Why does SD go crazy when you apply a high weight to a prompt term?

[CalculonPrime]

I've found that if you increase a prompt term too much, stable diffusion goes nuts. For example, if you have a LORA that's not being applied (let's say a LORA to draw an actress), so you try nudging it more, you can write <myLora:1.5> instead of the default weight of 1. But anything above 2 or so will cause the entire image to grossly distort. In my case, if I write <myLora:10> the image looks like an explosion in a paint factory.

Can someone explain what's happening internally that causes such weird behavior with higher weight values?