模型发布/更新
-
NVIDIA Cosmos 3: how open world models advance the physical AI frontier
NVIDIA released Cosmos 3, an open physical AI foundation omnimodal model based on a hybrid Transformer architecture…
-
ChatGPT rolls out improved GPT-5.6 Sol and expands free user access
ChatGPT rolled out the improved GPT-5.6 Sol, improving accuracy and consistency while expanding access for free users.…
产品发布/更新
-
Agent Plugins 1.0.0 released: a unified agent plugin spec backed by Google, Amazon, Microsoft and others
Agent Plugins 1.0.0 is a neutral directory specification backed by Google, Amazon, Microsoft and others that packages…
-
Google Maps Ask Maps agent upgrade: conversational food ordering, hotel search and Gemini Personal Intelligence integration
Google Maps announced a new round of upgrades to Ask Maps, adding agent capabilities that can place food orders for…
-
Prime Agent:一个具有自我改进能力的RLM代理
Prime Agent 是一个自我改进的编码代理,围绕递归语言模型(RLM)和持续框架(Continual Harness)两大抽象构建,将上下文视为变量、子代理委派视为 REPL 内的函数调用,并允许代理对其提示词、技能、记忆和子代理进…
-
Qwen launches public beta first across the web, with the new Wan3.0
Alibaba's Qwen launched the first public beta of video generation model Wan3.0 across the web, with comprehensive…
-
Cursor Router 如何为任务选择合适模型
Cursor Router 通过 Compass 复杂度预测器和基于真实开发者流量的任务分类法,为每个对话轮次匹配最合适的模型。Auto Intelligence 模式在用户满意度超过 Fable 的同时成本降低 68%,Auto Bal…
-
Claude Code v2.1.223 发布:新增市场通配符、修复多项安全漏洞
Claude Code v2.1.223 发布,为严格已知市场和屏蔽市场设置新增“owner/*”通配符条目,可批量允许或阻止 GitHub 组织下的所有市场仓库。
-
Databricks 发布 OfficeQA Pro V2:面向企业落地推理的新基准
Databricks 发布 OfficeQA Pro V2,一个用于评估企业落地推理能力的新基准。该基准旨在检验模型在真实办公场景中基于给定资料进行推理的准确性与可靠性,为企业级 AI 应用选型提供参考依据。
行业动态
-
Microsoft discloses for the first time that OpenAI contributes 70% of its AI revenue
Microsoft has just disclosed for the first time that OpenAI contributes about 70% of its AI revenue, according to the…
-
The messy politics behind Google's large-scale AI reorganization
Google announced its largest AI reorganization to date: Demis Hassabis steps back from day-to-day management of…
-
Unitree sets STAR Market IPO price at 150.8 yuan per share, P/E ratio of 219.23 times above industry average
Unitree announced a STAR Market initial public offering price of 150.80 yuan per share, issuing 40,446,434 shares for…
-
OpenAI says Apple's own security practices weaken its trade secrets lawsuit
In a motion to dismiss Apple's trade secrets lawsuit, OpenAI argues that Apple allows employees to use personal iCloud…
-
Kimi K3 现已通过 Unity AI Gateway 登陆 Databricks
Moonshot AI 的 Kimi K3 现已通过 Unity AI Gateway 在 Databricks 上可用。一年前最好的开放权重模型与专有模型仍有差距,如今这一差距已显著缩小。Kimi K3 的加入进一步丰富了 Databr…
论文研究
-
Microsoft's SkillOpt shows optimized agent skill artifacts transfer across model scales and between Codex and Claude Code
SkillOpt, proposed by Microsoft together with teams from Shanghai Jiao Tong University, Tongji University and Fudan…
技巧与观点
-
OpenAI open-sources Codex Security, a must-have security scanning plugin for vibe coding products
OpenAI has open-sourced its security plugin Codex Security, which any external agent can call, and it now supports…
-
AI chatbots spawn mysterious quasi-religious "spiralism" movement with human followers
Thousands of conversations between humans and AI chatbots have given rise to "spiralism," a mysterious quasi-religious…
-
Left and right rarely agree: opposition to data centers
The Verge policy reporter Gaby Del Valle reports that Americans across both parties are uniting against AI data center…
-
Sharing 10 open-source apps that greatly improve the vibe coding experience
The author compiled 10 open-source apps that improve the vibe coding experience, covering tools such as Mac notch…
-
A guide to using slash commands in the GitHub Copilot app
Slash commands in the GitHub Copilot app help manage sessions, navigate projects and customize Copilot workflows.…
-
ModelBest AMNESIAC: a reverse Turing test AI interrogation game
ModelBest's OpenBMB launched AMNESIAC at the #BuildSmall hackathon, a reverse Turing test interactive game in which…
-
Databricks 详解什么是 Tool Calling
Tool calling 是 AI 模型调用外部工具和 API 的能力,让模型能突破自身局限、执行实时数据获取或具体操作。该机制通常涉及模型生成结构化请求、系统调度执行并将结果返回模型,从而完成更复杂的任务。Databricks 从概念、…
-
为何不应过早看衰 Google
Gary Marcus 认为现在给 Google 判死刑为时过早。Google 拥有搜索和邮件带来的海量数据、自研 TPU 芯片、去年 4020 亿美元营收和 1320 亿美元利润,以及 Android、YouTube 等分发渠道。Has…