The idea that that LLM reliability or bias can be solved with more LLM is... infuriatingly persistent.