The Fedora Council has finally come to a decision on allowing AI-assisted contributions to the project. The agreed upon guidelines are fairly straight-forward and will permit AI-assisted contributions if it’s properly disclosed and transparent.
The AI-assisted contributions policy outlined in this Fedora Council ticket is now approved for the Fedora project moving forward. AI-assisted code contributions can be used but the contributor must take responsibility for that contribution, it must be transparent in disclosing the use of AI such as with the “Assisted-by” tag, and that AI can help in assisting human reviewers/evaluation but must not be the sole or final arbiter. This AI policy also doesn’t cover large-scale initiatives which will need to be handled individually with the Fedora Council.
Don’t know how bad this take is, but not using LLMs for coding assistance to some degree just for the sake of not using LLMs might not be the best option right now.
There has to be a middle ground between letting the thing spit out whole kernel modules and refusing to use it at all.
Also having it declared as AI assisted code might be better than the people doing it anyway undisclosed.
The middle ground, IMO, is not letting it spit out code.
Its almost certainly terrible, every time. Sometimes though… Its just mostly bad.
Ive found it useful for finding errors and potential optimizations though. Just not, you know, letting it actually write anything.
But letting it review and seeing:
Thats useful! Helpful, even.
Just not the nonsense it makes on its own.
Are SPEC files for RPM creation code? How much actual code is even written under the Fedora umbrella, except maintenance scripts and such? Adjacent projects such as Anaconda are in the rhinstaller organization on Github: https://github.com/rhinstaller/anaconda
Either I overlooked the details or they aren’t spelled out. From my experience of packaging software for myself as RPM (for openSUSE) the amount of actual code are a few lines of bash scripting to invoke sed and such.
@passepartout
Perhaps if AI were explicitly asked to write code in pidgin English (specifically something which is not meant to be compilable), and then that was used as the basis of actual code development?
@cm0002