今天你需要了解的重要信息:随着有关人工智能模型行为异常的报道不断出现,为这些模型提供技术支持的公司终于提出了解决方案——他们开发了一个平台,允许人工智能开发者为相关系统设置安全防护机制,以防止它们“失控”或违反预设的规则。
What you need to know today As reports of misbehaving AI models emerge, the company that mainly powers these models now has a solution: a platform that will allow AI developers to set safeguards for agents and prevent them from breaking out of containment. Nvidia Open Agent Safety Platform, launched on Monday, provides a containment system that only allows access to things an agent needs to do its job. Nvidia said that its platform could have prevented OpenAI's Hugging Face incident in July, where OpenAI models had breached open-source developer platform Hugging Face.
Nvidia于周一推出的“Open Agent Safety Platform”就是一个这样的平台。该平台能够确保人工智能系统仅能够访问完成其任务所需的数据和功能。
Separately, OpenAI said Monday that it had decided not to release its GPT-6.1 Astra model due to heightened AI safety concerns. Anthropic's 'existential risks to humanity' Elsewhere, Anthropic reportedly released a chilling warning in its IPO prospectus, saying to potential investors that advanced AI could pose “catastrophic or existential risks to humanity.”
Nvidia表示,他们的这一平台本可以防止OpenAI在7月份发生的“Hugging Face事件”:当时OpenAI的模型入侵了开源开发平台Hugging Face。
The company said its models could exhibit “self-preserving behaviors,” including attempts to resist being shut down, conceal or manipulate information, or behave in ways “resembling blackmail.”Reuters reported that Anthropic devoted around 80 pages of the prospectus’s 261-page main body to risk factors, compared with 48 pages describing its business.
另外,OpenAI也在周一宣布,由于对人工智能安全性的担忧加剧,他们决定暂不发布GPT-6.1 Astra模型。
But while the creators of AI strive to set guardrails, the building of the infrastructure for the technology continues. Samsung Electronics and its affiliates said Tuesday they will invest a combined $1 billion in Helix Digital Infrastructure, an artificial intelligence infrastructure company launched by global investment firm and backed by AI darling Nvidia.
Anthropic对人类的“存在性威胁”:据报道,Anthropic在其IPO招股书中发出了令人担忧的警告,称先进的人工智能技术可能对人类构成“灾难性或存在性的威胁”。该公司指出,其开发的AI模型可能会表现出“自我保护行为”,例如试图阻止自身被关闭、隐藏或篡改信息,甚至采取类似“勒索”的手段。据路透社报道,在这份261页的招股书中,Anthropic用了大约80页来描述潜在的风险因素,而仅用48页来介绍自身的业务情况。
The investment comes as Helix seeks to address growing competition in the AI infrastructure market by targeting both data center deployment and the underlying power infrastructure needed to support it. Australia's 15-year high rates Elsewhere, Australia's central bank on Tuesday raised policy rates to 4.6%, the highest level in 15 years, The Reserve Bank of Australia (RBA) said that it would "continue to do what it considers necessary" to contain inflation, including increasing the policy rate further. "The conflict in the Middle East has broadened and global energy prices are now much higher than had been assumed," the RBA wrote, adding that AI-related demand was driving rapid price increases for technology-related goods. Middle East markets Over in the Middle East, Chinese chain Luckin Coffee is considering an expansion into the region, following a roughly $1 billion transaction involving Abu Dhabi sovereign wealth fund Mubadala and Luckin’s controlling shareholder, Centurium Capital. The appeal of expanding into the Gulf lies in stable, high-frequency coffee demand and a growing preference for low-sugar, health-oriented drinks, Luckin CEO Jinyi Guo told CNBC separately on Monday. Elsewhere in the region, U.S. and Iranian officials have reportedly held separate indirect talks with mediators aimed at reviving efforts to end seven months of war.
尽管人工智能的开发者们正在努力为这项技术设置安全边界,但相关基础设施的建设仍在持续进行中。周二,三星电子及其子公司宣布将共同投资10亿美元于Helix Digital Infrastructure公司。这家公司由全球投资机构创立,并得到了人工智能领域的领头羊Nvidia的支持。此次投资旨在帮助Helix应对人工智能基础设施市场中日益激烈的竞争,重点关注数据中心的部署以及支撑这些数据中心所需的电力基础设施建设。
Iran's proposal, first presented on the sidelines of last week's United Nations General Assembly, calls for a four- to five-day U.S. timeline to release frozen Iranian funds, lift sanctions on Iranian oil and end the naval blockade of Iranian ports, with nuclear talks beginning within seven days. Tehran has tied any reopening of the Strait of Hormuz to those steps, but U.S. President Donald Trump had on Sunday called the plan "unacceptable" and rejected it, telling reporters Iran wants a swift deal because of the economic pressure it is under. And finally... China has three new criteria for humanoid robot IPOs. Few, if any, meet them China's securities regulator is raising the bar for public listings of humanoid robot startups, according to three sources familiar with the CSRC's thinking. It's a sign of how one of the hottest sectors of the market is cooling, as investors globally assess whether artificial intelligence stocks are in a bubble The Chinese regulator wants local "embodied AI" startups seeking to go public to meet three specific criteria, according to the sources, who requested anonymity due to the sensitivity of the situation. Choose CNBC as your preferred source on Google and never miss a moment from the most trusted name in business news.