Apple researchers conducted a study on LLMs to evaluate their mathematical reasoning abilities and found that these models rely on probabilistic pattern-matching, not formal reasoning. They recorded ...