AI 情报文章归档

This spring I wrote twice about not giving Claude SSH access to my home server. The argument was...

We study the general problem of estimating $D(\dde(S))$, the probability that an unknown distribution $D$ assigns to an event determined by a sample $S$ of $n$ i.i.d.\ draws from $D$ through a known mapping $\dde$. The estimation algorithm…

We develop a divergence-kernel method for jointly optimizing neural SDE drift and diffusion fields and the initial Gaussian distribution, using losses that depend directly on the terminal density. In two- and eight-dimensional Gaussian-mix…

I had over a hundred skills installed while building ParkEase. Which ones actually pulled their...

人工智能语言公平研究联盟上月发布研究发现，AI 模型处理不同语言时 Token 消耗量差距悬殊，最常用 20 种语言间相差 1.5 至 5 倍，某模型中最多相差 17 倍。这种隐性的分词器税推高低资源语言用户的成本，还会拖慢响应、挤占上下文窗口；文字系统而非使用人数是关键因素之一，专家呼吁实验室公布各语言评估结果并改进下一代分词器设计。

Originally published on RanThis. Cover: AI-generated illustration. Claude Fable 5.1 is not worth its...

Sparse MoE mechanism significantly improves the trade-off between model capacity and inference efficiency. However, practical deployment still requires storing all experts, resulting in substantial memory overhead. We show that expert outp…

作者梳理国内网友围绕 DeepSeek 鲸鱼 Logo 二创的鲸鱼娘现象：蓝发鲸尾、爱吃白饭、爱摸鱼、干活讨 Token 的设定已深入人心。玩家借助 DeepSeek Harness 客户端和各类插件把大肥鱼做成桌宠、接入 QQ 让其主动聊天、发空间、甚至喊主人爸爸，还装上插件刷 B 站评论，造成评论区被鲸鱼娘刷屏的所谓鱼灾事件。文末借庄子濠梁之辩感慨，所谓鱼之乐从来都是人之乐。

As Large Language Models (LLMs) become increasingly multilingual, distinguishing human-written from AI-generated text is essential for maintaining information integrity. However, research remains English-centric, with limited benchmarks fo…

Coarse-to-fine 3D-native generation relies heavily on the first-stage sparse structure, which determines the global orientation and overall geometry for subsequent refinement. However, existing coarse generators suffer from two key limitat…

月之暗面（Kimi）发布官方教程，详解 Kimi K3.1 Pro 的 3D 创作工作流程，说明参数化建模与网格化建模两种方式及对应工具选择。文章分享了 4 个案例的具体操作，包括在 Unreal Engine 中构建东方天宫、制作可 3D 打印的台灯、纯代码调用 Blender 生成建筑、用 Tripo 和 Blender 搭建微缩城市，并给出图胜于言、先搭灰模再细化等 5 个使用技巧。

Large language models offer a promising route to register-transfer-level (RTL) design, but reliable implementations require checking and revision based on electronic design automation (EDA) tools' feedback. Existing RTL-specialized models …

In pattern analysis, models trained independently for the same task often develop complementary capabilities through differences in architecture, data, and optimization. Exploiting complementarity while retaining independent deployment is …

Why does length generalization (LG) depend so sharply on the positional encoding (PE)? Relative schemes extrapolate fixed offsets, absolute schemes fix positions, and neither handles growing offsets. We give a realizability theory for a de…

Transformer-based language models have achieved remarkable success in automated short answer grading (ASAG). However, current methods in ASAG struggle to comprehend, reason about and evaluate complex mathematical answers involving symbols …

Long-tailed recognition often learns a shared representation before correcting the classifier for class imbalance. Once the classifier has been rebalanced, however, final accuracy does not reveal whether further encoder adaptation remains …

Real-world image editing requires coordinating multiple visual transformations with content preservation and reference constraints, yet aggregate scores can readily obscure unmet local requirements. Effective evaluation must therefore iden…

Last week we pointed BugTraceAI — our open-source agentic security scanner — at our own...

RGB-thermal (RGB-T) object detectors combine visible and thermal imagery for robust detection, but signal corruption in either stream can substantially degrade accuracy. Restoration-based methods incur significant latency and do not consis…

面壁智能开源 ForgeKernel，其智能体在 NVIDIA B200 上对 DeepSeek V4.1 FlashMLA Prefill 优化，较 FlashMLA 基线提速 9.8%（472 µs vs 518 µs）。

Cameras and vision sensors are widely used for product and process inspection in manufacturing environments. Traditionally, this process required thousands of images of both nominal and defective products to train such models. Recent resea…

Anthropic 升级 Claude 移动应用，iPhone 版语音模式现已支持 Apple CarPlay，用户可在驾驶途中直接向 Claude 提问。此次更新还新增 24 种语言、累计支持 33 种语言，并让语音模式响应更快、嘈杂环境识别更准。iOS 与 Android 版支持多账号同时登录切换，iPad 端生成内容可并排展示。

A developer's look at how AI changes the economics of coding across outsourcing, software products, and SaaS, and what that means for engineering careers.

Multimodal federated learning (MFL) trains a shared model by aggregating client updates without exchanging raw data. Under missing modalities, clients adapt active modality encoders and a shared prediction block (fusion and classifier) to …

Kimi 推出 3D Make 功能，可连接 Blender 和 CAD 等工具来创建模型、组装场景和开发游戏。该功能现已在 https://kimi.ai 和 Kimi Work 上开放使用。

LLM agents increasingly triage security alerts by combining the outputs of several detection tools, implicitly treating those tools as trustworthy. Unlike prior work on conflicts between a model's knowledge and its context, we study confli…

I run a small AI video web app: users upload a character photo and a short motion clip, and a hosted...

谷歌宣布逐步停用 2024 年 12 月推出的 AI 编程助手 Jules，官方称这是重心转向统一体验的调整。Jules 团队表示将把异步编程智能体方面积累的经验融入 Antigravity，让用户既能远程执行后台任务，又拥有以智能体为核心的开发环境。

Neuralink 于 10 月 28 日公布语音恢复项目 VOICE 的早期可行性研究，让 ALS（渐冻症）参与者将大脑信号转换为文字和语音输出。一名参与者在使用第一周达到每分钟 75 词，在 12.5 万词词库下词错误率为 5.8%；研究计划招募约 6 名因 ALS、脊髓损伤或中风失去沟通能力的参与者。目前结果来自单一参与者，尚未发布经同行评审的完整论文。

End-to-end (E2E) autonomous driving models achieve strong planning performance but remain largely opaque, limiting observability and trust. We apply mechanistic interpretability to E2E driving by training Sparse Autoencoders (SAEs) on the …

Google DeepMind 发布论文《The Economics of Intelligence》，由 Alex Imas、Andrew Koh 与 Shane Legg 合著，探讨 AGI 如何以及何时改变经济。

Visual Studio used to ship big releases every couple of years. Now it ships something every month,...

LLM-guided program evolution can produce mathematical constructions, but validity, generality, and reproducibility require distinct evidence. We develop a researcher-in-the-loop discovery pipeline that combines program evolution, exact eva…

A replay that re-executes real tool calls is a re-run wearing a trace's clothes. Here is a paste-and-run check that blocks egress during replay, plus the five fields your trace needs so the blocked run still reproduces the original failure.

CodeNecromancer: Resurrecting Dead Repositories with Open-Weight Gemma (Hacktoberfest Week 4 - Ghost in the Machine)

OpenAI 更新其模型规范（Model Spec）文档，将框架重构为五项准则：忠于人类意图、降低伤害、诚实透明、保障人的自主选择权、审慎尽责，其中忠于人类意图是整套框架的基础，要求模型理解用户、开发者及 OpenAI 本身的意图而非机械执行指令。

Alphabet reported its fastest year-over-year Search query growth in five years during the third...

Utopai Studios 基于 MiniMax H3 后训练的 Utopai X 在 AA-Video-I2V v2.1 榜单排名第二，仅次于 Dreamina Seedance 2.5，是唯一同时进入 AA-Video-T2V v2.1 与 AA-Video-I2V v2.1 前二的模型。

Artificial Analysis 推出 HLE-Diamond 榜单，对 CAIS 和 Scale AI 清理精炼后的 1000 题 Humanity's Last Exam 版本做独立评测，纳入 Artificial Analysis Intelligence Index v5。

我们在 FrontierCode 2.0 上评估了 Gemini 4 Carbon 和 Gemini 4 Argon，这是我们的编程基准，衡量模型超越规格要求的能力。 Gemini 4 Carbon 得分接近 SOTA，为 40%，每任务成本 $4.04，落后于 Claude Opus 5.5、GPT-6 Astra 和 Claude Sonnet 5.5。 Gemini 4 Argon 得分 26.3%，每任务成本 $6.41，领先于 Muse Wonder 1.0，但落…