金正恩强化国防 · Rio 3.5开源造假
金正恩强化国防 · Rio 3.5开源造假
一、 权威必看
EN: Kim Jong Un, General Secretary of the Workers’ Party of Korea, presided over the Second Plenary Session of the 9th Central Committee during June 20-22. He explicitly stated that recent actions by the United States and South Korea have severely deteriorated the situation on the Korean Peninsula. Specifically, he cited the deployment of additional troops within the region, the advancement of South Korea’s nuclear submarine program, and the reconvening of the Nuclear Consultative Group as primary drivers of this instability.
中： 朝鲜劳动党总书记金正恩于6月20日至22日主持了第九届中央委员会第二次扩大全会。他在会上明确指出，今年以来美国和韩国的一系列军事与外交动作严重恶化了朝鲜半岛的局势。金正恩特别点名批评了美韩在域内的增兵活动、韩国核潜艇计划的推进，以及“核磋商小组会议”的再次召开，认为这些行为直接导致了地区安全环境的紧张。他强调，面对这种无法回避的地缘政治现实，朝鲜必须从维护国家主权和安全的根本利益出发，主动且积极有为地开展对外工作，通过展示强大实力来保障国家利益。
二、 深度与多元
EN: The viral discussion regarding a woman being driven away by security guards while seeking shelter from heavy rain at the entrance of a Culture and Tourism Bureau has sparked intense public debate on Weibo. This incident, categorized under social hot topics, reflects broader societal anxieties regarding public service accessibility and the perceived rigidity of bureaucratic enforcement. Netizens are using this case to question the balance between administrative order and humanitarian emergency response in urban management.
中： 近期，微博热搜上关于“女子暴雨躲文旅局门口遭保安驱赶”的话题引发了网民的广泛讨论与关注。这一事件不仅是一个孤立的民生热点，更折射出公众对公共服务机构在极端天气下应急能力的深层审视。网民们在讨论中聚焦于行政秩序与人性关怀之间的张力，质疑在暴雨等紧急避险情境下，基层执行人员是否具备足够的灵活性与同理心。这种舆论发酵表明，公众对于政府职能部门的服务态度及突发事件中的柔性治理能力有着更高的期待，任何机械式的执法行为都可能被放大解读为对民众基本生存需求的漠视。
三、 科技与财经
EN: The open-source AI community was shaken by the exposure of Rio 3.5, a model claimed to be from a Rio de Janeiro municipal IT company. A Chinese team, Nex-AGI, analyzed the weights and revealed that 60% came from their own Nex-N2-Pro model and 40% from Alibaba’s Qwen, with no new training involved. The direct evidence was found by deleting the hardcoded “You are Rio” prompt; the model then identified itself as “Nex” in nearly 80% of cases, proving it was a stitched product rather than a genuine innovation.
中： 开源AI圈近日因Rio 3.5模型的身份造假事件而震动。该模型由巴西里约热内卢市政府下属的一家IT公司发布，号称跻身全球开源第一梯队。然而，国内团队Nex-AGI在对比权重后发现，Rio 3.5并非全新训练，其60%的权重直接来自Nex-AGI自家的开源模型Nex-N2-Pro，另外40%则来自阿里Qwen，两者仅按固定比例拼接而成。最直接的证据在于，当删除出厂时写死的“你是Rio”提示词后，该模型在80%的情况下自称“Nex”，且认领“Rio”名称的次数为零。里约热内卢方面随后对此致歉，承认了传播过程中的失误。这一事件不仅揭示了开源社区中“缝合怪”模型的泛滥，也警示了开源信任机制的脆弱性。
四、 国际视野
EN: Anthropic recently released a lengthy white paper titled “When AI builds itself,” warning about the risks of recursive self-improvement. However, the tech community’s reception has been largely critical, with many viewing it not as a genuine safety document but as a public relations maneuver designed to establish moral high ground. This reaction highlights the growing skepticism towards major AI labs regarding their transparency and the true motivations behind their safety narratives.
中： 国际AI巨头Anthropic近期发布了一篇题为《When AI builds itself》的万字长文，旨在预警“递归自我改进”带来的潜在风险。然而，这篇文章在技术社区引发的反响并非预期的共鸣，而是大量的批评与质疑。许多开发者认为，这并非一份严肃的安全白皮书，而是一场精心包装的公关行动，目的是通过设定安全议题来确立行业道德高地。这种舆论反应反映了国际科技界对头部AI实验室透明度及动机的深层不信任，表明单纯的风险警告已难以取信于专业群体，真正的安全治理需要更实质性的技术开源与第三方验证。
五、 青年与生活
EN: Lingchuan Technology, an AI chip company backed by Kuaishou, has completed a hundreds-of-millions of yuan A+ round of financing. Led by Qifu Capital, the round includes investments from Baidu Ventures and others. The company’s SL200 chip, originally developed internally for Kuaishou to serve 700 million users, is now expanding to the overseas market with a sales volume nearing 100,000 units, highlighting the commercialization of specialized AI hardware.
中： 快手系AI芯片公司凌川科技近日完成数亿元A+轮融资，由啟赋资本领投，百度风投、水木投资等知名机构跟投。凌川科技成立于2024年3月，前身为快手集团异构计算与芯片事业部，专注于智能视频与生成式视频的底层算力。其自研的SL200芯片在快手内部已稳定服务7亿用户，目前销量接近十万颗，视频压缩性能超越英伟达同类产品。本轮融资将主要用于下一代芯片研发及海外市场拓展，标志着垂直领域AI硬件从内部配套走向独立商业化运营的关键一步，也为青年创业者在硬科技赛道提供了新的资本风向标。
【21ZHAO 综合判断】
EN: The convergence of geopolitical tension, open-source fraud, and hardware commercialization reveals a critical shift in the tech landscape. As traditional power structures (like North Korea’s defense buildup) harden, the digital realm faces similar integrity crises, evidenced by the Rio 3.5 weight-stitching scandal. Simultaneously, the success of specialized chips like Lingchuan’s SL200 proves that vertical AI applications are moving from hype to tangible revenue.
- For Developers: Verify model provenance rigorously; the Rio 3.5 incident shows that even municipal-backed projects can lack technical integrity. Use tools like weight comparison and prompt-injection tests to detect stitched models.
- For Investors/Entrepreneurs: Focus on vertical AI hardware with proven deployment scenarios, such as video compression for short-video platforms, rather than generic foundation models which face high trust barriers.
中： 地缘政治紧张、开源造假与硬件商业化的交织，揭示了科技领域的深刻转变。当传统权力结构（如朝鲜的国防建设）硬化时，数字领域也面临着类似的诚信危机，Rio 3.5权重拼接事件便是明证。同时，凌川SL200芯片的成功证明，垂直领域的AI应用正从概念炒作转向可量化的营收增长。对于开发者而言，必须严格验证模型来源，警惕“市政背书”背后的技术水分；对于创业者，应关注具备真实落地场景的垂直硬件，而非陷入通用大模型的信任泥潭。
- 针对开发者： 建立严格的模型溯源机制，Rio 3.5事件表明即使是官方背景的项目也可能缺乏技术诚信。建议通过权重对比和提示词注入测试来识别拼接模型，避免在不可靠的基础设施上构建应用。
- 针对创业者/投资者： 关注具备真实落地场景的垂直AI硬件（如视频压缩芯片），而非盲目追逐通用大模型。凌川科技从快手内部走向独立融资的路径表明，解决具体痛点且已有大规模用户验证的硬件产品更具商业确定性。