The GNU Compiler Collection (GCC) developers now have a need to set a policy whether AI / Large Language Model (LLM) generated patches will be accepted for this open-source compiler stack.
The GCC compiler doesn’t currently have a policy in place whether to permit AI/LLM-generated patches. But within a bug report today there is a patch posted by a user in trying to fix a GCC 16 compiler regression.
A lot of open source development now is done by employees at big corps like IBM, Red Hat, Google and Microsoft.
Those companies heavily push towards using AI for development, some even enforcing it.
Will be interesting to see what happens when more and more OSS projects outright ban AI contributions, while more and more companies mandate using AI.If it’s maintainable, I don’t think there’s an issue (aside from the usual license violations in LLM training). The big problem is when people send giant unmaintainable patches that they can’t even explain themselves.
No, the license violations are the main problem.
AI will steal all publicly visible code regardless of its license as “training data” and then put out the same code almost verbatim when asked to solve the same problem. And projects need to prevent that to avoid costly legal battles they can’t afford.
Don’t.
Enforce the usual quality rules and most all llm results will be banned by default. Technically, you could try banning llm work by default, but that just results in people lying, and having to check over llm work anyways.
Well… I’d be curious to know how anyone would argue that AI generated code is GPL safe, especially considering they are often using black box, binary blob models trained on mystery code from all over the internet, with zero attribution or license information, and with some of it almost certainly being proprietary.
Are we going to start ignoring the very licenses that we created.
Alternatively, I would argue that all LLM code output must be GPL, since it was trained partially on GPL code.
Either that, or LLM code output cannot be used for any purpose at all, by anyone.
IMO I think it’s fine as long the submitter understand all their changes. This is easiest done if the submitter writes all the code themselves though.
In this case I would reject this. The submitter doesn’t seem to be qualified to understand the code change, and just let ChatGPT explain the code change for them. ChatGPT is a master in being confidently incorrect.
Size limit?
…
out of context joke
640KB ought to be enough for everybody - - Bill Gates
