Is there any research using multiple expert llms rather than one general purpose

[intro] It seems to me that general purpose models will always suffer from confusion. One can clearly make specialist LLMs. Maybe one that is expert at math, Another that is expert at art history, and another on pop culture etc. [verse 1] Then have a coordinator AI that can take requests, Determine which model or models need to be used To produce the most accurate result. It seems to me that the human brain does this. [chorus] We train how to walk. Then that part of the brain takes over walking And the rest of the brain doesn’t need to get involved. LLMs the way they are now will always require guardrails And censoring to keep them from going off the deep end. [verse 2] How do we get to reliability without all that? The quest for an AI that's both wide and deep, Mimicking the human brain's own neat trick, Separate yet integrated, specialized yet coordinated. [bridge] Looking for the path to an AI that's refined, A coordinator, a specialist, intertwined. Merging the best of all