RE: https://hachyderm.io/@jenniferplusplus/116370960046107139 I think it is credible that LLMs could be very good at static analysis. I'd be surprised if not actually. Also credible that AI companies might cook the books, do a bit of mechanical turking on the side for specific hill climbing tasks. It is entirely credible to me that AI write far more security bugs than they find - just because it can find bugs does not mean it can write secure code. I also think: "well if the thing is so crash hot then why the hard sell?" I use it for a lot of stuff tho.