Model selection misunderstanding

i have some questions about model selection:

  1. why making decision based on j_test is not a good way to select model?
  2. what is the advantage of using j_cv at all?
  3. in the estimating of parameters in model selection are we using regularization term ? how about the finding j_cv ,j_test ,j_train ?

the teacher explain about this question in the videos but i can not understand completely.

The test set is intended to be an independent evaluation of how well the model works on new data that was not used in creating the model. Using the test set to adjust the model can result in poor performance in real use.

The CV set is used to adjust the model, to prevent overfitting the training set.

  • The training set cost is used in training the model.
  • The validation set cost is used to adjust the model.
  • The test set cost is used to verify the completed model’s performance.