Posts
All the articles I've posted.
-
Why doesn't GPT ever ask me anything?
I noticed Codex never asks clarifying questions, even on complex tasks. Dug into why. Turns out OpenAI and Anthropic made opposite bets on the same problem, and both are now converging.
-
Evals for skills: are you doing it?
Tests are to code what evals are to skills. Same job, different artifact. A short field guide to evaluating Claude skills with cassettes, graders, and a CI gate that doesn't lie.
-
FHE vs Garbled Circuits: Making MLaaS possible through Secure Computation
A simple walkthrough of how Fully Homomorphic Encryption and Garbled Circuits help make Machine Learning private, based on my recent paper comparing both for secure neural network inference.
-
Indirect Prompt Injection: Lessons from testing VoxPilot
Tested indirect prompt injection on my AI assistant - VoxPilot, and saw how easily these attacks can scale. Here's why everyone should care.