Shapley Value Attribution in Chain of Thought — LessWrong