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

Isn't this basically the mythical man-months LLM edition?