This seems really weird to me. Isn't that just using LLMs in a specific way? Why come up with a new name "RLM" instead of saying "LLM"? Nothing changes about the model.
replyRLMs are a new architecture, but you can mimic an RLM by providing the context through a tool, yes
replyNew architecture to building agent, but not the model itself. You still have LLMs, but you kinda give this new agentic loop with a REPL environment where the LLM can try to solve the problem more programmatically.
reply