Built this after the kernel's Assisted-by: tag policy landed earlier this year. Three numbers I wanted to see:
- by commits: 166 of 26,044 mainline commits since 2026-01-01 (0.64%)
- by new code: 3,304 of 1,130,805 lines added (0.29%)
- against the entire 42.5M-line kernel: 78 parts per million
Methodology in detail at the bottom of the page. Submitted side comes from `lei q -d mid -f mboxrd 'b:"Assisted-by:" AND d:20260101..'` against lore.kernel.org/all, with v1/v2/v3 respins collapsed and bots filtered. Merged side is git log on a fresh shallow clone since 2026-01-01. Kernel-wide line totals from the same clone, with 245 shallow-boundary phantom commits filtered (they show as 90k-files-changed diffs against missing parents).
Three things I deliberately don't show, with reasons documented: merge rate per vendor (different humans use different tools for different patch types, the ratio is not a model quality signal), motivation behind any specific tag string, and patches that landed without disclosure (this measures policy compliance, not actual AI usage).
Would be nice to get context from the mailing list. GKH explains his Clanker fuzzers, but this (presumably generated) summary seems unaware of it.
https://www.phoronix.com/news/Clanker-T1000-AMD-Ryzen-AI-Max for some context
Good call, I did try to search for it but must have missed it. Updated the page, Thanks for catching it.
Built this after the kernel's Assisted-by: tag policy landed earlier this year. Three numbers I wanted to see:
Methodology in detail at the bottom of the page. Submitted side comes from `lei q -d mid -f mboxrd 'b:"Assisted-by:" AND d:20260101..'` against lore.kernel.org/all, with v1/v2/v3 respins collapsed and bots filtered. Merged side is git log on a fresh shallow clone since 2026-01-01. Kernel-wide line totals from the same clone, with 245 shallow-boundary phantom commits filtered (they show as 90k-files-changed diffs against missing parents).Source: https://github.com/snek-git/assisted-by
Built with Claude Code.
Three things I deliberately don't show, with reasons documented: merge rate per vendor (different humans use different tools for different patch types, the ratio is not a model quality signal), motivation behind any specific tag string, and patches that landed without disclosure (this measures policy compliance, not actual AI usage).
Well, we have Linus as an omnipotent dictator, but looks cool anyway :P
Looks interesting!