OpenAI下调算力支出目标 据最新消息， OpenAI已大幅下调算力建设支出目标至6000亿美元，该金额较此前宣称的1.4万亿美元基 础设施承诺大幅缩水。 另外，数据显示，2025年OpenAI的营收为131亿美元。OpenAI最新预计，其2030年总收入将超过2800亿 美元。 OpenAI正在敲定一轮总额可能超过1000亿美元的巨额融资，其中约90%来自战略投资者，包括英伟 达、软银集团、亚马逊等。这笔交易，可能使OpenAI的整体估值超过8500亿美元。 据悉，OpenAI计划将相当比例的新资金投入算力相关硬件采购，包括来自英伟达的芯片产品，用于数 据中心和基础设施建设。 据外媒报道，作为OpenAI酝酿已久的融资轮的一部分，英伟达向OpenAI投资接近300亿美元。报道称， 这项投资实际上将取代两家公司之前计划的1000亿美元投资合作协议。 据CNBC报道，数月前，OpenAI的创始人兼CEO萨姆·奥尔特曼曾吹嘘1.4万亿美元的基础设施投入承 诺。如今，OpenAI正告知投资者，其2030年总算力投入目标约为6000亿美元。 据报道，消息人士透露，由于外界普遍担忧其扩张野心远超潜在收入，这家AI巨头 ...

"现在我们已经不再使用这种冷却方式了，但网上还是有这类说法，比如'别用 ChatGPT，每次提问要消耗 17 加仑 （IT之家注：约 77.3 升）水'之类的言论。"奥尔特曼称，"这完全不符合事实，极其荒谬，与现实毫无关联。" 他还表示，人们担忧人工智能的能源消耗总量是合理的 —— 并非单次提问的能耗，而是全球如今大规模使用人工智 能带来的总能耗。在他看来，这意味着世界需要"尽快转向核能、风能和太阳能"。 IT之家 2 月 22 日消息，本周，OpenAI 首席执行官萨姆 · 奥尔特曼在《印度快报》举办的一场活动上，回应了外界 对人工智能环境影响的担忧。 奥尔特曼此次赴印度是为参加一场重要的人工智能峰会。他表示，有关人工智能耗水的担忧"完全是假的"，不过他 也承认，过去数据中心采用蒸发冷却技术时，耗水确实是个实际问题。 目前法律并未要求科技公司披露其能源与水资源使用量，因此科学家 们一直在尝试独立开展相关研究。数据中心也被认为与电价上涨存在 关联。 主持人援引此前与比尔 · 盖茨的对话提问：称 ChatGPT 单次提问的能 耗相当于 1.5 部 iPhone 的电池电量，这一说法是否准确。奥尔特曼回 应："能 ...

新浪科技讯 2月22日上午消息，针对GLM-5发布后规则透明度不够、灰度太慢等问题，智谱日前发布 GLM Coding Plan致歉信称："GLM-5发布后，流量超出预期，公司扩容节奏没有跟上，不得已将GLM-5 按照Max、Pro、Lite的顺序逐步开放。目前Max用户已经全面开放，Pro用户虽已开放，但高峰期可能 会因集群负载较高遇到限流，Lite用户将会在节后非高峰期逐步灰度开放。" 针对受到影响的Lite和Pro用户，智谱支持自主申请退款，退款遵循2026年1月1日至今天，智谱全部请客 的原则。 老用户（2026年1月1日前订阅，且2月12日后仍有效）：退2026年1月1日至今+剩余天数的全部金额。 新用户（2026年1月1日后订阅，且2月12日后仍有效）：全额退还当前订阅周期内全部实付金额。（文 猛） 『 『 V 』 『 』 『 』 『 』 『 』 『 』 『 』 『 』 『 』 『 』 『 』 』 GLM-5的参数规模是GLM-4.7的两倍以上,定位是效果逼 近Claude Opus 4.5的"大杯"模型,适合复杂任务。为了让 更多用户能把GLM-5用在真正需要的场景,我们设计了分 层使用策略:日 ...

Core Viewpoint - The company, Zhipu, issued an apology on February 21 regarding the GLM Coding Plan, acknowledging three main mistakes: insufficient transparency in rules, slow rollout of GLM-5, and poorly designed upgrade mechanisms for existing users [1][15]. Group 1: GLM-5 Launch and Performance - Zhipu officially launched its flagship model GLM-5 on February 12, achieving over a 20% performance improvement in programming scenarios compared to its predecessor [3][17]. - The model's performance is reported to be close to that of Claude Opus 4.5, and it achieved the best results in three agent evaluations in the open-source domain [3][17]. - Following the launch, Zhipu raised the prices of the GLM Coding Plan packages, with a 30% increase in China and over 100% for the overseas version, marking it as the first AI-native company in China to increase prices for large model commercialization services [3][17]. Group 2: User Experience and Rollout Issues - Due to unexpected demand, Zhipu had to gradually open GLM-5 to users in a tiered manner: Max users are fully open, Pro users may experience throttling during peak times, and Lite users will be gradually opened after the holiday [1][9][24]. - The company has acknowledged that the previous dashboard refresh rate of one hour was alarming for users and has since optimized it to a ten-minute refresh rate [6][22]. - Zhipu is currently facing challenges from malicious actors and resource hoarding, which have impacted service availability [9][24]. Group 3: Refund and Compensation Policy - Zhipu announced a refund policy for affected Lite and Pro users, allowing them to apply for refunds due to the service disruptions [1][9][24]. - The refund window will open one week after the holiday and will close on March 6, 2026, with full refunds for both old and new users based on their subscription periods [11][25]. - Users who have already utilized GLM-5 will receive a 15-day extension on their subscription period as compensation [11][26]. Group 4: Company Background and Market Position - Zhipu, founded in 2019, is recognized as a leading AI model company in China and is often referred to as "China's OpenAI" [3][18]. - The company has made significant advancements in AI model development, including the release of the GLM-130B model, which was the only Asian model selected in the Stanford evaluation of mainstream global models in 2022 [3][19]. - As of February 20, Zhipu's stock price surged by 42.72%, reaching a new high of 725 HKD, with a total market capitalization of 323.2 billion HKD, placing it among the top tier in the Hong Kong TMT sector [1][2][17].

Over the past two years, artificial intelligence (AI) stocks surged as investors rushed in, afraid of missing out on the next big tech boom. Valuations climbed quickly, and excitement around AI pushed many big names sharply higher. Recently, tech and AI stocks have taken a hit amid concerns about high valuation and questions over whether aggressive AI investments are delivering results fast enough. However, seasoned investors know that the real opportunity often appears when the excitement cools down. So ...

Core Viewpoint - The company issued an apology letter on February 21 regarding the GLM Coding Plan, acknowledging three main mistakes and outlining compensation measures for affected users [1][6]. Group 1: Mistakes Acknowledged - The company identified three key errors: insufficient transparency in rules, slow rollout of GLM-5, and poorly designed upgrade mechanisms for existing users [2][6]. - Following the release of GLM-5, demand exceeded expectations, leading to a staggered rollout of the model in Max, Pro, and Lite tiers, with Max users fully opened and Pro users facing potential throttling during peak times [2][11]. Group 2: Market Performance - On February 20, the company's stock price surged by 42.72% to 725 HKD, reaching a market capitalization of 323.2 billion HKD, marking a 524% increase from the initial offering price of 116.2 HKD [2][4]. - The company has positioned itself as a leading player in the TMT sector of the Hong Kong stock market, surpassing competitors like Kuaishou and Baidu in market capitalization [4]. Group 3: Product Details and Pricing - The GLM-5 model has shown over a 20% performance improvement compared to its predecessor, GLM-4.7, and is designed for complex tasks, closely matching the performance of Claude Opus 4.5 [4][5]. - Following the GLM-5 launch, the company raised the prices of its GLM Coding Plan packages, with a 30% increase for domestic users and over 100% for international users, becoming the first AI-native company in China to increase prices for large model commercial services [4][5]. Group 4: Compensation and User Support - The company will allow affected Lite and Pro users to apply for refunds, with a policy to refund all amounts paid from January 1, 2026, to the present [12][13]. - Users who mistakenly upgraded from old packages to new ones between February 12 and 16 will be allowed to revert to their previous plans, with the company absorbing any cost differences [15][16].

Core Insights - A recent Microsoft study confirms that even the most advanced large language models experience a significant decline in reliability during multi-turn conversations [1][3] - The phenomenon termed "lost conversation" reveals a systemic flaw in these models [3] Performance Metrics - The success rate of these models in single prompt tasks can reach 90%, but drops to approximately 65% when the same tasks are broken down into multi-turn dialogues [6] - While the core capabilities of the models decrease by only about 15%, their "unreliability" surges by 112% in multi-turn scenarios [7][8] Behavioral Mechanisms - Two primary behaviors contribute to performance decline: "premature generation," where models attempt to provide final answers before fully understanding user needs, leading to compounded errors [10] - "Answer inflation" occurs in multi-turn dialogues, where response lengths increase by 20% to 300%, introducing more assumptions and "hallucinations" that affect subsequent reasoning [10] Model Limitations - Even next-generation reasoning models equipped with additional "thinking tokens," such as OpenAI o3 and DeepSeek R1, did not significantly improve performance in multi-turn conversations [12] - Current benchmark tests primarily focus on ideal single-turn scenarios, neglecting real-world model behavior, posing challenges for developers relying on AI for complex dialogue processes [12]

今天关于 OpenAI 大幅下调支出目标的新闻被讨论得非常广，我们在星球中也第一时间对这个新闻 做了说明。 这个新闻的源头应该是 CNBC 发的这个新闻。 下午，国内的媒体也进行了转发，但很可惜，国内媒体并没有搞清楚始末就进行了转发。 1.4 Trillion 的由来 在去年 10 月份，奥特曼直播中表示，OpenAI 致力于投入 1.4 万亿美元，建设总计 30GW 的算力 资源。这里的 1.4 万亿可以理解成 Capex，但不仅包括 OpenAI，还有它的合作伙伴。 而 600 B 是指 OpenAI 到 2030 年的 OpEx，所以根本就不是一回事。 这里简单解释一下 CapEx 和 OpEx 的区别： 资本支出（CapEx） CapEx 是公司花大钱买或建造能用很多年的东西，比如买机器、建厂房、买服务器或升级大楼。这 些钱不是马上花完就没了，而是把东西记成资产，之后每年通过"折旧"一点点算进成本。简单说， 就是为了未来赚钱而做的长期投资，花钱的时候现金流出很大，但当期利润不会一下子掉很多，因为 成本被分摊了。 运营支出（OpEx） OpEx 是公司每天正常运转必须花的钱，比如付员工工资、交房租、水电 ...

Core Insights - Coinbase CEO Brian Armstrong has downplayed concerns regarding quantum computing's potential to compromise blockchain encryption, labeling it a "very solvable" issue [5][21] - The company has established an advisory board to evaluate the implications of quantum computing and prepare for future threats [4][3] - Industry leaders, including Vitalik Buterin, emphasize the urgency of implementing quantum-resistant technology to ensure long-term cryptographic safety for networks like Ethereum [2][3] Company Developments - Coinbase is actively engaging with major blockchains to strategize upgrades to post-quantum cryptography [4] - The advisory board will produce research on quantum computing's impact on blockchain and provide recommendations to organizations [3] - Coinbase's proactive stance positions it as a leader in addressing potential quantum threats in the cryptocurrency space [4][5] Industry Trends - The emergence of quantum computing poses significant challenges for blockchain security, prompting industry figures to advocate for immediate action [3][2] - The formation of advisory boards by companies like Coinbase reflects a growing recognition of the need to prepare for technological advancements that could disrupt current encryption methods [4][3] - The cryptocurrency sector is increasingly focused on ensuring the integrity of digital assets in the face of evolving technological threats [5][3]

②阿里千问宣布推出Qwen Coding Plan，支持更多模型； AI进化速递丨OpenAI推进新一轮巨额融资 ①OpenAI下调算力支出目标至6000亿美元，推进新一轮巨额融资； ④第二届国际人形机器人论坛19日至20日在瑞士比尔/比安创新园举行，探讨机器人应用发展趋势。 ③武汉出台十条措施，全力支持人工智能"超级个体"创新发展； ...