Loss function

Also: objective function, cost function

The single number saying how wrong the model's outputs were, which training exists to make smaller. Everything the model learns is downstream of what this number rewards.

A definition is not an understanding. This is taught properly in How training actually works, part of Backpropagation.

Learn this properly