Post

It's nice that real papers are talking about this stuff. I've read so many blog posts where the idea is "you just ..." and no, it doesn't actually work.

Hi, I'm not a programmer, but I’ve had a few ideas that ChatGPT and Deepseek helped me shape more clearly. Maybe you'll find something interesting here — this is my OSF profile: osf.io/profile/ There are a few concepts posted already, others I haven’t published yet.

… apropos of nothing the medium that ought to have benefitted from the concept of PageRank is academic papers and citations. Give me that visualization! Every extant scientific paper as a giant digraph!

Great read! Six design patterns to guard LLM agents from prompt injections. Pair them with CONFSEC’s zero-visibility inference layer for end-to-end safety—guardrails outside, cryptographic privacy inside.

That's what I like about this paper: it doesn't suggest using more prompts to try and detect and prevent prompt security problems, which is a very common bad solution simonwillison.net/2022/Sep/17/...