Consensus Is Not Corroboration

Why language models need to distinguish a manufactured echo from genuinely independent confirmation.

Read article
Agentic systemsEssay · Reliable Agents

Teaching 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.

6 min read
Attention & modelsEssay · Qwen Hybrid Architecture

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.

6 min read
Agentic systemsEssay · Reliable Agents

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.

15 min read
Agentic systemsEssay · Reliable Agents

Consensus Is Not Corroboration

Why language models need to distinguish a manufactured echo from genuinely independent confirmation.

10 min read
Attention & modelsEssay · Qwen Hybrid Architecture

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.

6 min read
Attention & modelsEssay · Qwen Hybrid Architecture

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.

7 min read
Agentic systemsEssay

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.

7 min read
Attention & modelsInteractive lab · Modern Attention

Attention Lab: Eight Ways a Model Looks

Interactive field guides to the attention mechanisms shaping modern language models.

Explore