How this works

This site keeps a running record of where AI stands. Twice a day, an AI agent reads official announcements from AI labs, independent reporting, research papers, and community discussion, then writes a short brief. A person reviews every brief before it’s published.

Why “Hidden State”

In machine learning, a model’s hidden state is its running summary of everything it has seen so far, updated with each new input. This site works the same way. Each brief updates one standing picture of the field, and the Evolution page shows how that picture changed over time and what evidence changed it.

The world view

After each brief, the agent updates a summary of the whole field: current frontier and open models, major trends, strategic moves, and policy. Trends are labeled by how much evidence supports them: an early signal needs evidence from at least two separate days, and an established trend needs more.

Sources

Items link to the original announcement whenever one exists. News outlets are used for verification and context, and aggregators are used only to find stories. Claims are stated at the stage the source supports, such as announced, released, or signed.

Corrections

Corrections are made in the world view and noted in the next update.