AI 情报文章归档

Thinking Machines Lab 发布第二代 Inkling 首款模型 Inkling 2 Small，为 594B 总参数、17B 激活的稀疏 MoE，支持文本与图像输入、1M token 上下文和可变推理强度。

Named entity recognition (NER) is challenging for social media data in a low-resource language. LLMs often lag behind pretrained language models (PLMs) in NER performance, but they show promise in handling cases that PLMs struggle with. Us…

Continuous music emotion recognition (MER) estimates how Arousal and Valence evolve within a song, yet the field's standard metric, pooled frame-level correlation, is dominated by between-song differences and cannot distinguish a model tha…

Weak supervisors are usually judged by their own errors, but what matters is what a student learns from them. For squared loss and any linear learner shared by all supervisors, we show that a supervisor's value is determined by its errors …

感谢 @arena！🎉 很高兴看到 Hy4 成为 Agent Arena 上排名第一的中国模型。

搜狐创始人张朝阳在 2026 世界互联网大会乌镇峰会上承认自己“错过移动互联网”，并称大模型需要拼资源、搜狐做不起，因此对 AI 投资显得冷静。他表示不要全被 AI 带跑，赚钱的机会恰恰在不被注意的地方，产品设计上也没有追逐算法推荐。

Most agent demos stop at "it answers in a notebook". A customer asked me for the other half: how do...

我很高兴与 @LaudeInstitute 合作（连同我的 @Sanjay_Ghemawat、John Hennessy、Dave Patterson、Ben Gomes 和 @andykonwinski），共同资助开放科学登月项目！

Translating parallel scientific codes between HPC programming frameworks is a critical but labor-intensive task. Existing LLM-based approaches rely on fixed sequential pipelines with undifferentiated repair strategies and correctness crite…

Frontier language models are increasingly deployed as agents, and how well an agent performs depends not only on its model but also on its harness: the prompts, tools, memory, and control loop through which the model acts. When an agent is…

淘宝 MNN 团队发布并开源面向手机应用的轻量模型 MNN-Pocket-0.3B，约 3 亿参数，在 iPhone 18 Pro 上通过 MNN 引擎完全离线运行，平均解码速度约 350 token/s。

The growing reliance on high-fidelity, state-of-the-art machine learning based models for engineering decision-making necessitates robust and systematic approaches for model selection. While high predictive accuracy at untrained points is …

Proximal Policy Optimization (PPO) stabilizes policy updates by clipping the importance ratio between the new policy and the data-collecting policy. The standard rule uses one fixed clipping range for all samples. While this choice offers …

Can frontier general-purpose agents understand client intent, explore creative directions, execute revisions, and prepare production deliverables? We introduce DRAFT-Bench (Design Reliability Across Feedback and Tasks), a lifecycle-oriente…

Probabilistic load forecasts for AI data centers need to represent both uncertainty under changing workloads and temporal dependence to support grid operations. This paper develops an open, grid-oriented benchmark using measured GPU power …

中国移动在 2026 移动网络大会上宣布，2027 年计划采购总金额超 5,000 亿元，其中算力投资超 1,000 亿元，终端采销超 1,000 亿元。董事长陈忠岳表示算力是投资最大增量，未来两年智算能力将连续翻番，吉瓦级计算园区明年将增至 7 个。

Quantization-aware training (QAT) is the most reliable way to obtain accurate low-bit large language models (LLMs), but training with AdamW keeps two dense FP32 states for every weight matrix. We find that low-rank memory-efficient optimiz…

Automated scientific discovery has the potential to advance understanding of disease mechanisms and accelerate identification of early disease markers. Manual data annotation is infeasible at population scale, motivating methods that autom…

曹操出行宣布已接入中国移动 12580 个人智能助理，成为中国移动首个打车生态合作伙伴。用户在 12580 内用自然语言说出出行需求，即可唤起曹操出行 AI 打车服务，在聊天界面完成行程确认、车型与价格选择及车费自动结清。此次为首期试点，通过 A2A 协议一次对接灵犀主控，后续可上线多个中国移动生态入口。

Many professional reports reuse the structure of prior editions while updating the content. We formalize this setting as form-restricted report generation: given a sample report and new grounding documents, generate a report that preserves…

Real-world multivariate time series rarely come with verified ground-truth causal structure, so research on time-series causal discovery and treatment-effect estimation depends on synthetic data. Existing tools are either fixed datasets th…

We study the problem of causal effect identification from an arbitrary set of observational and/or interventional distributions, which we call Universal Identifiability (uID). The uID problem generalizes existing classical identifiability …

Recently, Claude Code just... stopped. Not a bug in my code, not an outage. I had burned through my...

Recent progress in embodied robotics is bringing robots closer to deployment in homes, as vision-language models (VLMs) are demonstrating significant advances in reasoning, making VLM-driven robots promising home assistant agents. These ag…

Ricursive Intelligence 发布开放权重模型 Ripple-65B Preview，宣称在美国开放权重模型中智能体编程表现最佳，在 TerminalBench 2.1 上优于 Meta Muse Spark Open 816B、NVIDIA Nemotron 3 Ultra 550B 等更大规模模型。

Machine learning methods for solving partial differential equations (PDEs) significantly improve computational efficiency over traditional numerical solvers such as finite element method (FEM). However, PDE dynamics inherently span both lo…

百度发布开源基础模型 ERNIE 5.5 Flash Base，参数规模为 280B。该推文附上 the-decoder.com 的报道链接，未给出更多技术细节。

Artificial Analysis 推出 Endpoint Accuracy Index v1.1，新增 GLM-5.3 Flash 与 Meta Muse Spark Open 等六个开源权重模型覆盖。

Five ways a prediction-market model fooled me, and the one metric I trust now For five...

据 Business Insider 报道，Mira Murati 创立的 Thinking Machines Lab 正以 400 亿美元投前估值筹集 50 亿美元，此前其以超 100 亿美元估值完成 20 亿美元种子轮，500 亿美元估值的融资尝试谈判破裂。

Gemini Omni 1.2 Flash 已在 Higgsfield 上线。 15 秒片段，角色一致，最高 4K，带原生音频。 现已在 Higgsfield 可用。

在 2026 移动网络大会暨中国移动合作伙伴大会上，中国移动称 AI 驱动移动网络流量快速增长。其内网实测数据显示，今年手机上网流量增长 18%，增速是去年的两倍，其中 AI 生成内容带来流量增长 5 倍；AI 应用流量上下行为 1:6，区别于全网平均的 1:9。

Execution Truth Validator: Defeating Exchange API State Drift and Phantom Orders ...

Coding agents like Claude Code and Codex are fast, yet a lot of wall-clock time goes to them waiting...

CVE-2026-103572: Copilot in Visual Studio blocked .mcp.json by string. The agent wrote .mcp.json. and Windows saved it as the real file.

很高兴看到 MiniCPM-OCR 集成进 LiteParse！🚀💙 感谢 LlamaIndex 团队让文档解析对开发者更加易用。🙌

Meta CEO 扎克伯格 11 月 6 日在 Threads 宣布个人 AI 助手 Muse 正式上线实时语音对话与富有表现力的虚拟形象功能，率先面向美国用户，覆盖移动端、网页端和 Mac 应用。

From a multi-agent testing experiment to a broader discussion about how engineering teams should...

Dwarkesh Patel 花费 20,000 美元 API 额度让多个 AI 模型在社交欺骗游戏 Blood on the Clocktower 中对弈，并上线了 benchmark 网站 clocktowerbench.com，也开放让人类与 AI 对玩。他观察到模型今年在临场策划、虚张声势和协作推理拆穿阴谋方面明显进步，并称 OpenAI 新模型在该游戏上明显强于 Anthropic。Noam Brown 转发表示对该评测方法印象深刻，并邀请 Dwarkesh 到 …

World Labs 发布 Frameboy 演示项目，让网友在网页端用高度优化的 Atlas 版本实时探索 AI 生成世界，以最高 15 帧每秒从云 GPU 流媒体传输。Atlas 是其两个月前发布的自回归世界模型，拥有像素级摄像机控制。IT之家实测称其类似游戏机，可通过点击“游戏卡带”选择世界并自由探索，目前需运行在云端 GPU 服务器上，成本不低。