New Paper 🎉: Can language models relearn removed concepts?