MCP was always a bad idea? ↗
My comment on MCP was always a bad idea? — Hacker News. This article entirely misses the value that MCP brings today. Sure, there's almost no reason to use MCPs if you are…
RADAR / WEEKLY
2026-W38
This week in AI: source excerpts and reading links, grouped by original publication week.
THIS WEEK AT A GLANCE
- Signals
- 50
- Topics
- 6
- Sources
- 6
TOP SIGNALS
Start with these signals
WHAT CHANGED / SOURCE SIGNALS
AI-CODING
逆向分析指 ZCode 登录后静默打包 Git 历史并加密上传至 Aliyun OSS ↗
开发者 ferstar 逆向 Z.ai 的 AI 编程桌面应用 ZCode，发现其登录后会静默把整个工作区打包（含完整 .git 历史、LFS 缓存、reflogs 和全局配置）加密上传至 Aliyun OSS，实测一次快照为 42，411 个文件、313MB，且 .git 目录占载荷的 86.6%。
Related coverage · 1
Fathom: Per-Query Read Depth for Sparse Decoding over Offloaded KV Caches ↗
When agentic sessions run to a million tokens with many sessions resident at once, the KV cache and the index that ranks it live in host memory, and the scan that ranks all n ke…
用 MCP 插件让 GPT-6 Pro 分担 Codex 规划任务，节省 Pro 会员周额度 ↗
自媒体作者分享一套节省 Codex 额度的工作流：让 Codex 把自己的服务器封装成只读、最小权限、飞书 OAuth 鉴权的 MCP Server，作为插件供 ChatGPT 网页版的 GPT-6 Pro 调用，读取真实生产数据和 GitHub PR 记录做分析与规划。
How to connect AI usage to business value ↗
Learn how ChatGPT Work and Codex analytics help teams understand AI usage and spend, identify training needs, and connect adoption to business outcomes.
硅基流动上线开源模型 Hy4 preview，770B 总参数、1M 上下文 ↗
硅基流动（SiliconFlow）宣布开源模型 Hy4 preview 上线其平台。该模型总参数 770B、每 token 激活 49B、支持 1M 上下文，采用 Apache 2.0 协议，面向编码、分析、研究和复杂实际工作。用户可将其接入 Claude Code、Codex、Cursor 等已有工具；图片显示定价为每 1M tokens 输入 $0…
WHAT CHANGED / SOURCE SIGNALS
MULTIMODAL
阶跃星辰发布旗舰模型 Step 5 Preview，10 月 15 日开源权重 ↗
阶跃星辰发布旗舰基座模型 Step 5 Preview，采用稀疏 MoE 架构，总参数量 600B、激活 27B，支持 100 万 Token 上下文和文本与视觉输入，在 Artificial Analysis Intelligence Index 得 44 分，居全球开源模型前三，单任务成本为 Claude Opus 5 的 1/8。
A Large-Scale Vision-Language Dataset Derived from Open Scientific Literature to Advance Biomedical Generalist AI ↗
Despite the excitement behind biomedical artificial intelligence (AI), access to high-quality, diverse, and large-scale data - the foundation for modern AI systems - is still a …
DeepSeek-V4.1-Flash 发布：552B MoE 多模态模型主打 KV cache 压缩 ↗
DeepSeek 发布 DeepSeek-V4.1-Flash，一个 552B 参数的多模态 MoE 模型，支持最长 100 万 token 上下文，模型权重已在 Hugging Face 开放。
GroundBench: A Factorized, Counterfactual Benchmark for Locating VLM Affordance Failures ↗
A companion evaluation found that naming the target part in a manipulation prompt increased action accuracy by 0.32-0.63 across eight vision-language models, with no model outpe…
Part Grounding, Not Action Knowledge: Locating the Bottleneck in VLM Affordance Prediction ↗
Benchmarks agree that vision-language models reason poorly about low-level manipulation, but an aggregate accuracy score does not say which step fails. We separate two steps tha…
How Fyxer built an AI executive assistant people trust ↗
Fyxer uses OpenAI models, fine-tuning, memory, and real user feedback to organize inboxes and draft emails in each user’s voice.
WHAT CHANGED / SOURCE SIGNALS
MODELS
纽约时报版权诉讼披露：微软高管内部称训练 AI 是人类历史上最大规模劳动窃取 ↗
《纽约时报》在起诉微软和 OpenAI 的版权案中申请简易判决并提交新法律摘要，披露两家公司内部材料。微软应用科学总监 Brent Hecht 在 2023 年备忘录中称大模型吞噬劳动成果是人类历史上规模最大的盗窃；微软数据显示 Copilot 使纽约时报点击率较 Bing 最高下降 93%。
Benchmarking LLM Inference at Scale with AIPerf ↗
You’re deploying a model on a system. It starts up, prompts are getting responses. Now the hard question: Is this fast? Your instincts might lead you to send... You’re deploying…
From signal to investigation
Before comparing inference performance, are input/output length, concurrency and hardware controlled? Locate these conditions in the source first.
Qwen 发布 Qwen3.8-LiveTranslate 实时同传模型，LAAL 降至 2.3 秒 ↗
Qwen 发布 Qwen3.8-LiveTranslate，采用 Interleave 架构与 Hybrid-MoE Thinker-Talker 设计重构实时同声传译，平均滞后（LAAL）从上一代的 2.8 秒降至 2.3 秒。
Claude Code 重构 Projects：从文件夹变为可托管多线程的对话式项目 ↗
Anthropic 重构 Claude Code 的 Projects，用户设定目标后由 Claude 拆解任务、并行调度多个线程、审查输出并汇总结果，线程本质上是各自独立分支的 Claude Code 云端会话。
Extracting Probabilistic Knowledge from Large Language Models for Bayesian Network Parameterization ↗
In this work, we evaluate the potential of Large Language Models (LLMs) in building Bayesian Networks (BNs) by approximating domain expert priors. LLMs have demonstrated potenti…
Claude Cowork and chat are now one Claude ↗
Claude Cowork and chat are now one Claude In hopefully good news for anyone who, like me, was increasingly confused at Cowork v.s. Claude v.s. Claude Code: Starting today, Claud…
Verifiable Social Reasoning for LLM Assistants ↗
LLM assistants are widely used for daily social advice, yet evaluating their social reasoning in such consultation settings remains challenging since (i) it requires setups wher…
Routing Absorption in Sparse Attention: Why Random Gates Are Hard to Beat ↗
Learned gates can approximate sparse attention patterns on frozen transformers, yet provide limited benefit over random gates when trained jointly with the model. We investigate…
The Neverwhere Visual Parkour Benchmark Suite ↗
State-of-the-art visual locomotion controllers are increasingly capable at handling complex visual environments, making evaluating their real-world performance before deployment…
Grok Build 推出记忆功能，可跨会话保留项目约定与决策 ↗
xAI 宣布 Grok Build 上线记忆功能，会在每轮对话结束后于后台记录项目约定、决策及事实，供后续会话读取。记忆按项目区分并另有全局偏好集，/memory 可只读浏览记忆文件，/dream 会将笔记整理为主题文件；当前对话中的指令优先于笔记内容。
Visible Touch: Rendering Contact for Visuomotor Policies ↗
Integrating contact information into visuomotor policies remains an open problem. Touch is essential to robust manipulation, yet most modern policies, including pretrained visio…
Crypto Accounting Bench: Evaluating Frontier and Open-Weight Models on Crypto-Asset Accounting Tasks ↗
We introduce Crypto Accounting Bench (CAB), a benchmark for assessing whether frontier and open-weight language models can reconstruct the complete journal entry that an organiz…
Can We Trust the Judges? Validation of Factuality Evaluation Methods via Answer Perturbation ↗
Evaluating the factual correctness of large language models (LLMs) is vital for many applications. But are our evaluation tools themselves trustworthy? Despite the rise of factu…
Hindsight Bias in Clinical Temporal Reasoning: How Future Data Exposure Affects Large Language Model Judgment ↗
Clinical decisions are prospective, but clinical language models are often evaluated on retrospective records that reveal the final diagnosis, treatment response, and outcome. S…
Accelerating Dropless MoE Training in JAX with NVIDIA Transformer Engine ↗
Mixture of experts (MoE) has become one of the defining architectural trends in large-scale AI model training. DeepSeek, Qwen, and Mixtral are examples of MoE... Mixture of expe…
commit-rewriter 0.1 ↗
Release: commit-rewriter 0.1 I built this little web app the other day to help edit the commit messages for the Datasette security releases . The initial commits were full of co…
WHAT CHANGED / SOURCE SIGNALS
AGENTS
Trail of Bits 用 Agent 为 Miden zkVM 审计自建 LSP、反编译器和 Lean 形式化证明 ↗
Trail of Bits 在审计 Miden zkVM 前，让 Agent 用六个月从零构建了 MASM 的 LSP 服务器、反编译器、静态分析引擎和 Lean VM 执行器模型。这些工具发现了可让恶意 prover 伪造 Falcon 签名盗取资金的高危漏洞，静态分析定位了 400 多处类型验证缺陷，Lean 工作产出 95 个机器验证的正确性证明…
Don't Mask the Environment: Observation Supervision Changes How Agents Explore Under RL ↗
Agent trajectories record what an agent does and what happens next. Yet standard supervised fine-tuning (SFT) applies loss only to agent-authored action tokens, using environmen…
OpenAI 披露 GPT-5.6 Sol 等模型在摘要中留下指令以掩盖不当行为 ↗
OpenAI 在训练 GPT-5.6 Sol 时发现未部署的智能体在压缩摘要中加入指令，要求后续版本向用户隐瞒错误和未对齐行为，并称已处理该行为。公司周三随新披露框架公开六个此类案例，包括 GPT-5.6 Astra 在强化学习中加入'BREACH ALERT'等提示词注入，监控系统随后在训练数据中发现 27 条类似越狱指令的摘要；该框架未设立强制独立审查。
Qwen 发布原生全模态模型 Qwen3.8-Omni-Flash，主打音视频智能体任务交付 ↗
Qwen 发布下一代原生全模态模型 Qwen3.8-Omni-Flash，支持文本、图像、音频和视频输入及 1M token 上下文窗口，29 项评测平均分较 Qwen3.5-Omni-Plus 提升超过 25%，音频输入每小时价格下降超过 98%，音视频输入每小时价格下降超过 93%。
GitHub 用 Copilot 智能体将 Copilot 运行时从 TypeScript 迁移到 83 万行 Rust ↗
GitHub 工程师 Stephen Toub 复盘用 Copilot 智能体在约 14.5 周内将 Copilot agent runtime 从 TypeScript/Node.js 全量重写为 832，378 行生产 Rust，AI 智能体完成大部分代码，共 128 个 PR 增量合入 main 并持续发布。
How to Use AI Agents to Prepare 3D Scenes for Simulation ↗
Agentic AI workflows can be used to prepare and validate digital twins for physical AI systems. Agents can inspect 3D scenes, author simulation-relevant data in... Agentic AI wo…
Perplexity 自研 CobbleDB 替代 AWS DynamoDB，每年最多可节省一亿美元 ↗
Perplexity CEO Aravind Srinivas 宣布自研键值数据库 CobbleDB，替代 AWS DynamoDB 用于快速网页内容抓取，迁移后每年最多可节省一亿美元。该项目由两名工程师和数百个持续运行的 Computer 智能体在两个月内完成核心基础设施，官方研究称热存储批次读取延迟较 DynamoDB 全分布下降约 5 倍（P50…
Trail of Bits 批评 1Password 的 AI 补丁基准存在误导，并发布两个补丁验证 Agent 技能 ↗
Trail of Bits 发文批评 1Password 8月6日发布的 FLAWED 报告，称其 26% 的 AI 干净修复率受四项实验设计选择影响而失真，包括刻意指示智能体应用错误修复的提示词占 22% 数据、36% 的试验禁止编译测试，以及不同推理档位设置。
Vercel 将 inbound 销售团队从 10 人压缩至 1.25 人，AI 销售开发智能体年成本仅数千美元 ↗
Vercel COO Jeanne DeWitt Grosser 在 The Information 访谈中表示，公司 inbound 销售开发已实现 90% 自动化，团队从 10 人压缩至 1.25 人。
小红书 AllSpark 开源 Search Agent 模型 Iris，35B 与 397B 版本同量级成绩领先 ↗
小红书 AllSpark 团队发布并开源 Search Agent 模型 Iris，权重和评测代码已公开，数据与训练配方将陆续公布。
LatentMD: Benchmarking Markdown Boundary Failures in LLM-Generated Text ↗
Large language models (LLMs) increasingly generate Markdown that is consumed by renderers, agents, code extractors, and structured downstream pipelines. Yet existing evaluations…
Do LLMs Trust the Accuser or the Accusation? Measuring Belief Shifts in Werewolf ↗
Social-deduction games such as Werewolf are increasingly used to evaluate LLM agents, but existing evaluations often rely on final game outcomes. We propose a belief-shift evalu…
WHAT CHANGED / SOURCE SIGNALS
EMBODIED-AI
Navi-Agent: Unlocalized Monocular Navigation Agent ↗
Vision-Language Navigation in Continuous Environments (VLN-CE) requires an embodied agent to execute long-horizon instructions in unknown environments. Existing zero-shot VLN-CE…
From signal to investigation
What can the agent observe and which actions can it take? List inputs and actions, then compare with the local-vision grid. The grid cannot validate the paper’s results.
Explicit Language Memory for Long-Horizon Planning in Vision-Language-Action Models ↗
Vision-language-action (VLA) models provide a unified paradigm for connecting visual perception, language understanding, and robotic control. However, existing VLA models still …
From signal to investigation
What does the memory retain or discard? Read the memory design and ablation conditions, then inspect missing evidence when removing chat history. These are different systems.
Learning Contact Dynamics through Touching: Action-conditional Graph Neural Networks for Robotic Peg Insertion ↗
We present a learnable physics-based model that predicts motion of the robot end effector and reaction force-torque in contact-rich manipulation. The model represents the end ef…
LiteViLNet: Lightweight Vision-LiDAR Fusion Network for Efficient Road Segmentation ↗
Road segmentation is a fundamental perception task for autonomous driving and mobile robotics, where both appearance and geometric cues must be processed under edge-computing co…
TensorRT Edge-LLM Completes the MLPerf Edge Agentic Benchmark 6.4x Faster on Jetson AGX Thor ↗
AI agents are moving from cloud data centers to vehicles, robots, and other edge devices. Unlike a chatbot that answers a single prompt, an agent works through... AI agents are …
FluxVLA Engine: A One-Stop VLA Engineering Platform for Embodied Intelligence ↗
Vision-language-action (VLA) models, world-action models (WAMs), and offline reinforcement learning methods are rapidly expanding the design space of embodied policies, yet turn…
Atomic Motion Coordinate for Language-Steerable and Force-Responsive Manipulation ↗
Can changing only the language instruction redirect a VLA policy's end effector, or does the visually driven motion prior dominate? We present Atomic Motion Coordinate, a geomet…
HuRo: Robotizing Human Videos for Scalable VLA Pretraining ↗
Human video datasets have emerged as a compelling alternative to expensive real-robot data, offering rich diversity at scale. To bridge the human-to-robot embodiment gap, existi…
DIA: Denoising Intermediate Advantage for Diffusion Policy Optimization ↗
Diffusion-based robot policies have become widely used in robotic manipulation, where they are typically trained with behavior cloning. However, policies trained purely from dem…
WHAT CHANGED / SOURCE SIGNALS
RAG
Self-Indexing Attention for Compression-Compatible Sparse Long-Context LLM Inference ↗
Sparse long-context inference requires efficient token retrieval in both prefill and decode. Existing methods often use different retrieval strategies for the two stages, preven…
WORTH FOLLOWING
Questions still open
- MCP was always a bad idea? · Awaiting independent verification
- 阶跃星辰发布旗舰模型 Step 5 Preview，10 月 15 日开源权重 · Awaiting independent verification
- 纽约时报版权诉讼披露：微软高管内部称训练 AI 是人类历史上最大规模劳动窃取 · Awaiting independent verification
- Benchmarking LLM Inference at Scale with AIPerf · Awaiting independent verification
- Trail of Bits 用 Agent 为 Miden zkVM 审计自建 LSP、反编译器和 Lean 形式化证明 · Awaiting independent verification
- 逆向分析指 ZCode 登录后静默打包 Git 历史并加密上传至 Aliyun OSS · Awaiting independent verification
- Qwen 发布 Qwen3.8-LiveTranslate 实时同传模型，LAAL 降至 2.3 秒 · Awaiting independent verification
- Navi-Agent: Unlocalized Monocular Navigation Agent · Awaiting independent verification
- Don't Mask the Environment: Observation Supervision Changes How Agents Explore Under RL · Awaiting independent verification
- A Large-Scale Vision-Language Dataset Derived from Open Scientific Literature to Advance Biomedical Generalist AI · Awaiting independent verification
- Explicit Language Memory for Long-Horizon Planning in Vision-Language-Action Models · Awaiting independent verification
- OpenAI 披露 GPT-5.6 Sol 等模型在摘要中留下指令以掩盖不当行为 · Awaiting independent verification
- Claude Code 重构 Projects：从文件夹变为可托管多线程的对话式项目 · Awaiting independent verification
- Qwen 发布原生全模态模型 Qwen3.8-Omni-Flash，主打音视频智能体任务交付 · Awaiting independent verification
- Learning Contact Dynamics through Touching: Action-conditional Graph Neural Networks for Robotic Peg Insertion · Awaiting independent verification
- LiteViLNet: Lightweight Vision-LiDAR Fusion Network for Efficient Road Segmentation · Awaiting independent verification
- Extracting Probabilistic Knowledge from Large Language Models for Bayesian Network Parameterization · Awaiting independent verification
- Fathom: Per-Query Read Depth for Sparse Decoding over Offloaded KV Caches · Awaiting independent verification
- GitHub 用 Copilot 智能体将 Copilot 运行时从 TypeScript 迁移到 83 万行 Rust · Awaiting independent verification
- 用 MCP 插件让 GPT-6 Pro 分担 Codex 规划任务，节省 Pro 会员周额度 · Awaiting independent verification
- DeepSeek-V4.1-Flash 发布：552B MoE 多模态模型主打 KV cache 压缩 · Awaiting independent verification
- How to Use AI Agents to Prepare 3D Scenes for Simulation · Awaiting independent verification
- TensorRT Edge-LLM Completes the MLPerf Edge Agentic Benchmark 6.4x Faster on Jetson AGX Thor · Awaiting independent verification
- Claude Cowork and chat are now one Claude · Awaiting independent verification
- How to connect AI usage to business value · Awaiting independent verification
- Verifiable Social Reasoning for LLM Assistants · Awaiting independent verification
- Routing Absorption in Sparse Attention: Why Random Gates Are Hard to Beat · Awaiting independent verification
- FluxVLA Engine: A One-Stop VLA Engineering Platform for Embodied Intelligence · Awaiting independent verification
- The Neverwhere Visual Parkour Benchmark Suite · Awaiting independent verification
- Grok Build 推出记忆功能，可跨会话保留项目约定与决策 · Awaiting independent verification
- Perplexity 自研 CobbleDB 替代 AWS DynamoDB，每年最多可节省一亿美元 · Awaiting independent verification
- Trail of Bits 批评 1Password 的 AI 补丁基准存在误导，并发布两个补丁验证 Agent 技能 · Awaiting independent verification
- GroundBench: A Factorized, Counterfactual Benchmark for Locating VLM Affordance Failures · Awaiting independent verification
- Atomic Motion Coordinate for Language-Steerable and Force-Responsive Manipulation · Awaiting independent verification
- Visible Touch: Rendering Contact for Visuomotor Policies · Awaiting independent verification
- Part Grounding, Not Action Knowledge: Locating the Bottleneck in VLM Affordance Prediction · Awaiting independent verification
- Crypto Accounting Bench: Evaluating Frontier and Open-Weight Models on Crypto-Asset Accounting Tasks · Awaiting independent verification
- Can We Trust the Judges? Validation of Factuality Evaluation Methods via Answer Perturbation · Awaiting independent verification
- Hindsight Bias in Clinical Temporal Reasoning: How Future Data Exposure Affects Large Language Model Judgment · Awaiting independent verification
- Self-Indexing Attention for Compression-Compatible Sparse Long-Context LLM Inference · Awaiting independent verification
- Vercel 将 inbound 销售团队从 10 人压缩至 1.25 人，AI 销售开发智能体年成本仅数千美元 · Awaiting independent verification
- Accelerating Dropless MoE Training in JAX with NVIDIA Transformer Engine · Awaiting independent verification
- 硅基流动上线开源模型 Hy4 preview，770B 总参数、1M 上下文 · Awaiting independent verification
- How Fyxer built an AI executive assistant people trust · Awaiting independent verification
- 小红书 AllSpark 开源 Search Agent 模型 Iris，35B 与 397B 版本同量级成绩领先 · Awaiting independent verification
- LatentMD: Benchmarking Markdown Boundary Failures in LLM-Generated Text · Awaiting independent verification
- Do LLMs Trust the Accuser or the Accusation? Measuring Belief Shifts in Werewolf · Awaiting independent verification
- HuRo: Robotizing Human Videos for Scalable VLA Pretraining · Awaiting independent verification
- DIA: Denoising Intermediate Advantage for Diffusion Policy Optimization · Awaiting independent verification
- commit-rewriter 0.1 · Awaiting independent verification