“Axiom believes that simply scaling informal language models is not a viable path to achieving Artificial General Intelligence (AGI) in mathematics.”