But that was kind of my point
Maybe it'd be easier to train the models on key parts of the legal code and give it a hard aversion to breaking the law - rather than training on vague value judgements and then hope the model doesn't break the law
Maybe it'd be easier to train the models on key parts of the legal code and give it a hard aversion to breaking the law - rather than training on vague value judgements and then hope the model doesn't break the law