I wonder if this approach can be used to determine whether a text was generated by a specific LLM.