idk the field, what about JanAI & a Mistral model or LAION model outputing machiavellian biases?