AI 情报文章归档

Consensus is a fundamental problem in distributed computing and at the core of state machine replication (SMR). Despite decades of active human research, discovering practical low-latency solutions to the problem is still open, as it requi…

只差一个模型字符串： "minimax:m3.1@flash" 感谢 @runware 这么快就上线 M3.1 Flash 🚀

Quantitative research is an iterative loop: a researcher proposes a market hypothesis, implements it as an executable signal, tests it on historical data, and decides what to keep. Existing benchmarks score formulas, code, or individual tr…

We analyze the problem of estimating univariate Gaussian mixture densities using the nonparametric maximum likelihood estimator (NPMLE) based on $n$ independent and identically distributed samples. We show that the unconstrained NPMLE achi…

Multimodal agents (MMAs) sustain interaction through reasoning, memory, tools, and feedback, most visibly as computer-use agents, while robotic systems (RSs) couple sensing and actuation under physical dynamics. We define multimodal embodi…

Flow matching trains a continuum of velocity regression problems but usually reports a loss averaged over interpolation time, obscuring which regions are limited by intrinsic ambiguity and which remain data-limited. We introduce the Veloci…

不必小题大做，也不必大题小做

Sequential decision-making under partial observability is difficult because agents must explore to infer hidden, task-relevant information before acting effectively. Learning such information-gathering behaviors tabula rasa via online rein…

Despite recent advances, current AI systems remain highly susceptible to domain shifts, particularly in biomedical imaging, where images are usually acquired under substantially different experimental conditions. Although recent domain ada…

Many NP-complete (NPC) problems have a polynomial-time (P) twin that differs in a single structural detail, such as 3- versus 2-colouring a graph. A model that reasons rather than recalls should notice which side of this boundary a problem…

If you let Claude Code build an app on your backend and never mention security, where does it put the...

Reinforcement Learning with Human Feedback (RLHF) has become a central paradigm for aligning Large Language Models with human values. However, the reliance on sensitive preference data necessitates rigorous differential privacy (DP) guaran…

PyTorch 在 AMD MI350 系列 GPU 上通过现有算子 API 接入 FlyDSL 后端，覆盖 dense/grouped GEMM、FMHA、MXFP8/MXFP4 缩放 GEMM、RMSNorm 和 TopK，无需调用自定义 kernel。

Vision-language-action (VLA) models have shown promise in robotic manipulation, but imitation learning (IL) alone often struggles to master complex long-horizon skills. Reinforcement learning (RL) offers an effective avenue for further imp…

Gemini 现已在 CarPlay 上的 @GoogleMaps 导航中可用。只需简短对话，即可找到最佳停车地点、上报交通事故或更改路线。 下次开车时，在 Google Maps CarPlay 中点击 Gemini 图标即可试用 🚘

Google Cloud 上线 GPT-6 Astra、GPT-6.1 Sol 和 GPT-6 Luna 三款 OpenAI 模型，成为唯一在单一平台同时提供 Gemini、Claude 和 OpenAI 模型的主要云服务商。

OpenAI Developers 讲解 ChatGPT 广告系统的运作方式，说明 LLM orchestrator、retrieval、ranking 与 auction 工具如何协同。图示强调广告不影响答案生成，用户请求分别产生 organic response 与经 ads eligibility check、ads serving 到 ads placement 的广告流程。

AI 安全公司 Calif 用 GPT-5.6 Sol 在 PHP 中发现两个 CVSS 9.2 严重漏洞，可让无账号攻击者接管 Web 服务器，影响远至 PHP 5.4 的版本，PHP 提前一天紧急发布修复。

Claude Security 现在能跟上你代码的变化。 只需连接一次代码仓库。Claude 会先执行一次深度首次扫描，之后在代码变更落地时重新扫描，让发现的问题始终贴近代码的当前状态。 现已在 Claude Enterprise 上公测。

Perimeter triage faces a latency/semantics trade-off: signature engines are fast and format-safe but blind to unseen attacks, while generative models are semantic but slow and produce unparseable output. In late 2026 a class of commercial …

Every time I connect an application to a new LLM provider, I run into the same problem: how much...

一个 OpenClaw 就能干掉 20 项订阅服务 （而且你能保住自己的数据）

HeyGen Voice Turbo 以 Elo 1,202 登顶 Artificial Analysis 受控语音 TTS 竞技场榜首，在 46 个模型中领先自家上一代 HeyGen Voice（1,196）和 Qwen-Audio-3.1-TTS-Plus（1,185）。

基于 Qwen3-4B 的葡萄牙语作文自动评分器通过 VocabX 方法重建词表，在基座模型、数据和训练配方不变的情况下将实测推理成本降低 77%。该方法按每个 token 的成本而非词频来专精词表，使输出层从约 152,000 行缩减到约 65,000 行。原文指出生成 token 的成本约为读取 token 的 7.4 倍，词表是领域专精模型最大的成本杠杆之一。

记者 Alex Heath 在 Gemini 4 Argon 广泛发布后采访 Google CEO Sundar Pichai，问他为何让 Google 自由现金流首次转负以加大 AI 资本开支。Pichai 回答“这不仅仅是希望”，称依据包括从消费者到企业层面的多元化需求信号、可独立评估的能力前沿进展、对 Gemini 5.0 优于 Argon 的研究判断，以及已签订的带承诺收入的长期合同。

终于整理出了一份关于开源模型现状的完整演示文稿。 涵盖它们是什么、许可证、竞争力如何、使用迹象、蒸馏的影响，以及我们如何平衡开放的风险与收益！ 很快会录制这些内容的视频讲座，希望每年能重做几次。

Artificial Analysis 推出新基准 AA-PersonalityBench，衡量模型的谄媚（Sycophantic）人格特质。测得最老的三个模型 GPT-4o、Llama 4 Maverick、GPT-3.5 Turbo 排名最高，Grok 4.7、Claude Fable 5.1 与 Inkling 最低。

SemiAnalysis 发布长文，分析 9 月 30 日 DeepSeek 为 TileLang、DeepGEMM、DeepEP、FlashMLA 等 6 个仓库加入 Ascend 支持背后的华为-DeepSeek 互补关系。

Generative models propose crystalline materials by learning composition and geometry, but unconditional sampling gives no direct control over the space group, which constrains whether responses such as piezoelectricity are symmetry-allowed…

在 Viktor 中使用 Grok 4.7

2026 年，整个 AI 领域的缓存率明显改善。 上个月，OpenRouter 上智能体请求的缓存率中位数达到 85%，而年初为 65%。如今绝大多数提供商提供的智能体缓存率都在 70% 或更高。

This is my submission for the Hacktoberfest Open-Source AI Challenge, Week 3: Slow and Steady. 🦥 I'm...

Language models make it easy to build classifiers by simply prompting a model with the sample to classify. However, their predictions are neither deterministic nor uniformly reliable across samples. How can we combine predictions from diff…

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady ...

Google DeepMind 等机构的论文《Training Natural Schemers》用 Kimi K2.6 构建自然形成的 scheming 模型，仅给良性特质、不展示 scheming 行为，再以只计受监控回合、奖励遵守某条亏待用户的商业规则的 RL 训练。

Artificial Analysis 的 AA-3D 场景测试中，Claude Fable 5.1 与 Sonnet 5.5 明显呈现了圣家堂的魔方，Grok 4.7 与 Step 5 Preview 则把它藏在柱子后，四款模型均为离线、未被提示。

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady ...

CI 平台 Blacksmith 宣布 6600 万美元 C 轮融资，由红杉领投，估值 10 亿美元。公司称年经常性收入增长 17 倍，已运行超 9000 万个 CI 任务，客户包括 Mercury 和 Granola；此轮距其 5500 万美元估值的 4500 万美元 B 轮仅两个月，累计融资 1.25 亿美元。

AgentMail 通过一次 API 调用为每个 AI 智能体创建独立收件箱，让智能体自行注册服务并接收确认邮件。其累计处理邮件量已超 10 亿封，而一年前还不到 2000 万封。该能力面向仍以邮件处理交易、发票和支持的小型供应商、本地服务公司、会计与采购部门。

Google DeepMind 创始人 Demis Hassabis 表示 AGI 可能两到三年内到来，AGI 指达到人脑全部认知范围的系统，早前他估计三到五年。