Ok, this article is kind of neat, even I think it's poorly written and generally with a shakey thesis, if only to aid in reminding people that larger models are getting us nowhere. What we need are good TOOLS. New deterministic TOOLS. Not one big model that is supposed to replace every tool on a computer.
Though, I, like many, were surprised to see openssl on this "no found vulns" list. Like, what? Doesn't open ssl have a vuln like every two weeks? Didn't tom7 JUST write a paper on this? (https://tom7.org/httpv/httpv.pdf)
Then I saw openssl survived the pipeline because the llm couldn't figure out how to compile it. all is right in the world.
https://theweatherreport.ai/posts/symbolic-execution-and-llms/