LLMs are the ultimate amplification algorithm.

They lock in whatever things were dominant when they were trained.

A gravity well pulling any line of discourse back to it.