A new study put ChatGPT to the test by asking it to judge whether hundreds of scientific hypotheses were true or false--and the results were far from reassuring. While the AI got it right about 80% of the time on the surface, its performance dropped ...