← Zongyang (Zane) Qiu

If GPT-6 Is Worth Following, Why Not Follow the Human Brain?

Zane Qiu · October 25, 2026

A couple of days ago I was chatting with a friend about where our research is headed next. I told him I want to figure out how to build intelligence that is purely vision-based, or, put another way, intelligence that doesn't depend on language.

He said his belief is different from mine. He thinks language and code can solve every problem.

Honestly, I don't think of this as a "belief." This kind of intelligence is right in front of every one of us, and I can't help being interested in it. I'm also very optimistic that LLMs can solve a lot of problems really well, and the evidence is already here: with language and code, large language models can do a huge range of things. Have one write a few hundred lines of code and you get a gorgeous 3D scene; hook it up to a robot and it can write control code, call a low-level policy, and finish real manipulation tasks.

The clearest example is GPT-6 Astra. It came out in September, and OpenAI didn't even position it as a robotics model, but once people plugged it into robots, it came first on several benchmarks by a wide margin. Now lots of people are trying to reproduce what it can do, even though we still don't know exactly what tricks or data went into it.

But however well this road goes, there is one question, the most central one scientifically, that it won't answer.

When we work on Artificial Intelligence, at the end of the day it's a science of intelligence. We want to understand the underlying principles of how intelligence comes about, and try to reproduce it with human technology.

Lately I've been reading about Core Cognition theory: humans (and many animals) come equipped very early with a few very basic cognitive systems. Babies who can't talk yet already know that a hidden object hasn't simply vanished. A cat has never studied physics, yet it knows exactly how hard to push off to land on top of a cabinet. Crows use twigs to hook bugs out of holes. None of them have language, but all of them are highly intelligent: they react to their environment, and they have cognition in the real sense of the word.

So in the end, no matter how strong LLMs get, the question of how this kind of intelligence forms won't get answered on its own.

Think about what made Hinton great. It wasn't some fancy demo, or how many points his model gained on a benchmark. It's that for decades he held onto one idea: building intelligence with networks that resemble the brain's neurons. For a long time few people believed in that path, but he found a way to train these networks, and in the end something like biological intelligence emerged from them. That is a scientific contribution. It tells us that learning structure from biology really can lead to intelligence.

Today's next-token prediction is one mode of intelligence, and a very powerful one. But it is still different from the way living things think.

So why is biological intelligence worth pursuing?

Back to GPT-6 Astra. It's so strong at embodied tasks that everyone has started following it, trying to replicate it. But let's not forget: the most capable embodied agent is the human being. Picking up a cup, tying shoelaces, walking down a crowded street without bumping into anyone: the human brain does all of this effortlessly, on about 20 watts, and most of the time language isn't involved at all. On this front, the human brain beats every robot model out there.

GPT-6 Astra has shown us a path that works. We don't fully understand how it works yet, but we're still trying to reproduce it. God has also shown us that biological intelligence works, with extremely high performance and high efficiency. I don't see any reason not to follow it and reproduce it too.

As an AI researcher, I'd rather answer the deeper question: from nothing but seeing, hearing, and interacting with the world, can intelligence emerge? And how does it emerge? I'd love to hear what you think!

If you'd like to talk about this, reach me through the contact options on my homepage.