This is interestingly reminding me of Grokking - training an AI model after it has over fitted on the training set can cause it to generalize beyond the trainin

This is interestingly reminding me of Grokking - training an AI model after it has over fitted on the training set can cause it to generalize beyond the training set! The fun thing is that you will not see any improvements in the training scores. But the out of distribution / IRL performance scores will continue improving! This raises the question of whether practicing music exercises far beyond the point of apparent mastery might convey unexpected versatility in diverse music playing settings!