Can someone explain what "controlling reasoning with system prompt" means?

The model's reasoning capabilities can be controlled via a system prompt.

I tried by putting "Do not think" in the system prompt but that doesn't seem to help either, the model still thinks.