“Adam D'Angelo does not believe a brand new architecture is needed for continued LLM progress, although he notes continuous learning and memory are difficult challenges for current architectures.”