LLMs could have finance applications, but data limitations and market noise persist; a key data point is ~177 billion market tokens per year vs GPT-3’s 500 billion tokens in training.