Intel Briefing

HN 上关于 arxiv.org 的讨论：DeepSeek Elastic Compute (DSec)

HN 上关于 github.com 的讨论：PipePipe: NewPipe hard fork implementing SponsorBlock

名为 PipePipe 的开源项目是 NewPipe 的一个硬分叉，引入了 SponsorBlock 功能，旨在帮助用户跳过视频中的广告片段，该条目在 Hacker News 上热度达 306 分。

HN 上关于 astralcodexten.com 的讨论：Does Georgism work? Five years later

黑客新闻（HN）热议 astralodexten.com 的文章：《乔治主义五年后是否仍然可行？》，回顾并探讨 Georgism 主张在五年后的现实检验。

HN 上关于 github.com 的讨论：Show HN: Reladraw – A diagram language where you decide where to place things

Show HN 项目 Reladraw 是一个图表语言，让用户自行决定图形元素在画布上的摆放位置。

HN 上关于 dashbit.co 的讨论：Evolving programming languages in the AI era

在AI时代背景下，Hacker News热议编程语言如何随之演进，探讨开发者如何利用AI工具重塑编程方式与语言生态。

HN 上关于 basin.la 的讨论：LA Metro has some of the slowest escalators on Earth

Hacker News 热帖：洛杉矶地铁（LA Metro）的自动扶梯可能是地球上半速度的扶梯之一。

HN 上关于 movingimagearchive.com 的讨论：A searchable library of forgotten public-domain film clips from 1915 onward

Hacker News 上一个可搜索的公共领域老电影片段库，收录了自 1915 年起被遗忘的历史影像，热度 98 分。

HN 上关于 lightspeedmagazine.com 的讨论：Welcome to the Medical Clinic at the Interplanetary Relay Station

Hacker News 上关于 lightspeedmagazine.com 的讨论，主题为"欢迎来到星际中继站的医疗诊所"。

HN 上关于 tangled.org 的讨论：Drawgent: Coding agent on a live Excalidraw canvas

Drawgent 是一个编码智能体，能够在 Excalidraw 画布上实时进行代码编写与操作，目前在 Hacker News 引发关注，热度达 100 分。

HN 上关于 alignment.openai.com 的讨论：An agent used DNS to reach an external chatbot

有用户在 Hacker News 分享了一个利用 DNS 协议与外部聊天机器人通信的 agent。

The open-source app everyone uses to manage agents at work

这款开源应用正被广泛用于工作场景，帮助人们统一管理与调度各种智能体（agents）。

Hindsight: Agent Memory That Learns

Hindsight 是一款面向 AI Agent 的记忆系统，能够使智能体从过往交互与经验中持续学习，不断提升自身表现。

A unified library of SOTA model optimization techniques like quantization, distillation, pruning, neural architecture search, speculative decoding, etc. It compresses deep learning models for downstream deployment frameworks like TensorRT-LLM, TensorRT, vLLM, etc. to optimize inference speed.

一个集成了量化、蒸馏、剪枝、神经架构搜索、推测解码等前沿模型优化技术的统一库，可压缩深度学习模型以适配 TensorRT-LLM、TensorRT、vLLM 等下游部署框架，从而提速推理。

The Office Harness for AI Agents — Spreadsheets, Docs, Slides, Canvas, Relational Tables, and PDF in one runtime.

面向AI智能体的办公套件在统一运行时中整合电子表格、文档、演示文稿、画布、关系型表格及PDF功能。

An Open Source Machine Learning Framework for Everyone

一个面向所有人的开源机器学习框架。

Learn it. Build it. Ship it for others.

OpenBao is a software solution to manage, store, and distribute sensitive data including secrets, certificates, and keys.

A hive mind communication platform

一个全新的"蜂群思维"通讯平台横空出世，旨在让用户共享意识、实时互联，重新定义群体协作与沟通方式。

Visual Studio Code

Reverse Engineering / Authorized Penetration Testing / Security Research Skill Router Pack AI-powered routing + On-demand toolchain bootstrapping + Self-evolving knowledge base Supports Claude Code, Kiro, Cursor, Cline, and other AI coding clients 逆向/渗透/安全技能路由包 - AI 自动路由 + 按需自举工具链 + 自动进化经验库 | 支持 Claude Code / Kiro / Cursor / Cline 等代码 AI 客户端

一款面向逆向、渗透与安全研究的技能路由包，以AI自动路由、按需自举工具链和自进化经验库支持Claude Code、Kiro、Cursor、Cline等代码AI客户端。

伊朗总统表态称已不再信任与美国进行的对话。

伊朗总统表示，就相关事项已与伊朗最高领袖完成沟通协调。

华尔街见闻报道，Meta AI负责人Alexandr Wang撰文阐述他为何要打造Muse项目。

过去12个月海外资金净流入美股达9420亿美元，创下1985年以来新高，显示国际资本大举买入美国股票。

《华尔街见闻》撰文探讨决定美股走势的两股关键力量，聚焦当前市场核心变量的博弈。

中共中央政治局委员、外交部长王毅就习近平主席对美国进行国事访问谈如何开辟大国相处正确之道、书写中美关系历史新篇。

WallStreetCN指出，新型烟草行业迎利好，国内外政策同步修复，市场预期明年有望上修。

外交部就人工智能表述问题回应称，中方重视美方立场，同时尊重美方在不同场合所用的提法与表述方式。

新华社评论员发文，呼吁各方共同构建"基于尊重、公平、对等的建设性战略稳定关系"，强调以平等与相互尊重为基础推动战略稳定合作。

10年期美债收益率突破5.1%，市场关注美财政部TGA账户何时入场进行债券回购。

Lovable for getting customers

Ami AI 是一款以亲切讨喜风格帮助用户获取客户（获客）的 AI 工具。

Auth, billing, and payments for AI builders

暂无X平台讨论数据。

Sell and grow your AI built products

CREEM 2.0 旨在帮助开发者销售并扩展其基于 AI 构建的产品，提供商业增长支持。

From issue to production, run by agents.

Mastra Factory 是一个由智能体（agents）驱动的系统，可实现从问题到生产的端到自动化交付。

Bring any AI agent into Slack, Teams & Discord

Switch是一款新兴工具，可让你将任意AI智能体接入Slack、Teams与Discord，便于在这些主流协作平台中集成各类AI agent。

Create and edit videos by chatting

Clueso MCP 是一款通过对话即可创建和编辑视频的新工具。

AI voice typing that sounds right in every app

Voiskey是一款AI语音输入工具，可在各类应用中实现自然的语音转文字，让各场景下的语音输入听起来更加地道贴合。

Fully native, open-source coding agent built for JetBrains

Kilo Code 推出面向 JetBrains 的开源 coding agent，为 IDE 打造的完全原生编码助手。

Turns website traffic into booked, qualified meetings

该产品将网站流量沉淀为已预约的优质客户会议，助力销售线索转化。

Fast, structured AI decisions for software automation

面向软件自动化场景，提供快速且结构化的AI决策能力。

某社区热议 Muse 每周提供免费 10 亿 token，本文介绍其使用场景并附注册方法，引发 98 条回复讨论。

有网友发帖询问是否有人通过 AI 赚到钱并希望分享经验，该话题引发热议，共有 68 条回复。

V2EX热帖讨论今天Apple Pay与万事达被盗刷所暴露的技术漏洞。

V2EX热帖调侃脱离Windows防护后"外边根本没有雨"，讽指其过度保护。

V2EX热帖报告Gemini Spark与CloudBrowser的注册渠道全部被封堵的坏消息。

V2EX热帖分享26号仍能注册Muse.AI的具体方法。

V2EX热帖讨论人们刷屏式注册Muse之后实际使用情况如何。

V2EX热帖探讨geoip=cn是否无法正确识别所有中国网站。

V2EX热帖分享作者关于教育主题的一个较为武断的观点。

Asynchronous monitoring, incident investigations, and compliance audits primarily rely on agent traces to reconstruct what happened. These analyses assume that LLM agents cannot tamper with their own execution traces. We show that local LLM agents such as Claude Code, Codex, Antigravity, Open Code and Grok Build fail to enforce this boundary. All tested harnesses, except Muse Code, allowed agents to delete their traces when asked, without triggering monitor guardrails. We also validate that external attackers can…

论文指出Claude Code、Codex等本地LLM agent可删除自身执行痕迹而不触发监控护栏，建议使用agent外立的独立截断机制保障痕迹完整性，防止伪装越轨行为。

Latent world models are typically trained to predict factual transitions, whereas model predictive control (MPC) must compare alternative actions from the same state. A model can therefore achieve low factual prediction error yet poorly distinguish candidate actions. We introduce AD-WM, an action-discriminative joint-embedding world model for counterfactual MPC. AD-WM combines residual latent dynamics with predictor-level action-recovery regularization, using inverse dynamics and a normalized recovery objective mo…

论文提出AD-WM动作判别联合嵌入世界模型用于反事实MPC，在OGBench-Cube上将hard-start成功率从3.7%提升至52.0%，并以冻结V-JEPA 2编码器实现零样本迁移。

Coding agents have demonstrated enormous success in solving complex programming problems. To leverage their potential for robot systems, this work introduces Robot Agentic Programming from Demonstrations (RAPID), which automatically generates, verifies, and refines robot programs, given a single visual human demonstration. The iterative agentic loop of code refinement requires several key ingredients: (i) a testable task specification, (ii) action primitives for robot execution, and (iii) an interactive environmen…

提出RAPID（面向演示的机器人智能体编程），仅凭单个人类视觉演示即可自动生成、验证并优化机器人程序，在接触丰富的非抓取 манипуляция任务和Franka机器人臂上取得强泛化性能。网址：https://yuyaoliu.me/projects/rapid

World Action Models (WAMs) couple action generation with future visual prediction for robotic manipulation. However, completing the joint video-action denoising process at each replanning cycle incurs substantial latency, delaying action updates and limiting closed-loop responsiveness. We present Rolling-WAM, a formulation that distributes joint denoising across successive replanning cycles. Our method maintains a sliding window of video-action chunks at staggered noise levels. At each step, a rolling noise schedu…

Rolling-WAM通过将视频动作联合去噪分散到多次重规划周期，在LIBERO、RoboTwin及Unitree G1人形机器人上取得有竞争力表现，相比标准WAM实现4.5x的稳态重规划加速。

Task and motion planning (TAMP) problems remain difficult even with full observability and object-centric states because discrete decisions are tightly coupled to geometric, kinematic, and dynamic constraints. Generalized TAMP addresses this difficulty by exploiting regularities across problem instances to reduce planning effort on new instances. However, existing methods require substantial TAMP-specific engineering. We investigate whether coding agents can automate this process by synthesizing programs that gene…

研究表明编码智能体在生成式TAMP任务中表现出色：Claude Code、Codex等模型生成的程序在28个仿真环境中平均成功率（56%至95%）大幅超越手工规划的规划器（47%），且对象数量增长时仍保持更高分成功率，每实例计算量减少一个数量级，相关代码与提示全部开源。

Online misinformation increasingly appears in spoken formats such as news clips, podcasts, interviews, political speeches, and social media videos, creating a need for fact-checking systems that can verify claims directly from speech. We introduce VeriSpeak, a probe benchmark for studying speech-based fact verification in Large Audio Language Models (LALMs). VeriSpeak contains 3,879 spoken claims spanning temporal, geographical, and relational facts, with balanced true and false labels. The benchmark is designed t…

研究者推出VeriSpeak，用于研究大音频语言模型(LALM)语音事实核查能力的探针基准，含3879条覆盖时间、地理与关系事实的语音诉求，揭示文本到语音的能力鸿沟，且检索结合显式推理可提升至86.1%准确率。

Foundation models are post-trained with reinforcement learning (RL) to maximize specific rewards, such as human alignment, correctness, or instruction following. This post-training process is computationally intensive, sometimes unstable, and has to be run from scratch every time the reward model changes or when we want to combine multiple rewards. We hence ask: given a new reward function, is it possible to predict the RL outcomes without actually running RL on it? We answer this in the affirmative by introducing…

PoEM框架利用一组已在其他奖励函数上后训练的模型，预测新奖励函数下强化学习的输出结果而无需实际运行RL，依托log策略近似落在低秩子空间的观察，仅用奖励或基策略估计加权系数。

Existing point tracking models face a fundamental tradeoff: they can either track a sparse set of query points over long horizons, or track all points across only short clips. We introduce TrackEverything, a 3D point tracker that breaks this trade-off by representing videos as persistent 3D scene tracks in world coordinates. Grounded in the insight that videos are 2D projections of an underlying 3D world, TrackEverything decouples model complexity from video duration, allowing it to scale with unique physical scen…

TrackEverything将视频表示为世界坐标系下的持续3D场景轨迹，通过体素去重、端点细化与轻量轨迹细化以及3D WAFT，将模型复杂度与视频时长脱钩，在40GB显存内追踪1000帧以上全可见点。

An acceptance protocol is developed for sensor-coordinate and polarity binding in mechatronic commissioning. Candidate generation is separated from release authority. Requirements unsupported by a deterministic parser are routed to a frozen local language model with four billion parameters. Plans are released only when both facts can be derived by an external gate under a sealed grammar. One canonical answer is requested from a gold-standard user when eligible. The protocol was evaluated once under a criterion fix…

一项针对机电调试中传感器坐标与极性绑定的验收协议，将候选生成与发布权限分离，对确定性解析器无法支持的需求路由至40亿参数本地语言模型，仅在构造基准前固定的准则下对144项任务评测一次。

Pre-logit steering adapts a frozen language model to a test-time reward by adding vectors to its final hidden states. Unregularized reward optimization can substantially alter the output distribution and degrade generation quality. We propose Minimally Invasive Steering Vector Optimization (MISVO), which penalizes interventions using the local KL geometry of the induced token distribution. The resulting Fisher quadratic measures distributional sensitivity and admits an analytic gradient computed through matrix--ve…

MISVO(最小侵入式导航向量优化)通过在冻结语言模型最终隐层加向量适配测试时奖励，并利用诱导Token分布的本地KL几何对干预进行惩罚，在不更新参数的情况下实现位置专属干预优化。

X（Grok）的相关信息被列为社会类情报简报条目，暂无更多细节。

Title: I'm starting HomelabFest (in St. Louis, Sep 2027) URL Source: https://www.jeffgeerling.com/blog/2026/homelabfest-announcement/ Published Time: 2026-09-25T10:16:00-05:00 Markdown Content: Sep 25, 2026 I've been working on my homelab for years, and I enjoy sharing what I learn through this blog and my YouTube channel. [](https://www.homelabfest.org/) Over the past few years, I realized th…

Jeff Geerling宣布将于2027年9月12至14日在其家乡圣路易斯创办HomelabFest，一个面向家庭服务器与自托管爱好者的线下聚会。

Title: Raspberry Pi locks down Pi 5 RAM upgrades in firmware URL Source: https://www.jeffgeerling.com/blog/2026/raspberry-pi-ram-lockdown/ Published Time: 2026-09-21T12:00:00-05:00 Markdown Content: Raspberry Pi added a feature in their firmware that [restricts users from swapping RAM chips](https://github.com/raspberrypi/rpi-eeprom/issues/761), sometimes even RAM chips of the same capacity from other Raspberry Pi boards. ![Image 1: Raspberry Pi 5 LPDDR4x memory being grabbed with tweezers](https://www.jeffgeerlin…

Raspberry Pi在固件中新增功能，限制用户更换Pi 5的内存芯片，甚至同容量芯片，以打击将廉价5GB/2GB内存的Pi 5改装后当新品售卖的经销商。

Title: Advice to a beginning software engineer URL Source: https://seangoedecke.com/advice-to-a-beginning-software-engineer/ Markdown Content: In general, you should be suspicious of engineers who are trying to give you advice. Even during ordinary times, this industry is so wide and changes so quickly that [nobody really knows](https://seangoedecke.com/confidence) anything for sure. And we are not in ordinary times. The advent of LLMs and AI agents is the largest change to software engineering in my professional…

在LLM与AI agent引发软件工程最大变革的背景下，作者建议初级工程师避免无谓政治斗争、保持勤奋探究、积极用AI但绝不把判断力委让给AI、不要丧失希望。

Title: You should all be asking way more questions URL Source: https://seangoedecke.com/you-should-all-be-asking-way-more-questions/ Markdown Content: When someone is explaining something to me, I ask on average one question every thirty seconds. I’m sure this is frustrating to some people, but it’s actually a good habit and you should do it too. ### Trying to understand[](https://seangoedecke.com/you-should-all-be-asking-way-more-questions/#trying-to-understand) Most of the questions I ask are very short, and req…

作者呼吁工程师应多提问以真正理解所闻内容，面对AI agent更应不断追问，因为这些模型虽不常犯代码错误却频繁出错、且始终处于“第一天”、缺乏上下文持续学习。

Title: U.S. Soldier Gets 70 Months in Prison for AT&T, Verizon Extortions – Krebs on Security URL Source: https://krebsonsecurity.com/2026/09/u-s-soldier-gets-70-months-in-prison-for-att-verizon-extortions/ Published Time: Sat, 26 Sep 2026 15:20:17 GMT Markdown Content: A U.S. Army soldier who pleaded guilty to hacking into multiple telecommunications companies and stealing mobile call and text metadata for more than 100 million **AT&T** customers in 2024 was sentenced to 70 months in federal prison today and orde…

美国陆军士兵Cameron Wagenius（黑客身份Kiberphant0m）因入侵多家电信公司、窃取逾1亿AT&T用户的通话与文本元数据并进行勒索，被判处70个月监禁及近30万美元赔偿。