For all its knowledge it can't solve even the most basic problems accurately - but what do you expect from a language model?