It actually did better than I expected with the paper "Bypassing memory safety mechanisms through speculative control flow hijacks" [0]
I copied and pasted the text from sections 3-7 (Case Studies - Conclusion) and Section 2 on its own (describes the attack)
It did pull out some important statements, better than I expected. Probably won't save me much time, but I was quite disappointed by the fact that the Advanced and Basic versions were the same for both which kinda felt a bit cheap to get the same results especially since it still cost to get that advanced result. Maybe including information about how the basic version is restricted and what the advanced does better would make it easier to know when the advanced version won't be useful.
I also tested with a random write-up I'll be covering tomorrow "Breaking the Competition" [1] I had higher hopes for this since it was more of a blog-ish post. I did get different results for basic and advanced with this one, but the result was basically non-sense, worse than expected, and worse than the paper summary.
Overall, probably not something that I'll end up using, but technical content also isn't the intended use-case which is totally fair. I'll also add that one feature that I looked for immediately was API access as I'd have wanted to integrate this into an app I use to plan episodes.
- [0] https://arxiv.org/pdf/2003.05503v1.pdf
- [1] https://medium.com/ctf-writeups/breaking-the-competition-bug...