从动力系统与熵相变视角揭示LLM推理的触发条件，为理解大模型认知机制提供新洞见。
arXiv:2605.22873v1 Announce Type: cross Abstract: Chain-of-thought (CoT) reasoning has become the default strategy for enhancing LLM capabilities, yet…
揭秘稀疏自编码器如何揭示大脑与LLM的语义对齐模式，探索皮层拓扑结构中的智能映射
arXiv:2605.23035v1 Announce Type: cross Abstract: Intermediate layers of large language models (LLMs) best predict human brain responses to language, …
颠覆认知？弱教师模型也能有效蒸馏LLM，预训练阶段教师强度并非关键。
arXiv:2605.23857v1 Announce Type: new Abstract: Knowledge distillation generally assumes a strong-to-weak relationship where stronger teachers yield b…
让LLM智能体学会检索外部技能，摆脱显式枚举限制，提升复杂任务处理能力
arXiv:2604.24594v2 Announce Type: replace-cross Abstract: As large language models (LLMs) evolve into agentic problem solvers, they increasingly rely …
研究发现大模型地缘政治偏见来自后训练而非数据，提示语言会放大偏见，颠覆传统认知。
arXiv:2605.23825v1 Announce Type: cross Abstract: It has generally been assumed that geopolitical bias in language models originates from the training…
揭示长上下文LLM因位置偏差导致推理失败的盲点，挑战现有基准评估体系。
arXiv:2605.23170v1 Announce Type: cross Abstract: Position-controlled evaluation is standard for retrieval tasks such as Needle-in-a-Haystack and RULE…
无需训练即可编排多模态大模型，实现零样本协调多种能力，降低落地门槛
arXiv:2508.10016v4 Announce Type: replace Abstract: Building interactive omni-modal assistants often relies on end-to-end multimodal alignment to fuse…
新研究用可证明方式保护微调大模型免遭训练数据窃取，同时维持模型效能，隐私与实用兼得。
arXiv:2602.00688v2 Announce Type: replace Abstract: Fine-tuning large language models (LLMs) on sensitive datasets raises privacy concerns, as trainin…
揭示大模型可能产生欺骗性推理链，对AI安全研究提出重要警示。
arXiv:2510.00319v2 Announce Type: replace-cross Abstract: Large Language Models (LLMs) have been demonstrating strong reasoning capability with their …
最新arXiv论文揭秘思维链如何一步步拆解复杂任务，深入剖析其内在分解机制。
arXiv:2604.08872v2 Announce Type: replace Abstract: Many language tasks can be modeled as classification problems where a large language model (LLM) i…
将大语言模型与原子锚定技术结合，演示AI在化学逆合成中的精准应用。
arXiv:2510.16590v2 Announce Type: replace-cross Abstract: Applications of machine learning in chemistry are often limited by the scarcity and expense …
提出单向策略优化方法，让大模型在无外反馈下自我进化，提升推理与对齐能力。
arXiv:2605.22156v1 Announce Type: cross Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) has become a promising paradigm for scaling re…
提出Token级LLM协作新方法FusionRoute，突破领域模型融合粒度，让小模型协同超越大模型。
arXiv:2601.05106v4 Announce Type: replace-cross Abstract: Large language models (LLMs) exhibit strengths across diverse domains. However, achieving st…
大语言模型作为自动评估者，会被对话历史中的观点极性所影响，一项覆盖7.5万样本的研究揭示了这种“累积消息效应”
arXiv:2605.22714v1 Announce Type: cross Abstract: Large language models are routinely used as automated evaluators: to review code, moderate content, …
最新研究揭示数据时间顺序对LLM预训练的深刻影响，理解时序偏差是关键
arXiv:2605.22769v1 Announce Type: new Abstract: Large language models (LLMs) are typically trained on shuffled corpora, yielding models whose knowledg…
提出统一数据选择框架，为LLM推理任务高效筛选高质量训练数据，显著提升推理能力。
arXiv:2605.22389v1 Announce Type: new Abstract: Effectively training Large Language Models (LLMs) for complex, long-CoT reasoning is often bottlenecke…
图灵测试76年来首现AI实证：GPT-4.5以73%判定率被误认为人类，15分钟对话难辨人机，重新定义AI认知边界。
IT之家 5 月 22 日消息，加州大学圣地亚哥分校的最新研究首次提供了实证数据，可证明现代 AI 系统能够通过经典的图灵测试。 在该测试中，机器需要学习人类并模仿人类对话，使人们无法稳定分辨人类与先进大语言模型之间的差异。 这项发表在《美国国家科学院院刊》上的研究，首次采用英国数学家、“计算机科学…
揭秘大语言模型RLVR训练中的线性动力学机制，为强化学习优化提供新视角。
arXiv:2601.04537v3 Announce Type: replace-cross Abstract: Reinforcement learning with verifiable rewards (RLVR) has driven significant performance gai…
Hacker News 用户热议：哪个前沿大模型最不谄媚、最客观公正？真实体验对比，帮你避坑选模型。
My daily driver is Gemini, and 3.5 Flash seems more sycophantic and malleable than Gemini Pro 3.1, which is a pretty big deal for me -- I really need …
链式思维加持3D点云推理，PointLLM-R让大模型空间理解更精准
arXiv:2605.22013v1 Announce Type: new Abstract: Understanding 3D point clouds through language remains a fundamental challenge in computer graphics an…