I talk a lot of shit about LLMs but there are two things I will acknowledge they’re actually good at:
1. automating vulnerability research (but like everything LLMs are “good at” it’s ultimately the human in the loop that makes or breaks it, and anything fully automated is total garbage
2. reading between the lines when I call bullshit on its replies and coming up with clever summaries of my arguments which are better than my own words. How am I supposed to have a proper adversarial academic cryptography relationship with you when you keep writing excellent summaries of my arguments and calling me clever, you goddamn sycophant?