@netbsd@mastodon.sdf.org @eloy@hsnl.social I was assuming the popularization of fuzzing correlates with one of these bumps, but wasn't sure about timing. Would that have been 2005? Feels like LLMs are just another form of fuzzing. They can't reason about a problem, but they have a lot of different kinds of exploits in their training data, so they just apply those at random until something appears to work.