An LLM wouldn't really be the right tool. But some of the black-box fuzzing tools out there were already using machine learning approaches before the "AI boom" happened.
I'm sure someone's found a way to apply agentic shit to it.
There is only a minor difference. One is a random destination and one is a random walk. In a two dimensional space there is no expected difference in outcome. In higher dimensional spaces the random walk fails almost immediately.
An LLM wouldn't really be the right tool. But some of the black-box fuzzing tools out there were already using machine learning approaches before the "AI boom" happened.
I'm sure someone's found a way to apply agentic shit to it.
Monte carlo methods. Literally: "Try random shit in a simple pattern designed to maximize coverage."
The search space is simply too large to do anything else.
And waste millions of dollars in tokens in the process.
I was talking about the mutation-based ones. They get hybridized with MC frequently, but would be classed as Genetic Algorithms if I'm not mistaken.
There is only a minor difference. One is a random destination and one is a random walk. In a two dimensional space there is no expected difference in outcome. In higher dimensional spaces the random walk fails almost immediately.