I would argue that “info retrieval” is also something the LLM space has yet to yet to solve to a human level of reliability, but I think your comment is right on. I see this all as part of the greater symbolic vs. stochastic (/neat v scruffy) dynamics