https://blog.qife122.com/tags/%E6%80%9D%E7%BB%B4%E9%93%BE/ Recent content in 思维链 on 办公AI智能小助手 Hugo zh-cn qife Fri, 03 Oct 2025 00:06:01 +0800 https://blog.qife122.com/p/%E5%A4%9A%E6%99%BA%E8%83%BD%E4%BD%93ai%E7%94%9F%E6%88%90%E6%80%9D%E7%BB%B4%E9%93%BE%E8%AE%AD%E7%BB%83%E6%95%B0%E6%8D%AE%E6%8A%80%E6%9C%AF%E8%A7%A3%E6%9E%90/ Fri, 03 Oct 2025 00:06:01 +0800 https://blog.qife122.com/p/%E5%A4%9A%E6%99%BA%E8%83%BD%E4%BD%93ai%E7%94%9F%E6%88%90%E6%80%9D%E7%BB%B4%E9%93%BE%E8%AE%AD%E7%BB%83%E6%95%B0%E6%8D%AE%E6%8A%80%E6%9C%AF%E8%A7%A3%E6%9E%90/ <h1 id="多智能体ai生成思维链训练数据">多智能体AI生成思维链训练数据</h1> <p>思维链推理要求大语言模型不仅执行多步操作，还要解释采取每个步骤的原因，这已被证明能提升大语言模型的推理能力。思维链推理的一个有前景的应用是确保大语言模型遵守负责任AI政策。</p> https://blog.qife122.com/p/%E5%A4%9A%E6%A8%A1%E6%80%81%E6%8E%A8%E7%90%86%E8%87%AA%E5%AF%B9%E9%BD%90%E7%AD%94%E6%A1%88%E5%AF%BC%E5%90%91%E6%80%9D%E7%BB%B4%E9%93%BE%E6%8A%80%E6%9C%AF/ Tue, 09 Sep 2025 21:46:28 +0800 https://blog.qife122.com/p/%E5%A4%9A%E6%A8%A1%E6%80%81%E6%8E%A8%E7%90%86%E8%87%AA%E5%AF%B9%E9%BD%90%E7%AD%94%E6%A1%88%E5%AF%BC%E5%90%91%E6%80%9D%E7%BB%B4%E9%93%BE%E6%8A%80%E6%9C%AF/ <h1 id="从答案到推理依据基于答案导向思维链的多模态推理自对齐方法">从答案到推理依据：基于答案导向思维链的多模态推理自对齐方法</h1> <p>实现多模态大语言模型（MLLMs）类人推理能力一直是研究目标。当前方法主要侧重于合成正向推理依据，通常依赖人工标注或复杂系统，且往往忽略负向推理，这限制了模型在多模态推理中的泛化能力和鲁棒性。</p>