focused on a breakneck speed of development. Both are demoralizing to developers.
The whole point of this post is that you cannot get that kind of performance out of an LLM, because you cannot nurture and train them the way you can humans.
Then go make one. The problem with this idea is someone has to know more than an LLM to make software with it, and the people who fall into that category have no interest in doing that, because it’s a massive money sink with no ROI for a FOSS project. And it is soul-detroying.
Keep in mind the issue isn’t someone forking Godot and making LLM code, it’s people trying to leverage LLMs to change Godot into what they want with no idea of the impact their changes would make because they didn’t take the time to understand the system they’re trying to change.
Yes, so I’m not hoping someone does this on Godot’s mainline. It’s just that, since it’s open-source software, I’m curious if anyone has already made such a fork.
I completely agree with these rules. AI generates garbage if the human using it doesn’t know what they’re doing. I find it especially outrageous that some people are replying using AI as well.
I’m glad they have taken such a sensible stance on this.
I just recently started donating to the foundation and feel reassured in doing so by this decision. I think their conservative approach is called for as the quality of the project has a very high priority - I think there’s a high interest for this engine to succeed in the long run.
At this point who knows if AI is just responding to AI, the ultimate feedback loop. For all we know some of these accounts on the forum are just AI anyway.
Why not make a seperate AI showcase folder ?. Everybody happy.
I love the way this policy update was written. I have referred my employer to it twice this week as an example of how you can communicate something like this:
I always try to remain kind as long as someone disclosed honestly that they used gen ai. An approach I learned in this community.
Just be open about it.
I’m not sure. It’s probably easier to migrate stuff from the Asset library to the new store front if you’re good at generating marketing material semi-automatically.
I spent a week on trying to eek out some artsy stuff at new resolutions (without automated help of course).
It’s worth it, though. The new asset store is proving to become a big improvement. There is also a disclosure form field with an encouragement to use it.