AI 情报文章归档

If you run long Claude Code sessions you know the moment. The context hits the auto-compact...

Graph-based retrieval-augmented generation (GraphRAG) answers questions by retrieving over a knowledge graph built from a corpus. Yet retrieval strategies are designed around the structure their own construction pipeline produces, and whet…

Setting. This is an empirical characterization study; we propose no new method and run no comparisons against alternative objectives. We fine-tune Qwen3.5-4B with sparse LoRA and length-normalized DPO plus an SFT retention term (the RPO ob…

Let $d_1,\dots,d_m$ be directions in $\mathbb{R}^k$, where $k\ge 2$. A multiset $X$ of $n$ points of $\mathbb{R}^k$ projects to the $m$ multisets of numbers $\{\{x\cdot d_j : x\in X\}\}$, or equivalently to the $m$ sorted lists of these nu…

百度在深圳AI Day上发布面向物理AI的世界模型Apollo World，可将复杂道路交互与罕见驾驶场景转化为可重复、可控的仿真，用于自动驾驶系统训练与测试。该模型支持自主决策的AI交通智能体、实时生成的摄像头与LiDAR等传感器数据，以及虚拟城市中任意两点间的完整载客行程。Apollo World已用于Apollo Go的测试验证。

Microfluidic droplet generation is critical for many applications, yet efficient device design in high-dimensional parameter spaces remains computationally prohibitive due to complex multiphase flow dynamics. This work presents a unified m…

During my vacation, I decided to spend some time learning more about the Model Context Protocol, or...

魔芯科技在世界互联网大会发布基于华为昇腾原生的 4D 世界模型 MoWorld 2.0，将重力、弹力、浮力、光照、热力等物理要素融入空间建模与交互，输出可仿真、可实时渲染、可反复调用的 4D 高斯资产。

Background/Objectives: CNNs perform well in retinal disease classification from spectral-domain optical coherence tomography (SD-OCT), but whether EfficientNet's natural-image scaling hierarchy transfers to medical-image transfer learning …

Traditional document parsing systems use Markdown as their output representation, which is effective for linear text but does not fully preserve fine-grained spatial structure, table geometry, visual hierarchy, or rendering semantics. We s…

Long-horizon language-model agents accumulate increasingly large interaction histories, making context compaction essential for efficient execution. Existing learned compaction methods often rely on downstream task outcomes or solver behav…

Skills let language agents reuse experience across tasks as natural-language documents. Recent work evolves skills from interaction, but existing methods have limitations when skills evolve online. Prior methods update a skill offline from…

I built an autonomous scientific discovery engine called RUMI. I let it run for up to 21 hours per...

If you use a terminal coding agent on more than one project, you probably know my old routine. A...

An AI application can produce a clear, confident response after sending a request to the wrong...

Multivariate time series anomaly detection (MTSAD) relies on modeling both temporal dynamics and inter-sensor dependencies to identify complex system faults. Although graph-based detectors explicitly model sensor topologies, they train tas…

Every holiday season we sit in the same queue and wonder if we picked the wrong day. So I made Home Assistant watch public traffic cams and let an LLM score what it sees.

Chain-of-Thought (CoT) reasoning substantially improves Large Language Models (LLMs), yet its out-of-distribution (OOD) generalization mechanism remains theoretically underexplored. We study CoT as compositional OOD generalization: the tar…

There's a draft spec going around for giving AI agents real undo — not a confirmation dialog before...

We show that a hardwired distributed processing unit (HDPU) can take a neural decision at the sensor and so remove the inference from the host. Built for zebra-crossing detection on a low-cost Artix-7 FPGA, the unit decides on every camera…

Active embodied question answering requires agents to acquire evidence through viewpoint changes and physical interaction. The central challenge is to avoid undirected interaction: taking plausible actions without specifying what evidence …

Reliable channel estimation and symbol detection in 5G New Radio (NR), as well as emerging 6G radio systems, rely on demodulation reference signals (DMRS). The density and placement of these pilots on the orthogonal frequency-division mult…

据引述报道，中国 AI 和云厂商已订购约 600 万张 RTX Pro 5500，总额约 700 亿美元，但 Nvidia 未来 4 个月内只能供应约 1%。

Partially interpretable models extend simple interpretable models with the option to defer complicated inputs to an accurate but uninterpretable black-box model. By deferring more inputs, the model gets more accurate; by deferring fewer, i…

Target-conditioned diffusion drafters predict candidate tokens in parallel, yet each new block must wait for updated target-model hidden states. This forces drafting and verification to run sequentially, limiting speedup even though each b…

I'm not an engineer. I come from the sales world. But between June 18 and July 16, I pushed 172...

I maintain Claude Monitor, an open-source browser extension that shows claude.ai's usage limits: the...

Brains and large language models need not be considered equivalent systems to reveal an interpretable commonality: both exhibit high-dimensional, nonlinear dynamics whose behavior can only be modeled imperfectly. This creates a shared chal…

Most AI leaderboards ask whether the model got it right. I wanted to know something else: when it's...

There's a good argument going around right now that an agent's log tells you what it did but keeps no...

404 Media 的 Jason Koebler 撰文认为，本次选举中选民在红州、蓝州和紫州全面拒绝了特朗普与硅谷的科技寡头路线。

Batch inference costs 50% less than real-time inference at Anthropic, OpenAI and Google, in exchange...

I'm a cybersecurity student, and I wanted to see how well an AI agent could help with the first,...

LLM inference digest (Oct 28–Nov 4, 2026): elastic KV-cache offloading, MoE-tuned speculative decoding, and day-0 serving for Mistral Large 4.

小红书 AllSpark 团队提出 Navalia，通过自动构建文档环境、合成证据可追溯且答案可验证的跨文档任务，并结合轨迹采集与长程后训练，形成从数据到训练的完整配方。

每部电影都始于一个创意。用 PixVerse Canvas 创作你的创意。 https://app.pixverse.ai/canvas?utm_source=Social&utm_medium=X

The Verge 分析 Alphabet Q3 财报：搜索查询增速达五年最快，但搜索收入同比增长从约 17% 放缓至 12%，付费点击增速从 13% 降至 8%，财报后股价下跌约 14%。

James Shore 在其系列文章第二部分提出测量 AI 对交付速度影响的方法。他指出交付速度不等于生产力，代码行数和 PR 都不可靠，建议采用随机分组、先估时后分配的组内重复测量实验，并提醒速度提升需按任务占比折算为整体收益，引用 METR 研究称工程师高估 AI 提速近 50%。

美团 LongCat 团队发布 LongCat-2.5，总参数 1.6 万亿、每 token 约 480 亿激活，主打长时程智能体任务，结合视觉理解与智能体编码，权重以 MIT 协议开源。

华为云智果 AgentArts 智能体平台通过中国信通院可信 AI 企业级智能体平台评估，成为首个获当前最高等级 4+ 级的产品。评估依据《企业级智能体技术要求》，覆盖技术、应用与服务能力，考察智能体认知、规划、执行、业务集成、知识管理、运维及安全合规。平台支持 LangChain / LangGraph 或自研框架，具备生产级长程任务、企业级安全环境、行业知识深度和全链路可观测四大能力。