An observatory for intelligence

Why this exists

The question of when — and whether — machines reach general intelligence is being argued in fragments: benchmark announcements, laboratory blog posts, conference papers, and timeline claims that are made constantly and scored almost never.

The fragments are not the problem. The absence of a structure that holds them together is. Agitology is that structure: a single framework where a paper, a deployment result and a regulatory decision can each be assessed for the same thing — whether they change what we should believe about progress toward general intelligence.

What we publish

Four kinds of thing, never blurred together:

• News — what happened, with the source named and its Index impact scored.
• Research — what was discovered, with limitations printed as prominently as findings.
• Analysis — what it means, written by the desk and revised when scores move.
• Perspectives — what other people think, published under their name and not ours.

Underneath all four sits the AGI Index: twelve weighted capability dimensions, each with published evidence, published counter-evidence, and a confidence rating that is separate from its score.

What we refuse to claim

That AGI progress can be objectively measured. It cannot. There is no agreed definition, no unit, and no instrument. The Index is an argued framework that produces a number, and its value is that it is explicit enough to be wrong in a specific way rather than vague enough to be unfalsifiable.

We publish the threshold being scored against, the weights, the evidence ordering, and the ways the number misleads. If you disagree with the result, the framework is laid out clearly enough for you to say exactly where.

On this preview build

Everything currently on this site is illustrative sample content, written to demonstrate the product rather than to report anything. Specifically:

• The news items describe invented developments. Organisation names are real; the developments attributed to them are not.
• The papers are invented, with fictional authors at real institutions.
• Every named contributor is fictional.
• Source and paper links are deliberately disabled so nothing can be followed as though it were sourced.
• The Index scores are illustrative values demonstrating the scoring framework, not estimates of anything.

The preview strip at the top of every page says so, and every content card carries a sample marker. Both disappear automatically once real editorial material replaces the seed content.

Contributing

We publish arguments, not announcements. The pitch most likely to be commissioned names a specific Index dimension, says the score is wrong, and states what evidence would settle it. Practitioners writing about the gap between what their field claims and what it can demonstrate are especially welcome.

Less likely: product announcements, capability claims without evidence, and timeline predictions without preconditions.

Contact

Corrections, disputes, pitches and data requests all go to the same place. A claim that a score is wrong, with the evidence, gets read first. See the editorial policy for how corrections and conflicts are handled.