New paper: : Can language models relearn removed concepts.