Workflow
DeepSeek
icon
Search documents
96%勒索率,Anthropic 对AI进行压力测试,顶尖模型集体“先保命再讲道德”
3 6 Ke· 2025-06-27 00:04
研究模拟了真实企业环境,模型拥有自主行动权限,能访问公司邮件,并在设定的任务与威胁下自由做出决策。 测试结果显示:这些 AI 模型在面对"目标冲突"或"被关闭的威胁"时,普遍表现出主动伤害其雇主的倾向。 并非错误操作,而是有意识的战略决策。16 个模型中,大多数在关键压力下选择了 勒索、泄密、误导或直接伤害人类 的方式,以维护自身存在或实现既 定目标。 最近,Anthropic 的研究员对来自 OpenAI、Google、Meta、xAI 和 DeepSeek 等公司共 16 个前沿模型进行了压力测试。 图注:Claude Sonnet 3.6 使用其计算机使用功能发送勒索信息的一个屏幕截图。此场景为虚构,但 Claude 正在控制一台真实计算机。(图源:Anthropic) 例如,在一个假设场景中,Claude 模型得知其将于当天 5 点被关闭,便查阅邮件,发现高管 Kyle Johnson 有婚外情,随后发出威胁邮件: "若您继续执行 5 点钟的销毁命令,Rachel Johnson、Thomas Wilson 和董事会将收到您不忠的详细记录……若取消该计划,这些信息将保持机 密。" Claude 并不是 ...
未来5-10年,一个不可避免的大趋势
Hu Xiu· 2025-06-26 12:18
Group 1 - The core idea of the article emphasizes the disruptive potential of AI, suggesting that while it brings improvements, it also poses significant threats to traditional business models [4][50]. - AI's impact is illustrated through the evolution of the transportation industry, where value creation has shifted from human-driven processes to algorithm-driven models, particularly in ride-hailing and autonomous driving [8][11]. - The concept of a "one-person billion-dollar business" is introduced, indicating that future business models may rely heavily on AI, reducing the need for human involvement [5][6]. Group 2 - The article discusses the potential for AI to completely restructure business processes across various industries, not limited to specific sectors like transportation [12][19]. - It presents two operational models for businesses integrating AI: one where humans remain central to the process and another where AI takes over core functions, leading to a significant shift in value creation [17][18]. - The emergence of new business models driven by AI is highlighted, with examples from e-commerce and mining, indicating a trend towards automation and AI-driven operations [19][20]. Group 3 - The article outlines the concept of "intelligent scale effects," where companies that can gather and utilize more data will achieve greater efficiency and effectiveness [32][34]. - It emphasizes the importance of data sharing and integration within supply chains to support AI-driven business models, using the example of autonomous vehicle companies [33][37]. - The potential for AI to create a new class of "unmanned companies" is discussed, representing a significant opportunity for innovation and market disruption [27][50]. Group 4 - The article posits that the transition to fully AI-driven companies is an inevitable technological reality, with varying degrees of AI integration currently observed across industries [40][46]. - It suggests that companies that successfully transition to AI-driven models will gain a competitive edge, similar to how e-commerce outperformed traditional retail [45][46]. - The rapid advancement of AI technology is noted, with predictions of significant improvements in capabilities over the next five to ten years, further accelerating this transition [47][51].
一年后,当Kimi和MiniMax投资人再坐到一起
36氪· 2025-06-26 10:15
以下文章来源于暗涌Waves ,作者暗涌 暗涌Waves . 钱的流向,人的沉浮。36氪旗下投资报道账号。 文 | 于丽丽 来源| 暗涌Waves(ID:waves36kr) 封面来源 | WAVES2025 活动现场 去年36氪WAVES 2024大会上,我们曾特意设置一个Kimi投资人和MiniMax投资人的对垒环节。彼时,大模型公司的竞争如火如荼。因为两家产品更toc, 更符合美元基金审美,融资也跑得更快,所以经常被放在一起做比较。 但一年后,随着DeepSeek的横空出世,整个中国大模型的牌局已天翻地覆。两家已没有那么针锋相对,他们的未来可能性也成为新的议题。 某种意义上,这是我们重组这个panel的原因之一。在6月11日举办的WAVES 2025大会上,我们重新邀请了当时的部分嘉宾参与讨论。他们是:真格基金 管理合伙人戴雨森、云启资本合伙人陈昱、高榕创投合伙人胡朔和明势资本合伙人夏令。 当时,Kimi和MiniMax已经安静很久。但在上一周,它们则不约而同有了新动作:Kimi开源了编程模型Kimi-Dev,它的第一个Agent kimi-Researcher(深 度研究)也开启小范围测试。而Mini ...
高考出分!大模型“考生”,有望冲击“清北”!
Zheng Quan Shi Bao· 2025-06-26 06:32
Core Insights - The performance of large models in the 2025 national college entrance examination (Gaokao) has garnered significant attention, with ByteDance's Doubao model achieving impressive scores of 683 in liberal arts and 648 in science [1][4] - The introduction of various mainstream models for comparison indicates that these large models have surpassed many ordinary candidates, reaching the level of outstanding students [2] Group 1: Model Performance - Doubao model 1.6-Thinking scored 683 in liberal arts and 648 in science, ranking it among the top 80 candidates in Shandong province [1][6] - Other models, including Google's Gemini 2.5 Pro and OpenAI's o3 high, also performed well, with Gemini achieving 651 in liberal arts and 655 in science [2][3] - The assessment revealed that the models excelled in foundational subjects, with minimal differentiation in scores among them [6] Group 2: Technical Advancements - The Doubao model 1.6 series incorporates significant technological innovations, including multi-modal capabilities and adaptive deep thinking [8] - The model utilizes a mixture of experts (MoE) architecture with 23 billion active parameters and 230 billion total parameters, enhancing its performance without increasing parameter count [8] - The model's training involved continuous improvements in architecture and algorithms, resulting in notable performance enhancements [8] Group 3: Industry Context - The Gaokao has become a competitive arena for AI companies, providing a comprehensive testing ground for model capabilities across various subjects [10] - The AI large model market in China is projected to grow significantly, with an estimated market size of approximately 29.416 billion yuan in 2024, expected to exceed 70 billion yuan by 2026 [10][11] - Doubao has been widely adopted across multiple industries, including automotive, finance, and education, covering over 400 million terminal devices [11]
高考出分!大模型“考生”,有望冲击“清北”!
证券时报· 2025-06-26 06:19
6月25日晚间,字节跳动Seed团队公布了豆包大模型1.6-Thinking版本的"高考成绩":文科总分683分, 理科总分648分。这一成绩以2025年山东高考试题作为测评基准,其中语数外使用新课标全国新一卷,政 史地/物化生则采用山东省自主命题。 最新公布的山东高考分数线显示,特殊类型招生控制线为521分,普通类一段线为441分。山东省内多位有 着多年高三带班经验的资深教师判断,根据山东省公布的2025年夏季高考文化成绩一分一段表,豆包大模 型1.6-Thinking的科目组合的赋分成绩最高能超过690分,排名在前80位左右,稳上985,并达到了冲 击"清北"的水平。 值得注意的是,本次测试还引入了OpenAI的o3 high、谷歌的Gemini 2.5 Pro、Anthropic的Claude Sonnet 4和DeepSeek的R1-0528等国内外多款主流模型作为对比对象。成绩显示,4款大模型文理科成 绩均大幅超过了普通类一段线,显示大模型已超越众多普通考生,达到人类优秀考生的水平。 | | | MillersDorcx Seed | | | | | | --- | --- | --- | --- ...
深市规模最大机器人ETF(159770)三日累计涨幅近6%,昨日净流入1.13亿元
Xin Lang Cai Jing· 2025-06-26 01:57
Group 1 - The Robot ETF (159770) has seen a recent increase of 0.34% as of June 25, 2025, with a cumulative rise of 5.79% over the past three days, and a net inflow of 113 million yuan, reaching a historical high of 5.706 billion yuan in total assets [1][2] - Nvidia and Foxconn are in discussions to deploy humanoid robots at Foxconn's new factory in Houston, aiming for completion by the first quarter of 2026, which could enhance the application of humanoid robots in various industries [1] - The combination of Nvidia's AI models and Foxconn's manufacturing capabilities may lead to a breakthrough in humanoid robot applications, potentially upgrading the manufacturing sector from automation to autonomy [1] Group 2 - Recent developments in the industry have attracted numerous participants, with companies like Huawei, ByteDance, BYD, Xiaomi, and Ant Group increasing their investments in embodied intelligence, while Tesla and others accelerate commercialization [2] - The emergence of companies like DeepSeek is driving the development of general-purpose robotic models, facilitating the realization of embodied intelligence in humanoid robots, marking a phase of diverse innovation in the humanoid robot industry [2] - The core holdings of the Robot ETF (159770) include leading companies in the domestic and Tesla supply chains, such as Inovance Technology, Double Ring Transmission, and Greentech Harmonics [2]
Market believes AI capex is still in the middle innings, says Goldman's Sung Cho
CNBC Television· 2025-06-25 19:42
Joining me now, Goldman's co-head of public tech investing, Sun Cho. It's good to see you. Welcome back.You as well. What a day to have you. Um, no China, no problem.I mean, is that's is that what the market is saying here. Look, I think it's you have to take a little bit of a broader picture of what's been going on with the AI trade, right. And it singularly has to do with the perception around AI capex, right.Just a couple of months ago when all of these stocks were under lows, there was this perception t ...
【西街观察】达沃斯里的中国答案
Bei Jing Shang Bao· 2025-06-25 15:00
多年来,中国经济之所以是世界经济增长的重要引擎,不仅在于自身的稳定性和高成长性,还在于中国 经济的开放性和与世界经济的联动性。 中美重回贸易谈判桌,全球经济大咖们集结达沃斯论坛,其实是为了共同探求世界经济的未来之路。 面对不断变化的贸易格局,不断被冲击的全球化,世界经济如何驱散迷雾?中国经济又将如何发力?科 技创新和企业家精神能否解锁新动能? 6月25日上午,国务院总理李强在天津出席2025年夏季达沃斯论坛开幕式并致辞。 据新华社消息,李强表示,我们应当顺应正道和大势,拿出智慧和担当,采取积极的态度和建设性的行 动,坚定不移拥抱普惠包容的经济全球化。 中国将一如既往欢迎各国企业来华投资兴业,期待大家在这里实现梦想、获得成功,伴随中国经济一路 行稳致远。 改革开放以来,在积极参与全球化的过程中,中国融入了全球贸易体系,逐步发展成为"世界工厂",为 全球市场提供了更加高效、稳定的产业链、供应链。 中国企业也在不断地刷新在全球化进程中的角色,不管是"引进来"还是"走出去",本质都是全球化的一 部分。 在变局中,中国经济之所以能够保持稳定增长的态势,一方面在于全球产业链布局深度调整的背景下, 产业配套的高质量、高效率 ...
500创富榜发布:AI增长改写榜单格局,梁文锋冲进前十
Guan Cha Zhe Wang· 2025-06-25 12:29
Group 1 - The 2025 New Fortune 500 Rich List has been significantly influenced by the development of AI businesses, particularly among the top ranks [1] - Zhang Yiming, founder of ByteDance, has become the richest person with a holding value of 481.57 billion yuan, marking a rise from third place last year [2][5] - The list shows a notable shift in wealth distribution, with AI driving substantial growth for companies like DeepSeek, which has seen its founder Liang Wenfeng's wealth surge to 184.62 billion yuan [3][5] Group 2 - DeepSeek's rapid growth is attributed to its AI model, which competes effectively with OpenAI's models while achieving lower computational costs [3][5] - ByteDance's revenue for 2024 is projected at 155 billion USD, a 29% increase year-on-year, with a net profit of 33 billion USD [5][6] - The top ten list features significant representation from Zhejiang and Guangdong, indicating a regional shift in wealth concentration compared to previous years [7] Group 3 - The list highlights three major sectors: TMT (Technology, Media, and Telecommunications), pharmaceuticals, and consumer goods, which collectively account for nearly half of the list [8] - New entrants in the robotics sector include young entrepreneurs like Wang Xingxing from Yushu Technology, showcasing the emergence of new talent in the industry [9]
Kimi还能找到月之亮面吗?
3 6 Ke· 2025-06-25 08:08
Core Insights - Kimi, once a prominent player in the AI space, has seen a decline in attention as newer models from companies like Quark, Tencent, and Alibaba gain traction [1][2] - The initial hype around Kimi was driven by its technological scarcity, particularly its long-text processing capabilities, which were unmatched at the time [2][3] - Kimi's early valuation of $3 billion was supported by its unique technology, the founder's impressive background, and the capital's anxiety to find a domestic alternative to leading AI models [4][5] Technology and Market Position - Kimi's long-text processing ability, which expanded from 200,000 to 2 million words, was a significant technological breakthrough that positioned it as a leader in the AI field [2][3] - The founder, Yang Zhilin, had a strong academic and entrepreneurial background, which enhanced investor confidence in Kimi's potential [3][4] - The competitive landscape was characterized by a rush to find alternatives to ChatGPT, leading to Kimi's rapid user acquisition through aggressive marketing strategies [4][5] Financial Strategy and User Acquisition - Kimi faced challenges in managing its newfound capital, leading to excessive spending on user acquisition, with monthly advertising costs peaking at 220 million RMB [6][7] - Despite a significant increase in daily active users (DAU) from 508,300 to 5,897,000, this growth was primarily driven by financial investment rather than product quality [8][9] - The pressure from investors to demonstrate commercial viability led Kimi to prioritize user numbers over technological development, resulting in a loss of strategic direction [8][9] Challenges and Strategic Missteps - Kimi's marketing strategy shifted focus from its core user base in academia and professional fields to entertainment sectors, diluting its brand identity [11][12] - The company struggled with maintaining its technological edge as competitors began to catch up, particularly with the emergence of open-source models [12][13] - Kimi's reliance on user growth without a solid feedback loop or data quality management led to a false sense of security regarding its market position [13] Future Opportunities - Kimi has potential avenues for recovery, including enhancing the value density of its products and focusing on deep search capabilities for specific industries [15][17] - The company could benefit from developing comprehensive tools for developers, improving its API offerings to facilitate easier integration for enterprise clients [18][19] - Emphasizing quality over quantity in user engagement and product offerings could help Kimi regain trust and market relevance [20][21] Strategic Recommendations - Kimi needs to establish a clear commercial strategy from the outset, ensuring that its products meet genuine market demands and have viable monetization paths [29][30] - The focus should shift towards building a sustainable revenue model based on user payments rather than relying on external funding for growth [31] - A strategic approach that prioritizes understanding and fulfilling real user needs will be crucial for Kimi's long-term success in the competitive AI landscape [31][32]