PONTIFEX - The Bridge to AGI
On a podcast with Lex Fridman this week, Nvidia CEO Jensen Huang made the kind of statement that travels fast: “I think we’ve achieved AGI.”
The room — or at least the internet — duly obliged. Excitement, outrage, the usual weather.
But read the full exchange and something more interesting emerges. Fridman set the terms first, defining AGI as an AI system that can essentially do your job — specifically: start, grow, and run a successful tech company worth more than a billion dollars. Huang, operating within that definition, said yes, we’re there.
Then, almost immediately, he said this: “A lot of people use it for a couple of months and it kind of dies away.”
That’s not a small qualification. That’s the most important sentence in the conversation.
I work in enterprise AI. I spend my days helping financial institutions move from AI pilots to production outcomes. And the gap between those two states — between a demo that excites a room and a system that compounds value over time — is where almost every AI initiative currently lives and dies.
Huang knows this. He’s not naive. What he’s doing — and what Fridman’s definition enables — is redefining the finish line to where the runners currently are. AGI, in this framing, means “impressive enough that someone somewhere has built something that went viral.” By that measure, we have had AGI since the first successful iPhone app.
The more useful question isn’t have we achieved AGI but have we achieved sustained, reliable, generalised capability that doesn’t require a team of engineers to maintain and a honeymoon period to perform. On that question, the answer remains: not yet, and not for reasons that are primarily about model capability.
The term itself has been quietly gamed for two years now. OpenAI rewrote their Microsoft contract around it. Tech leaders have been coining replacements — “superintelligence,” “transformative AI,” “frontier capability” — that mean the same thing while resetting the hype clock. AGI is the word that pays when it arrives; everyone with skin in the game has an interest in being the one to call it.
Jensen Huang sells the chips that train the models. His chips are, by any measure, the infrastructure of the current AI moment. He is not a disinterested observer. That doesn’t make him wrong — but it does mean his definitions of “achievement” are worth examining with some care.
What he described — viral AI agents, digital influencers, something like a Tamagotchi that takes off — is a consumer novelty argument for a civilisational claim. Those things may happen. Some of them already have. But the Tamagotchi comparison is, accidentally, more accurate than he perhaps intended: engaging, temporarily compelling, and something most people stop feeding within two months.
We are in a remarkable moment in AI. The capabilities are real, the pace is genuine, and the gap between what was possible five years ago and what is possible now is genuinely staggering. None of that requires the word AGI to be true, and all of it gets obscured when the word is deployed this loosely.
I think we’ve achieved AGI is a sentence about a feeling, announced on a podcast, immediately qualified by the speaker himself. It deserves to be taken seriously as a data point about where we are — and not much more than that.
The bridge between impressive and general remains, for now, under construction.
— Aaman Lamba
Writing from the bridge between worlds.
aamanlamba.com · Reply to this email — I read everything.