Login
You're viewing the functional.cafe public feed.

Replies

  • 💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • May 5, 2026, 6:23 AM

    @malwaretech this was an informative video. Thanks for being honest about the costs both in tokens and experience needed to get anywhere with the LLM.

    You mentioned "fuzzers?" in your video. I assume there are other automated tools that the vulnerability research community is using Today.

    How do they compare to the LLM pipeline in terms of outcomes, potential and costs?

    Disclaimer: I am no fan of LLMs but I am always a fan of understanding what is true and what is fiction/marketing.

    💬 0🔄 0⭐ 0
  • May 5, 2026, 7:06 AM

    @malwaretech tech makes sense as a tool(but not as a service/standalone, triage is a must).

    Do you know there false positive rate, like keeping stats.
    It would be interesting getting a number on LLMs tricking user vs. doing the job.

    💬 0🔄 0⭐ 0
  • May 5, 2026, 7:22 AM

    @malwaretech out of curiosity, wouldn't it have been efficient / effective to craft your prompt using a known and blacklisted vulnerable driver? That seems like that might have saved some tokens.

    💬 0🔄 0⭐ 0
  • May 5, 2026, 10:56 AM

    @malwaretech I was very interested in the part you described as "skipping a step" where the AI found a perfect vuln and you ran with it. But on reflection it found a shortlist and it was your skills in reviewing the shortlist that found the perfect vuln. Reminds me of how people reviewing a cold reading session recalled the accurate information the fake psychic had delivered but were then shown where in the transcript that information was first mentioned and realised that the techniques simply gave them a framework to supply information that the coldreader could then claim credit for knowing.

    💬 0🔄 0⭐ 0