Self-Reflection Makes Large Language Models Safer, Less Biased, and Ideologically Neutral

Fengyuan Liu, Nouar AlDahoul, Gregory Eady +2

Previous studies proposed that the reasoning capabilities of large language models (LLMs) can be improved through self-reflection, i.e., letting LLMs reflect on their own output to…