Text this: From Explicit to Implicit: A Theoretical Framework and Transfer Method for Preference Internalization in Language Models