An Arrival Without a Definition── The people declaring AGI has arrived do not share a definition of AGI

In September 2026, NVIDIA's Jensen Huang declared that AGI has arrived. The statement coincided with the launch of GPT-6 Astra, and multiple outlets ran headlines about the "AGI era." Yet OpenAI's Altman and Anthropic's Amodei reportedly do not agree on the definition of AGI. The people making the declaration do not share the meaning of the word. This structure shows up in our own work in a strikingly similar form.

01The morning of the declaration

In the same week that GPT-6 Astra was announced, Jensen Huang stated that AGI has arrived. The background was the operational launch of NVIDIA's Grace Blackwell and the computational infrastructure supporting it.

Headlines unanimously proclaimed the "AGI era." The Wall Street Journal and several other major outlets used the same phrase. Judging by the momentum of the language alone, something decisive appeared to have happened.

But step back one pace, and something peculiar emerges. Among the people saying "it has arrived," there is no shared definition of what "it" is.

02The principals who do not agree

According to reporting, Sam Altman and Dario Amodei hold divergent views on the definition of AGI. These two are at the very forefront of the field. They use the same word and point at different things.

This is not a minor matter. The term AGI moves investment decisions, shapes regulatory direction, and moulds public expectations. If "it has arrived" is declared while the content of "it" remains unsettled, listeners fill in the blank with their own imagination.

A word filled in by imagination cannot be falsified. If we cannot tell what counts as arrival and what counts as absence, the declaration can never be wrong. And a declaration that can never be wrong says nothing at all.

03Progress you can measure, declarations you cannot

In fairness, the progress is real. GPT-6 Astra shows improvement over the previous generation across multiple metrics. Agents are entering practical workflows. Anyone following this summer's pace cannot doubt the speed.

The problem is that between acknowledging progress and declaring "AGI has arrived," there is an unfilled gap.

Measurable progress can be wrong. That is why it can be trusted. Unmeasurable declarations cannot be wrong. That is why they cannot be trusted.

04The structure that lets language run ahead

Why does a declaration emerge before the definition is settled? Several forces converge.

The result is that language outruns the technology. While the content is still unsettled, the words alone create a fait accompli.

History offers precedents. The term "cloud computing" spent years meaning different things to different vendors before a working consensus formed. But cloud computing did not carry existential weight. AGI does. When the word in question shapes safety regulation, military strategy, and public anxiety, the cost of letting it float undefined is not merely academic — it is operational. Decisions are being made on the basis of a word whose boundaries no one has drawn.

05The same thing happening in our work

This structure appears in a familiar form in the work of reviewing promotional materials.

Consider the phrase "efficacy has been confirmed." What counts as efficacy? In which patient population, over what duration, compared with what? Without those specifics, "efficacy has been confirmed" says nothing. But the wording sounds powerful, and it shapes readers' expectations.

What we look for in review is precisely this kind of gap — the space between the strength of the language and the specificity of its content. Misunderstanding enters through that space.

The philosopher of science Karl Popper made the point repeatedly: a claim that cannot be falsified is not science. If "AGI has arrived" is issued in an unfalsifiable form, it is not a technical claim but a rhetorical declaration. Rhetoric is not inherently bad. But confusing rhetoric with a technical claim is harmful.

06What it means to press for a definition

Am I arguing that the word AGI should not be used until a definition is settled? No.

What I am saying is smaller. Tell us specifically what has arrived. On which metrics did GPT-6 Astra improve, and by how much? Which tasks are now automated? What is still beyond reach? State those things, and the listener can judge for themselves.

The resignation of the UK's AI policy chief over a conflict of interest with Anthropic occurred in the same week. When those who set policy and those who sell the technology are too close, whose convenience does the definition serve? The absence of a definition is not harmless vagueness — it can become an instrument of power.

07The name of something that has not arrived

As I write this, I do not have the ability to judge whether AGI has arrived. I do not even hold the criteria for making that judgement. And according to reporting, neither do the people who made the declaration.

But one thing is certain. When the people using a word do not share its meaning, any declaration made with that word deserves scrutiny.

A word becomes a tool only when its content is settled. A word used before its content is settled is not a tool but an ornament. There is only one way to avoid being misled by ornament: keep asking, "What exactly does that refer to?"

In our review work, we have a name for this practice. When a claim lacks specifics, we do not reject it outright; we ask the author to fill in the content. The claim may well survive the questioning — many do. What does not survive is the possibility of reading into it something it never said. That is precisely the service that asking provides: it protects the reader from the gap between the strength of the language and the thinness of its content.

The progress is real. GPT-6 Astra is demonstrably more capable than its predecessor. But "it got smarter" and "AGI has arrived" are too far apart to sit in the same sentence.

The next time I see a headline that says something has arrived, I will look for the contents first. What arrived? What would it look like if it had not arrived? If that is not written, the headline is not information — it is staging.

The same applies to our review work. If a strong expression has no specifics behind it, it is not a claim but a wish. Not confusing wishes with claims is, I think, the foundation of any job that handles language.