On 4/5/2026 22:59, Mikael Magnusson wrote:
I think it would be a mistake to treat all AI-generated content the same. They're tools and the quality of the work is the result of both the quality of the tool and the person wielding it.Recently, we've started gettitng more and more patch submissions that use undisclosed AI, in some cases nonsensical, in other cases not. Should we add a policy like many other projects have, and if so, what is our position? Personally I have noticed that people who use AI to generate patches to send to mailing lists tend to not review them very carefully themselves before firing them off.
Yes, a disturbingly significant number of people using AI surrender their thinking to it, but a lot don't. I have used it myself to generate code for my own projects and as long as you're willing to argue with it and do the hard work of architecting/software engineering, the code is decent if not quite as optimized as a human would make it if they put in the time.
People are just as capable of producing crap as AIs. AIs just do it faster. I recently got a look at something my company paid a contractor to write several years ago and quite frankly I was shocked that anyone had so little professional pride as to take money for that crap. (I'm not sure anyone had heard of LLMs at the time. I know ze didn't use one, AIs are too stuck on "best practices" even when they don't fit the need. An AI would never create executable config.)
Either way, patches need to be scrutinized for quality and tested.