Collateral damage [LWN.net]

Collateral damage

Posted Aug 29, 2026 14:24 UTC (Sat) by gray_-_wolf (subscriber, #131074)
In reply to: Collateral damage by snajpa
Parent article: Ryabitsev: Creepy crawlies

Why can't the llm just clone the repository and investigate in that? Same for mailing lists, many offer an archive for download. Both (the cloned repository, the mailing list archive) can be cached locally and reused, no need to hit the web interface every single time.

to post comments

Collateral damage

Posted Aug 29, 2026 14:37 UTC (Sat) by dskoll (subscriber, #1630) [Link] (1 responses)

Because the scrapers are probably vibe-coded?

/me ducks...

Collateral damage

Posted Aug 29, 2026 15:01 UTC (Sat) by mb (subscriber, #50428) [Link]

Actually, I think the problem is that the crawlers are probably not vibe-coded or the crawler programming is not coupled back to the crawler execution. If crawler programming was coupled back to a tiny fraction of crawler execution logs, then an LLM agent could easily find out where requests are wasted (such as going endless cycles in generated content that is easier to access in another way. e.g. git clone) and implement counter measures automatically.