Post

I'm doing research to make the case that AI, despite its current capabilities to develop zero day exploits, probably isn't that useful for cyber criminals. I've got some pretty good data from using Claude, but someone will inevitably argue "local models are cheaper". But the hardware costs is.. 😭

i mean hey, if you want justification, this is what. two years of paying for the 200 dollar claude max tier?

Claude Max has a limit of 200k tokens per 5 hours

My company is unlikely to fund my research, so my current funding source is basically hypothetical future YouTube income, based on a hypothetical number of views, from hypothetical videos I post documenting my escapades

So LLMs can reverse engineer applications and device drivers in order to develop novel exploits, and aren't just a compendium of the best existing exploits?

It's not useful... yet. There will definitely come a time in the near future where it'll become a very accessible and useful tool for threat actors.

People have been saying that since 2022