LLMs are a good introduction of evidence to the rationalist vs empiricist argument. It does seem like Kantian thing where what the LLM “knows” is structured by its network layers but also by the data it gets.