It’s a hard problem. QEMU is currently reviewing it’s AI policy and we’re mindful to how helpful LLMs can be too someone new to the codebase while also capable of getting confused as the scope extends more widely over the repository.
We’re open to wider changes being made with pre-approval but I suspect they will be special cases like guided re-factoring where they can remove the tedium from the engineer.
We’ve been experimenting with using models for patch review but again it can be a time sync if the models go off on a wild adventure. If they could reliably catch the common gotchas that would at least alleviate some of the pressure on maintainers.
The most important part of Godot’s policy is the humans responsibility and ownership of the code.
It’s a hard problem. QEMU is currently reviewing it’s AI policy and we’re mindful to how helpful LLMs can be too someone new to the codebase while also capable of getting confused as the scope extends more widely over the repository.
We’re open to wider changes being made with pre-approval but I suspect they will be special cases like guided re-factoring where they can remove the tedium from the engineer.
We’ve been experimenting with using models for patch review but again it can be a time sync if the models go off on a wild adventure. If they could reliably catch the common gotchas that would at least alleviate some of the pressure on maintainers.
The most important part of Godot’s policy is the humans responsibility and ownership of the code.
qemu devs are ml members?