“Jürgen Schmidhuber argues that current large language models are heavily biased towards human perspectives because they are trained exclusively on human-generated data from the World Wide Web.”