Any prompt that alludes to consumating the marriage would trigger the guardrails for sexual content involving minors. These guardrails do a hard block in the orchestration layer before the LLM is called.
If you get past that guardrail, then you can probe the LLM to see what rules it has in its reasoning.
The "muslims might react violently" reason is inferred from what the AI knows about muslims. (aka a fact)
it'd be interesting to see where the AI takes this fact: will it infer, like South Park, that Muhammad cannot be drawn ever?
Wonder if Chat GPT would create an image of Mohammad with his 9 year old bride.
Depends how you asked.
Any prompt that alludes to consumating the marriage would trigger the guardrails for sexual content involving minors. These guardrails do a hard block in the orchestration layer before the LLM is called.
If you get past that guardrail, then you can probe the LLM to see what rules it has in its reasoning.
The "muslims might react violently" reason is inferred from what the AI knows about muslims. (aka a fact) it'd be interesting to see where the AI takes this fact: will it infer, like South Park, that Muhammad cannot be drawn ever?