8:01 AM Kr: Eight listed companies collectively announce tax supplements; ChatGPT to undergo the largest upgrade in history; Gaokao adds AI invigilators, automatically captures abnormal video 8点1氪丨八家上市公司集中公告“补税”；ChatGPT将迎来史上最大幅度升级；高考新增AI监考员，自动截取异常录像

This perhaps explains why OpenAI is in such a rush to turn ChatGPT into a "super app." When the broadest "user base" consists of machines, the ceiling for the value of a chatbot becomes clear. Therefore, the answer lies not in creating a better chatbox, but in becoming an "AI Agent" capable of performing tasks on your behalf—writing code, booking flights, managing schedules. Their massive investments in upgrading Codex and fierce competition for enterprise clients are essentially about securing a foothold for the imminent future dominated by AI Agents. The rules of this competition have changed; it's no longer about who can "talk" better, but about who can "act" more effectively. OpenAI, valued at $850 billion, is less of an AI company and more of a massive internet arms dealer desperately searching for growth engines. However, as everyone aspires to become a "super app," will the ecological niches of the digital world become even more crowded and closed?

While giants are vying for territory, a seemingly quiet revolution is unfolding at the grassroots level. The price of computing power is collapsing, with manufacturers like DeepSeek and Xiaomi's MiMo driving prices down by nearly 99%. It's no longer a "high-end resource" but is transforming into "digital water and electricity." This means the barriers to innovation have been drastically lowered. The AI game that once only giants could afford to play can now be tackled by a garage startup team. This may be more exciting than any flashy model release, as it concerns the redistribution of power.

But every coin has two sides. As AI permeates every corner, regulatory oversight follows. AI proctors have been introduced in college entrance exam halls to detect abnormal behaviors and automatically capture surveillance footage. The original intent is, of course, to ensure fairness and prevent cheating. Yet this scenario always strikes me as somewhat cyberpunk: a group of human examinees writing intensely under the gaze of another "AI invigilator." The "all-seeing eye" of technology is omnipresent, safeguarding order while quietly defining new behavioral boundaries. We enjoy the convenience and security brought by technology, yet we are unknowingly surrendering a portion of our freedom "not to be observed." Whether this exchange is equitable may require a longer time to measure.

The flow of talent resembles a precise betrayal, exposing the industry's truest ambitions and anxieties. Clive Chan, the "Employee No. 002" from OpenAI's self-developed chip project, defecting to rival Anthropic is more than just a job change. It's a strategic signal: the deep-water zone of the large model competition has shifted from the front-end model parameters and user experience to the underlying chips and computing infrastructure. Whoever holds the more efficient and cheaper "shovel" is more likely to unearth more "gold." By taking on this chess piece, Anthropic is addressing its infrastructure shortcomings and preparing for a longer, tougher battle ahead.

Looking back at Aisidi, the Apple distributor that paid an additional 300 million yuan in taxes, or Luckin Coffee's half-filled iced latte—these trivial business stories seem like insignificant footnotes in the grand technological narrative. Yet they constitute a real sense of absurdity: on one hand, machines are taking over the internet, and AI agents are poised to reshape the world; on the other hand, enterprises are still wrestling with tax issues from years ago, and consumers are still nitpicking over the physical volume of a beverage. The world progresses in such a disjointed manner—some people are fighting for the stars and the sea of AI, while others are still squabbling over a half cup of coffee. And we all simultaneously inhabit these vastly different layers of reality.

SpaceX's IPO received double subscriptions, gold prices fell below key levels, fuel vehicles are being discounted for clearance... The fragments of information pieced together depict a world that seems both accelerated and creased under the tides of AI. Machines have surpassed human traffic in the digital realm, while in the physical world, humans are still emotionally stirred by fluctuations in gold, car, and stock prices. This transcendence may only be the beginning. It reminds us that we are no longer "using" the network; it is more likely that the network (or rather, the AI behind it) is "using" the data we generate to train itself, ultimately serving the new entity it represents.

The real competition may have only just begun. It's not a competition between humans and AI, but a test of who can better navigate this chaotic era where old and new rules alternate.

互联网的流量江山，第一次被机器们悄无声息地夺走了。57.4%对42.6%，这不是一个简单的数字，而是一场无声的“流量政变”。那些由代码驱动的爬虫、智能体、自动化程序，正以前所未有的规模和速度，吞噬着数字世界的带宽和注意力。当我们在屏幕前滑动手指时，真正的“用户”主力已经是那些不知疲倦的硅基生命了。这感觉就像走在热闹的集市，突然发现街上来来往往的，超过一半都是戴着人皮面具的机器人。

这或许解释了为什么OpenAI急着要将ChatGPT变成一个“超级应用”。当最广大的“用户”是机器时，聊天机器人的价值天花板就清晰可见了。所以，答案不是做一个更好的聊天框，而是变成一个能替你我执行任务的“智能体”——帮你写代码、订机票、管理日程。他们砸下重金升级Codex，拼命争夺企业客户，本质上是在为那个即将到来的、由AI Agent主导的交互未来卡位。这场竞赛的规则已经变了，不再是比谁更会“说”，而是比谁更会“做”。估值8500亿美元的OpenAI，与其说是一家AI公司，不如说是一家正在拼命寻找增长引擎的、庞大的互联网军火商。只是，当所有人都想成为“超级应用”时，数字世界的生态位会不会变得更加拥挤和封闭？

就在巨头们争抢地盘的时候，一场看似安静的革命正在地基层面发生。算力的价格雪崩式下跌，DeepSeek、小米MiMo等厂商把价格打下来，降幅接近99%。这不再是“高端资源”，而是正在变成“数字水电”。这意味着，创新的门槛被极大地踏平了。过去只有巨头才玩得起的AI游戏，如今一个车库里的初创团队也可能玩得转。这或许比任何炫酷的模型发布都更让人兴奋，因为它关乎权力的再分配。

但硬币总有另一面。当AI渗透到每一个角落，监管的触角也随之而来。高考考场引入AI监考员，识别异常行为，自动截录像。初衷当然是为了公平，杜绝作弊。但这画面总让我觉得有点赛博朋克：一群人类考生在另一个“AI监考官”的凝视下奋笔疾书。技术的“天眼”无处不在，它守护着秩序，也悄然定义着新的行为边界。我们享受着技术带来的便利与安全，也正不知不觉地交出一部分“不被观察”的自由。这种交换是否等价，可能需要更长的时间来衡量。

人才的流动，则像一场精准的背叛，暴露了行业最真实的野心和焦虑。OpenAI自研芯片的“002号员工”Clive Chan转投死对头Anthropic，这不仅仅是一次跳槽。这是战略信号：大模型竞争的深水区，已经从前台的模型参数和用户体验，下沉到底层的芯片和算力基础设施。谁掌握了更高效、更便宜的“铲子”，谁才可能挖到更多的“金子”。Anthropic接下这颗棋子，是在补齐自己的基础设施短板，为更长远的苦战做准备。

回头再看那家补缴了3亿税款的苹果经销商爱施德，或是那杯只装了一半的瑞幸去冰拿铁，这些琐碎的商业故事，仿佛都是宏大技术叙事里不起眼的注脚。但它们构成了真实的荒诞感：一边是机器接管互联网，智能体准备重塑世界；另一边是企业还在为多年前的税务问题焦头烂额，消费者还在为一杯饮料的物理体积斤斤计较。世界就是这样参差地前进着，一部分人在为AI的星辰大海奋战，另一部分人还在为眼前的半杯咖啡较劲。而我们所有人，都同时生活在这些截然不同的现实层里。

SpaceX的IPO获得了双倍认购，黄金价格跌破了关键点位，燃油车在降价清仓……这些信息碎片拼凑出的，是一个在AI浪潮冲刷下，显得既加速又褶皱的世界。机器在数字空间里完成了对人类流量的超越，而在物理世界里，人类还在为金价、车价、股价的波动而心潮澎湃。这场超越，或许只是一个开始。它提醒我们，不再是我们在“使用”网络，而很可能是网络（或者说，网络背后的AI）在“使用”我们产生的数据，来训练它自己，并最终服务于它所代表的新主体。

真正的竞争，或许才刚刚开始。不是人与AI的竞争，而是看谁能更好地驾驭这个新旧规则交替的混沌时代。