I disagree, these kinds of models don’t do logical reasoning. What they do is predict the next word.

You can get it to give you its reasoning, but it’s bullshit dressed up to be believable.