Comment Re:AI should be banned in the court room * (Score 2) 65
A language model produces language based on its training data. There is no way for it to produce something that is true, it just produces tokens which have a high probability of following the previous tokens. If your training data is legal documents then it will just be very good at producing legal documents that *look* right.
So maybe an AI can do what you say, but it won't be one that is based on the language models we're seeing today.