What's this about?
Konstantin Ryabitsev put numbers to a complaint kernel developers have made for a while. Across five geo-distributed nodes, at any given moment, 14 CPU cores are doing nothing but rendering git commits as HTML for scrapers. That is more CPU than all legitimate access combined, including git clones. He calls it "background radiation" — a permanent tax on capacity, spent producing output useful for exactly one purpose.
Why is kernel.org such a target?
Because it is a guaranteed pre-AI corpus. Every kernel commit and every LKML message was written by a human before LLMs existed, which makes it unusually valuable training data — training a model on model-generated text gives it, in his phrase, "the equivalent of a digital prion disease." The bitter irony is the method: everything on kernel.org is a `git clone` away, complete and cheap. The scrapers instead walk it commit by commit as rendered HTML, which is the most expensive possible way to get data that is already free.
What's the community saying?
- The economic framing that lands hardest: how expensive would AI access be if every user paid their fair share instead of shoving the cost onto the people doing the actual work? One reply extends it to Uber's pricing.
- Site operators report the same pattern from their own logs — traffic from tiny countries, obscure browsers and outdated operating systems, consistent with scrapers behind VPNs and header anonymisers, which is what makes blocking hard.
- A lighter sub-thread over the xkcd-style plots, and the author noting drily that no, he does not write with LLMs.