Researchers searched for a signal of pain inside large language models—and they may have actually found one.