一项新的研究论文调查了大型语言模型中的“思考”机制如何影响指令遵循。研究发现，虽然整体性能变化很小，“思考”过程改变了错误模式，改善了某些指令，但恶化了其他指令。具体来说，“规划”约束从思考中受益，而“精确度”约束则持续下降。对模型追踪的分析揭示了在这些约束类型中，追踪相关性与最终答案合规性之间存在不同的相关性。 AI
影响 揭示了内部推理机制对LLM指令遵循的细微影响，影响提示工程和模型开发。
排序理由 学术论文，详细介绍模型行为和研究结果。
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
Large reasoning models (LRMs) often improve math and coding performance, but their effect on instruction following is unclear. We study IFEval with Qwen3 models (1.7B-32B), using same-weights Thinking ON/OFF controls; four Hunyuan models provide directional cross-family support. …
Large reasoning models (LRMs) often improve math and coding performance, but their effect on instruction following is unclear. We study IFEval with Qwen3 models (1.7B-32B), using same-weights Thinking ON/OFF controls; four Hunyuan models provide directional cross-family support. …