AI 论文与研究

LlamaIndex 为 LlamaParse 上线数据连接器，首批支持 SharePoint 和 Google Drive，开发者可直接连接源文件夹而无需手动上传文档。同步文件可用于 Parse、Extract 和 Index，连接会每日自动同步新增与更新文件，也可用 Sync now 手动拉取；通过 Index 配置抽取时，后续索引更新只处理新增或变更文件。

LimiX-2 迈向通用结构化数据智能的上下文机制网络 论文：https://huggingface.co/papers/2609.17488

NVIDIA 发布开源合成数据工具 NeMo Data Designer（NDD）技术报告，采用声明式配置定义数据集各列，支持生成文本、代码、结构化输出、图像、嵌入向量及控制多样性的统计采样器，并可通过插件扩展类型。

NVIDIA 论文提出 Agora，把多个研究智能体的共享记忆记录为 Git 中只增不改的 DAG，每个结果、假设和验证都是不可变 commit，父边标明依赖关系，索引列出开放分支与验证状态。

NVIDIA 论文提出 Agora，把研究智能体的结果、假设和验证记录为不可变的 Git commit，形成 append-only DAG，避免多个编码智能体重复实验并让彼此复用已验证结果。

Pew Research 调查 37 个国家 42，151 人后发现，其中 34 国多数受访者认为 AI 未来 20 年更可能导致失业而非创造新岗位，澳大利亚和韩国各有 76%、美国 71% 的人持此看法。受访者还普遍认为 AI 会拉大贫富差距，且美国、瑞典等国年轻人的担忧在过去一年明显上升。全球中位数 41% 的人对 AI 既担忧又兴奋，37% 以担忧为主，仅 13% 以兴奋为主。

The symbiotic scaling of artificial intelligence models and high-performance computing systems continually creates algorithmic challenges in their convergence. Foundation models (FMs) are a crucial example, requiring months-long training o…

Characterizing the physical properties of clay and cementitious materials matters across many fields, from materials science to geological waste disposal. Property simulation typically calls for 3D imaging, which is expensive, not always a…

We study repeated contract design when a principal observes outcomes but not the actions that generate them. The principal may use any bounded outcome-contingent payment vector, and the agent's best response can make expected profit discon…

Large-scale routing often requires visiting clusters of nodes in a prescribed order, giving rise to the Ordered Clustered Traveling Salesman Problem (OCTSP). Optimizing each cluster independently seems natural, but misses non-local depende…

Spatial reasoning abilities correlate strongly with performance in STEM fields. Games offer a compelling medium for training these critical skills in developing children who have a natural proclivity for play. However, to facilitate human-…

LEO satellite networks feature dynamic topologies, time-varying links, and diverse service requirements, which make conventional routing schemes difficult to support fine-grained quality-of-service (QoS) provisioning. Existing studies main…

Magnetic Resonance Imaging (MRI) acquired at different field strengths exhibits pronounced variation in noise, resolution, homogeneity, and contrast, which limits comparability across acquisition settings and complicates downstream analysi…

Deceptive online job advertisements have emerged as a primary pathway into forced labour, yet systematic detection methods remain underdeveloped due to data scarcity and absence of empirically validated indicators. We formalise this detect…

Traffic regulations are written for human interpretation and therefore rely on shared background knowledge and flexible phrasing, which inherently introduce ambiguity, context dependence, and semantic underspecification. These linguistic c…

We study reconstruction in autoencoders that apply the same forward map before and after setting the observed coordinates to zero. For equal odd input and hidden dimensions $d\geq 3$, among orientation-preserving diffeomorphisms whose Jaco…

We study smooth strongly convex--strongly concave minimax optimization with general nonlinear coupling in the deterministic unconstrained setting. We propose a pure single-loop damped extragradient method with fixed parameters and two new …

开发者 @AlicanKiraz0 基于 VoxCPM2 微调出 Kahya-TTS，用近 100 小时自然土耳其语语音数据，将其语音生成能力适配到土耳其语 TTS。该模型已在 Hugging Face 开源，用于探索开源语音模型向新语言和专用数据集的适配。

Three-dimensional reconstruction from unorganized point clouds remains a challenging problem in computer vision, geometric modeling, and computer-aided design. While neural implicit methods achieve impressive reconstruction accuracy, geome…

Speculative Decoding (SD) has significantly accelerated Large Language Model (LLM) inference, yet existing approaches face a fundamental tradeoff between two drafting strategies: neural drafting and context-based copying. Neural drafts (e.…

We introduce a matrix generalization of the cone of Harvey and Lawson and prove that it is an exact special Lagrangian manifold. We further show that it belongs to a family of exact special Lagrangian manifolds that foliate the balanced ma…

Ubiquitous time series data across diverse domains enables critical applications in areas such as transportation systems and power grids. Recently, training foundation models on massive datasets to achieve accurate zero-shot forecasting ha…

Generally, most voice agents are cascaded systems, i.e., an ASR model transcribes the caller's audio, a language model reads the transcript and decides what to say and which backend tools to call, and a TTS model speaks the reply. Nearly a…

Conventional satellite remote sensing transmission follows a reconstruct-then-infer paradigm that optimizes pixel-level fidelity, creating an objective mismatch with downstream tasks such as classification and detection, especially at low …

Cortical hypometabolism measured by Fluorodeoxyglucose Positron Emission Tomography (FDG-PET) is a highly sensitive biomarker for dementia diagnosis. However, high costs, radiation exposure, and limited accessibility constrain its clinical…

Cognitive offloading to AI can reduce opportunities to practice skills, creating risks of deskilling. However, it remains unclear how to prevent deskilling without restricting access to AI. Here, we design two interventions to reduce offlo…

Vision-language models (VLMs) are fragile under image corruption. We find that the wording of the question affects VLMs in two opposite ways. Verbose questions make VLMs substantially more robust---e.g., rephrasing "Is there a cat?" into "…

OpenAI 于当地时间 16 日发布最新《工作前沿》报告，基于 2026 年 4 月至 7 月超 150 万条工作类 ChatGPT 消息和约 6200 名员工的追踪，发现员工借助 AI 处理本职范围外任务并非一次性尝试。

东北大学硕士生 Soham Padia 基于 Olmo 3-32B 开发了众包评测游戏 Steering Arena，让玩家提交短文本前缀来引导模型产生更亲社会（prosocial）的回应。

UC Berkeley 团队在 SWE-bench Lite 和 Terminal-Bench 2.0 上评测 21 个模型与 harness 组合（7 个模型。

Mozilla 发布 91 页报告称开放权重 AI 目前仅落后前沿约 4 个月。OpenRouter 上 8 月 token 量前十的模型有 8 个是开放权重。

IDC 发布《全球智能眼镜市场季度跟踪报告》，2026 年第二季度全球智能眼镜出货量 354.7 万台，同比增长 35.3%；其中音频及音频拍摄眼镜出货 249.4 万台（增长 54%），AR / ER 眼镜出货 50.5 万台（增长 75.7%），VR / MR 头显出货 54.8 万台（下降 23.3%）。

冰岛初创公司 Treble 完成 1800 万美元 A 轮扩展融资，由 Paladin Capital Group 领投，KOMPAS VC、Frumtak Ventures、EIC 和 Omega ehf 参投，累计融资超 4000 万美元。其语音模拟平台提供合成数据生成与模型评测，客户包括 Amazon 和 Logitech，今年早些时候还与 Hugging Face 合作推出语音识别基准。

Mozilla 发布 91 页报告称开源权重模型仅落后前沿约 4 个月；8 月 OpenRouter 按 token 量计前 10 模型中 8 个为开源权重、7 个为中国造，DeepSeek 成为首个在周请求数上居首的开源模型。

这篇 Google 论文里展示的 AI 发现 RSI 循环很有意思 但话说回来，Google 很擅长写学术论文，却不知为何在把研究转化成优秀 AI 产品方面差得多。我觉得他们需要多一点 OpenAI/Anthropic 的文化 或者回归 2000 年代那个创新的 Google

百灵发布面向 PB 级大模型数据治理与探索的统一宽表系统 OmniTable，相关论文获 VLDB 2026 最佳工业论文奖。在一项真实 SFT 数据治理任务中，端到端处理周期从约 14 天缩短至约 2.5 天，缩短约 82.1%，人工操作步骤从约 45 个减至约 12 个。

GPT-6 Astra 在 BrokenArXiv 与 ArXivMath 基准上以 88.6% 对 87.7% 击败 Fable 5.1，单题成本仅 $3.63，而后者为 $12.77，性能相近但成本不到三分之一。该基准聚焦近一个月 arXiv 上被推翻的猜想，模型在 harness 内执行而非直接调用 API。

Google 近期连续发布多篇认真对待 AI 并考量政策影响的论文。其中一篇分析指出，没有单一政策是万能解，各维度均存在权衡；研究识别出位于帕累托前沿的"最小后悔"政策序列，按经济情景触发：轻度情景扩大失业保险、现代化 EITC 与员工主导的再培训，中度情景 EITC 转向负所得税，极端全面转型情景则以 Universal Basic Capital 兜底。

研究提出 prediction-powered smoothing（PP-S）及跨报告分类体系借力的扩展 PP-TS，用于在标注样本有限时更准确地估计各领域评估均值。该方法还给出近似无偏的基于设计的交叉验证分数，用于在直接估计与平滑估计之间选择。在可验证评分的基准和人工标注的已部署智能体流量上，新估计器在点估计与区间估计上均优于直接估计器，覆盖率接近名义水平。

研究发现在线策略蒸馏（OPD）中，基础学生模型与后训练教师模型即使声明的停止集合相同，也可能将停止概率放在不同的 EOS token 上，这种终止 token 不匹配会抑制学生模型偏好的终止动作，导致回复过长甚至耗尽生成预算。