AI x Infosec Researcher @RevEng_AI • Binary Program Analysis • PhD Candidate @TUBerlin • prev. built @VulHuntRE at Binarly • Capturing Flags for @ENOFLAG
Our work on exploring new ways for efficient firmware fuzzing will be published at @USENIXSecurity 2023!
With SAFIREFUZZ, we introduce near-native rehosting, drastically improving the performance of fuzzing embedded targets.
Super happy about this work w @domenuk@nSinusR
there is little moat in harnessing if you don't care about tokens/money spent. just point the latest model at the right target.
however, if you want to optimize for speed and cost, there are interesting things left to work on.
@ky1ebot this is frankly insane, I'm actually curious to see more of this model.
do you think this (scale) could've been achieved with earlier models but more compute/tokens, too?
@Fox0x01 don't you think that's rather a psychological than a technical/legal problem?
at some point, these people will have to (re-)learn that blaming their agent does _not_ mean they are not responsible. the person orchestrating/overviewing the agent stays accountable.
understanding and benchmarking vulnerability detection capabilities is an interesting challenge and I can recommend @niklas2484's recently published PhD thesis on that topic.
it includes "Top Score on the Wrong Exam", one of my favorite (pre-LLM) ml-for-vuln-detection papers.
as the final contribution of his dissertation, Niklas discusses the question "How and Why Agents
Can Identify Bug-Introducing Commits" in a joint work with his supervisor @mboehme_ , again finding that well-designed experiments and ablations are crucial to develop understanding.
for a little bit this morning, I was pondering whether to start the fresh week by switching to Opus 5.5 as my go-to.
how is anyone remotely serious about code quality/hygiene living like this?