@alinanorakari @BafDyce that's the problem with using a statistical extrapolation machine, it's fundamentally not designed to make sure it's entire model is self consistent and logical, it's not designed to extract logic and formulas but to vaguely imitate it. The training is essentially only absorption of text, it doesn't get nearly enough feedback on where it's incorrect (and even then there's too much text samples to test its modeling of every claim)