A new Apple research paper finds that large reasoning models (LRMs) currently offer limited reasoning abilities beyond standard large language models, highlighting significant challenges in developing truly human-like AI reasoning.