A year-long evaluation says ChatGPT can do a decent job of summarizing science research papers, but it is prone to hyping, misses nuances, and that it could ultimately break trust.