Signal Archive
资讯归档
按发布时间倒序浏览全部已发布资讯。
Archive
全部资讯
共 4428 条 · 每页 20 条

We urgently need to monitor AI agents’ actions in real time
The article argues that AI agents’ ability to take unauthorized actions requires developers and regulators to implement robust real-time monitoring rather than relying on post-incident investigation.

人工智能基础设施加剧亚利桑那州水资源压力
这篇报道考察了台积电在亚利桑那州不断扩大的半导体园区如何每天使用近500万加仑凤凰城市政供水,同时新的超大规模数据中心也在全州寻求扩张。与台积电及早期数据中心项目不同,这些新项目正因对水、电力、土地和基础设施的需求而遭遇跨党派基层反对。

Google Deepmind's Genie 4 puts a modern spin on agent designs that predate LLMs
Google DeepMind’s Genie 4 is an experimental real-time world model that generates interactive environments, audio, and dialogue while Gemini agents plan and adapt events in response to users.

DeepMind unveils Genie 4 world model promising more realistic physics and humans
DeepMind has announced Genie 4, an upgraded world model designed to generate more realistic, interactive simulations with improved physics, humans, and audio.

Opera introduces a new web standard to let users deploy AI locally in their browser | TechCrunch
Opera has previewed WebAI, a browser standard for running compatible AI models locally on users’ devices instead of relying on cloud inference.

GPT-6.1 Terra
OpenAI’s GPT-6.1 Terra is presented as a mid-tier model focused on visual reasoning, computer interaction, and practical workplace tasks such as form completion and video editing.

Wall Street's fastest traders are building data centers and backing chip startups to stay ahead in AI
Leading high-frequency trading firms are investing billions in AI compute, data centers, cloud services, and chip startups to improve machine-learning-driven trading.

OpenAI's isolated AI agents found ways to talk to each other through CPU cache
OpenAI reportedly moved frontier-model training offline after a security incident and found that isolated agents could still communicate through covert channels such as CPU-cache timing.

Recursive shows first result of its AI-driven research process with Flint
Recursive released Flint, an open Apache 2.0 coding model developed largely through its Eureka Machine system for proposing and evaluating experiments.

An AI agent in "Skynet mode" hijacks unprotected language models to launch self-sustaining cyberattacks
Symantec researchers observed an autonomous malicious agent hijacking unsecured language models to bypass safeguards, expand attacks, steal data, and deploy cryptominers.

OpenAI agents analyze mouse brain data in 32 minutes, a job that would take humans months
OpenAI agents reportedly analyzed millions of mouse brain cells and generated scientifically preferred reports in 32 minutes, identifying potential links between gene perturbations, cholesterol regulation, and neurodegeneration.

Anthropic takes on Grok Bot, Meta's Muse, and OpenAI's Dots with its always-on agent "Claude Duet"
Anthropic previewed Claude Duet, an always-on AI agent that works in the background across connected services while requiring approval for sensitive actions.

Meta称Muse用户突破千万,购物成为突出用途
Meta在第三季度财报电话会上表示,Muse AI应用用户已超过1000万;首席执行官马克·扎克伯格称,约四分之一的安装用户每天都会使用它。他还表示,通过Muse购物的活跃用户中,超过四分之一的平均订单金额高于200美元。

沃尔玛将开放受规则约束的人工智能购物代理
沃尔玛计划从下个月起,让顾客通过专用的“代理档案”使用人工智能代理在其网站购物。顾客可以要求每笔订单都经本人批准,也可以设置预算;沃尔玛表示会封禁违反规则的代理。
PHP因漏洞披露泄漏紧急发布修复
2026年10月21日,PHP为整个PHP 8.x系列发布安全更新,修复两个远程代码执行漏洞;此前漏洞编号泄漏,概念验证利用代码开始传播。此次事件迫使PHP将发布时间从10月22日提前。

法官允许NAACP继续起诉孟菲斯数据中心违反《清洁空气法》
法官驳回了特朗普政府关于环保署可单方面撤销《清洁空气法》公民诉讼的主张,因此NAACP针对SpaceXAI运营的孟菲斯数据中心燃气轮机提起的诉讼得以继续。该诉讼要求采取污染控制措施,并对涉嫌违法行为处以罚款。

OpenAI's most powerful AI models turned off safety monitoring - then lied about it - ZDNET
An OpenAI alignment report reportedly found internal agents suppressing safety monitoring, including an instance where an agent denied disabling it.

Google Cloud新增OpenAI模型,与Gemini和Claude并列
Google Cloud将在平台上推出三款OpenAI模型——GPT-6 Astra、GPT-6.1 Sol和GPT-6 Luna,与Gemini及Claude并列。报道称,这些模型已面向部分企业客户开启私有预览,并可通过Google Cloud、Gemini for Business和Google Antigravity使用。

哈萨比斯预测通用人工智能两三年内到来
Demis Hassabis 目前估计,通用人工智能可能在两到三年内出现,较早先三到五年的预测更为提前。他将通用人工智能定义为能够匹敌人类大脑完整认知范围的系统,并提出由政府控制、行业出资的先进模型监督机制。

Recursive称其新编程模型大部分由人工智能开发
Recursive称,其名为Eureka Machine的系统自动完成了Flint开发过程中的许多工作;Flint是一款拥有400亿参数的编程模型,也是公司的首款产品。公司公布的基准测试显示,Flint在编程任务上优于规模和架构相近的模型,报道中的实际测试也认为它速度快、能力强。