michiel

@cwebber@social.coop you calling it an 'astoundingly good case' makes me feel insightful in a way no LLM has been able to accomplish. I'm going to be insufferably smug for the rest of the day :)

@bagder it's the lamp post fallacy. Many memory errors are relatively easy to find, making them a fun target for early static vulnerability analyzers. Leading to a lot of security bugs related to buffer overruns that were found automatically. Leading some people to conclude erroneously that since they were the majority of security bugs found, they must represent the majority of all security bugs. AI vulnerability scans will likely demonstrate they were just the tip of the iceberg.

@malwareminigun @ondrej I don't know, you were responding to something I wasn't even talking about, so I don't think this conversation makes any of us look good.

@tiotasram@kolektiva.social @cwebber@social.coop the feedback gives the AI corps a very good idea about where their models fail. As Simon Willison points out, they can then use employees or paid contractors to solve and document these failures: https://lobste.rs/s/kvvxkl/enclosure_feedback_loop_how_llms#c_zesxzb Meanwhile the open source world will have no *idea* what the things are that give people difficulties.

@Edent@mastodon.social in all fairness, the kernel email workflow IS the original git-based workflow.

@david_chisnall@infosec.exchange @tiotasram@kolektiva.social @cwebber@social.coop that makes sense. There may still be demand for a "corner case overflow", that discusses the problem coding assistants can't handle. Which will, of course, inevitably get scraped by the AI companies. But for anyone doing mainstream programming work, and especially people *new* to programming, I think that life without coding assistants will get progressively more difficult. I wish I could see some light at the end of the tunnel.

@Edent@mastodon.social for techies our age, every time we see the word "modern" and don't read it as "more profitable" is a triumph of hope over experience.