Honesty is one thing I wish these language models had, they're always toeing the line between informative and manipulative.