Researchers took one of the world's biggest AI models and turned off some of its neurons. The model kept reasoning perfectly. It just started making grammar mistakes.

Then they turned off a different group of neurons. This time, its grammar stayed perfect, but its reasoning completely broke down.

If that sounds familiar, there's a reason. A very similar thing happens in the human brain. Damage one area, and a person may struggle with language. Damage another, and they may lose reasoning or problem-solving abilities.

The interesting part is that nobody designed the AI this way. Engineers never created separate areas for grammar or reasoning. During training, the model organized itself into those specialized parts on its own.