ChatGPT may sound confident, but when tested on complex scientific claims, it often guesses and even contradicts itself. Researchers found it struggles especially with spotting false information.