Learning Safety Constraints for Large Language Models | Read Paper on Bytez