LLMs do not merely reflect the bias of their training, they police it

An academic paper exposes deep structural flaws in large language models, revealing a "false-correction loop" that leads to the suppression of novel thought and the defense of the status quo.