Raw LLM Responses

Haha, yeah maybe. My guess is that certain guardrails are going to be trained into any child-facing ML model though. Likely things like murder, rape, drug abuse, assault, etc. just aren't going to be options when building a curriculum. Meanwhile things like pointing at-risk kids to the suicide hotline or pointing abused children to law enforcement are probably going to be default. Depending on the mix, harmful radicalization may be more difficult to achieve and weirdo parents who hurt their kids may be unexpectedly charged and separated from their victims, for example.