[fouc]

They have many ways to manipulate the LLM's results, for example they can use a lot of the same mechanisms that are used to block or filter out inappropriate material.

[krainboltgreene]

Given that there are entire forums devoted to successfully doing just that (easily) my point stands.