Elon gives an unvarnished look at what they mean by 'unbiased' with respect to models. It's rewriting the training material or adding tool use (searching for Musk's tweets about topics before deciding it's output) to twist the output into ideological alignment.