就在宣布离开月之暗面（Moonshot AI）仅仅几天之后，中国青少年人工智能天才陈光宇似乎已找到下一站。
Just days after announcing his departure from Moonshot AI, teen Chinese artificial intelligence prodigy Nathan Chen Guangyu seems to have found his next destination.
周二，广受关注的匿名社交媒体账号 Arfur Grok 发帖称，这位前月之暗面研究员将加入 Discovery Loop。该 AI 初创公司由四位前谷歌高级研究人员创立，其中包括前首席科学家杰夫·迪恩（Jeff Dean）。
On Tuesday, Arfur Grok, a widely-watched anonymous social media account that tracks AI talent movements in the US, posted that the former Moonshot AI researcher was joining Discovery Loop, an AI start-up founded by four former Google senior researchers including former chief scientist Jeff Dean.
该初创公司成立仅两个月，凭借其明星创始团队以及雄心勃勃的使命——利用 AI 实现科学研究自动化——迅速吸引了业界关注。
Launched just two months ago, the start-up has quickly attracted industry attention due to its star-studded founding team as well as its ambitious mission: to automate scientific research with AI.
Discovery Loop 已招募多位来自美国大型科技巨头的明星研究人员，其中包括其他前谷歌工程师。但年仅 17 岁的陈光宇很可能是其最年轻的成员之一。
Discovery Loop has already recruited several star researchers from US big tech giants, including other former Google engineers. But at just 17 years old, Chen is likely to be one of its youngest members.
陈光宇周二未立即回复置评请求。Discovery Loop 表示不评论人事事务。
Chen did not immediately respond to a request for comment on Tuesday. Discovery Loop said it does not comment on personnel matters.
作为中国最具潜力的年轻 AI 人才之一，陈光宇上周宣布离开中国前往美国，引发了关于他下一站去向的广泛猜测。
One of China’s most promising young AI talents, Chen’s announcement last week that he was leaving China for the US sparked widespread speculation about his next destination.
尽管在月之暗面任职不足一年，陈光宇却帮助开创了技术突破，为其旗舰 Kimi 模型带来了成本效益提升，使其具备了与美国巨头 OpenAI 的 GPT 和 Anthropic 的 Claude 竞争的实力。
Despite a stint at Moonshot that lasted less than a year, Chen helped pioneer technical breakthroughs that delivered cost-efficiency gains for its flagship Kimi models, making them competitive with US heavyweights such as OpenAI’s GPT and Anthropic’s Claude.
月之暗面已崛起为中国最强大的 AI 模型实验室之一，其 K3 模型凭借强大的能力和开放权重特性获得了全球关注。
Moonshot has emerged as one of China’s strongest AI model labs, with its K3 model gaining global traction due to its powerful capabilities and open-weight nature.
在陈光宇宣布离职后，月之暗面高级研究员杜宇伦在社交媒体发文称，他与陈光宇曾就其离职决定进行过“多次长谈”，但最终认为“在他这个年纪，尽可能去探索并追随他新的好奇心，是最好的选择”。
In a social media post after Chen’s announcement, Moonshot senior researcher Du Yulun wrote that he and Chen had “several long conversations about his decision” to leave, but it was ultimately “best for [Chen] to explore and follow his new curiosity as much as possible at his age”.
据《商业内幕》上月报道，尽管 Discovery Loop 尚未推出任何产品，但已在另一轮融资中寻求约 500 亿美元的估值。
Though Discovery Loop has yet to release any products, it is already seeking a valuation of around US$50 billion in another funding round, Business Insider reported last month.
其首轮融资由 Radical Ventures 和 Khosla Ventures 领投，Google 母公司 Alphabet 也参与了投资。
Its initial funding round was led by Radical Ventures and Khosla Ventures, with Google parent Alphabet also an investor.
在 8 月初加州大学伯克利分校举办的 Agentic AI Summit 上，Dean 离开 Google 后首次公开露面之一，他表示其初创公司的工作将类似于实现递归自我改进（RSI）——这是美国和中国领先 AI 公司的既定目标。
In one of his first public appearances after leaving Google, Dean said at the Agentic AI Summit at the University of California, Berkeley, in early August that his start-up’s work would be similar to efforts to achieve recursive self-improvement (RSI) – the stated goal of leading AI companies in both the US and China.
正如 RSI 旨在通过 AI 本身实现 AI 开发的自动化，Dean 解释说，AI 也许有一天能自主并行运行数千个科学实验，并利用结果确定下一步最佳实验，从而将科学过程从数年缩短至数小时甚至数分钟。
Just as RSI looks to automate AI development through AI itself, Dean explained that AI could one day autonomously run thousands of scientific experiments in parallel and use the results to identify the best experiments to do next, thereby shortening the scientific process from years to hours or even minutes.
他说：“通过构建擅长理解许多、许多不同科学和工程领域的模型，你可以在一个模型中获得跨许多不同领域的博士级专业知识。”
“By building models that are really good at understanding many, many different domains of science and engineering, you can get PhD-level expertise in a model across many different domains,” he said.