Since we last covered the way that the Linux Kernel project has been dealing with the influx of code created using LLM-based coding assistants, all the way back in April, things have changed a lot, with most recently the Linux maintainers deciding to fully drop the requirement to mark LLM-assisted commits with the Assisted-by tag.
While not a fully uncontroversial decision – as also reported by Phoronix – with key maintainers already having made this decision it seems certain to proceed. This is not the only change made fairly recently, with a ‘Sashiko’ bot introduced that is supposed to handle code reviews.
In a July LKML post [Linus Torvalds] defends this decision, with a follow-up post in August by [Jakub Kicinski] it being stated that budget has been secured for using Meta’s LLMs, using said LLM-based resources for reviewing patches, as well as creating and editing commit messages. As noted by [Jakub] they’re running multiple frontier models on each patch to try and weed out confabulations.
Whether this will prove to be the right decision remains to be seen. Whereas most open source projects have firmly banished or heavily curtailed LLM-based contributions on slop, copyright as well as licensing grounds, the Linux Kernel project may be an interesting experiment in what the outcome of the exact opposite approach will be.
Someday there will be retrocomputing articles about people going back and programming in BASIC or whatnot “the old-fashioned way,” where you have to memorize syntax and commands and stuff. People will do it for nostalgia’s sake.
Like my grandpa punching tapes near the fireplace, drinking Five Alive, sure.
Feels like it was a contractual decision and that it was more of a compromise rather than a desire.
It’s really sad to see even the smartest people throwing more and more money into big tech pockets. No one can run these frontier models on their own independently, they have manufactured a need, and everyone is becoming more and more dependent on them. It’s so depressing.
They’ll democratize because of two factors:
– optimization of models and specialty models
– cheaper and ever more powerful mass-produced AI chips
The advantage of big AI companies will not stay for long. One does not need a cutting-edge, god-powerful model for programming task. “Good enough” models will even run on the edge soon.
I don’t understand the purpose of this. Wouldn’t the Assisted-by tag at least be useful for bookkeeping purposes?
It’s meaningless.
Define “Assisted-by”
The BDFL is now 56. At best he has maybe 10 more years of productivity left in him.
Once he’s out of the picture, Linux is likely to become a design by committee shitshow.
This.
But the question is whether AI will be able to ruin the kernel fast enough, before gen-z will rewrite it in rust?
gen-z is capable of coding ? that’s a bold claim !
He won’t have a problem working another 25 years in his role.
Unless he’s hit by a bus, or someone pulls a Luigi on him.