LLMs can reduce their bias in multi-turn conversations (b-score.github.io)