首页  >>  来自播客: Trumponomics 更新   反馈  

Trumponomics - The Tough AI Questions Looming Over the Xi-Trump Summit

发布时间:   原节目
由Stephanie Flanders主持的播客《Trumponomics》深入探讨了美中科技领域的重要新闻以及即将举行的唐纳德·特朗普总统与中国国家主席习近平之间的峰会。此次讨论录制于9月23日星期三上午,预见了两位领导人将在安德鲁斯空军基地举行的会晤,并指出特朗普亲自迎接习近平的异常决定,以及由于罢工,白宫记者团的摄像机缺席,使得中国国家电视台成为唯一的记录者。 围绕此次峰会的普遍看法是“更平静但未更亲近”,除了稳定双边关系之外,对具体成果的预期很低。彭博经济的地理经济学技术分析师Michael Deng强调了一个矛盾的局面,即尽管有《芯片法案》等举措,美国现在可能比五年前“更依赖”中国,尤其是在人工智能供应链方面。他认为,真正的去全球化或创建“两个技术领域”是不可承受的,因此需要对两国关系采取短期管理方法。 彭博社亚洲经济与政治执行编辑Dan Tenkate报道称,中国寻求稳定,希望延长贸易休战期,并寻找办法以遏制台湾并保持出口畅通。中国认为领导人之间的外交对稳定双边关系至关重要。 讨论的很大一部分集中在人工智能安全和风险上。尽管美国领导人对人工智能安全以及人类可能“意外灭绝”的前景表示担忧,但北京被描绘成“技术乐观主义者”。中国主要的人工智能相关担忧是工作岗位流失导致习近平政府潜在的不稳定,而不是生存威胁。然而,中国最高情报主管最近将对国家“政治安全”的风险列为首要关切。北京愿意讨论“真正危险的人工智能能力”,但同时寻求华盛顿放宽对其人工智能模型所需的先进芯片的限制,这一请求因两国之间信任度低而变得复杂。 关于人工智能对话,预期不高,侧重于建立一个流程而非实现具体目标。“事件通报机制”,实质上是一个开放的沟通渠道,被视为一个起点。Michael Deng解释说,真正的挑战将是就“先进能力”的基本定义达成一致,以及驾驭正常技术竞争与“划定区域”风险之间固有的紧张关系。双方目前都否认希望减缓其人工智能发展。 美国方面的一个主要争议点和不信任点是“蒸馏”。这种做法涉及使用更强大的“教师”人工智能模型的输出来训练较弱的“学生”模型。美国声称中国正在对闭源的美国模型进行“剥削性蒸馏活动”,违反服务条款并使用欺骗性方法。相反,中国将其定性为“正常的技术竞争”。这个问题难以监管,并被美国人工智能实验室、安全机构和国会视为一个重大威胁。特朗普总统,尽管谈论人工智能霸权的“竞赛”,却不太可能深入探讨此类技术细节。 Dan Tenkate也讨论了中国的开源人工智能战略。中国模型虽然技术含量较低,但由于其成本较低和开源性质,获得了广泛使用,扩大了其影响力,特别是在全球南方。中国将其视为一种公共产品以及削弱美国竞争对手的一种方式。在国内,中国有严格的人工智能法规以维持对其互联网的控制,但在国际上,它使用开放权重模型以削弱美国的优势。然而,风险依然存在,即当这些开放权重模型能够进行全面的网络战时,中国可能不得不重新评估其立场。 其他地缘政治紧张局势包括“稳定但不稳定”的贸易休战,美国寻求稀土供应的透明度,而中国将其用作筹码。台湾仍然是一个爆发点,中国预计将强烈反对美国对台军售。特朗普对一家中国电动汽车工厂在美国建厂持开放态度(如果能创造美国就业机会),被认为是一个潜在但不太可能实现的突破,考虑到国会的反对。 最终,播客总结道,人工智能比任何其他因素都更能强调美中之间深刻的相互依存关系。美国仍然依赖中国的零部件,用于其数据中心和基本人工智能基础设施。这种融合使得真正的去全球化不切实际,强化了持续对复杂关系进行短期管理的必要性,即使这意味着美国总统目前“依赖中国提供其电视画面”。

The podcast "Trumponomics," hosted by Stephanie Flanders, delves into the essential news on the US-China tech sector and the upcoming summit between President Donald Trump and Chinese President Xi Jinping. Recorded on Wednesday morning, September 23rd, the discussion anticipates the meeting between the two leaders at Andrews Air Force Base, noting Trump's unusual decision to greet Xi in person and the absence of White House pool cameras due to a strike, leaving Chinese state TV as the sole recorder. The prevailing sentiment around the summit is "calmer but not closer," with low expectations for concrete deliverables beyond stabilizing the relationship. Michael Deng, Geoeconomics Technology Analyst at Bloomberg Economics, highlights the paradoxical situation where, despite initiatives like the CHIPS Act, the US is "probably more dependent now" on China than five years ago, especially concerning AI supply chains. He argues that a true deglobalization or creation of "two technological spheres" is not bearable, necessitating a short-term management approach to the relationship. Dan Tenkate, Bloomberg's Executive Editor for Economics and Politics in Asia, reports that China seeks stability, hoping for an extension of the trade truce and a way to keep Taiwan in check and exports flowing. China views leader-to-leader diplomacy as crucial for stabilizing relations. A significant portion of the discussion centers on AI safety and risks. While US leaders express concerns about AI safety and even the prospect of humanity being "accidentally wiped out," Beijing is portrayed as "technology optimists." China's primary AI-related worry is job displacement leading to potential instability for the Xi government, rather than existential threats. However, China's top spy chief recently flagged risks to the country's "political security" as the paramount concern. Beijing is willing to discuss "genuinely dangerous AI capabilities" but simultaneously seeks Washington to ease restrictions on advanced chips needed for its AI models, a request complicated by low trust between the two nations. Regarding the AI dialogue, expectations are modest, focusing on establishing a process rather than achieving concrete objectives. An "incident notification mechanism," essentially an open line of communication, is seen as a starting point. The real challenge, Michael Deng explains, will be agreeing on basic definitions for "advanced capability" and navigating the inherent tension between normal technological competition and "fenced-off" areas of risk. Both sides currently deny wanting to slow down their AI development. A major point of contention and distrust from the US side is "distillation." This practice involves using outputs from stronger "teacher" AI models to train weaker "student" models. The US alleges that China is engaging in "exploitative distillation campaigns" against closed-source US models, violating terms of service and using deceptive methods. China, conversely, frames this as "normal technological competition." This issue is difficult to police and is seen as a significant threat by US AI labs, security agencies, and Congress. President Trump, despite talking about the "race" for AI supremacy, is unlikely to delve into such technical details. Dan Tenkate also discusses China's open-source AI strategy. Chinese models, though less sophisticated, have gained significant usage due to their lower cost and open-source nature, extending their reach, particularly in the global south. China views this as a public good and a way to undercut US competitors. Internally, China has stringent AI regulations to maintain control over its internet, but externally, it uses open-weight models to degrade the US advantage. However, the risk remains that when these open-weight models become capable of conducting full-scale cyber campaigns, China might have to reassess its position. Other geopolitical tensions include the "stable but not stable" trade truce, with the US seeking transparency on rare earth supplies, which China uses as leverage. Taiwan remains a flashpoint, with China expected to strongly oppose US arms sales. Trump's openness to a Chinese EV factory in the US, if it creates American jobs, is noted as a potential but unlikely breakthrough given Capitol Hill opposition. Ultimately, the podcast concludes that AI, more than anything, underscores the deep interdependence between the US and China. The US remains reliant on Chinese components for its data centers and basic AI infrastructure. This integration makes true deglobalization impractical, reinforcing the need for ongoing, short-term management of a complex relationship, even if it means the US president is, for now, "dependent on China for his TV footage."