I'm not so sure. LLMs are more like human simulators and are pulling this "reasoning" from what they've seen in the training data. It therefore might be possible that they cannot be more intelligent than those who wrote the source material.