GPT-5.1 was shown a folded piece of paper with holes punched through it. The task was simple. Imagine unfolding the paper. Count the holes.

It scored 6%.

Gemini 3 Pro got 27%. o3 got 13%.

These are the models running multi-billion-dollar AI labs. The ones doing math olympiads. The ones writing production code. The ones we keep calling "PhD-level."

And they can't do something a 4-year-old does with a pencil and a piece of paper.

Just think about that for a second.

Now try this. Spill a glass of water on the table. Don't actually do it. Just imagine it.

You didn't think it in sentences. You SAW it happening. The glass tipping. The water spreading. Your hand reaching for a towel before the rest of the scene even finished.

That whole scene played out in your head as pictures. Not words.

This is the thing AI cannot do.