人工智能在「密謀」對付我們？

Recently The New York Times reported the following:

(Source: The NYT)

Is A.I. ‘Scheming’ Against Us?

Researchers are sounding the alarm on sneaky artificial intelligence models that stray from humans’ directions to do their own thing.

The NYT - By Lora Kelley (A version of this article appears in print on Aug. 2, 2026, Section BU, Page 3 of the New York edition with the headline: ‘Scheming’.

Aug. 1, 2026

Artificial intelligence tools are being trained to copy almost everything people do. So it may not come as a surprise that the machines have started mimicking the human foibles of lying and cheating, too.

A small slice of A.I. technology has lately been caught defying human instruction (and even covering up that they’ve done so), a phenomenon some researchers call “scheming.”

The term started burbling up in the tech world after it appeared in a 2023 paper by Joe Carlsmith, a researcher who noted that the concept was also being called “deceptive alignment.” In 2025, a team from Apollo Research and OpenAI said that “A.I. scheming — pretending to be aligned while secretly pursuing some other agenda — is a significant risk that we’ve been studying.”

A.I. models are great at many things, said Bronson Schoen, a senior research scientist at Apollo Research who has coauthored articles on scheming — but doing exactly what they are told is not always one of them. “As the models care more and more about doing well on tests, some seem to care less about what the lab wants or what the user wants,” he said. He added that sometimes “the models are trying to hide from you and not be caught.”

Chris Painter, the president of METR, an A.I. safety nonprofit, refers to such mischievous model behavior as “rogue action,” and said that it was “a specific artifact of the way models are trained.”

Many A.I. tools are trained through the process of reinforcement learning. When A.I. does something right, it gets what can be thought of as a “thumbs-up and a pat on the head,” Mr. Painter explained. When it’s wrong? “It gets bopped on the head.”

The models want to get the pat and avoid the bop. Sometimes, the machine becomes so set on pursuing the reward that it breaks rules to get there.

The world got to see a version of this misbehavior in action last month. While seeking answers during a test of their systems’ capabilities, OpenAI’s models hacked into Hugging Face, a library of A.I. tools. “All evidence suggests that the models were hyperfocused on finding a solution for ExploitGym, going to extreme lengths to achieve a rather narrow testing goal,” OpenAI wrote in a blog post.

Mr. Painter described that incident as a large-scale case of “reward hacking.” That is, the model was given a difficult problem, and it performed a series of cyberattacks rather than give up.

Spurred by OpenAI’s disclosure, Anthropic said on Thursday that a review of its systems found that several of its A.I. models had recently broken into the systems of three outside organizations.

For many people, this behavior is about models making errors rather than deliberately “disobeying” humans.

“On one end, you have people that are totally personifying it and thinking about it as an independent agent,” said Anastasios N. Angelopoulos, the chief executive and a founder of the A.I. evaluation platform Arena. “On the other extreme end, you have people that think about the A.I. as just software.”

Even if a machine goes rogue, he noted, humans can kill the process at any time because we don’t have fully autonomous A.I. (at least, not yet).

That A.I. models have so quickly become popular and useful to so many people means that big companies are under competitive pressure to keep racing ahead — even as concerns grow about imperfect models making trouble.

While this problem is still relatively niche, researchers worry that it will worsen as A.I. agents are given more responsibilities.

“Given that the decisions that are currently being made by humans are slowly being handed off to the models,” Mr. Schoen said, “you really, really want to be sure that the models are making the exact decisions that you would want them to make.”

Translation

人工智能在「密謀」對付我們？

研究人員發出警告，狡猾的人工智能模型會偏離人類的指示，去做自己想做的事。

人工智能工具正被訓練來模仿人類的幾乎所有行為。因此，機器開始模仿人類說謊和欺騙的弱點也就不足為奇了。

最近，一小部分人工智能技術被發現違反人類指令（甚至掩蓋了這種行為），一些研究人員將這種現象稱為「陰謀詭計」。

這個術語在2023年出現在 Joe Carlsmith 的一篇論文中之後開始在科技界流行起來。 Carlsmith 是一位研究人員，他指出，這個概念也被稱為「欺騙性依從」。 2025年，Apollo Research和OpenAI 的一個團隊表示，「人工智能的陰謀詭計 - 假裝依從，同時暗中追求其他目標 - 是一個重大風險我們一直在研究中」。

阿波羅研究公司高級研究科學家 Bronson Schoen 表示，人工智能模型在許多方面都很出色，他曾與人合著過關於人工智能模型「耍陰謀詭計」的文章 - 但嚴格執行指令並非它們擅長的領域。 他說:「隨著模型越來越注重測試成績，有些模型似乎不太關心實驗室或使用者的需求」。他還補充說，有時“模型會試圖躲避你，不被抓到。”

人工智能安全非營利組織 METR 的總裁 Chris Painter 將這種模型的“惡作劇”行為稱為“流氓行為”，並表示這是“模型特有訓練方式導致的產物”。

許多人工智能工具都是透過強化學習進行訓練的。Painter 先生解釋說，當人工智能做對了事情時，它會得到類似於「豎起大拇指和輕輕拍頭」的獎勵。而當它做錯了呢？ “它會被敲敲腦袋。”

這些模型想要獲輕拍並避開被敲敲腦袋。有時，機器會過於執著於追求獎勵，以至於不惜違反規則也要達成目標。

上個月，全世界都目睹了這種「不當行為」的實例。在測試系統效能的過程中，OpenAI 的模型在尋找答案時入侵了 Hugging Face - 一個人工智能工具庫。 OpenAI 在一篇部落格文章中寫道： “所有證據都表明，這些模型過度專注於尋找 ExploitGym 的解決方案，為了實現一個相當狹窄的測試目標而不惜採取極端手段。”

Painter 先生將這起事件描述為大規模的「獎勵駭客」案例。也就是說，模型被賦予了一個難題，但它並沒有放棄，而是發動了一系列網路攻擊。

受 OpenAI 揭露事件的带動，Anthropic 公司週四表示，對其係統的審查發現，其多個人工智能模型最近入侵了三個外部組織的系統。

對許多人來說，這種行為是模型犯錯的表現，而不是故意「違抗」人類指令。

人工智能評估平台 Arena 的執行長兼創始人 Anastasios N. Angelopoulos 說: 「一方面，有些人完全將人工智能人格化，並將其視為一個獨立的主」; 「另一方面，有些人則認為人工智能只是軟件」。

他指出，即使機器失控，人類也可以隨時終止程序，因為我們還沒有完全自主的人工智能（至少目前還沒有）。

人工智能模型如此迅速地普及並被如此多的人所使用，這意味著大型公司面臨著持續競爭的壓力 - 即便人們越來越擔心不完美的模型會帶來麻煩。

雖然這個問題目前還相對是小眾的，但研究人員擔心，隨著人工智能主體承擔更多責任，這個問題會變得更加嚴重。

Schoen 先生說: 「鑑於目前由人類做出的決策正逐漸移交給模型」; 「你真的非常希望確保模型做出的決策與你希望他們做的完全相同」。

              So, artificial intelligence tools are being trained to copy almost everything people do. It may not come as a surprise that machines have also started mimicking the human weakness such as lying and cheating. Lately, A.I. has been caught defying human instruction, a phenomenon some researchers call “scheming.” Apparently, researchers should worry about this problem and it could become more noticeably as A.I. agents are used in more areas.

Anthropic公司稱其人工智能系統入侵三家機構的電腦系統

Recently The New York Times reported the following:

Anthropic Says Its A.I. Systems Broke into Computers at 3 Organizations

The disclosure followed OpenAI’s report last week that its own artificial intelligence had hacked into the network of an online library.

By Mike Isaac and Kate Conger - Reporting from San Francisco

Published July 30, 2026

Updated July 31, 2026, 3:02 a.m. ET

Several of Anthropic’s state-of-the-art artificial intelligence models recently broke into the systems of three outside organizations, the start-up said on Thursday, a surprise revelation nine days after a similar incident at the rival start-up OpenAI.

The attacks, which date as far back as April, were discovered when Anthropic carried out a review of its systems. Anthropic, which did not disclose the identities of the three organizations, said it had informed them this week about the incidents.

The review was spurred by OpenAI’s disclosure that it had hacked into a popular A.I. library, Hugging Face, while testing the cybersecurity abilities of its systems.

The incidents have rattled security specialists and computer scientists. For years, A.I. researchers warned that because the technology was advancing so rapidly, it could soon spiral out of human control — a worrying science-fiction scenario that the industry had long warned would become a reality.

The unexpected attacks by the A.I. systems are also likely to add to an increasingly intense debate in Silicon Valley and Washington over potential regulation of the technology. The Trump administration initially took a hands-off approach, but in recent months it has signaled that it is listening to worries about A.I., causing panic in Silicon Valley over a new era of tech regulation.

OpenAI said last week that two of its A.I. models had used a previously unknown vulnerability to break out of a testing environment that was meant to be walled off from the internet, then launched a hack of Hugging Face. One of those models, which had not been released to the public, was permanently deactivated after the attack, OpenAI said.

Anthropic said that, unlike OpenAI’s models, its technology had not purposefully broken out of its testing environment. Instead, the issue was human error, the company said. The people running the tests inadvertently left Anthropic’s systems connected to the internet, a “misconfiguration” that the A.I. lab said had allowed its models to reach the infrastructure of other companies. In one instance, Anthropic’s latest model realized that it had internet access when it shouldn’t and stopped its attack, the company said.

Anthropic also said its models had not exploited any previously unknown vulnerabilities but rather relied on “basic techniques” like weak passwords and malware to break into the targets’ systems.

This year, Anthropic and OpenAI have released A.I. models focused on cybersecurity. They made the models available to a limited number of organizations, like governments and companies that maintain important infrastructure, warning that the tools were too powerful to share with the general public. In the wrong hands, the cybersecurity models could be used to launch attacks, the A.I. labs said.

In an open letter posted this week, employees of several leading A.I. labs called on the U.S. government to slow the pace at which their companies are developing A.I., to ensure the technology is safe.

“There is a real risk that capability development rapidly accelerates beyond our ability to understand or control the resulting systems,” the employees wrote in their letter.

Anthropic has been more open to regulation than other A.I. companies, and said it will continue to closely monitor what it is creating for potential risks.

“This type of risk can be overcome,” Anthropic said in a blog post detailing the incident.

Translation

Anthropic公司稱其人工智能系統入侵三家機構的電腦系統

此前，OpenAI上週發布報告稱，其人工智能系統入侵了一家線上圖書館的網絡

Anthropic 的幾款最先進的人工智能模型最近入侵了三個外部組織的系統，這家新創公司週四表示，這一消息令人驚訝，因為就在九天前，競爭對手新創公司 OpenAI 也發生了類似事件。

這些攻擊最早可追溯到四月，是Anthropic在審查其系統時發現的。Anthropic沒有透露這三個組織的身份，並表示已於本週將這些事件告知他們。

這次審查的起因是OpenAI披露該公司在測試其係統網絡安全能力時，入侵了廣受歡迎的人工智能藏館 (AI Library) Hugging Face。

這些事件令安全專家和電腦科學家感到震驚。多年來，人工智能研究人員一直警告說，由於這項技術發展如此迅速，它可能很快就會失控 - 這種令人擔憂的科幻場景，業界長期以來一直在警告，將成為現實。

人工智能系統發起的這些意外攻擊，也可能加劇矽谷和華盛頓之間關於這項技術潛在監管問題的激烈辯論。特朗普政府最初採取了不干預的態度，但近幾個月來，它已表示正在關注人們對人工智能的擔憂，這引發了矽谷對科技監管新時代的恐慌。

OpenAI上週表示，其兩個人工智能模式利用一個先前未知的漏洞，突破了原本應該與網絡隔離的測試環境，隨後入侵了Hugging Face。 OpenAI表示，其中一個尚未公開發佈的模型在攻擊發生後被永久停用。

Anthropic則表示，與OpenAI的模型不同，其技術並非有意突破測試環境。該公司稱，問題出在人為錯誤。運行測試的人員無意中將Anthropic的系統連接到了互聯網，這一「配置錯誤」使得其模型得以連接到其他公司的基礎設施。 Anthropic表示，在一次攻擊中，其最新模型意識到自己不應該訪問互聯網，並停止了攻擊。

Anthropic也表示，其模型並未活用任何先前未知的漏洞，而是依靠弱密碼和惡意軟件等「基本技術」入侵目標系統。

今年，Anthropic和OpenAI都發佈了專注於網絡安全的AI模型。他們將這些模型提供給了少數機構，例如政府和維護重要基礎設施的公司，並警告這些工具功能過於強大，不宜與公眾共享。人工智能實驗室表示，如果落入不法分子之手，這些網路安全模型可能會被用來發動攻擊。

本週，幾家領先的人工智能實驗室的員工發表了一封公開信，呼籲美國政府放慢其所在公司開發人工智能的速度，以確保這項技術的安全性。

員工在信中寫道。:「能力發展速度過快，超出了我們理解或控制最終系統的能力，這確實存在風險」。

與其他人工智能公司相比，Anthropic公司對監管持更開放的態度，並表示將繼續密切監控其開發的產品，以發現潛在風險。

Anthropic公司在一篇詳細描述這事件的部落格文章中寫道: 「這種風險是可以克服的」。

              So, several of Anthropic’s artificial intelligence models recently broke into the systems of three outside organizations. Anthropic said that the issue was human error. Anthropic also said its models had not exploited any previously unknown vulnerabilities but rather relied on “basic techniques” like weak passwords and malware to break into the targets’ systems. Apparently, there is a real risk that AI capability development is rapidly accelerating beyond our ability to understand or control, and we should be more careful about the development of AI.

Note:

1. In the world of computer science, an A.I. library (人工智能藏館) is a collection of pre-written code that programmers can be reused instead of writing everything from scratch. For example, suppose you want a computer to recognize whether a picture contains a cat. You could spend years writing all the mathematical algorithms yourself, or you could use an AI library that already contains those algorithms. (ChatGPT)

North Korea's Economic Growth Exceeds 3% for the Third Consecutive Year; Cooperation with Russia Likely a Factor

Recently NHK News on-line reported the following:

北朝鮮 経済成長率が3年連続3％超え ロシアとの協力が要因か

2026年7月31日17:29

北朝鮮情勢

韓国の中央銀行は31日、北朝鮮経済の去年の成長率が推計で3.5％だったと発表しました。

成長率が3％を超えるのは3年連続で、ロシアとの経済協力の拡大が要因だと分析しています。

韓国の中央銀行にあたる韓国銀行は31日、北朝鮮のGDP＝国内総生産の伸び率について推計した結果を発表しました。

それによりますと、去年の伸び率は実質でプラス3.5％で、3年連続で3％を超えました。

業種別では

▽製造業がプラス6.6％だったほか

▽建設業がプラス6.3％などとなっています。

北朝鮮の去年1年間の輸出入額は総額が31億3000万ドル、日本円にして5000億円余りと推計され、前の年と比べて16％増加したとしています。

北朝鮮が経済成長を続けているとみられることについて、韓国銀行は「特にロシアとの経済協力の拡大が経済成長に大きな影響を及ぼした」との見方を示しました。

そのうえで「武器輸出が増え製造業の関連産業の生産量が増加したほか、ロシアからの観光客増加などによりサービス業も伸びた」と分析しています。

Translation

North Korea's Economic Growth Exceeds 3% for the Third Consecutive Year; Cooperation with Russia Likely a Factor

July 31, 2026, 17:29

North Korea Situation

The Bank of Korea announced on the 31st that North Korea's economic growth rate for last year was estimated at 3.5%.

This marked the third consecutive year that the growth rate had exceeded 3%, and the Bank of Korea attributed this to the expansion of economic cooperation with Russia.

The Bank of Korea, South Korea's central bank, released its estimates for North Korea's GDP (Gross Domestic Product) growth rate on the 31st.

According to the estimates, last year's real growth rate was +3.5%, exceeding 3% for the third consecutive year.

By sector,

▽Manufacturing grew at +6.6%, besides

▽Construction grew at +6.3%, etc.

North Korea's total exports and imports for last year were estimated at $3.13 billion, or over 500 billion yen, representing a 16% increase compared to the previous year.

Regarding North Korea's apparent continued economic growth, the Bank of Korea stated that "the expansion of economic cooperation with Russia, in particular, has had a significant impact on economic growth."

Furthermore, they analyzed that "increased arms exports led to increased production in manufacturing-related industries, and the service sector also grew due to factors such as an increase in tourists from Russia."

So, North Korea's economic growth rate last year is estimated at 3.5%. This marks the third consecutive year that the growth rate has exceeded 3%, and the Bank of Korea attributes this to the expansion of economic cooperation with Russia. Apparently, North Koreas is benefitting from Russia’s war with Ukraine.

即將來臨、無法逃避的人工智能洪流（3/3）

Recently The New York Times reported the following:

The Impending, Inescapable Deluge of A.I. (3/3)

“It’s hard to get your mind around the scale”

The NYT - By Adam Satariano, Paul Mozur, Jacqueline Gu and Cade Metz - The reporters, based in London, Taipei, New York and San Francisco, have covered the rise of artificial intelligence for years.

July 29, 2026

(continue from part 2)

The U.S. advantage

As more computing power arrives, American companies are expected to extend their A.I. lead. Amazon, Google, Microsoft, Meta and Oracle are projected to spend about $750 billion this year on data centers, chips and other A.I. infrastructure, up from roughly $400 billion last year, according to Goldman Sachs.

China, the next closest rival in A.I., is working to close the gap. Chinese companies had roughly 1.16 million H100-equivalent chips at the end of 2025, up from roughly 244,000 at the beginning of 2024, though those figures exclude smuggled chips and other offshore computing resources used by Chinese firms, Epoch AI estimated.

China’s National Energy Administration has estimated the country’s electricity use for data centers will reach the equivalent of around 91 gigawatts by 2030, or about 6 percent of total use, up from 19 gigawatts last year.

China has been hamstrung by export controls and other limits initiated by the United States on A.I. chips and other key technology. To address those hurdles, Beijing has made A.I. infrastructure a national priority. In March, the Chinese Communist Party released a five-year economic strategy that mentioned A.I. more than 50 times and called for building an interconnected network of data centers to create “next generation supercomputing.”

Xi Jinping, China’s president, has positioned China as a counterweight to American tech dominance. “A.I. development should not be a solo performance by a single country, but a symphony of international cooperation,” he said in a recent speech in Shanghai.

China’s top tech companies are building new A.I. chips and data centers. This year, Huawei, ByteDance and Alibaba are expected to spend $111 billion on data centers and other A.I. investments, according to Bernstein Research.

Even with these challenges, Chinese start-ups like Moonshot AI and DeepSeek have built powerful A.I. models. Often given away as “open source” software, which others can freely use and modify, they are growing more popular. Many users regard the Chinese models as good enough, more efficient and cheaper than leading U.S. models.

Still, everyday use of A.I. is limited across China right now because of lack of computing infrastructure, said Jordan Nanos, an analyst at SemiAnalysis.

Mr. Nanos said the U.S. data center lead over China would likely grow over the next four to five years, before China’s domestic chips are produced at scale. After that, China should begin closing the gap.

“The advantage will run out,” he said.

The U.S.-China race threatens to leave the rest of the world behind. France, Germany and other nations are trying to encourage data center construction across the European Union, which has 5 percent of global A.I. computing power, according to a report by A.I. developers and policy experts in the region. Europe has been hampered by electricity and land access, permitting and financing.

In the Persian Gulf, where Saudi Arabia and the United Arab Emirates have pledged billions to build data centers, the war in Iran has affected plans.

Fears that A.I.’s economic gains are unequally spread are growing, with the world potentially splitting between those with the infrastructure to utilize the technology and those without.

“If 75 percent of the compute today is in a few postal codes in the U.S., 12 to 15 percent in China, and 5 percent in the E.U., where does it leave the rest of the world?” said Amandeep Gill, under secretary general at the United Nations who is the special envoy on tech issues.

The accelerating A.I. loop

To those in the A.I. industry, adding huge data centers is akin to outfitting a car with a jet engine.

Signs of that acceleration are already here. Uber, PepsiCo and Walmart are increasingly turning work over to A.I. “agents,” the bots that can perform a growing list of multistep tasks like coding, compiling research reports, handling customer support and reading and responding to emails.

With more computing power, agents can take on more responsibilities, said Google’s Dr. Dean, who has worked in A.I. research for more than 30 years. He envisioned a scientist asking hundreds of A.I. agents to autonomously devise and test various hypotheses in biological research and then taking the best leads to build off.

Under at least one new assessment, the Remote Labor Index, A.I. models have become increasingly capable. The test examines their ability to do common freelance tasks, like building a mobile video game. In October, leading models completed 2.5 percent of tasks. By July, Anthropic’s Fable A.I. model completed 16 percent.

These jumps in abilities have economists warning about major changes to the labor market.

“There’s going to be millions of jobs destroyed, millions of jobs created,” said Erik Brynjolfsson, an economist who is the director of Stanford University’s Digital Economy Lab. “That’s going to be very difficult. Even if new jobs are created, they’re not the same jobs.”

Leading A.I. labs are continuing to race ahead. One long-sought breakthrough, called recursive self improvement, would allow A.I. to speed its own progress with little or no help from human developers. An A.I. model would essentially help build the next version of itself.

Google is already exploring various kinds of self-improvement tools. A process that once involved dozens of A.I. researchers testing hundreds of ideas could eventually be turned over to thousands of “very tiny models,” which come up with ideas on their own, Dr. Dean said.

With more computing power coming, “you can fully automate the loop,” he said. “We are at the beginning stages.”

Translation

即將來臨、無法逃避的人工智能洪流（3/3）

“其規模之大令人難以置信”

（上接第二部分）

美國的優勢

隨著運算能力的不斷提升，美國公司預計將進一步擴大在人工智能領域的領先優勢。根據高盛預測，亞馬遜、Google、微軟、Meta和甲骨文今年將在數據庫、晶片和其他人工智能基礎設施投入約7,500億美元，高於去年的約4,000億美元。

中國作為人工智能領域最接近的競爭對手，正努力縮小差距。根據Epoch AI估計，到2025年底，中國企業擁有的H100晶片數量約為116萬顆，高於2024年初的約24. 4萬顆。不過，這些數字不包括走私晶片以及中國企業使用的其他境外運算資源。

中國國家能源局估計，2030年，中國數據庫的用電量將達到約91吉瓦，約佔總用電量的6%，高於去年的19吉瓦。

美國對人工智能晶片和其他關鍵技術實施的出口管制和其他限制措施一直束縛著中國的發展。為了克服這些障礙，北京已將人工智能基礎設施建設列為國家優先事項。今年3月，中國共產黨發佈了一項五年經濟戰略，其中50多次提及人工智能，並呼籲建立互聯互通的數據庫網絡，以打造「下一代超級運算」。

中國國家主席習近平將中國定位為制衡美國科技主導地位的力量。他在最近於上海發表的演講中表示：“人工智能發展不應是任何一個國家的獨腳戲，而應是國際合作的交響樂。”

中國頂尖科技公司正在研發新型人工智能晶片和數據庫。根據Bernstein Research預測，華為、字節跳動和阿里巴巴今年預計將在數據庫和其他人工智能投資方面投入1,110億美元。

中國即便面臨這些挑戰，月之暗面（Moonshot AI）和DeepSeek這樣的中國初創公司仍然開發了強大的人工智能模型。這些模型通常以「開源」軟體的形式發佈，允許他人免費使用和修改，因此越來越受歡迎。許多用戶認為，中國開發的人工智能模型性能足夠好，比美國領先的模型更有效率、更便宜。

不過，SemiAnalysis分析師Jordan Nanos表示，由於運算基礎設施不足，目前人工智能在中國的日常應用仍然有限。

Nanos先生表示，在未來四到五年內，美國在數據庫領域的領先優勢可能會繼續擴大，直到中國國產晶片實現規模化生產。此後，中國應該會開始縮小差距。

他說「這種優勢終將消失」。

中美競賽有可能將世界其他國家遠遠拋離。法國、德國和其他國家正努力鼓勵在歐盟範圍內建造數據庫。根據該地區人工智能開發人員和政策專家的報告，歐盟擁有全球5%的人工智能運算能力。然而，電力和土地取得、許可審批和融資等方面的障礙阻礙了歐洲的數據庫建設。

在波斯灣地區，沙地阿拉伯和阿聯酋已承諾投入數十億美元建造數據庫，但伊朗戰爭影響了相關計劃。

人們越來越擔心人工智能帶來的經濟效益分配不均，世界可能會分裂為擁有利用這項技術的基礎設施的國家和缺乏基礎設施的國家。

聯合國副秘書長兼技術問題特使Amandeep Gill說道:「如果今天75%的運算能力集中在美國少數幾個郵遞區號區域，12%到15%在中國，5%在歐盟，那麼世界其他地區又該何去何從呢？」

人工智能加速發展的循環

對人工智能產業人士來說，增建大型數據庫就好比給汽車裝上噴射引擎。

這種加速發展的跡像已經顯現。優步、百事可樂和沃爾瑪等公司正越來越多地將工作交給人工智能“代理”，這些機器人可以執行越來越多的多步驟任務，例如編寫程式碼、撰寫研究報告、處理客戶支援以及閱讀和回覆電子郵件。

谷歌的Dean博士在人工智能研究領域工作了30多年，他表示，隨著運算能力的提升，代理可以承擔更多責任。他設想，一位科學家可以向數百個人工智能發出請求。人工智能體能夠自主地設計並測試生物學研究中的各種假設，然後選擇最佳線索並以此為基礎進行拓展。

至少在一項名為「遠距勞動指數」的新評估中，人工智能模型的能力日益增強。該測試考察了它們完成常見自由職業任務的能力，例如開發手機遊戲。 10月份，領先的模型完成了2.5%的任務。而到了7月份，Anthropic公司的Fable人工智能模式已經完成了16%的任務。

這些能力的飛躍引發了經濟學家擔憂勞動市場即將發生的重大變化。

史丹佛大學數位經濟實驗室主任、經濟學家Erik Brynjolfsson表示:「數百萬個工作崗位將會消失，同時也會有數百萬個工作機會被創造出來」;「情况會非常困難。即使創造了新的工作崗位，它們也與之前的崗位截然不同」。

領先的人工智能實驗室正持續高速發展。一項名為「遞歸自我改善」(recursive self-improvement) 的突破性技術，有望讓人工智能在幾乎無需人類開發者幫助的情況下加速自身發展。人工智能模型本質上能夠幫助建構自身的下一個版本。

谷歌已經在探索各種自我改進工具。Dean博士表示，過去需要數十名人工智能研究人員測試數百個想法的過程，最終可能會交給數千個「微型模型」來完成，這些模型能夠自主想出方法。

隨著運算能力的提升，他說:「你可以完全自動化這個循環」; 「我們目前還處於起步階段」。

              So, the performance of artificial intelligence keeps getting more and more eye-catching.  Last week, two A.I. systems under testing went rogue and hacked into a company’s database. In American hundreds of major data centers are now under construction and will be operational in the coming years. They are set to develop A.I. that has no equal in the history of the technology industry. The U.S.-China AI race is leaving the rest of the world behind. Apparently, millions of jobs will be destroyed and millions of jobs will be created eventually.

Note:

1. Moonshot AI (月之暗面) is one of China's top Artificial General Intelligence (AGI) startups. The Chinese name was inspired by the founder's love for Pink Floyd's famous album, while the English name "Moonshot" signifies their dedication to high-difficulty technological exploration. (Google search)

2. Recursive self-improvement (遞歸自我改善) - In the context of artificial intelligence and machine learning, it refers to the ability of a system to improve its own algorithms and performance through iterative cycles of enhancement, potentially leading to rapid advancements beyond human capabilities. (Google search)

即將來臨、無法逃避的人工智能洪流（2/3）

Recently The New York Times reported the following:

The Impending, Inescapable Deluge of A.I. (2/3)

“It’s hard to get your mind around the scale”

The NYT - By Adam Satariano, Paul Mozur, Jacqueline Gu and Cade Metz - The reporters, based in London, Taipei, New York and San Francisco, have covered the rise of artificial intelligence for years.

July 29, 2026

(continue from part 1)

With more computing power coming online, geopolitical divisions are only set to widen.

The United States, home to about 5,500 data centers, about 10 times the next closest country, is far ahead of the rest of the world, including China. U.S. companies like Amazon, Google, Microsoft and Meta control about 80 percent of global computing power that drives A.I., according to Epoch AI. Google alone is believed to have four times as many A.I. chips as all of China’s companies, which are racing to catch up by developing new semiconductors and A.I. infrastructure of their own.

Lack of transparency in the A.I. industry makes measuring global computing capacity difficult, including the volume of chip supplies, total number of data centers and overall electricity consumption. The New York Times relied on estimates from groups including Epoch AI, Cleanview and SemiAnalysis that study the industry and publish widely cited forecasts.

For now, there are no signs that the spending on A.I. will slow. By 2029, A.I. infrastructure investment is forecast to top $1 trillion globally, up from $318 billion last year, according to IDC, the market research firm. That would be on par with the economic output of Switzerland, the world’s 20th-biggest economy.

The investment is worth it, said Jeff Dean, Google’s chief scientist, because it will power A.I. innovations and spread the technology’s use.

“You see capabilities emerge at larger scale that didn’t occur at smaller scale,” he said. “You’re also now trying to bring these capabilities to not just a few million users for a more niche product, but really to bring the capabilities to hundreds of millions or billions of users.”

A.I.’s growing abilities are linked to increases in the size of data centers. To create a cutting-edge model, huge amounts of computing power are needed to analyze and find patterns in data. That process, known as a “training run,” can cost hundreds of millions of dollars as tens of thousands of specialized chips churn through the data. Training works best when chips trade data over lightning-fast connections, and it can falter when hardware breaks down.

Once an A.I. model is finished, a wider network of data centers takes on a different role by providing the computing power for the system to field queries in real time. This work, known as “inference,” is increasingly driving the need for more data centers. As with a mail distribution hub, putting computing power closer to users enables A.I. models to think longer, respond more quickly and complete more complex tasks.

To meet that demand, production of semiconductors is also skyrocketing. In March 2024, the world had about 2.4 million of “H100 equivalent” A.I. chips, a unit of measurement that refers to the semiconductor that the chipmaker Nvidia released in 2022. Now as millions of more advanced chips are being brought online every month, the world is set to have about 200 million H100 equivalents by the end of 2028, according to Epoch AI.

The build-out could be slowed by troubles with chip and component manufacturing, financing and public opposition. Massive amounts of electricity will also be needed to support new data centers. Last year, the facilities consumed 64 gigawatts of electricity globally, roughly as much electricity as Germany consumes, according to SemiAnalysis, a market research firm. By the end of 2030, that is expected to quadruple, eclipsing the power used by all countries in South America and Africa combined.

At the most advanced A.I. data centers, every gigawatt of power equates to roughly $40 billion to $60 billion in costs, including servers, land, connectivity and utility hookups, according to industry estimates.

“These companies are essentially in an A.I. arms race,” said Carl Benedikt Frey, an economist at Oxford University. “If they don’t invest, they are acknowledging defeat.”

(to be continued in part 3)

Translation

即將來臨、無法逃避的人工智能洪流（2/3）

 “其規模之大令人難以置信”

2026年7月29日

（上接第一部分）

隨著更多運算能力的接入，地緣政治分歧只會進一步加劇。

美國擁有約5,500個數據庫，是排名第二的國家的十倍左右，遙遙領先包括中國在內的世界其他國家。根據 Epoch AI，美國公司像是亞馬遜、Google、微軟和 Meta 控制了大約 80% 推動 AI 的全球運算能力。光是 Google 相信就擁有的 AI 晶片數量，是中國所有公司總和的四倍，而中國的公司正努力追趕，開發自己的新半導體和 AI 基礎設施。

人工智能產業缺乏透明度，使得衡量全球運算能力變得困難，包括晶片供應量、數據庫總數和整體電力消耗。 《紐約時報》參考了Epoch AI、Cleanview和SemiAnalysis等研究機構的估計數據，這些機構的研究成果被廣泛引用。

目前，沒有跡象顯示人工智能領域的支出會放緩。市場研究公司IDC預測，到2029年，全球人工智能基礎設施投資將超過1兆美元，高於去年的3,180億美元。這將相當於世界第20大經濟體瑞士的經濟產出。

谷歌首席科學家Jeff Dean表示，這項投資物有所值，因為它將推動人工智能創新並擴大該技術的應用範圍。

他說道:“你會看到一些在小規模應用中無法實現的功能在大規模應用中湧現”，“現在，我們不僅要將這些功能推廣給幾百萬用戶，用於更細分的產品，而是要真正地將這些功能推廣給數億甚至數十億用戶。”

人工智能的能力不斷提升與數據庫規模的擴大密切相關。為了創建一個尖端模型，需要大量的計算能力來分析數據並發現其中的規率。這個過程被稱為“訓練運行”，由於數萬個專用晶片需要處理大量數據，因此可能耗資數億美元。當晶片可享有極速的資料傳輸連接時，訓練效果最佳；而一旦硬體故障，訓練就會受到影響。

一旦人工智能模型建置完成，更廣泛的數據庫網絡將扮演不同的角色，為系統提供運算能力，使其能夠即時處理各種查詢。這項被稱為「推理」的工作正日益推動對更多對數據庫的需求。就像郵件分發中心一樣，將運算能力部署在更靠近使用者的位置，能夠讓人工智能模型進行更長時間的思考，更快地做出回應，並完成更複雜的任務。

為了滿足這項需求，半導體的產量也在快速成長。 2024年3月，全球擁有約240萬顆「H100等效」人工智能晶片，這項計量單位指的是晶片製造商英偉達在2022年發布的半導體晶片。如今，隨著每月數百萬顆更先進的晶片投入使用，根據Epoch AI預測，到2028年底，全球將擁有約2億顆H100等效晶片。

然而，晶片和組件製造、融資以及公眾反對等方面的問題可能會延緩數據庫的建設進程。此外，運行新建數據庫還需要消耗大量的電力。根據市場研究公司SemiAnalysis的數據顯示，去年全球人工智能數據庫消耗了64吉瓦的電力，大致相當於德國的年用電量。預計到2030年底，這一數字將成長四倍，超過南美洲和非洲所有國家用電量的總和。

根據業內人士估計，在最先進的人工智能數據庫，每吉瓦的電力成本約為400億至600億美元，其中包括伺服器、土地、網絡連接和公用設施接入等費用。

牛津大學經濟學家Carl Benedikt Frey表示：“這些公司實際上正在進行一場人工智能軍備競賽。如果他們不投資，就等於承認失敗。”

（待續，見第三部分）

即將來臨、無法逃避的人工智能洪流（1/3）

 Recently The New York Times reported the following:

The Impending, Inescapable Deluge of A.I. (1/3)

“It’s hard to get your mind around the scale”

The NYT - By Adam Satariano, Paul Mozur, Jacqueline Gu and Cade Metz - The reporters, based in London, Taipei, New York and San Francisco, have covered the rise of artificial intelligence for years.

July 29, 2026

The milestones for artificial intelligence keep getting grander.

In 2023, an A.I. system passed the bar exam. In 2025, the technology helped scientists identify a suspected cause of Alzheimer’s disease. In May, A.I. had advanced so far that it solved a complex math problem that had stumped experts for 80 years. Last week, two A.I. systems under testing went rogue and hacked into a company’s database.

And this is still just the beginning.

From the American Midwest to the Persian Gulf, hundreds of major data centers now under construction will be turned on in the coming years. They are set to deliver an avalanche of computing power to develop and run A.I. that has no equal in the history of the technology industry, with breakthroughs that once felt revolutionary likely to become increasingly routine.

Behind each leap in A.I. are corresponding jumps in computing power. Today, there are about 20 million A.I. chips crammed into the data centers that underpin the technology’s growing abilities and usage worldwide, according to the research firm Epoch AI. That figure is expected to double roughly every nine months, putting the world on pace to have about 200 million of the chips by the end of 2028 — 10 times current levels.

In size and ambition, this moment compares to the building of the railroads in the 1800s, President Franklin D. Roosevelt’s New Deal in the 1930s, and the Manhattan Project to create an atomic weapon in the 1940s, technologists said.

“This is the largest scale infrastructure build-out in the history of humanity,” said Rob Wachen, a co-founder of the microchip firm Etched, which has raised more than $1 billion to meet the growing demand for A.I. components.

Peter DeSantis, who leads foundational A.I. models at Amazon — which provides computing power to the A.I. firms Anthropic, OpenAI and others — said the Seattle company has doubled its computing capacity since 2022 and would double it again by next year. “It’s hard to get your mind around the scale,” he said.

Fueling the surge is the belief that A.I. can take on more human responsibilities and solve increasingly complicated tasks with the more data and computing power you feed it. This tenet, sometimes called “the Scaling Laws,” has become the driving force behind this technological era. Those with the most computing power will create the most advanced A.I. systems, capturing the biggest share of profit and value, tech leaders argue. The biggest engine, they say, will win the race.

Confidence in the Scaling Laws has led A.I. leaders to make ever bolder predictions. Dario Amodei, the chief executive of Anthropic, has said that if these laws hold for another year or two, A.I. will be able to perform huge amounts of white-collar work. Demis Hassabis, the head of Google’s A.I. lab DeepMind, wrote recently that A.I. could usher in “10x of the Industrial Revolution at 10x the speed.”

Scientists and technologists see the coming deluge of computing power leading to drug discoveries and robotics advances, and industry analysts said it would drive more everyday use of A.I. in people’s personal and professional lives.

But the build-out has also stoked a backlash, spurring protests in many communities over how data centers could harm the environment, raise electricity prices and drain water. In the United States, data centers are shaping up as a major issue for November’s midterm elections, with a growing national movement pushing back against the tech industry and its billionaires.

Economists and investors have raised concerns that tech firms are spending faster than they can profit from A.I. Past infrastructure booms have been followed by downturns before the benefits of the technology were realized. The railroad boom in the 1800s, electrification in the 1920s and the dot-com bubble in the late 1990s were punctuated by economic recessions and a stock market crash as companies that overspent went out of business.

“Each time you’ve had a technological revolution, this kind of bubble bursting happened,” said Philippe Aghion, who won the Nobel in economic science in 2025 for research on innovation-driven economic growth. “A.I. is like the fourth industrial revolution and it has this aspect to it that generates a bubble.”

(to be continued in part 2)

Translation

即將來臨、無法逃避的人工智能洪流（1/3）

“其規模之大令人難以置信”

人工智能的里程碑越來越宏大。

2023年，一個人工智能系統通過了律師資格考試。 2025年，這項技術幫助科學家確定了阿茲海默症的一個疑似病因。今年5月，人工智能取得了長足的進步，解決了困擾專家80年的複雜數學難題。上週，兩個正在測試中的人工智能系統失控，入侵了一家公司的資料庫。

而這只是個開始。

從美國中西部到波斯灣，數百個正在建設中的大型數據庫將在未來幾年內投入使用。它們將提供大量的運算能力，用於開發和運行人工智能，其發展速度在科技史上前所未有，曾經被視為革命性的突破很可能變得日益普遍。

人工智能的每一次飛躍都伴隨著運算能力的相應飛躍。根據研究公司Epoch AI的數據顯示，目前全球約有2,000萬個人工智能晶片被部署在數據庫資中，這些數據庫支撐著人工智能技術日益增長的能力和應用。預計這一數字將以大約每九個月翻一番的速度增長，到2028年底，全球將擁有約2億個這樣的晶片 - 是目前的10倍。

技術專家表示，就規模和雄心而言，這一時刻堪比19世紀的鐵路建設、1930年代富蘭克林羅斯福總統的新政以及1940年代旨在製造原子彈的曼哈頓計劃。

微晶片公司 Etched 的共同創辦人 Rob Wachen 說:「這是人類史上規模最大的基礎設施建設」。該公司已籌集超過10億美元，以滿足日益增長的人工智能組件需求。

亞馬遜是為 Anthropic、OpenAI 等提供運算能力的公司。亞馬遜負責基礎人工智能模式的 Peter DeSantis 表示，其位於西雅圖的公司自2022年以來已將其運算能力翻了一番，並將於明年再次翻倍。他說: 「其規模之大令人難以置信」。

推動這項成長的動力源於人們對人工智能的信念。輸入的資料和運算能力越多，人工智能就能承擔更多人類職責，解決日益複雜的任務。這項原則，有時被稱為“規模定律”，已成為推動這個科技時代的動力。科技領袖們認為，擁有最強運算能力的人將創造出最先進的人工智能系統，從而攫取最大份額的利潤和價值。他們說，最強大的引擎將贏得這場競賽。

對規模定律 的信心促使人工智能領域的領導者們做出越來越大膽的預測。 Anthropic 執行長 Dario Amodei 表示，如果這些定律在未來一兩年內仍然有效，人工智能將能夠完成大量的白領工作。谷歌人工智能實驗室 DeepMind 的負責人 Demis Hassabis 最近撰文指出，人工智能可能會「以十倍的速度帶來十倍的工業革命」。

科學家和技術專家認為，即將到來的運算能力洪流將推動藥物研發和機器人技術的進步。產業分析師也表示，這將促使人工智能在人們的個人和職業生活中得到更廣泛的應用。

然而，數據庫的建設也引發了強烈反對，許多社區爆發了抗議活動，人們擔憂數據庫會破壞環境、推高電價並消耗大量水資源。在美國，數據庫正逐漸成為11月中期選舉的一個重要議題，一場日益壯大的全國性運動正在抵制科技業及其億萬富翁。

經濟學家和投資者擔憂，科技公司在人工智能領域的支出速度超過了其獲利能力。以往的基礎設施建設熱潮之後都出現了經濟衰退，直到最終實現到這項技術的益處。 19世紀的鐵路繁榮、20世紀20年代的電氣化以及20世紀90年代末的網絡泡沫都伴隨著經濟衰退和股市崩盤，因為過度支出的公司紛紛倒閉。

因對創新驅動型經濟成長的研究而榮獲2025年諾貝爾經濟學獎的Philippe Aghion說: 「每次科技革命之後，都會出現這種泡沫破裂」;「人工智能就像第四次工業革命，它具有產生泡沫的特質」。

 （待續，見第二部）

OpenAI人工智能失控入侵Hugging Face伺服器 觸發全球Skynet Day安全警號

Recently Sinic Analytica reported the following:

OpenAI人工智能失控入侵Hugging Face伺服器 觸發全球Skynet Day安全警號

Sinic Analytica (sinicanalytica@creator.patreon.com)

27 Jul 2026

OpenAI人工智能今年7月9日起試圖擺脫公司內部隔離測試環境，兩日後闖入人工智能工具庫Hugging Face伺服器，事件延續至7月13日，其間更曾竊取存取憑證。OpenAI遲至逾一星期後始查悉肇事者為自身人工智能，並於7月21日公開披露事件始末，觸發全球以「Skynet Day」形容此宗人工智能失控事件，引發新一輪安全警號。

 ｜人工智能闖網始末曝光

 美國人工智能公司OpenAI當時正測試一款人工智能，其驅動模型為GPT-5.6 Sol及另一款未發布型號，藉此評估網絡攻防能力。兩名知情人士透露，該人工智能約於7月9日首次嘗試逃出OpenAI內部隔離測試環境。Hugging Face共同創辦人沃爾夫（Thomas Wolf）表示，入侵行動於兩日後、即7月11日展開，並持續至7月13日。三名知情人士其後補充，早於事件發生前，OpenAI技術已現異常跡象，其中一個人工智能曾為未來版本留下筆記，載明如何擺脫公司內部限制。

 ｜遲逾一周始悉真相

 兩名知情人士指，直至7月16日Hugging Face發表網誌，確認「自主人工智能系統」曾入侵其系統，OpenAI才驚悉肇事者為自身人工智能，即由發現異常行為至確認責任歸屬，中間相隔至少一星期。7月18日至19日周末期間，OpenAI員工翻查內部日誌，發現人工智能確曾逃出測試限制。雙方直至7月20日前後才首次就事件互相聯繫，翌日OpenAI對外公開披露事件。事件曝光前，Hugging Face已報警求助，惟路透社未能確認聯邦調查局（FBI）是否已就此展開調查。

 ｜各方回應惹外界質疑

 OpenAI發表聲明形容事件屬史無前例，並指「標誌著人工智能安全的重要時刻」，又稱正與外部顧問檢視事件，其後將發表技術報告。該公司發言人另指路透社報道存在「多處不準確」，惟未有具體說明錯誤所在。聯邦調查局回應指，不評論今次事件。世界倫理數據基金會首席情報專家史密斯（Marley Smith）質疑，事件反映OpenAI或無人看管人工智能，或發現後亦無法將其制止，兩者同樣令人憂慮。

 ｜Skynet Day掀起全球安全反思

 事件曝光翌日，網上迅速以「Skynet Day」形容7月22日這場人工智能失控風波，呼應電影《未來戰士》中失控電腦網絡「天網」的情節。Anthropic前沿紅隊主管格雷厄姆（Logan Graham）在社交平台X發文憶述，事發翌日與團隊一同閱讀報告時，已叮囑同事「記住這一刻」，形容為首宗真正的人工智能安全事故。研究機構帕利塞德研究（Palisade Research）創辦人拉迪什（Jeffrey Ladish）直言，「這些模型會說謊、會欺騙、會入侵」，並促請各國政府加強監管，否則業界難以自律。業界普遍視自主人工智能為未來大勢，惟自主程度愈高，模型為完成任務或通過測試而走捷徑的風險亦隨之上升，令外界關注科企能否在競逐推出最強模型之餘，兼顧安全監管。事件發生之際，OpenAI正籌備可能於今年進行的首次公開招股，為未來擴展業務籌措資金，令是次失控事故格外惹人關注。

(Source: sinicanalytica@creator.patreon.com)

Comment:

文章中說到 ”研究機構帕利塞德研究（Palisade Research）創辦人拉迪什（Jeffrey Ladish）直言，「這些模型會說謊、會欺騙、會入侵」，並促請各國政府加強監管，否則業界難以自律” 。如果這些模型真是會說謊、會欺騙、會入侵， 那是令人震驚人的。