Plus: Google DeepMind's watermark. AI language models are not humans, and yet we evaluate them as if they were, using tests like the bar exam or the United States Medical Licensing Examination ...