Researchers are working toward more transparent language models