Ultimately it’s predicting words based on the context given for all the good and bad of that. It’s a very good compression and search mechanism, and capable of far more than we thought we’d be able to do with statistics on language. But talk of it thinking or being biased or whatever is a...