What is interesting about LLMs is that they are probabilistic, however much they are refined by RLHF, which is not how thoughts/words are put together. There has to be a more efficient path for making logical distinctions.