AI 情报文章归档

"It learns over time" is the most common claim made about AI agents with memory, and the hardest one...

Small language models (SLMs) running on mobile devices can keep user data private and work without connectivity. But they perform well only when fine-tuned for the target task and sized for the target device, a process that today requires …

Adobe 宣布与华特迪士尼达成多年战略合作，Adobe 成为迪士尼官方 AI 创新合作伙伴，计划将 AI 工具推广至主题乐园、内容制作和体育内容传播等领域。

This question came up in Chinese AI developer circles recently, sparked by a talk Wang Jian gave at...

SDR-to-HDR reconstruction is an ill-posed problem in which clipped highlights, crushed shadows, and quantization artifacts must be recovered while preserving the original content. Latent diffusion models provide powerful generative priors …

Keeping the same URLs during a website migration is valuable, but it is not a guarantee that organic...

Compositional generalization is an important form of out-of-distribution generalization, defined as the ability to train on some combinations of modules (e.g., functions such as sort, filter) and then generalize to unseen combinations. Des…

Open-Vocabulary Scene Graph Generation (OV-SGG) requires models to recognize visual relationships beyond the training vocabulary, yet existing methods often rely on dataset-specific object--predicate co-occurrence patterns. Under long-tail…

Elon Musk 转发 @SawyerMerritt 引述 Ars Technica 报道，SpaceX 发射副总裁 Dontchev 表示已用 Grok 自动检测航空航天零件中的异物碎片（FOD）。同一段内窥镜视频中，人工检出率约 94%，Grok 达 99.9%。原文链接：https://arstechnica.com/space/2026/11/spacex-now-fundamentally-an-infrastructure-company-senior-off…

Simulation optimization typically returns a single estimated optimal point. When decisions also depend on criteria evaluated independently of simulation, such as resource cost, the decision maker instead needs the set of designs whose expe…

Author: Cyrus Wong (黃俊彥) — Google Developer Expert - Cloud (AI), Senior Lecturer at HKIIT ...

Existing Chinese educational corpora often suffer from fragmented coverage and a lack of systematic organization. In this paper, we propose WoW (\textbf{Wo}rld Kno\textbf{w}ledge), a large-scale dataset designed for the full-cycle training…

Judge an AI testing tool by what survives a manual check, not by how many bugs it reports. I build...

Reliable inference for rankings of large language models based on human pairwise comparisons must preserve the privacy of individual preference records. To meet this need, we develop methods for score and rank inference under the Bradley-T…

蚂蚁灵波科技开源触觉基础模型 LingBot-Touch，基于近 5000 小时真实触觉数据预训练，覆盖磁触觉、视触觉和压阻式三类传感原理，输出统一触觉表征，可跨传感器、跨灵巧手、跨模型和跨任务复用。

KL-regularized reinforcement learning, the objective behind maximum-entropy RL and RL from human feedback, is variational inference in the graphical model of "control as inference": the regularized optimal policy is a posterior over trajec…

Integrating human feedback into language models has gained increasing attention as a means to align model outputs with human preferences, with Reinforcement Learning from Human Feedback (RLHF) serving as a prominent framework for preferenc…

在 @opencode 上用 LongCat-2.5 构建你的下一个项目吧。现已免费开放！

The Information 援引两名现任美国商务部官员的话称，Nvidia 尤其大力游说反对任何针对远程访问的规则，而远程访问是中国 AI 实验室经由新加坡和马来西亚数据中心获取受限芯片的途径。

A residential solar installer — 22,000 accounts across three states — found this one in a state...

魅族 Flyme AIOS 宣布「超级 Aicy」已上线，提供 24 小时在线的智能助手服务。其功能包括指定人、指定场景代接通话，高铁、飞机、酒店查询对比预定与打车导航，以及接入飞书套件的办公助手和股票数据查询。该助手还支持定时服务、问屏总结、视频生成，并深度集成 130+ 项手机原子能力，可主动备车调节车内环境。

OpenAI 宣布已在内部研究设施部署控制室系统，可在收到停止指令后最长 35 秒内阻止失控 AI 智能体继续输出。任务需先向控制室请求唯一任务 ID 并附加到推理请求，由控制室确认任务能否运行，停止时拒绝下一次推理请求。延迟源于推理授权请求的缓存层，公司正努力缩短至 0；所有前沿强化学习任务还配备独立模型监控器，发现潜在问题时发出警报。

As AI agents become increasingly capable, users must still recognize when and how to ask for help, leaving many opportunities for useful support unrealized. Proactive agents hold the promise to close this gap, but effective proactivity req…

Azeem Azhar 认为当前 AI 板块是"向优质资产转移"而非崩盘：截至 10 月的十二个月内 AI 收入达 1880 亿美元，年化 3450 亿美元，同比增长 4.2 倍。

As AI-based products permeate society, specifically systems using generative models, a belief system is emerging about what these products are and what they can and cannot do. One common belief concerns the perception of the form of langua…

We introduce SafeRoute-MAS, a coverage-aware, risk-calibrated procedure that keeps all specialist messages or removes one using only pre-integration information. On a sealed cache of 5,328 committees, its source-compatible operating point …

Analog integrated circuit (IC) design is a multi-turn, simulator-grounded engineering task that demands iterative reasoning over coupled performance tradeoffs, yet proprietary process design kits (PDKs) make hosted large language model wor…

Active learning aims to reduce annotation cost by selecting the most informative examples for labeling. In the low-budget regime, where uncertainty-based methods often fail, coverage-based active learning methods such as ProbCover have bee…

腾讯首款专属 AI 游戏陪伴 Agent“鹅次元”于 11 月 10 日开启内测，面向 PC 玩家提供拟人化 AI 陪伴和实时语音互动。该产品能识别用户游戏画面并据此互动，支持 PC / 手机 QQ 陪伴聊天，还具备桌宠、一键更换壁纸等功能。目前提供“夏夏”“沈露白”“谷雨”“贝拉”四个角色，擅长《无畏契约》《英雄联盟》《反恐精英》等游戏。

很高兴看到 MiniCPM-OCR 集成进 Docling！🚀💙 感谢让开源文档处理更加易用！🙌

Parallel drafters accelerate LLM decoding by proposing a whole block of draft tokens in a single forward pass, making them attractive for LLM serving. Recent systems optimize goodput (i.e., accepted tokens per second) because rejected draf…

Most language models deployed outside the United States and China are fine-tuned descendants of a small number of pretrained families. We argue that fine-tuning leaves the decisive properties of a model inherited from its base: vocabulary,…

据 The Wall Street Journal 报道，Anthropic 将在本周设定估值区间，计划下周四或周五开始交易。部分机构投资者正推动估值低于上周讨论的 1.5 万亿美元目标，该数字已低于几个月前传闻的 2 万亿美元。报道还提到其主要竞争对手 OpenAI 上周融资 110 亿美元后估值约为 1.25 万亿美元，低于其最初寻求的 200 亿美元融资规模。

You kick off three agents on the Mac at home, then head to the office. Somewhere on the way one of...

Why Soulcast pull requests and zero-shot roadmaps collapse when requirements shift—and why the engineering journey still matters.

Meta 推出迄今最小、效率最高的模型 Muse Ember，在 Artificial Analysis 排行榜中表现媲美 Claude Haiku 5.5，单任务成本仅为其 1/3，并在多项测试中超越 GPT-6 Luna。

Claude Fable 5.2 在 Artificial Analysis 的 AA-3D 交互式 3D 体验基准上以 1271 Elo 登顶，领先 Claude Sonnet 5.5（1204）和 Claude Opus 5.5（1202）67 分，95% 置信区间 1246–1297 与其他模型不重叠。

Paramount Skydance CEO David Ellison 在财报电话会上称，AI 对电影行业的影响可能超过 CGI，是"现代史上最大的创意解锁"。他以《终结者》为例称当年 400 万美元的拍摄成本如今需 6000 万至 7000 万美元，AI 有望压低预算、释放原创故事空间。他表示公司正自研 AI 工具，但强调 AI 是艺术家的工具，"永远不会取代"艺术家。

据 The Information 报道，推理服务商 Fireworks 遭黑客入侵，已告知 Harvey 等客户。入侵经一家托管其研究系统的第三方云服务商进入开发机（10 月 24 日），触及客户账单记录，部分账户的数据集和不完整模型权重也受影响，部分客户考虑迁走业务。

How I built a JavaScript visualizer by instructing Claude Code in plain language, including what I got wrong, how I checked the work, and where this approach falls short.