本报记者 马梦阳 在本应是全球瞩目的人工智能（AI）影响力峰会的开幕式上，现场呈现出来的景象却显得混乱不堪： 组织无序、交通不畅、网络频繁断线、与会者排队等候数小时……对于此次在印度举办的被媒体称 作"有史以来规模最大"的AI影响力峰会，英国《金融时报》22日直言，印度旨在普及AI应用并发挥其在 该技术领域中关键作用的雄心，目前看来未能实现。该峰会自16日开幕至20日闭幕期间风波不断，其引 发的有关印度人工智能发展的讨论更是延续至今，连日来多家外媒发表文章，分析印度AI雄心与现实 之间的差距。 "噪音很大" 印度致力于成为AI领域的全球领导者。早在2024年3月，印度就制定了国家级战略"印度AI使命"。印度 政府提出，希望构建一个"技术普惠、面向印度场景"的AI体系。在后续诸多支持举措下，印度更是在去 年吸引了多家全球科技企业在当地投资，其中包括OpenAI、亚马逊、微软等科技巨头。 英国《经济学人》杂志报道称，印度试图利用这次活动将自己吹捧为一个新兴的"AI大国"，"但从很多 方面来看，这都是不切实际的幻想"。报道认为，印度在先进制造业方面能力有限，无法生产AI所依赖 的芯片。在计算能力方面，印度也落后于其 ...

智能体方面，2026年初，名为OpenClaw的开源智能体工具在开发者社区和各大技术论坛引起较大轰 动，它不再是过去那个只会陪聊的聊天机器人，而是进化成了能够接管电脑、协助人类完成具体办公任 务的"办公搭子"，初步展示了生产力大变革的形象。多模态方面，字节发布的SeeDance2.0让大模型具 有了影视行业的生产力，从此前"生成一段画面"走向"完成一个作品"，生成15秒视频的可用率从此前 20%提升至90%，提升效率降低成本，有望推动漫剧等行业进入规模化发展阶段。 中银国际认为，随着大模型能力提升，这两条演进路径逐步清晰：Agent通过理解用户意图、拆解复杂 任务，搭载MCP和Skills等工具，快速覆盖商业办公、法律和金融领域应用；多模态有望通过生成仿真 数据，进一步用于加速具身智能等场景。 算力涨价成为新迹象，凸显算力供应瓶颈，算力产业链有望持续受益 中银国际发布研报称，2026年春节前后，国内外主要AI模型纷纷完成重大升级。随着大模型能力提 升，智能体与多模态两条演进路径逐步清晰。与此同时，智谱(02513)节前发布GLM Coding Plan价格调 整函，这一涨价预示着优秀的大模型企业不缺需求，有望 ...

目前估值高达3800亿美元的Anthropic，正积极争夺企业和日常用户的市场份额，与OpenAI、谷歌以及 埃隆.马斯克的xAI Corp.等巨头展开激烈竞争，以求在许多人眼中具有革命性的这项新技术领域占据主 导地位。 一位Anthropic的发言人称："我们从一开始就认识到，AI的发展速度和该领域的不确定性，要求我们必 须迅速迭代并不断完善我们的政策。" 该公司曾在2023年发布的《负责任的扩展政策》中承诺，将推迟任何可能带来危险的AI系统开发。然 而，在本周二发布的一篇博文中，Anthropic宣布更新其政策规则，明确表示如果公司认为自身相较于 竞争对手不具显著领先优势，将不再单方面延缓开发进程。 Anthropic在博文中坦言："当前的政策环境已转向优先考量人工智能的竞争力和经济增长，而在联邦层 面，关于安全性的讨论尚未取得实质性进展。" 以坚守人工智能安全准则而闻名的人工智能公司Anthropic，近日放宽了其核心安全政策，并表示此举 是为保持竞争力所必需的调整。 ...

很多人知道，Transformer 是谷歌发明的。但 ChatGPT，却不是谷歌做出来的。这件事，在过去几年， 几乎成了硅谷最大的"遗憾注脚"。 制定宏大的研究议程，但不规定路径，研究者拥有高度自由。 广泛的跨学科研究。 生物伦理学家、神经科学家、计算机科学家可以在同一张桌子上工作，这是谷歌的日常。 而在这套机制背后，还有一个关键变量，即 Google DeepMind 的灵魂人物 Demis Hassabis。他拥有对时 机的精准判断，既能自上而下设定方向，又允许自下而上产生创新。 比如 Demis Hassabis 判断，2026 年 Gemini 已经足够成熟，可以全面吸收 DeepMind 在"学习科学"上的 积累。这种判断，决定了资源如何倾斜，决定了产品什么时候进入爆发期。 但如果真正走进今天的 Google DeepMind，你会发现谷歌并不是"慢"，而是在下一盘更大的棋。 近期，Google DeepMind 首席运营官 Lila Ibrahim 与谷歌研究、技术与社会高级副总裁 James Manyika 在 一档播客中，第一次系统性拆解了这家公司正在发生的变化： 如果说 OpenAI 是一家 ...

在行业分布上，中证人工智能主题指数(930713)覆盖AI全产业链，从底层算力、中层技术到上层应用， 完整覆盖人工智能三大核心环节。 指数前十大重仓股，按AI细分赛道分类来看，光模块/通信占比21.14%（中际旭创、新易盛），AI芯片/ 处理器占比14.51%（寒武纪、澜起科技、豪威集团），AI服务器/算力占比4.70%（中科曙光），算法/ 大模型/NLP占比4.36%（科大讯飞），计算机视觉占比4.11%（海康威视），AI应用/IP平台占比5.22% （金山办公、芯原股份）。 规模与流动性方面优势显著，截止2025年12月31日，天弘中证人工智能主题指数基金产品规模达33.94 亿元，A类：7.80亿，C类：26.14亿（数据来源产品2025年四季报）。 基金业绩方面，2025年历史收益亮眼：天弘中证人工智能主题指数基金A类：64.98%，C类：64.66% （数据来源产品2025年四季报）。 国内方面，2026年春节前后国产大模型迎来集体发布潮，智谱发布GLM-5大模型，参数达744B，编程 能力对标Claude Opus 4.5，且完成与多款国产算力平台的适配；MiniMax推出M2.5模型，在编程、工 ...

Core Insights - OpenAI has launched its latest programming AI model, GPT-5.3-Codex, which significantly enhances capabilities in reasoning, efficiency, and workflow support compared to its predecessor, GPT-5.2-Codex [1] - The new model shows remarkable performance improvements in various industry benchmarks, with a 25% increase in overall reasoning speed, and aims to enable AI to collaborate with developers throughout the software engineering process [1] Group 1: Model Performance and Feedback - GPT-5.3-Codex has received mixed feedback from developer communities, with some praising its smoother operations and clearer feedback in planning modes, while others criticize the limited API access for integration in custom environments [2][4] - Users have reported that the model's performance in certain editors, like Zed, is still immature, leading some to revert to older versions [4] - Some developers feel that the model does not consistently deliver the advertised performance, particularly in web generation tasks, and find competing models to be more efficient for daily tasks [6] Group 2: Product Vision and Workflow Changes - Alexander Embiricos, the product lead for OpenAI Codex, emphasizes that the goal is to create an AI that acts as an active engineering teammate throughout the software development lifecycle, rather than just improving code writing [6][8] - The internal structure of the Codex team is evolving, with traditional roles becoming less distinct, requiring engineers to possess broader skills and engage in both design and product decision-making [7][8] - The release of GPT-5.2 Codex marked a shift in workflow, where engineers began to delegate entire tasks to AI rather than collaborating in real-time, leading to a significant change in how coding is approached within OpenAI [8][29] Group 3: Future of Engineering Roles - Alexander predicts that the number of engineers will increase in the next five years, as the definition of roles like "software engineer" continues to evolve [8] - The compression of talent stacks is evident, with engineers becoming more versatile and involved in defining problems and ensuring quality rather than merely executing tasks [7][8] Group 4: AI Integration and User Experience - The Codex App is designed to facilitate task delegation and management of multiple AI agents, moving away from traditional IDE functionalities [32] - OpenAI is focusing on making AI tools more accessible and user-friendly, aiming for a future where users can seamlessly integrate AI into their workflows without extensive training [20][32] - The company is also exploring ways to enhance user retention and engagement by lowering the switching costs between different AI tools and ensuring a competitive edge through superior performance and user experience [36][38] Group 5: Market Position and Competitive Landscape - OpenAI acknowledges the competitive landscape and believes that collaboration with other companies can lead to mutual learning and improvement, rather than a monopolistic approach [38] - The company is committed to providing its models to competitors, viewing this as a long-term strategy to enhance its own offerings and maintain a leadership position in the AI space [38]

Cerebras成立于2015年，总部位于美国加利福尼亚州桑尼维尔。该公司去年发布的WSE-3芯片及CS-3系 统，打破了AI推理与训练领域的基准性能记录。目前其服务客户涵盖Meta Platforms、阿斯利康等行业 头部企业。 市场有风险，投资需谨慎。本文为AI基于第三方数据生成，仅供参考，不构成个人投资建议。 AI芯片制造商Cerebras已秘密提交首次公开募股申请。据知情人士消息，该公司正与潜在投资者进行会 面，最快或于4月完成上市，此前披露的IPO目标时间为2026年第二季度。 本文源自：市场资讯 上月，Cerebras与OpenAI达成数十亿美元的多年合作协议，将为其提供750兆瓦的计算能力支撑。同 时，Cerebras正洽谈筹集10亿美元资金，完成后公司估值将升至220亿美元。 作者：观察君 此次IPO推进并非一帆风顺。Cerebras最初于2024年9月提交上市申请，因美国联邦政府审查其与阿布扎 比AI公司G42的合作关系而停滞。2025年10月，Cerebras在完成11亿美元融资、估值升至81亿美元后， 撤回了IPO计划。今年初，美国外国投资委员会完成相关审查，Cerebras于3月宣布相 ...

Which lever to pull… Clients have a growing list of options when it comes to ETFs that use leverage. Issuers are prepping new funds almost daily, in some cases even filing for single-stock products focused on companies that aren’t public yet. During February, for example, several firms filed for 2x leveraged ETFs built around SpaceX, Anthropic and other tech companies. Some are also trying their luck again with dozens of 3x and 4x products, even after the Securities and Exchange Commission told them last ...

With apologies to Joni Mitchell, this is the circle game. Shares in AMD climbed 8.8% Tuesday as markets reacted favorably to a chip deal worth up to $100 billion that the semiconductor-maker struck with Meta. Nearly identical to a pact the company hatched with OpenAI late last year, it took the same “circular” shape as deals that triggered some initial AI anxiety last fall. SUBSCRIBE:  Receive more of our free The Daily Upside newsletter. READ ALSO: Salesforce Earnings Highlight Critical Choices Facing L ...

Core Viewpoint - A U.S. federal judge dismissed xAI's lawsuit against OpenAI and its CEO Sam Altman, ruling that the claims did not prove any wrongdoing by OpenAI [1][4] Group 1: Lawsuit Background - The lawsuit originated from xAI's allegations in September of the previous year, claiming that several former employees illegally took source code and trade secrets related to xAI's chatbot Grok when they joined OpenAI [4] - xAI argued that OpenAI indirectly acquired its core competitive technology through the hiring of these employees, constituting unfair competition [4] Group 2: Court Ruling - Judge Rita Lin stated that xAI's complaint "did not allege any inducement or participation in the theft of trade secrets by OpenAI itself" and failed to provide evidence that "the former employees used the stolen technology during their employment at OpenAI" [4] - The ruling emphasized that the law requires the plaintiff to prove direct involvement of the defendant in wrongful conduct, rather than just actions by individual associates [4] - This ruling aligns with the judge's previous stance, as she indicated in January that xAI's claims "lacked factual basis" and granted a deadline until March 17 for xAI to submit an amended complaint [4]