Generative AI models with “reasoning” may not actually excel at solving certain types of problems when compared with conventional LLMs, according to a paper from researchers at Apple.