Here is an investigation of how different queries are classified as hateful vs not hateful in ChatGPT: https://davidrozado.substack.com/p/openaicms