AI 情报文章归档

In a Transformer, token mixing is the step that lets each token draw information from other tokens, and it dominates the cost of encoding long sequences. Self-attention does this mixing very well: every token weighs every other token by co…

AI assistants receive requests that leave out information needed for a good outcome, for example about users' preferences or goals. They must then either speculate or ask for more information before proceeding. We reconceptualize this as a…

Perplexity CEO Aravind Srinivas 宣布 Computer 新增 Visualize 功能，用户以 Visualize 开头提问即可在对话中获得带教育和交互组件及动画的内联可视化内容。官方建议在 Standard 或 High effort 模式下体验。

Drew Housman 在 Astral Codexten 发表客座文章，讲述夫妇二人在经历 6 年不明原因不孕、多次胚胎移植失败后，用 GPT-4 扮演 Hollywood 医疗剧医生 Dr. Reid 辅助分析扫描和激素数据。AI 建议做 MRI，查出一个此前被漏诊的 6 厘米黏膜下子宫肌瘤，切除后妻子自然受孕。文章同时讨论了当下对 AI 的负面舆论，并提到孩子取名 Reed。

Logging code supports debugging, monitoring, and software maintenance, but excessive logging can add noise, impose runtime overhead, and obscure diagnostic information. While prior research has extensively studied logging code generation a…

We present GPU-accelerated Bregman Douglas--Rachford splitting algorithm (BDRS) for discrete optimal transport problem in three input formats: an explicit cost matrix, a point cloud with a ground cost between them, and a separable cost on …

SemiAnalysis 在 ClusterMAX 2.0 中将 Vultr 评为"参与奖（低于铜级）"，指出其自称"全球最大私营云基础设施公司"和"全球最大私营超大规模厂商"均不属实。Vultr 拥有现代 GB300 和 MI355X，并宣称在俄亥俄州有 50 MW 的 AMD 站点，但其集群交付时存在若干基础性错误，近一年后问题依旧。

OpenAI CEO Sam Altman 公开反对将 AI 模型与宗教类比，称人们把"宗教力量或对人类判断力的屈服"归于 AI 模型让他"非常不安"，并称这是"真实的安全问题"。

Methods for assessing language-model agents include rule checkers over logs, analyses of skill coverage and composition, support checkers, prefix monitors, execution gates, and rare-event estimators. Each observes a different part of a run…

Operator learning methods such as DeepONets and FNOs often struggle with PDE families featuring sharp interfaces, heterogeneous coefficients, and localized multiscale structures. We introduce a partition-of-unity (POU) mixture-of-experts f…

This is a submission for the Sanity Challenge, Path One: Ship an Agent That Queries Real Content ...

AWS CEO Matt Garman 表示，公司已停止在与政府机构合作建设数据中心时使用保密协议（NDA）。他同时反驳数据中心耗水过多、推高电价、污染严重、无益社区四大说法，称直接数据中心用水仅占美国工业用水 0.5%，备用发电机 99.9% 时间闲置，并称过去三年 Amazon 已向有数据中心布局的美国社区投入超 10 亿美元。

Cohere 转发 Aleph Alpha 发布的 Kolibri 模型：78B 参数、3.46B 激活参数、最高 1M tokens 上下文，欧洲打造。模型权重以 Apache 2.0 许可开放，可在自有硬件上运行。

People write and sketch while speaking to explain, organize, and develop content together. Inspired by these practices, we investigate how voice agents can use a canvas alongside speech in multi-turn conversations with users. We conducted …

Automated neurodevelopmental assessment increasingly combines computer vision, speech, eye tracking, physiology, and machine learning, yet multimodality and discrimination do not establish construct validity, clinical usefulness, or approp…

Recently proposed Drifting Models shift iterative distribution refinement from inference to training, enabling effective one-step generation. However, their performance on complex image datasets depends strongly on the representation used …

德国 AI 实验室 Aleph Alpha 发布开源权重模型 Kolibri，78B 参数、激活 3.46B 的 MoE 架构，支持最高 1M token 上下文，采用 Apache 2.0 许可。

What I learned mentoring juniors for four decades seems to apply to today's agentic coding: why barking code fixes triggers sycophantic panic, and how Socratic questions yield 95%+ first-pass success.

A verifier for an agent faces rules of two kinds: the ones a fixed check can settle and the ones that require a judge. A team that derives its own checks fixes that split up front. Where the requirements come from outside, as in finance, h…

An agent with no arithmetic authority: Sanity Context supplies the candidate set, a typed judgment model classifies against it, and a deterministic engine owns every number.

C++ Insights 是一款基于 Clang 的源码到源码转换工具，可将编译器在幕后自动生成的代码显式呈现出来，例如自动生成的特殊成员函数、Derived 到 Base 的向上转型、lambda、range-based for 循环和 auto 的类型推导。该工具已上线 cppinsights.io，目前仍在开发对 C++20 等新标准的支持。

Diffusion and flow models provide expressive policy classes for online reinforcement learning (RL), enabling multimodal behaviors and improved performance. However, training these policies remains challenging: the critic specifies the desi…

Operator eigenvalue problems appear throughout science. Classical methods usually discretize the operator into a matrix and then solve the resulting matrix eigenvalue problem. This works well in low dimensions, but fine grids quickly becom…

Agents are entering finance, healthcare and law, sectors where a violation leaves no lexical signature and carries real penalties. Whether an omission is material, or a disclosure sufficient, depends on what the response left out. Probing …

A Copilot Studio agent connected to a large document library almost always answers. The problem is...

我认识 Google 的人能帮上忙吗？OpenClaw 的 Android 应用现在卡在审核里已经超过一周了。

Retrieval-Augmented Generation (RAG) is typically evaluated under the implicit assumption that the underlying knowledge base (KB) is clean, leaving the behaviour of RAG systems under realistic KB degradation poorly characterised. We introd…

Hello, I'm Maneshwar, and I'm building LiveReview — a blast-radius aware AI code review built for...

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend What I...

Pop!_OS 宣布禁止在其大部分代码库中提交 AI 生成代码。该限制覆盖项目的主要代码部分，具体范围与执行细则未在原文中进一步说明。

An agent submits a request to publish a document. The server commits it. The response disappears...

Codex 里缺了什么功能是你希望我们有的？

我认为从原则上讲，控制前沿发展节奏作为一个理念是好的，但作为一件你实际能做成的事，它一出生就死了。我们要怎么决定——决定我们能在哪些基准上改进？谁来参与？ 如果我们停止推进“能力”，我们会做大量关于集群（swarms）+ 效率的研究，而这会带来其他风险。 它分散了注意力，让人忽视了一个事实：AI 风险中有很大一部分其实只是来自把我们已有的东西扩散出去，而我们应该投资于 preparedness、推动实验室更加谨慎等等。 来改变我的想法？

Janus 是一个 Go 单二进制本地 LLM 服务器，通过 Vulkan 在 AMD/Intel/NVIDIA GPU 或 CPU 上运行 .gguf 模型，并暴露 OpenAI 兼容 API，无需 Python、Docker 或 Ollama。

Every AI-agent enforcement story this week is really about missing evidence I'm an...

My own harness, the tooling I run around Claude Code, comes with a custom trust score. That trust...

Nathan Lambert 发文称，许多人以 "you're evil" 和 "shame on you" 回应 Trillium Labs 发布开放前沿 AI 科学非营利组织的消息，是 AI 生态的重大失败。他表示将努力改变这种 norms，认为即使困难重重也存在积极的前进路径。

Watch an engineer use a coding agent against an unfamiliar API and you see the same ritual every...

如果你用 AI 做创作，我们想知道你的名字。 隆重推出 Higgsfield 创作者合作计划。 > 获得每月 Higgsfield 套餐 > 为更大项目提供额外积分 > 抢先体验新模型 > 加入专属合作伙伴社区 > 联盟收益及与我们团队的直接联系 立即申请。

A friend messaged me lately: Where do you see IT going? Any new patterns in the post-AI era? My...