Sam Altman posted something on X yesterday that I haven’t been able to stop thinking about:
“I am approximately as amazed by this cognitive feat as I am by GPT-5.6 discovering new math.”
The cognitive feat in question was his older kid putting two words together for the first time.
That’s it. That’s the bar. “More milk.” “Dada up.” The kind of thing every parent on earth experiences and thinks “aw, that’s cute” before going back to not sleeping through the night.
Sam Altman is equally amazed by that as he is by a multi-trillion-parameter neural network discovering a new mathematical theorem.
The Accidental Honesty of Sam Altman’s Tweet
I mean this genuinely, not just as setup for the jokes. Sam Altman accidentally told you something real about how AI executives think.
The entire AI industry runs on a framework where “model did something impressive” and “human did something developmentally normal” are measured on the same scale. They talk about “capabilities” and “milestones” as if GPT-5.6 solving an abstract math problem and a child learning object permanence are comparable events.
They’re not. But that’s the language they’ve built for themselves, and Sam just admitted he speaks it fluently.
Sam Altman is the CEO of the most valuable AI company on earth. He spends his days staring at benchmark charts and capability curves. Of course his brain has been rewired to see everything through that lens. His kid said two words for the first time, and somewhere in his neural pathways, that fired the same dopamine signal as watching Sol return its first inference.
I’m not mocking the guy for being a proud parent. I’m pointing out that living inside the AI hype cycle for seven years has reshaped his brain’s reward system to the point where basic human development and superhuman model performance occupy the same emotional register.
The Version Numbers Keep Getting Weirder
Also worth noting: this tweet references “GPT-5.6.” Not GPT-5. Not GPT-6. GPT-5.6. The version numbering has become software-release calendar cancer, and Sam Altman is out here tweeting about his kid in the same post he name-drops a point-release of his AGI pursuit.
GPT-5.6 sounds like a minor patch to fix a memory leak in the attention mechanism, not a “model discovering new math.” At what point did everyone just accept that the model names would be this ugly? Picture Isaac Newton discovering calculus and calling it Principia Mathematica v1.4.2. Doesn’t have the same ring.
The Real Joke Is on Sam
I don’t think Sam realized what he was actually saying when he hit post.
He’s right.
A human brain learning language for the first time IS more impressive than anything GPT-5.6 has done. A two-year-old wiring syntax, semantics, and social context from raw immersion in a chaotic environment (no labeled dataset, no RLHF, no goddamn system prompt) is a genuinely more remarkable cognitive feat than a statistical model that ate the entire internet and learned to pattern-match at scale.
Sam compared his kid’s achievement to GPT-5.6 discovering new math, and he framed it as GPT-5.6 being the impressive one. But he accidentally made the opposite argument. That kid will be having full conversations in six months. GPT-5.6 will be obsolete in twelve. The kid’s model is better.
Maybe the takeaway isn’t that Sam is weird for being amazed by his kid. Maybe it’s that everyone’s weird for being amazed by language models doing something a three-year-old does for breakfast.
Anyway. Congrats to Sam’s kid on the language milestone. That’s a real achievement. When does GPT-5.7 ship?
Previously on Tony Reviews Things: GPT-5.6 Is Here: Sol, Terra, and Luna Launch in Government-Approved Limited Preview
