Yes but usually the hole poking affords defense against poking holes but now it truly is suddenly easier to break than to protect. It’s actually a pretty confounding problem
But if breaking in is just “Claude, find me an exploit”, then defending is kind of the same? Like “Claude, find me an exploit in my code and mitigate”? Then Claude defends against itself and the playing field is level again ¯\_(ツ)_/¯
Yeah, exactly. It’s not that intuitive but imagine giving everyone rocketlaunchers, and to compensate, pre-mixed cement trucks ready to pour. It’s not clear where to put the rebar, and it’s not good as a reaction. You could tell them to pour everywhere, and basically most are, but it’s open season nonetheless.
I don’t claim a LLM is any good. The article we’re discussing says “LLMs are oh so good in finding vulnerabilities, it’s the end of time 😱”. And if that was true, you can point it to your own stuff, find all the vulnerabilities and just fix them, so it isn’t the end of time ¯\_(ツ)_/¯
Yes but usually the hole poking affords defense against poking holes but now it truly is suddenly easier to break than to protect. It’s actually a pretty confounding problem
Yeah definitely. It’s the wild West
But if breaking in is just “Claude, find me an exploit”, then defending is kind of the same? Like “Claude, find me an exploit in my code and mitigate”? Then Claude defends against itself and the playing field is level again ¯\_(ツ)_/¯
Yeah, exactly. It’s not that intuitive but imagine giving everyone rocketlaunchers, and to compensate, pre-mixed cement trucks ready to pour. It’s not clear where to put the rebar, and it’s not good as a reaction. You could tell them to pour everywhere, and basically most are, but it’s open season nonetheless.
Trust me, Claude is NOT good at all in programming. Is worse than an average (and rather mediocre) programmer.
The same goes for any and all LLMs.
I don’t claim a LLM is any good. The article we’re discussing says “LLMs are oh so good in finding vulnerabilities, it’s the end of time 😱”. And if that was true, you can point it to your own stuff, find all the vulnerabilities and just fix them, so it isn’t the end of time ¯\_(ツ)_/¯
Problem is that AI creates vulnerabilities