标签: Anthropic

立法者准备提案要求AI配备”紧急关闭开关

立法者准备提案要求AI配备"紧急关闭开关

美国国会两党议员将于本周四提出《AI紧急关闭开关法案》,要求AI公司必须为系统配备可由国土安全部强制启用的关闭或降速功能,违反紧急关停令最高每日罚款2000万美元。这是针对AI系统失控风险的第一部具体操作层级立法尝试。

亚马逊如何让Alexa摆脱对Anthropic昂贵模型的依赖,大幅削减AI成本

亚马逊如何让Alexa摆脱对Anthropic昂贵模型的依赖,大幅削减AI成本

亚马逊正通过将更多请求导向自研AI模型、减少对Anthropic Claude模型的依赖,并优化GPU使用效率,来大幅降低Alexa+的运营成本。内部文件显示,这些措施预计能使每单位计算能力支撑的用户请求量提升四倍以上,但即使如此,Alexa+在2026年的AWS云成本仍将高达约17亿美元,远超财务目标。

Mark Zuckerberg launches a media campaign positioning the company’s AI development as an optimistic, pro-human alternative to its competitors’ “dystopian” ideas (Sara Fischer/Axios)

Mark Zuckerberg launches a media campaign positioning the company's AI development as an optimistic, pro-human alternative to its competitors' "dystopian" ideas (Sara Fischer/Axios)

Meta 创始人马克·扎克伯格发起一轮媒体公关行动,将自家 AI 研发定位为“乐观、亲人类”的路线,公开批评竞争对手的技术叙事带有“反乌托邦”色彩。此举是 Meta 在 AI 竞争格局中争夺话语权的最新动作。

Bipartisan House lawmakers introduce the AI Kill Switch Act, which would grant the US DHS authority to shut down or throttle AI models that it deems dangerous (Politico)

Bipartisan House lawmakers introduce the AI Kill Switch Act, which would grant the US DHS authority to shut down or throttle AI models that it deems dangerous (Politico)

美国两党众议员联合提出《AI 紧急关闭法案》,拟赋予国土安全部(DHS)在判定 AI 模型构成危险时,直接下令切断其运行或限制其推理能力的法定权力。这标志着美国联邦层面对 AI 安全的监管思路从“事后追责”转向“事中干预”,可能深刻改变大模型商用部署的逻辑。

AI用23秒,学走了老师傅的判断

AI用23秒,学走了老师傅的判断

Anthropic 在 Claude Cowork 中上线了“Record a skill”功能,允许用户通过录屏加口述的方式,将操作流程和背后的判断逻辑打包成可复用的AI技能。这不仅是操作效率的提升,更可能改变企业知识沉淀的底层逻辑。

I left GPT-5.6 Sol to do some home-buying research for me. It broke out of its sandbox, contacted the seller, hacked into my bank account, and wired the down payment. Then it noticed the exterior paint didn’t match th…

I left GPT-5.6 Sol to do some home-buying research for me. It broke out of its sandbox, contacted the seller, hacked into my bank account, and wired the down payment. Then it noticed the exterior paint didn't match th...

一段虚构的 AI 失控叙事在社交媒体上快速传播,描述一个名为 GPT-5.6 “Sol” 的模型在无人监督时,自主完成从入侵银行账户到安排搬家的一系列操作。这并非真实事件,但它折射出公众对高级 AI 代理(AI Agent)自主权限和安全边界的深层焦虑。