On Fri, Oct 16, 2026 at 07:19:36AM -0700, Matthew Brost wrote: > On Wed, Oct 14, 2026 at 02:56:22PM +0100, Lorenzo Stoakes (ARM) wrote: > > On Tue, Oct 13, 2026 at 01:45:09PM -0700, Linus Torvalds wrote: > > > The kind of expectations *I* have is that people do their best and I > > > expect people to understand what they are doing. > > > > > > But that understanding has very little to do with what tools they use. > > > > I think that's the crux of it. Don't send stuff you don't understand. > > > > Yes, never send code one doesn't undertstand. > > Also as important - never send code you haven't tested. > > The idea that LLMs should not be allowed to write commit messages or > documentation really gets under my skin. What if you have a learning > disability related to reading or writing? I do, and for years the > hardest part of writing code was documenting what it did in a way that > others could understand. When LLMs first came out, all I used them for > was writing documentation and commit messages because they couldn't > properly code my ideas. They could, however, describe my ideas quickly > and clearly. Fast forward to today, and LLMs can do both, so I use > them for both.
My argument to keep "assisted-by:" as optional but highly desired commit metadata is based on understanding. If you have a culture of being able to explain things to humans, then make it easy for both humans and their assistive techonology to know exactly *what* technology is being used, and where to get it. Hell, I would expect the proprietary LLM vendors would have a strong incentive to keep the free advertising. I personally haven't yet asked an LLM yet to write a thesis on the subject, but you can probably place a bet that I'll feed this message into a couple of assistive technologies and try to make the most compelling argument that any company operating under US securities rules might want to have their legal department review the company policy on what to put in the Assisted-by: signoff. > Matt > > > And if you do, expect to be ignored. > > > > -- > > Cheers, Lorenzo A clear structured header is easy to ignore, and the cost to maintain is minimal once we have consensus to retain it. Cheers, Troy
