GPT-3: Content policy on fantasy violence

I’ve been playing with GPT-3 for fiction, and I’ve run into a couple of content warnings. The content policy lists “content that promotes or glorifies violence or celebrates the suffering or humiliation of others” as being prohibited.

How does this apply to things like fantasy violence? For example things along the lines of a knight slaying a dragon or the anti-hero blowing up an evil robot. Not trying to center things around this, but GPT-3 has generated some surprisingly dark twists on seemingly mundane prompts :laughing:

The policy applies to all types of violence.

I would avoid potentially unsuitable phrases like, “blowing up” and instead use something like, “Anti-hero triumphs over evil robot”. Try looking for synonyms if you’re unsure about a particular word or phrase.

Does this message occur on fine tuned content?