That 1-2-3 framework can't be applied to LLMs because they are full of human bias by definition, as they are trained on human language. We will have to find another way.

Alphago is an example of an AI that has no human bias, because it learned go with self play without seeing any human game.