Featured article
Consensus Is Not Corroboration
Why language models need to distinguish a manufactured echo from genuinely independent confirmation.
Read article ↗All writing
8 entriesTeaching an Agent Fleet to Distrust Documents: Zero-Trust M&A Diligence on Google Cloud
How Diligence Room uses isolated agents, hostile-document screening, and evidence gates to make autonomous deal diligence defensible.
The Tiny Draft Model Hidden Inside Qwen
Qwen ships a one-layer multi-token predictor that vLLM and SGLang can use for native speculative decoding.
Do LLM Agent Societies Adapt Their Values, or Eventually Die by Them?
What an evolutionary multi-agent simulation reveals about cultural persistence, selection, migration, and the danger of values that never bend.
Consensus Is Not Corroboration
Why language models need to distinguish a manufactured echo from genuinely independent confirmation.
Why Qwen Rotates Only a Quarter of Each Attention Head
Partial RoPE gives Qwen a compact relative-position channel alongside a larger RoPE-free similarity subspace.
Attention Should Be Allowed to Say No
Qwen's Gated Attention separates where an attention head reads from whether its output should influence the model.
The Case for Planner–Executor Architectures in Agentic Coding
Why serious coding agents should separate judgment from execution—and spend frontier-model capability where it matters most.
Attention Lab: Eight Ways a Model Looks
Interactive field guides to the attention mechanisms shaping modern language models.