Google DeepMind于周三推出DeepMind研究院，意在推进关于通用人工智能（AGI）的全球对话1。该研究院由Shane Legg、James Manyika和Demis Hassabis担任主任，其中Legg担任主编1。研究院发布了四篇论文集，涉及AGI经济政策、模型透明度保护、人类繁荣原则和前沿AI模型评估框架等议题1。
Rohin Shah和Anca Dragan的论文主张AI模型的透明度下降并非必然，呼吁开发者和监管机构直面安全权衡1。Hassabis提议建立由美国领导的前沿AI标准评估机构，开发者在发布前30天内自愿提交模型接受评估1。该评估框架在被证明有效后，可能成为美国部署前沿模型的强制要求，并可能包括前沿AI开发者的协调放缓1。这一倡议反映了业界安全辩论从笼统关切转向具体提议（披露、外部审查、协调放缓）的转变1。
Google DeepMind unveiled the DeepMind Institute on Wednesday, a new initiative designed to broaden the global conversation around artificial general intelligence (AGI) governance and safety.1 Led by directors Shane Legg, James Manyika, and Demis Hassabis—with Legg serving as editor-in-chief—the institute released a collection of four research papers addressing critical questions in AGI policy, including economic implications, model transparency, principles for human flourishing, and frameworks for evaluating frontier AI systems.1
A key proposal from the institute calls for the establishment of a U.S.-led frontier AI standards assessment body to independently evaluate the safety of cutting-edge models before deployment.1 Under this framework, developers would voluntarily submit models for assessment within 30 days of publication.1 Researchers Rohin Shah and Anca Dragan argue that transparency in AI models need not inherently decrease, and that developers and regulators should directly address the tradeoffs involved in safety measures rather than accepting reduced transparency as inevitable.1
If proven effective, the proposed evaluation framework could evolve into a mandatory requirement for deploying frontier models in the United States, and might eventually include coordinated slowdowns among frontier AI developers.1 The institute's launch reflects a broader industry shift from general safety concerns toward concrete policy proposals, including disclosure requirements, external audits, and collaborative development practices.1
评论
还没有评论，欢迎留下第一条。