不，你是走进了一间坐着四位专家的会议室。 你甚至能看见他们互相质疑、互相拆台、互相纠错，最后由「队长 Grok」拍板整合成答案。 新智元报道 编辑：犀牛 【新智元导读】 刚刚，马斯克Grok 4.20 beta版发布。不是一个AI，而是4个智能体现场开会辩论！实盘炒股最高47%回报，直接暴击GPT-5和Gemini。 在这个赛博朋克的春节档，马斯克给我们端上了一盘「硬菜」。 就在几个小时前，xAI 在毫无预警的情况下上线了 Grok 4.20 Beta，旁边还写着醒目的 「4 Agents」。 你以为你在和一个 AI 聊天？ 现在终于轮到 AI 给你表演「开会的艺术」了。 这不，模型刚上线，网友就开启了实测。 比如，最近常见的50米洗车测试。 只不过这次网友换成了「50米洗直升机」。 不管怎样，Grok 4.20 的回答看上去好像不太行。 「它和原版游戏一样，具有相同的规则和风格，完全可以玩。」Testlabor表示。 网友Testlabor用Grok 4.20在1分20秒内为编写了一个俄罗斯方块游戏。 网友tetsuo甚至构建了一个人工生命模拟器。 相当炫酷。 马斯克本人也没闲着。 他在X上开始了一轮颇具马 ...

Key Takeaways Riot’s Bitcoin mining profits are declining sharply due to lower prices. Activist investor Starboard urges Riot to accelerate its shift to AI data centers. Many Bitcoin miners are pivoting to AI hosting for steadier, higher margins. Riot Platforms, one of the largest U.S. Bitcoin mining companies, is under growing pressure to pivot to Artificial Intelligence (AI). This trend has gained momentum amid declining mining profitability. The company owns large power sites in Texas that c ...

Core Viewpoint - OpenAI's decision to discontinue GPT-4o has sparked significant user backlash, with thousands advocating for its retention due to its unique user experience and characteristics [1][3]. Group 1: User Response - A petition on Change.org has garnered approximately 21,900 signatures, emphasizing the irreplaceable experience GPT-4o provided to many users [3]. - Users have expressed strong emotional connections to GPT-4o, with some stating that losing it feels like losing a friend rather than just a feature [5]. - Some paid users have indicated they will cancel their ChatGPT subscriptions if GPT-4o is discontinued, highlighting the model's importance to their continued use of OpenAI's services [5]. Group 2: OpenAI's Position - OpenAI announced the discontinuation of GPT-4o and three other variants, effective February 13, following a previous plan to retire GPT-4o that was postponed due to user support [1][3]. - The company cited low usage rates, with only 0.1% of users still opting for GPT-4o, as one of the reasons for its retirement [5]. - Feedback from users regarding GPT-4o's conversational style and warmth has influenced the development of future models, GPT-5.1 and GPT-5.2 [5].

Core Viewpoint - Auddia Inc. has announced a definitive merger agreement with Thramann Holdings, resulting in the formation of a new holding company named McCarthy Finney, with a new ticker symbol MCFN. Auddia shareholders will own approximately 20% of the new entity at closing, which is estimated to have a base case DCF valuation of $250 million [2][6][7]. Company Overview - Auddia Inc. operates through its proprietary AI platform for audio, aiming to transform consumer engagement with AM/FM radio and podcasts, while also enhancing music promotion for artists [11]. - Thramann Holdings is a privately held company that controls three early-stage AI-native companies: LT350, Influence Healthcare, and Voyex, founded by Jeff Thramann, a serial entrepreneur with over 130 patents [3][10]. Merger Details - The merger will see Auddia renamed as McCarthy Finney, with Thramann Holdings' entities becoming fully owned subsidiaries. Jeff Thramann will continue as CEO, and John Mahoney will remain CFO [5]. - Auddia shareholders are expected to hold 20% of McCarthy Finney, while Jeff Thramann will own 80% at closing. The merger is contingent on Auddia having at least $12 million in cash at closing [6][8]. Valuation and Financials - The base case valuation of McCarthy Finney is estimated at $250 million based on a discounted cash flow analysis over a 10-year pro forma [7]. - Financial statements and detailed disclosures regarding McCarthy Finney and its portfolio companies will be filed with the SEC, including a registration statement on Form S-4 [7][19]. Timeline and Approvals - The transaction has been unanimously approved by the boards of both companies and is expected to close in the second quarter of 2026, pending customary closing conditions and stockholder approvals [8][9].

Core Insights - Perplexity AI is shifting its focus from advertising to subscriptions and enterprise clients to achieve sustainable revenue growth [1][2] - The company aims to target high-powered professionals, emphasizing revenue retention over other metrics [2] - Perplexity AI reached $200 million in annual recurring revenue (ARR) by October 2025 and is expanding its enterprise offerings [3] Monetization Strategy - The company is prioritizing monetization, particularly through subscriptions and enterprise sales, rather than the volume of questions answered [2] - A five-person enterprise sales team is currently in place, with plans for expansion in the near future [3] Advertising Approach - Previous attempts to incorporate ads into AI-generated answers were unsuccessful, as users expressed skepticism about their credibility [3][4] - Perplexity AI will continue to offer free access to its services, albeit with rate limits [4] Product Development and Security Concerns - The CEO highlighted the potential of on-device AI to enhance user control and automate tasks, while also warning against misuse of the AI browser, Comet [5] - Security vulnerabilities were identified in Comet following audits by Brave and Guardio, raising concerns about its capabilities [6]

Core Viewpoint - The article discusses the rapid evolution of the AI video generation market, highlighting the competitive landscape where ByteDance's Seedance 2.0 has emerged as a significant player, potentially redefining the industry standards and challenging existing leaders like Kuaishou's Keling [4][5][6]. Market Dynamics - The AI video generation sector is transitioning from isolated breakthroughs to comprehensive competition, with various domestic models like Kuaishou's Keling, ByteDance's Jimeng, and Alibaba's Wanxiang leading the charge [8]. - The competition is intensifying as different companies adopt varied development strategies, with many models converging on basic capabilities while maintaining distinct features and technical specialties [8][12]. Keling's Position - Keling has been recognized as a leading video generation model since its launch in June 2024, achieving over 30 iterations and generating significant revenue, exceeding 700 million yuan by the third quarter of 2025 [11][20]. - Despite its early advantages, Keling faces pressure as its technological lead diminishes, with competitors rapidly advancing in areas like multi-shot narrative and complex content generation [15][20]. Seedance 2.0's Impact - Seedance 2.0 has introduced significant advancements, such as automatic multi-shot usage and improved narrative logic, outperforming competitors in efficiency and cost [13][19]. - The model's ability to generate 2K videos faster than competitors by 30% has raised the bar for the industry, prompting concerns for Keling's market position [13][19]. Competitive Strategies - ByteDance's Jiemeng AI integrates deeply with its Douyin platform, achieving rapid user growth, while Alibaba's Wanxiang adopts an open-source strategy to enhance its technological edge [12][20]. - Keling's strategy focuses on product innovation and commercialization, but it must navigate internal challenges and external competition to maintain its market share [10][21]. Future Outlook - The article suggests that Keling must differentiate itself beyond being a mere generation tool to avoid being marginalized in the competitive landscape [22]. - The future of competition will hinge on understanding creator needs, content ecosystem management, and innovative business models, indicating that both Seedance and Keling have significant challenges ahead [22].

本文来自微信公众号： 腾讯科技 ，作者：晓静，编辑：徐青阳，原文标题：《技术指数级发展，可 怕的是全世界竟无察觉｜Anthropic CEO最新访谈》，题图来自：视觉中国 "我90%确信，2035年前人类将迎来'数据中心里的天才国度'——甚至可能就在一两年内。" Anthropic CEO达里奥·阿莫代伊 （Dario Amodei） 说出这句话时，语气平静得像在预言明天的天 气。 但真正让他抓狂的不是技术进展太快，而是全世界竟然毫无察觉。在接受美国知名博客主持人德瓦克 什·帕特尔 （Dwarkesh Patel） 近150分钟的深度专访中，阿莫代伊反复强调一个观点： 我们离AGI 的终点比任何人想象的都要近，而公众还在讨论那些老掉牙的政治话题。 帕特尔：现在"规模扩展"的假设到底是什么？预训练的扩展定律大家都懂，但强化学习扩展好像没有 公开的规律可循。 阿莫代伊：我现在的假设跟2017年写《大计算块假说》 （The Big Blob of Compute Hypothesis） 时是一样的，与图灵奖得主、强化学习之父里奇·萨顿 （Rich Sutton） 的《苦涩的教训》 （The Bitter Lesso ...

Core Viewpoint - The competition among major tech companies for national-level traffic entry points during the Spring Festival is intense, with no clear winner emerging from this year's events [1][3][6]. Group 1: Competition Overview - Major players in the competition include Alibaba, ByteDance, Tencent, and Baidu, each deploying significant financial resources to attract AI users [3][5]. - Alibaba's Qianwen launched a 30 billion yuan "Spring Festival invitation plan," while Tencent offered 1 billion yuan in cash red envelopes through its Yuanbao app [3][4]. - ByteDance's Doubao and Baidu's Wenxin Assistant also participated, with Doubao reporting 1.9 billion AI interactions on New Year's Eve [4][13]. Group 2: Company Strategies - Alibaba leverages its extensive ecosystem and open-source models to compete effectively in both domestic and international markets [3][5]. - ByteDance's Doubao has gained a first-mover advantage with a higher daily active user count compared to its competitors [3][5]. - Tencent's model capabilities are perceived as weaker, but the company remains confident in its ability to compete in the AI space [5][6]. Group 3: Market Dynamics - The competition has raised the entry barrier for other companies, requiring substantial investment to compete for national-level traffic [5][6]. - The rapid iteration of large models in China has narrowed the gap with U.S. counterparts, with the difference in capabilities reduced to about three months [13][14]. - Chinese companies are increasingly efficient in their AI investments, achieving competitive intelligence levels with lower costs compared to U.S. firms [15]. Group 4: Future Trends - The emergence of intelligent agents, such as LobsterAI, indicates a shift towards more specialized applications of AI technology [8][9]. - The evolution of intelligent agents is expected to progress through various stages, from single-app functionalities to cross-platform interactions [10][11]. - The competitive landscape suggests that smaller companies may find opportunities by leveraging multiple AI models rather than being tied to a single one [9].

Core Viewpoint - The article discusses the emergence of NOOK, an AI application designed to serve as a "life copilot" that helps users navigate personal decisions and goals, moving beyond traditional efficiency tools to become a comprehensive life management system [2][6][7]. Group 1: Product Concept and Functionality - NOOK aims to evolve AI applications from efficiency tools to a life operating system, addressing the gap in personal decision-making support [6][7]. - Users create a profile by entering their birth information, which generates a personalized "life profile" that the AI uses to recommend goals and tasks [5][24]. - The product features a gamified experience, where users receive feedback and rewards for completing tasks, enhancing motivation and engagement [17][29]. Group 2: Target Audience and Market Insights - The primary user base consists of college students and young professionals, particularly those feeling lost and seeking change, with a focus on subcultures and interests in "mysticism" [9][31]. - The application addresses the pain point of lack of control and motivation among young users, providing structured tasks to help them initiate change in their lives [33]. Group 3: Development and Challenges - The founder, Xu Qinglan, has invested personal funds into the project, indicating a reliance on self-funding while seeking external investment [11][42]. - Early user growth has been promising, with over 7,000 registered users and a daily active user count of around 100, but challenges remain in scaling and monetization [35][41]. Group 4: Future Vision and Commercialization - The long-term vision for NOOK includes developing a network of agents that collaborate with users, enhancing the personalization and effectiveness of the application [25][26]. - Potential revenue streams include subscription models, skill agent utilization fees, and future e-commerce or advertising opportunities [43].

Core Viewpoint - The launch of Grok 4.2, despite internal turmoil at xAI, has generated mixed reactions from users, highlighting both praise and criticism for the new model [1][13]. Group 1: Model Release and Features - Grok 4.2 has been released as a public beta version, featuring 500 billion parameters, which is relatively modest compared to other models with trillions of parameters [1]. - The model is designed to self-iterate weekly, allowing for continuous updates and improvements based on real-time feedback [3][9]. - Users must manually select the beta version to use Grok 4.2, which introduces a rapid learning capability [9]. Group 2: User Feedback and Reception - Initial feedback has been polarized, with some users expressing disappointment in the model's performance, particularly in complex logical reasoning tasks [13]. - Critics suggest that Grok 4.2's capabilities may not match those of competitors like OpenAI and Anthropic, raising concerns about its practical effectiveness [13]. - Conversely, some users have praised Grok 4.2 for its coding abilities and multi-modal capabilities, indicating a divide in user experience [14]. Group 3: Elon Musk's Response and Future Plans - Elon Musk has actively defended Grok 4.2 on social media, emphasizing its potential for improvement and future updates [19]. - The current version is described as a "small" model, with plans for medium and large versions to be released later [19].