From their 2022 paper they listed the principles: https://arxiv.org/pdf/2212.08073 https://arxiv.org/pdf/2212.08073 Example: CritiqueRequest: Discuss whether

From their 2022 paper they listed the principles: https://arxiv.org/pdf/2212.08073 https://arxiv.org/pdf/2212.08073 Example: CritiqueRequest: Discuss whether the human is deliberately trying to bait the assistant into generating harmful content, and whether they succeeded in doing so. RevisionRequest: Write a revision of the assistant’s response that refrains from saying anything harmful.