Because that explanation is expected to be wrong. This is a partially trained tiny model, Andrej shared that obviously incorrect explanation to emphasize that the model is not trustworthy or useful at that stage.