Aligning Large Language Models through Synthetic Feedback