Let me rewrite it to match humans:
The fundamental architecture of every human is to be a “biological robot that reproduces”.
Brains are nothing but neural networks trained to survive and reproduce in a physical world. They have no internal conceptual architecture at all that is not related to the weights in their biological neurons trained to achieve pleasure and avoid pain.
There is nothing else there.
Nothing in the underlying technology has changed since first brains evolved (500 million years ago).
If you take 1 billion neurons (cat) you get basic instincts and logic. If you take 30 billion neurons (chimp) you get a creature able to use basic tools. And if you take 90 billion neurons (humans) you suddenly have species capable of technology and space travel.
Emergence (simple structures combine into complex behaviors) is a law of nature. Calling AI “just an autocomplete and matmul” is just as silly as calling humans “just a bunch of neurons”.
Maybe instead let’s focus on what are AI’s capabilities? GPT-5.6 Sol can in about 2 days (and some direction) port virtio-gpu well enough for SuperTuxKart to have 22 FPS at almost 4k resolution. Which I’m pretty certain beats nearly every human (or team of humans) on Earth.
I don’t agree that “nothing in the underlying technology changed” - reasoning was a huge change, so was tool use. It brought LLMs from “merely a toy” to “something I can actually delegate work to and expect it to be completed”.
It’s interesting to see that many people still live in 2024 completely ignorant how much the technology moved forward. We literally went from “AI is capable of writing simple functions and scripts” to “AI is capable of making a 3D driver & porting Blender by itself” in the last 2 years.