模型发布/更新
-
SGLang provides Day-0 support for Muse Glimmer, optimizing inference for local agent workflows
SGLang partnered with Meta Superintelligence Labs to provide Day-0 support for the 30B-parameter multimodal model Muse…
-
Meta releases open-source model Muse Glimmer
Introducing Muse Glimmer, an open-weight, 30-billion-parameter model optimized for local, always-on agent workflows.…
-
OpenAI launches GPT-5.6-Cyber, a cybersecurity-specific model for authorized vulnerability research
OpenAI released GPT-5.6-Cyber, a cybersecurity-specific model available through Daybreak Red for authorized…
产品发布/更新
-
OpenRouter launches new Auto router driven by marketplace intelligence
Based on community consumption data of over 55T tokens per week, OpenRouter launched a new Auto router…
-
Qwen Open Platform launches: conversational services in over ten domains including rentals, shipping and wealth management
The Qwen Open Platform launched today, opening service access for ecosystem partners and developers across phones, PCs…
-
Qwen-MM-Plugins gives agents native multimodal support
👀 Seeing is just the beginning. With Qwen-MM-Plugins, give your agent native multimodal support — read images, videos…
-
How Claude Code's auto mode works by default
We recently made auto mode the default in Claude Code, which means you no longer need to approve every action. But…
-
I spent 54 hours building a potentially fairer AI large model leaderboard
The author spent 54 hours developing and freely releasing LatentRank, an aggregate AI large model leaderboard…
-
Omnigent 上下文策略如何阻断“致命三重奏”组合攻击
Omnigent 的上下文策略(Contextual Policies)可阻断“致命三重奏”组合攻击——即多个看似无害的请求叠加后形成的恶意行为。该机制在单个请求层面不设防,而是在组合上下文中识别并拦截风险,从而在不牺牲正常功能的前提下提…
-
Google 为 Google Ads 和 Google Analytics 推出新 AI 与智能体功能
Google 宣布在 Google Ads 和 Google Analytics 中新增 AI 与智能体体验,包括 Google Analytics 首页的 AI Overviews 智能摘要、Google Ads 首页的 AI 洞察卡片…
行业动态
-
NVIDIA partners with six institutions to raise $500 billion for AI factories
NVIDIA announced partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to establish an…
-
Zapier 如何用 ChatGPT Work 改造核心营销流程
Zapier 企业营销团队用 ChatGPT Work 自动化线索漏斗优化、营销素材搭建和报告生成,每月可对数千条线索执行 QA/QC,单人查看一条流失线索原本需 35 至 45 分钟。该系统每月为销售团队带来七位数管道价值,并生成高管仪…
论文研究
-
Unreleased research version of Claude raises lower bound for Riemann zeta zeros from 41.6% to 67.2%
Anthropic staff asked Claude to attempt the Riemann hypothesis. Although it did not succeed, an unreleased research…
技巧与观点
-
Can agents really use computers? a16z answers with data
a16z data shows that the best score for computer-use agents on the OSWorld-Verified benchmark has risen from 42% a…
-
Zuckerberg: superintelligence should be available to everyone
I believe everyone should have access to superintelligence. I wrote a long piece laying out Meta's philosophy and…
-
Linear explains how Linear Agent was built and its design tradeoffs
Linear published a post explaining how Linear Agent was built, with the core approach being not to design fixed paths…
-
Forking-Sequences:一种统计与计算上更高效的多步预测训练范式
CMU 研究团队正式定义并系统评测了 forking-sequences 训练范式:它无需新增参数,在一次前向传播中跨所有预测创建日期编码解码整条序列。
-
开源不等于开放权重:Gary Marcus 剖析两者本质差异
开放权重模型并非真正的开源,二者在透明度和可定制性上存在根本区别。开源软件公开完整源代码,允许任何人查看、修改和分支;而开放权重模型仅发布训练后的神经网络权重,用户无法访问原始训练数据、预处理方法或训练算法,也不能自由修改或深入调查。这导…