I disagree.

AI is very good at conforming to your own biases and pulling out the subtext of a prompt.

If your prompt goes along the lines of "I think x is healthy plan a meal for x", grok (and other AI) will happily affirm that you are correct and really smart for recognizing that "x" is the healthiest diet and then it'll give you that diet.

That's a biased answer. AI biases to your own biases.

Or maybe said another way. AI starts with the baseline assumption that you are an expert and correct in your prompt. It can be hard to get an AI to call you out for being wrong about something.