Academic Papers
Deep dives into the research papers powering modern AI. Reviewed and explained for working engineers.
The Signal Protocol: The Encryption Behind Two Billion Conversations
A technical deep dive into the X3DH key exchange, Double Ratchet algorithm, and the cryptographic design that protects Signal, WhatsApp, Google Messages, and Facebook Messenger.
Every message on WhatsApp, Google Messages RCS, Signal, and Facebook Messenger uses the Signal Protocol. This article walks through the full protocol stack, from the initial key exchange to per-message encryption, with interactive diagrams.
Heretic, Obliteration, and the Growing Toolkit for Removing Safety From LLMs
A security research review of abliteration, the Heretic framework, OBLITERATUS, and every other method people are using to strip safety alignment from open-weight models.
Open HuggingFace and search for abliterated. You will find over 5,000 models. This article reviews the techniques, tools, and debate behind removing safety alignment from language models, from the foundational NeurIPS 2024 discovery to one-click tools like Heretic.
REAP: How Cerebras Is Cutting Trillion-Parameter Models in Half
Router-weighted Expert Activation Pruning is reshaping how we deploy frontier MoE models
REAP removes up to half the parameters from Mixture-of-Experts language models while keeping most of their capability intact. The method has gained rapid traction, with pruned checkpoints appearing across GLM-5, DeepSeek-V3, Qwen3, and Kimi-K2.