On Tue, 2024-02-27 at 16:43 -0800, Linus Torvalds wrote: > I personally would find it more intriguing still if it actually > reacted to the code instead of just the discussion, but I suspect > that's still some way away.
So this bit may be closer than you think. The Software Heritage Org announced a LLM trained on all FOSS code in the archive: https://www.softwareheritage.org/2023/10/19/swh-statement-on-llm-for-code/ So it should be possible for Konstantin to plug into it. The question, as with all static checkers, would be whether it produces a decent enough signal to noise ratio for us to find it useful. I bet in the early days it won't so someone will have to help curate it and improve the model as a full time job, but I'm sure the SW Heritage folks would love to help. James
