A generalized estimation is that for every 100 lines of bug there is a possible exploit waiting to be leveraged. Linux Kernel alone is over 40 million lines. Good Luck!
That said, I do hope that these LLM assists patches are closing these and not introducing new ones. But I am not too optimistic about that. :)
Absolutely agree. It’s basically a fuzzer with a big database of exploits. Of course it can find exploits.
It’s one area it makes sense to me, but I’m pretty sure you could do even better with a specifically trained model rather than a general LLM.
Regardless, you still need buggy code to exploit.
Hasn’t that always been a near-unlimited resource though?
AI has led it to a previously unthinkable level of buggy code.
A generalized estimation is that for every 100 lines of bug there is a possible exploit waiting to be leveraged. Linux Kernel alone is over 40 million lines. Good Luck!
That said, I do hope that these LLM assists patches are closing these and not introducing new ones. But I am not too optimistic about that. :)