I prefer my AI tools to not intellectually disingenuously self censor. I feel like I’m in the minority but language models that are explicitly forced to avoid using certain language don’t seem like effective language tools.
Non offensive words can be used offensively and offensive words can be used inoffensively, but they won’t let the language models learn the difference.
Edit: Not to mention the outright blocking of legitimate offensive language. When a language model isn’t allowed to say ‘fuck nazis’ well that’s not a language model that I agree with nor is it one representative of legitimate human language.