I love how people come to me and are like “oh rituals so quiet” “oh are you guys still working on stuff” And then we drop stuff like privacy-preserving LLMs w feasible inference times (with an acceptance to ICML) and continue to put out serious AI research and it’s quiet
Ritual
RitualJul 11, 00:15
In our last blog post, we showed how to break privacy-preserving LLM schemes — decoding permuted model states with near-perfect accuracy. Today, we present our defense: Cascade 🕵️‍♂️✨
5.53K