Framing
On artificial general intelligence
Five different things get called the same word. This page separates them and states which one applies here, which is the weakest.
Five things that are not the same
| Kind of statement | What it would take | Status here |
|---|---|---|
| Research objective | Someone deciding what to work towards. | Applies. A direction, and nothing more. |
| Architecture target | A design intended to reach the objective. | Not published. |
| Hypothesis | A falsifiable statement with registered conditions that would refute it. | None published. |
| Experimental evidence | Controlled observation with negative controls and a stopping rule fixed in advance. | None published. |
| Qualified capability | Evidence, a tested scope, executed falsifiers and independent replication. | None. Zero published findings of any kind. |
| Achieved conclusion | All of the above, replicated independently, surviving sustained attempts to refute it. | Not claimed, and nowhere near. |
Why the distinction is enforced rather than promised
A north star is easy to convert into a product badge by a slow drift of adjectives. The defence here is mechanical: a deterministic rule set rejects public prose that exceeds its evidence state, including a fixed restricted vocabulary — proves, guaranteed, conclusively, state of the art, solves — and causal verbs are refused below an interventional causal level. These run in the build, and a violation stops publication rather than producing a warning.
The rules → · How qualification works →
What would change this page
A published claim record. Nothing else — not a demo, not a benchmark, not a capability announcement. If you ever read a strong claim here, check its coverage map first: it will tell you which parts of the sentence are carried by nothing.