Memory on LLMs is a big moat. My ChatGPT answers even using less power models give me better answers to my questions than all other models because of memory/context... This has been stronger than I originally imaged it to be!