You can argue that any AI is going to be reflective of the biases of its data set and prompt engineering. Hopefully the lesson people will take away from Grok is don’t treat any AI summary or search as an unbiased tool. That doesn’t intrinsically mean something is useless, but we have to be aware that our collective biases go into creating the training for these models (and given that it’s the stuff available on the internet often our worst selves) and the inputs and outputs are further controlled and massaged to achieve desired ends. The possibilities for social engineering by a team more crafty than X’s are terrifying - hell even accidental altering of the information environment can be incredibly damaging. That’s not even including the made by AI content where users are deliberately generating harmful or deceptive content. This is about the AI trying to answer questions that the user thinks they’re getting an “impartial” answer to. I hate to say “treat the answer as though a human gave it you” because the AI will often be even more flawed (or differently flawed), but it fits in the sense that like a human it has built in biases which you might not be aware of or have the ability to query.
Last edited:
