人工智能公司Anthropic于周一向美国证券交易委员会(SEC)提交了修订后的S-1A注册文件,披露其内部预测显示该公司可能很快符合其负责任扩展政策中关于“由AI驱动的显著加速进步”的定义,即“递归自我改进”(RSI) 1。尽管文件警告AI系统可能加速自身能力发展,但Anthropic强调目前尚无模型跨越该阈值 1。此外,该公司表示正在考虑更新政策,以确保人类保持对前沿模型发展的控制 1。
在技术进展方面,截至2026年8月的数据显示,Claude已在26%的AI研究任务中实现端到端完成工作,但在所有测量领域中完全自主工作的比例为零 1。与此同时,超过80%的代码合并行现由Claude编写,工程师每日合并的代码行数约为2024年的八倍 1。根据METR数据,Anthropic的Mythos Preview在2026年4月的软件任务时间跨度指标约为17小时,且自2023年以来大约每四个月翻一番 1。文中引用了2026年3月一项针对130名研究人员的内部调查,估计Mythos Preview带来4倍产出提升 1。
此次披露正值Anthropic推进上市计划之际,该公司原计划于10月在纳斯达克上市,股票代码为ANT,估值约2万亿美元 1。这一动向引发了关于前沿实验室是否借机推动监管保护的争议 1。Anthropic CEO Dario Amodei发表文章呼吁“放慢前沿”,而OpenAI、Google DeepMind和xAI也表达了类似观点 1。然而,David Sacks批评这种协调一致的呼吁是“监管捕获”,旨在保护大型玩家免受竞争 1。
Anthropic disclosed in a revised S-1A registration filing submitted to the U.S. Securities and Exchange Commission (SEC) on Monday that internal estimates suggest the company may soon meet its definition of "recursive self-improvement" (RSI), characterized by "significant acceleration of progress driven by AI," under its Responsible Scaling Policy 1. While the document warns that AI systems could accelerate their own capability development, it clarifies that no current models have crossed this threshold 1. In response, Anthropic stated it is considering policy updates to ensure humans retain control over frontier model development 1.
The disclosure coincides with Anthropic’s plan to list on the Nasdaq in October under the ticker symbol ANT, with an estimated valuation of approximately $2 trillion 1. Data in the submission highlights that as of August 2026, Claude completed 26% of AI research tasks end-to-end, yet the proportion of work performed entirely autonomously across all measured domains remained at zero 1. Additionally, more than 80% of merged code lines are now written by Claude, resulting in engineers merging roughly eight times as many lines per day compared to 2024 levels 1. METR data indicates that Anthropic's Mythos Preview achieved a time span metric for software tasks of approximately 17 hours in April 2026, doubling roughly every four months since 2023 1. An internal survey of 130 researchers from March 2026 estimated that Mythos Preview delivered a fourfold increase in output 1.
The filing has sparked debate regarding regulatory motivations within the AI sector. Anthropic CEO Dario Amodei recently published an article calling to "slow down the frontier," a sentiment echoed by OpenAI, Google DeepMind, and xAI 1. However, David Sacks criticized these coordinated appeals as "regulatory capture," arguing they aim to protect established large players from competition 1.
评论
还没有评论,欢迎留下第一条。