实验结果显示， 使用 CoT 提示的模型表现显著下降，尤其是 OpenAI o1-preview 模型的准确率下降了 36.3%。 这表明当模型过度依赖逐步推理时，CoT 可能会抑制其对隐性统计模式的学习能力。

With the inaugural World Conference of Classics set to open in Beijing on Thursday, scholars from around the globe are ...

More than 40 foreign experts in classical studies from 13 countries visited Henan's heritage sites, Longmen Grottoes and the ...