Just asking, this seems very similar to the attention algorithm that powers LLMs?