Matteo Capucci comments on What’s up with LLMs representing XORs of arbitrary features?