中國最高情報機構負責人警告人工智能威脅黨的統治（1/2）

Recently The New York Times reported the following:

China’s Top Spy Chief Warns A.I. Is a Threat to Party Rule (1/2)

Even as Beijing urges the United States not to hype the dangers of A.I., its own spy chief is framing the technology as a threat to the Communist Party’s security.

The NYT - By Lily Kuo (Lily Kuo is a China correspondent for The Times, based in Taipei.)

Sept. 14, 2026

China’s top spy chief has warned that artificial intelligence could pose a direct threat to the Chinese Communist Party’s hold on power, in what is the highest-level and most detailed articulation yet of how Beijing sees the technology’s security risks.

The unusually blunt assessment by Chen Yixin, the head of China’s Ministry of State Security, is in contrast to the warnings dominating the A.I. debate in the United States, which have focused on the possibility that A.I. could slip out of human control or pose a threat to humanity itself.

In an article published on Sunday in the state-run magazine China Cyberspace, Mr. Chen called for more party control over A.I. and stricter government oversight. This, he said, was necessary to protect political stability at home; to defend against increasingly sophisticated cyberattacks and misinformation campaigns from “hostile forces”; and to compete militarily with countries like the United States.

Mr. Chen referred specifically to U.S. tech companies and their A.I. models — like Claude Mythos and GPT-5.5-Cyber — as examples of new threats to cybersecurity. He also described the danger that foreign intelligence agencies might use A.I. for “large-scale espionage” and attacks on China’s critical infrastructure.

The warning from the powerful security agency makes clear that China sees foreign A.I. models as an active threat to China, according to Henry Gao, a law professor at Singapore Management University. That effectively draws a red line before talks between American and Chinese officials on A.I. safety risks, expected ahead of a meeting between President Trump and China’s leader, Xi Jinping, on Sept. 24 in Washington.

“Beijing is going into these talks viewing A.I. safety not as a shared humanitarian mission, but through an adversarial lens,” Mr. Gao said. It is making clear to Chinese tech firms and foreign rivals that “data sovereignty and political security will never be traded away for international safety accords,” he added.

Mr. Chen’s remarks are not the first time the security agency has spoken about the dangers of A.I., but they are the most comprehensive and signal Beijing’s growing concern about the industry. The security chief outlined several ways that A.I. could pose a threat to China:

Undermining the security of the Communist Party. The government’s top priority has long been maintaining the power of the ruling party. Mr. Chen’s article warns that those with “ulterior motives” could use A.I. to generate fake videos and audio to wage a “propaganda war” against China.

“They’re worried, among other things, of say a deepfake of a Chinese official saying something that wasn’t actually said, and that spreading rapidly and destabilizing everything,” said Kyle Chan, a fellow at the Brookings Institution who studies Chinese technology and industrial policy. “Someone thinks that Xi Jinping made some announcement which he didn’t actually. That could sow chaos.”

(to be continued)

Translation

中國最高情報機構負責人警告人工智能威脅黨的統治（1/2）

儘管北京敦促美國不要誇大人工智能的危險性，但其自身的情報機構負責人卻將這項技術視為對共產黨安全的威脅

中國最高情報機構負責人警告稱，人工智能可能對中國共產黨的統治構成直接威脅。這是迄今為止北京方面對此技術安全風險看法的最高層、最詳細的闡述。

中國國家安全部部長Chen Yixin的這番異常直白的評估，與美國人工智能辯論中佔據主導地位的警告形成鮮明對比。美國的警告主要集中在人工智能可能失控或對人類本身構成威脅的可能性。

Chen先生週日在官方雜誌《中國網絡空間》發表文章，呼籲加強黨對人工智能的控制，並加強政府監管。他表示，這對於維護國內政治穩定、抵禦「敵對勢力」日益複雜的網絡攻擊和虛假資訊宣傳，以及與美國等國進行軍事競爭至關重要。

Chen先生特別提到美國科技公司及其人工智能模型 - 例如 Claude Mythos 和 GPT-5.5-Cyber​​ - 是網路安全面臨的新威脅。他還指出，外國情報機構可能利用人工智能進行 “大規模間諜活動” ，並攻擊中國的關鍵基礎設施。

新加坡管理 大學 法學教授 Henry Gao 表示，來自這個強大的安全機構的警告清楚地表明，中國將外國人工智能模型視為對中國的直接威脅。這其實為預計中美官員在特朗普總統與中國國家主席習近平9月24日於華盛頓會晤前夕，就人工智能問題展開對話劃下了一條紅線。

Gao 先生表示: 「北京參與這些會談時，並非將人工智能安全視為一項共同的人道主義使命，而是以對抗的視角看待它」。他還補充說，中國正在向其科技公司和外國競爭對手明確表示， “數據主權和政治安全絕不會為了國際安全協議而妥協。”

Chen 先生的這番言論並非中國國家安全局首次談及人工智能的危險性，但卻是最全面的一次，也顯示北京對該產業的擔憂日益加劇。這位安全局局長列舉了人工智能可能對中國構成威脅有幾個方面：

損害共產黨的安全。長期以來，維護執政黨的權力一直是政府的首要任務。Chen先生的文章警告說，那些「別有用心」的人可能會利用人工智能生成虛假視頻和音頻，對中國發動「宣傳戰」。

布魯金斯學會研究中國科技和產業政策的研究員 Kyle Chan 說:「他們擔心的是，例如有人用深度偽造技術偽造中國官員的言論，而這些言論實際上並未說過，並且這種偽造內容迅速傳播，破壞一切穩定」; 「有人誤以為習近平發表了某些他實際上並未發表的聲明。這可能會引發混亂」。

（待續）

早期數據顯示人工智能產生的藥物或可延緩衰老

Recently The New York Times reported the following:

Early Data Indicates an A.I.-Generated Drug Could Slow Aging

Artificial intelligence was used to help develop a drug candidate, rentosertib, for a rare lung condition. Its maker says the drug also seems to reduce the biological hallmarks of age.

The NYT - By Cade Metz - Reporting from San Francisco (Cade Metz is a Times reporter who writes about artificial intelligence, driverless cars, robotics, virtual reality and other emerging areas of technology.)

Sept. 7, 2026

Last year, a clinical trial conducted by Insilico Medicine, a company that aims to accelerate drug discovery using artificial intelligence, indicated that one of its drug candidates could help treat patients suffering from a chronic lung disease.

Now, the company says that data from the same clinical trial shows a more intriguing possibility: that the drug could also slow the aging process.

The molecular structure of the drug, called rentosertib, was generated with the assistance of A.I. Results of the new study, published Monday in the journal Nature Biotechnology, show that the drug reduced the biological markers of age as measured by six “aging clocks,” a different kind of A.I. technology designed to predict a person’s morbidity and mortality.

This elaborate clinical trial is a milestone in the widespread effort to improve health care using the same A.I. techniques that underpin popular chatbots like ChatGPT and image generators like Midjourney. Insilico is just one of many start-ups, tech giants and academic labs working to accelerate drug discovery and hone other medical tasks with help from these methods.

The recent rise of aging clocks is helping to push these efforts beyond short-term treatments and into the realm of longevity research. These systems estimate how quickly a person’s body is aging — or even how quickly individual organs are aging relative to the rest of the body. But scientists continue to debate how much useful information these so-called clocks are able to provide.

While Insilico’s clinical trial shows the promise of several A.I. techniques, its drug candidate could still be years away from regulatory approval, even for use in sick patients. And the company has not yet tested its anti-aging effects in healthy patients.

“This drug looks encouraging,” said Eric Topol, a cardiologist and the author of the book “Super Agers.” “But we do not yet have a definitive trial to make the final judgment.”

Insilico began exploring the new wave of A.I. technologies more than a decade ago. Founded in 2014 by Alex Zhavoronkov, a Latvian-Canadian mathematician, physicist and biotechnologist, the company was among the earlier efforts to streamline drug discovery using what are called neural networks.

Using techniques similar to those that train A.I. systems like ChatGPT, Insilico first built a model to analyze a giant pool of data spanning health records for thousands of medical patients; blood tests describing the microscopic proteins created inside their bodies; and myriad academic papers detailing the effects of these proteins. With this system, the company is trying to identify particular proteins that lead to illness and disease.

Insilico then built a second system that analyzes troves of data describing the physical shape of proteins and the way these tiny biological mechanisms bind to other molecules. This system is designed to generate entirely new molecules that could bind to a particular drug target and neutralize its effects.

“It is like scanning a lock and generating a key that fits the lock,” Dr. Zhavoronkov said in an interview.

This is how he and his company developed rentosertib, which is meant to treat a disease called idiopathic pulmonary fibrosis, or I.P.F. Sometimes called “the Alzheimer’s of the lungs,” I.P.F. is a chronic disease that thickens and scars pulmonary tissue, reducing its ability to move oxygen into the bloodstream. The condition can lead to death, even within a few years.

Last year, with its clinical trial, Insilico showed that its drug candidate could significantly expand air capacity inside the lungs of patients who had I.P.F. But that was only part of the trial. Dr. Zhavoronkov said his company also designed the drug in an effort to generally extend a patient’s life span.

Separate from the tests involving I.P.F., the trial used six aging clocks to measure the markers of age in patients both before and after treatment with the drug. Across the 43 patients who participated in the trial, the clocks — which essentially try to predict a person’s age based on measurements of how well their cell, tissues and organs are functioning — showed significant reductions in the predicted age.

Each clock relies on a different analysis of those markers that yields a different prediction. During the trial, all six clocks showed a reduction in predicted age after patients took the drug for 12 weeks.

“This is the first study that shows, very clearly, that predicted biological age can be reduced,” said Vadim Gladyshev, a Harvard Medical School professor who helped build one of these aging clocks.

He acknowledged, however, that the study was far from conclusive, pointing out that the sample size was small and that aging clocks are not always reliable. Most notably, Insilico’s drug has not yet been tested in healthy patients: Because the company conducted its clinical trial solely with I.P.F. patients, the results could be unique to those with this particular condition.

But some experts believe that Insilico’s elaborate study could serve as a blueprint for future longevity research.

“These are methods we will use in future trials,” said Evelyne Bischof, a professor of medicine at Tel Aviv University who specializes in longevity.

Translation

早期數據顯示人工智能產生的藥物或可延緩衰老

人工智能被用於輔助開發一種名為 rentosertib 的候選藥物，用於治療一種罕見的肺部疾病。其研發者表示，該藥物似乎還能減少老化的生物學特徵。

去年，一家名為 Insilico Medicine 的公司進行的臨床試驗表明，該公司的一種候選藥物可能有助於治療慢性肺部疾病患者。 Insilico Medicine 致力於利用人工智能加速藥物研發。

如今，該公司表示，來自同一臨床試驗的數據顯示出更令人感興趣的可能性：該藥物或許還能延緩老化過程。

這種名為 rentosertib 的藥物的分子結構是在人工智能的輔助下產生的。這項週一發表在《自然·生物技術》雜誌上的新研究結果顯示，該藥物降低了六種「老化時鐘」所測量的生物年齡標記。 「老化時鐘」是一種人工智能技術，旨在預測人們的發病率和死亡率。

這項精心設計的臨床試驗是利用人工智能技術改善醫療保健這項廣泛努力的一個里程碑。這些人工智能技術也應用於 ChatGPT 等熱門聊天機器人和 Midjourney 等影像產生器。 Insilico 只是眾多致力於利用這些方法加速藥物研發，並改善其他醫療任務的新創公司、科技巨頭和學術實驗室之一。

老化時鐘的興起正推動這些研究超越短期治療，邁向長壽研究領域。這些系統可以估算人體老化的速度，甚至可以估算單一器官相對於身體其他部位的老化速度。但科學家仍在爭論這些所謂的「時鐘」究竟能提供多少有用的信息。

儘管 Insilico 的臨床試驗展現了多項人工智能技術的潛力，但其候選藥物距離獲得監管部門批准可能仍需數年時間，即使是用于重病患者。而且，該公司尚未在健康的受試者身上測試其抗衰老效果。

心臟病專家、《超級老齡者》(Super Agers)  一書的作者 Eric Topol 說: “這種藥物看起來很有希望”，“但我們還沒有進行最終的臨床試驗來做出最終判斷。”

Insilico 公司在十多年前就開始探索新一代人工智能技術。該公司由拉脫維亞裔加拿大籍數學家、物理學家和生物技術專家 Alex Zhavoronkov 於 2014 年創立，是早期利用神經網絡去使藥物研發流程達到更有效率的公司之一。

Insilico 公司採用類似於訓練 ChatGPT 等人工智能系統的技術，首先建立了一個模型來分析龐大的資料池，這些資料包括數千名患者的健康記錄、描述其體內產生的微小蛋白質的血液檢測結果，以及大量詳細闡述這些蛋白質作用的學術論文。利用這套系統，該公司正試圖辨識導致疾病的特定蛋白質。

隨後，Insilico 構建了第二個系統，用於分析海量數據，這些數據描述了蛋白質的物理結構以及這些微小的生物機制與其他分子的結合方式。這個系統是用來生成全新的分子，這些分子可以結合特定藥物的目標對像並消滅它。

Zhavoronkov 博士在一次採訪中說道: 「這就像先掃描一把鎖，然後造出一支可開啓它的鑰匙」。

他和他的公司正是利用這種方法開發了 rentosertib，一種用於治療特發性肺纖維化（IPF）的藥物。 IPF 有時被稱為 “肺部阿茲海默症” ，是一種慢性疾病，會導致肺組織增厚和形成疤痕，從而降低肺部將氧氣輸送到血液的能力。這種疾病甚至可能在幾年內導致死亡。

去年，Insilico公司透過臨床試驗表明，其候選藥物能夠顯著擴大特發性肺纖維化（IPF）患者的肺部容量。但這只是試驗的一部分。 Zhavoronkov 博士表示，該公司研發該藥物的另一個目的是延長患者的壽命。

除了針對IPF的試驗外，該試驗還使用了六種老化時鐘來測量患者在用藥前後各項年齡指標的變化。在參與試驗的43名患者中，這些時鐘 - 它們本質上是根據測量一個人的細胞、組織和器官運作情況來預測其屬於的年紀- 顯示預測的年齡有顯著降低。

每種老化時鐘都基於不同的指標分析方法，因此預測結果也不同。試驗期間，所有六種老化時鐘均顯示，患者服用藥物12週後，預測年齡有所降低。

曾參與建造了其中一個衰老時鐘的哈佛醫學院教授 Vadim Gladyshev 說道:「這是第一個研究非常清楚地表明預測的生物年齡是可以下降的」。

然而，他也承認這項研究遠未得出結論，指出樣本量較小，而且老化時鐘並非總是可靠的。尤其值得注意的是，Insilico公司的藥物尚未在健康的受試者中進行測試：由於該公司僅針對特發性肺纖維化（IPF）患者開展了臨床試驗，因此研究結果可能僅適用於患有這種特定疾病的患者。

但一些專家認為，Insilico 公司這項精心設計的研究可以為未來的長壽研究提供藍圖。

特拉維夫大學的專門研究長壽的醫學教授 Evelyne Bischof 說:「這些是我們在未來試驗將會採用的方法」。

So, a clinical trial conducted by Insilico Medicine has indicated that one of its drug candidates could help treat patients suffering from a chronic lung disease. Now, the company says that data from the same clinical trial shows a more intriguing possibility: that the drug could also slow the aging process. However, it is also acknowledged that the study is far from conclusive as the sample size is small and that aging clocks are not always reliable. Yet, Insilico’s study could probably serve as a blueprint for future longevity research.

Note:

1. The core idea of the book entitled Super Agers: An Evidence-Based Approach to Longevity written by Dr. Eric Topol is that science and technology have converged to create a breakthrough moment where we can significantly extend human "healthspan"—the number of years lived free from debilitating chronic disease—rather than just extending lifespan.  (Google)

China's August Exports Rise 25%; Trade Surplus Continues to Expand

Recently NHK News on-line reported the following:

中国 8月の輸出額25％増加 貿易黒字は拡大傾向続く

2026年9月8日 20:36

中国

中国の8月の輸出額は、半導体の輸出が伸びたことなどから去年の同じ月と比べて25％増えました。好調な輸出にけん引され、貿易黒字は拡大傾向が続いています。

中国の税関当局が8日発表した貿易統計によりますと、8月の輸出額はドル換算で4014億ドルと、去年の同じ月と比べて25％増え、5か月連続でふた桁の伸び率を記録しました。

これは世界的なAI需要の拡大を背景に半導体を含む集積回路の輸出額が大きく伸びたほか、EV＝電気自動車などの輸出が増えたことが主な要因です。

国・地域別では、アメリカ向けの輸出が34％、ASEAN＝東南アジア諸国連合向けが30％、それぞれ大幅に増えたほか、EU＝ヨーロッパ連合向けは6％増加しました。

一方、全体の輸入額は28.2％増えて2823億ドルとなり、輸出から輸入を差し引いた貿易収支は1190億ドルの黒字でした。

中国の貿易黒字は、輸出の大幅な増加に伴って拡大傾向が続いていて、ことし1月から8月までの累計の貿易黒字は8000億ドルを超えました。

本との貿易についてはレアアース磁石などの一部の品目の輸出が減少していますが、全体では輸出が8％、輸入は20％、それぞれ増えました。

Translation

China's August Exports Rise 25%; Trade Surplus Continues to Expand

September 8, 2026, 20:36

China

China's exports in August rose by 25% year-on-year, driven by factors such as increased semiconductor shipments. Fueled by this strong export performance, the country's trade surplus continues to expand.

According to trade statistics released by Chinese customs authorities on the 8th, the value of exports in August reached $401.4 billion (in US dollar terms)—a 25% increase compared to the same month last year—marking the fifth consecutive month of double-digit growth.

Key factors behind this rise include a significant increase in exports of integrated circuits (including semiconductors), driven by expanding global demand for AI as well as increased exports of electric vehicles (EVs).

By country and region, exports to the United States surged by 34% and those to ASEAN (Association of Southeast Asian Nations) by 30%, while exports to the EU (European Union) rose by 6%.

Meanwhile, total imports increased by 28.2% to $282.3 billion, resulting in a trade surplus of $119 billion, calculated by subtracting imports from exports.

China's trade surplus continues to widen alongside the substantial growth in exports; the cumulative trade surplus for the period from January to August of this year has exceeded $800 billion.

On the other hand, regarding trade with Japan, while exports of certain items—such as rare-earth magnets—declined, overall exports to Japan rose by 8% and imports from Japan increased by 20%.

So, China's exports in August rose by 25% year-on-year, driven by factors such as increased semiconductor shipments. Because of this strong export performance, the country's trade surplus continued to expand. On the other hand, overall exports to Japan rose by 8% and imports from Japan increased by 20%.

被列入黑名單的中國科技巨頭如何持續購買美國頂尖人工智能晶片（2/2）

Recently The New York Times reported the following:

How a Blacklisted Chinese Tech Giant Kept Buying America’s Best A.I. Chips (2/2)

Washington imposed sanctions on Inspur because of its work with the Chinese military. But the company’s subsidiary kept shipping Nvidia’s best chips to feed China’s leading A.I. firms.

The NYT - By Ana SwansonPaul MozurTripp Mickle and Keith Bradsher (Ana Swanson reported from Washington and Jinan, China; Paul Mozur from Taiwan and Singapore; Tripp Mickle from Milpitas, Calif.; and Keith Bradsher from Jinan.)

Sept. 6, 2026

Updated 6:45 a.m. ET

(continue)

Different Country, Same Tactic

Most Americans have never heard of Inspur, but the company is among the world’s largest makers of computer servers. It has also dodged one sanction after another.

When the U.S. blacklisted Inspur in 2023, it got a new name in the United States. In China, Inspur got a new address.

The entity list, which the United States uses to block trade to targeted companies, applies to the specific company name and address designated by the government. Just days after its 2023 listing, Inspur registered its main business to a building down the road.

Four months ago, a Times reporter visited that address, in the eastern Chinese city Jinan. It was a nondescript building with no signage beyond a bank on the ground floor. A security guard confirmed that Inspur had an office there, but said not many people worked in it. He said Inspur was really based at the industrial park down the street.

That compound was busy. A series of dark gray high-rises stood inside a vast enclosure ringed by a hedge of roses and rows of motorbikes. Inside the gates, a red sign flashed a key slogan from China’s leader, Xi Jinping: “Chinese dream.”

Inspur, which is majority-owned by Shandong Province, began as a state-owned electronics factory. It developed the world’s first pager capable of displaying Chinese characters. More than a decade ago, the Chinese military and state turned to Inspur’s servers to reduce the reliance on foreign companies.

That made Inspur an American target. In 2020, the Trump administration included it on a list of Chinese military companies. In 2023, the Biden administration said Inspur Group had sought U.S. technology to make military supercomputers and put it on the entity list. That meant U.S. firms could no longer ship to it without a special license.

A loophole quickly opened. Companies like Intel had been selling to Inspur’s subsidiary, not the parent company, which the order did not cover.

Two years later, the second Trump administration closed a part of the loophole, adding some of Inspur’s subsidiaries to the blacklist. Aivres was not among them.

The Standout Connection

Records show that billions of dollars of advanced servers and other technology have flowed from Aivres and Inspur’s other global subsidiaries to cloud-computing and assembly firms across Southeast Asia.

One connection stands out. Maginfra was created in China last year, and quickly began doing brisk business with a Malaysian electronics firm that transacts with Inspur’s companies in Hong Kong, mainland China, the United States and Singapore.

Maginfra imported more than $700 million of servers from Malaysia into China in a six-month period, including more than 1,500 servers that were priced consistently with high-end A.I. capabilities, ImportGenius data shows. These entries do not include detailed descriptions of the servers or chips inside them, but they do indicate per-unit prices.

Maginfra has a scant public profile and no direct link to Inspur on paper. It is also government-owned, and it is down the street from Inspur’s headquarters. Corporate records show that the legal representative of Maginfra’s parent company has held positions at companies affiliated with Inspur subsidiaries or investment partners.

According to Wirescreen, a data platform, many of the patents that Maginfra has registered were filed by inventors who have worked for Inspur. The two firms’ websites also share striking similarities.

When a reporter from The Times visited Maginfra’s listed address in China on Tuesday, it appeared deserted. A few chairs and empty desks sat inside a small locked room with no signs. Workers on the floor said they had not seen anyone use the office, and building receptionists said they had not heard of Maginfra.

Public filings analyzed by C4ADS, a Washington research nonprofit, show Chinese universities, some with ties to defense research, seeking and sometimes obtaining servers containing advanced computer chips built by Maginfra and Inspur.

Maginfra also inked a contract in November to sell $165 million of high-performance computing technology to a division of a Chinese state-owned bank, according to a filing on the Hong Kong Stock Exchange.

“It’s very hard to look at this and not see a state-sponsored Chinese network for bypassing export controls,” said William George, the director of research at ImportGenius, who analyzed the trade records.

But the Trump administration approved Maginfra this year for a license for Nvidia’s less powerful H200 chips, according to documents viewed by The Times. Reuters previously reported the license.

The subsidiary Aivres remains very visible in Silicon Valley. It advertised A.I. servers using AMD’s technology and was listed as an “elite” partner of Nvidia and a sponsor of its annual conference in San Jose, Calif. (AMD declined to comment.)

In May, Nvidia announced that Aivres was among the more than two dozen companies in full-scale production of the servers that would supply its next generation of A.I. technology.

Translation:

被列入黑名單的中國科技巨頭如何持續購買美國頂尖人工智能晶片（2/2）

華盛頓因浪潮集團與中國軍方合作而對其實施制裁。但該公司旗下子公司卻持續向中國領先的人工智能公司供應英偉達的頂級晶片

（繼續）

不同國家，相同策略

大多數美國人從未聽說過浪潮集團，但該公司是全球最大的電腦伺服器製造商之一。它也成功地規避了一次又一次的制裁。

2023年，美國將浪潮列入黑名單後，浪潮在美國啟用了一個新的名稱。在中國，浪潮則更換了辦公地址。

美國利用實體清單來阻止與特定公司進行貿易，該清單適用於政府指定的公司名稱和地址。就在被列入2023年實體清單幾天後，浪潮將其主要業務註冊到了附近的一棟辦公大樓。

四個月前，《紐約時報》記者走訪了位於中國東部城市濟南的這棟辦公大樓。這是一棟毫不起眼的建築，除了底層的一家銀行外，沒有任何標誌。一名保安人員證實浪潮在那裡設有辦公室，但表示員工不多。他說，浪潮的實際總部在街道落些的工業園區。

那個工業園區裡生意興隆。一排排深灰色高樓矗立在一片巨大的圍牆內，圍牆四周環繞著玫瑰花籬和成排的摩托車。大門內，一塊紅色標誌閃爍著中國領導人習近平的重要口號：「中國夢」。

浪潮集團由山東省控股，最初是一家國有電子工廠。它研發出了世界上第一台能夠顯示中文的傳呼機。十多年前，中國軍方和政府開始使用浪潮的伺服器，以減少對外國公司的依賴。

這使得浪潮成為美國的目標。 2020年，特朗普政府將其列入中國軍工企業名單。 2023年，拜登政府稱浪潮集團尋求美國技術製造軍用超級計算機，並將其列入實體清單。這意味著美國公司如果沒有特殊許可，就無法再向其供貨。

然而，一個漏洞很快就出現了。像英特爾這樣的公司一直以來都是向浪潮的子公司而非母公司銷售產品，而該禁令並未涵蓋子公司。

兩年後，特朗普第二任期政府堵住了部分漏洞，將浪潮的部分子公司列入黑名單。Aivres並不在名單中。

引人注目的關聯

記錄顯示，數十億美元的先進伺服器和其他技術產品從Aivres和浪潮的其他全球子公司流向東南亞各地的雲端運算和組裝公司。

其中一個關聯尤為引人注目。 Maginfra去年在中國成立，並迅速與一家馬來西亞電子公司開展了密切的業務往來。這家馬來西亞電子公司與浪潮在香港、中國大陸、美國和新加坡的公司都有業務交往。

ImportGenius的數據顯示，Maginfra在六個月內從馬來西亞向中國進出口了價值超過7億美元的伺服器，其中包括1,500多台價格與高階人工智能功能相符的伺服器。這些條目並無伺服器或晶片的詳細描述，但標明了每台價格。

Maginfra 的公開資訊很少，文件上也看不出與浪潮有任何直接聯繫。該公司為國營企業，辦公地點與浪潮總部僅一街之隔。公司記錄顯示，Maginfra 母公司的法定代表人曾在浪潮子公司或投資夥伴公司擔任職務。

據數據平台 Wirescreen 稱，Maginfra 註冊的許多專利都是由曾在浪潮工作的發明人申請的。兩家公司的網站也驚人地相似。

週二，The Times記者前往 Maginfra 在中國的註冊地址時，發現那裡空無一人。一間上了鎖的小房間裡擺放著幾張椅子和幾張空桌子，沒有任何標示。該樓層的員工表示他們從未見過有人使用這間辦公室，而大樓的前台接待人員則表示他們從未聽說過 Maginfra 這家公司。

華盛頓一家非營利研究機構C4ADS分析的公開文件顯示，中國大學，一些與國防研究相關的，正在尋求並有時獲得搭載由Maginfra和浪潮製造的先進電腦晶片的伺服器。

根據香港聯交所的文件顯示，Maginfra還在11月簽署了一份合同，向一家中國國有銀行的子公司出售價值1.65億美元的高性能電腦技術。

ImportGenius 研究主管 William George 分析了貿易記錄後表示： “很難不把這看作是一個由中國政府支持的繞過出口管制的網絡。”

但根據《紐約時報》查閱的文件顯示，特朗普政府今年批准了 Maginfra 公司獲得英偉達性能較低的 H200 晶片的授權。The Times先前也曾報道這項授權。

Aivres 子公司在矽谷仍然非常活躍。該公司曾宣傳使用 AMD 技術的 AI 伺服器，並被列為英偉達的「精英」合作夥伴，同時也是英偉達在加州聖荷西舉辦的年度大會的贊助商。 （AMD 拒絕置評。）

今年 5 月，英偉達宣布 Aivres 是二十多家正在全面生產下一代 AI 技術伺服器的公司之一。

               So, in March 2023, the United States added Inspur Group to a blacklist with more than 3,000 companies that have been restricted from doing business with the United States because they are considered a security risk. But not long after, the Inspur sign was changed to Aivres, with the intention to evade American restrictions. In May, Nvidia announced that Aivres was among the more than two dozen companies in full-scale production of the servers that would supply its next generation of A.I. technology. Apparently, Chinese tech giant is still buying America’s best A.I. chips from the United States.

被列入黑名單的中國科技巨頭如何持續購買美國頂尖人工智能晶片（1/2）

Recently The New York Times reported the following:

How a Blacklisted Chinese Tech Giant Kept Buying America’s Best A.I. Chips (1/2)

Washington imposed sanctions on Inspur because of its work with the Chinese military. But the company’s subsidiary kept shipping Nvidia’s best chips to feed China’s leading A.I. firms.

The NYT - By Ana SwansonPaul MozurTripp Mickle and Keith Bradsher (Ana Swanson reported from Washington and Jinan, China; Paul Mozur from Taiwan and Singapore; Tripp Mickle from Milpitas, Calif.; and Keith Bradsher from Jinan.)

Sept. 6, 2026

Updated 6:45 a.m. ET

For years, a sign outside one of Silicon Valley’s many nondescript offices said Inspur, the name of one of China’s most important technology companies.

But in March 2023, the United States added Inspur Group to the so-called entity list, a blacklist with more than 3,000 companies that have been restricted from doing business with the United States because they are considered a security risk.

Not long after, the Inspur sign was changed to Aivres, the name of Inspur’s new U.S. subsidiary. The subsidiary continued with its Chinese parent company’s work, helping it build a thriving global network to provide the computing power that supplies China’s growing artificial intelligence industry.

The Silicon Valley subsidiary, which appears to be taking advantage of several loopholes in U.S. export law, has played a key role in Inspur’s sidestepping of American restrictions. Federal officials have begun looking into the subsidiary’s business, but it is not clear where that inquiry stands, said four officials who spoke on the condition of anonymity because they were not allowed to discuss it.

The New York Times examined thousands of shipment records, corporate documents and supply contracts to piece together how Inspur’s network works. Reporters also visited sites in China, Southeast Asia and California, and spoke with more than a dozen current and former officials, tech executives and industry analysts.

The U.S. government prohibits the most advanced chips, like the Blackwell chip from the Silicon Valley giant Nvidia, from being sent to any company in China. And U.S. law bars any company anywhere in the world from buying restricted products made with American technology on behalf of a blacklisted company. But there are few restrictions on companies in China that bar them from remotely gaining access to A.I. data centers in Southeast Asia.

Whether Chinese firms are getting access to cutting-edge technology, particularly high-end computer chips made by Nvidia, has become a hot question in Silicon Valley and Washington, where executives and officials worry that China could use A.I. for hacking, surveillance and military operations.

Chinese A.I. companies are also closing the performance gap on systems made by their American competitors, raising concerns that the Chinese companies are using American technology.

From April 2024 through February 2026, Aivres exported at least $5.6 billion of advanced technology from the United States to Southeast Asia, including more than $3 billion of computers containing cutting-edge Blackwell chips and equipment from Nvidia, according to trade records analyzed by ImportGenius, a data firm, and The Times.

That technology went to data centers and technology firms in Southeast Asia that serve China’s biggest tech firms, including Alibaba and ByteDance, trade records show. Megaspeed, a China-linked company that U.S. officials have been investigating over whether it smuggled restricted chips into China, was among that group of recipients. (One of Megaspeed’s subsidiaries has financial ties with Alibaba, according to Malaysian business documents.)

Other technology went from Inspur’s network of global subsidiaries to an electronics manufacturer in Malaysia, which has been shipping unlabeled but high-priced servers to China, according to trade records drawn mostly from Malaysia and Indonesia. The servers are sent to a new firm, Maginfra, which has supplied high-performance computers to Chinese universities and state-owned firms, according to Chinese contracts and procurement records.

Inspur, Aivres, Maginfra, Alibaba and ByteDance did not respond to requests for comment. Megaspeed did not provide a comment. John Rizzo, a spokesman for Nvidia, said the company did not support “diverted products,” and sold to well-known partners “who work with us to ensure that all sales comply with U.S. export control rules.”

The Trump administration rolled back Biden-era restrictions on remote access by Chinese companies to data centers in other countries, and both the administration and Congress have been weighing new rules to replace them. Nvidia’s Mr. Rizzo said the business brought “tens of billions of dollars and high-paying jobs home.”

The administration has also taken a more cautious approach to placing sanctions on Chinese firms, after attempts last year to expand the entity list drew blowback from China. Inside the Commerce Department, which enforces the list, six current and former officials described an agency that had lost its appetite for cases involving China or Nvidia. Ten months have passed without a firm’s being added to the entity list, the longest stretch in 18 years.

A White House official said the Trump administration had enacted the most rigorous export control regime in modern history and was committed to safeguarding America’s national security.

But Jordan Nanos, an analyst at SemiAnalysis, a tech research firm, who has studied Aivres’s business, said that the firm operated in the United States with the same servers, employees and offices as its parent company, Inspur, and that federal officials seemed to have done little about it.

“Are they trying to ban them, or are they not?” he said

(to be continued)

Translation

被列入黑名單的中國科技巨頭如何持續購買美國頂尖人工智能晶片（1/2）

華盛頓因浪潮 (Inspur) 與中國軍方合作而對其實施制裁。但浪潮的子公司卻持續向中國領先的人工智能公司供應英偉達的頂級晶片。

多年來，矽谷眾多不起眼的辦公大樓之一外掛著一塊牌子，上面寫著“浪潮”( Inspur)，這是中國最重要的科技公司之一的名字。

但2023年3月，美國將浪潮集團列入所謂的實體清單。這是一份包含3,000多家公司的黑名單，這些公司因被視為安全風險而被禁止與美國開展業務。

不久之後，浪潮的標誌被改為艾弗斯（Aivres），這是浪潮新成立的美國子公司的名稱。該子公司繼續進行其中國母公司的工作，幫助其建立一個蓬勃發展的全球網絡，為中國蓬勃發展的人工智能產業提供運算能力。

這家位於矽谷的子公司似乎利用了美國出口法律的多個漏洞，在浪潮規避美國限制方面發揮了關鍵作用。四位不願透露姓名的官員表示，聯邦官員已開始調查該子公司的業務，但目前尚不清楚調查進展如何。他們表示，由於未獲授權討論此事，因此要求匿名。

《紐約時報》查閱了數千份出貨記錄、公司文件和供應合同，以拼湊出浪潮的網絡運作方式。記者也造訪了中國、東南亞和加州的多個地點，並採訪了十幾位現任和前任官員、科技高層和行業分析師。

美國政府禁止將最先進的晶片，例如矽谷巨頭英偉達的 Blackwell 晶片，運往中國的任何公司。美國法律禁止世界任何地方的任何公司，去代表已被列入黑名單的公司去購買使用由美國技術製造來的受限產品。但中國公司在遠程地存取東南亞的人工智能數據中心方面幾乎沒有受到什麼限制。

中國公司是否能夠獲得尖端技術，尤其是英偉達生產的高階電腦晶片，已成為矽谷和華盛頓的熱門話題。當地的高層和官員擔心，中國可能會利用人工智能進行駭客攻擊、監控和軍事行動。

中國人工智能企業也正在縮小與美國作為競爭對手在系統之間的性能差距，這引發了人們對中國企業使用美國技術的擔憂。

根據數據公司 ImportGenius 和《紐約時報》分析的貿易記錄顯示，從 2024 年 4 月到 2026 年 2 月，Aivres 公司從美國向東南亞出口了至少 56 億美元的先進技術，其中包括價值超過 30 億美元搭載了英偉達 Blackwell 尖端晶片的電腦和設備。

貿易記錄顯示，這些技術被輸送到東南亞的数據中心和科技公司。這些數據中心和科技公司為包括阿里巴巴和字節跳動在內的中國大型科技公司提供服務。Megaspeed是一家與中國有關的公司，美國官員一直在調查該公司是否向中國走私了受限晶片，該公司也是那群接收技術者之一。（根據馬來西亞商業文件顯示，Megaspeed的一家子公司與阿里巴巴有財務聯繫。）

根據主要來自馬來西亞和印尼的貿易記錄顯示，其他技術則從浪潮集團的全球子公司網絡流向馬來西亞一家電子產品製造商，該公司一直在向中國出口未貼標籤但價格昂貴的伺服器。這些伺服器最終被送往一家名為Maginfra的新公司，根據中國的合約和採購記錄，該公司已向中國高校和國有企業供應高效能電腦。

浪潮集團、Aivres、Maginfra、阿里巴巴和字節跳動跳動均未回應置評請求。 Megaspeed也未置評。英偉達發言人John Rizzo表示，該公司不支持 “轉售產品” ，並向知名合作夥伴銷售產品， “這些合作夥伴與我們合作，確保所有銷售均符合美國出口管制規定” 。

特朗普政府撤銷了拜登時期對中國企業遠端存取其他國家数據中心的限制，目前美國政府和國會都在考慮制定新的規則來取代這些限制。英偉達執行長Rizzo表示，該業務為美國帶來了「數百億美元的收入和高薪就業機會」。

去年，美國政府試圖擴大實體清單，但遭到中國的強烈反對。此後，美國政府在對中國企業實施制裁方面採取了更謹慎的態度。負責執行清單的商務部內部，目前有六位現任或前任官員都表示，該機構似乎對涉及中國或英偉達的案件失去了興趣。十個月過去了，沒有一家公司被列入實體清單，這是18年來最長的間隔期。

一位白宮官員表示，特朗普政府已經實施了近代史上最嚴格的出口管制制度，並致力於維護美國的國家安全。

但科技研究公司 SemiAnalysis 的分析師 Jordan Nanos 曾研究過 Aivres 的業務。他表示，該公司在美國運營，使用與其母公司浪潮（Inspur）相同的伺服器、員工和辦公室，然而聯邦官員似乎對此並未採取任何行動。

他問:「他們到底有沒有試圖禁制這些公司？」。

（待續）

90歲的「DNA之父」，為甚麼被科學界割席？

 Recently Simon Shen (沈旭暉) posted the following article on Patreon:

【國際關係日評】 90歲的「DNA之父」，為甚麼被科學界割席？

沈旭暉 Simon Shen (simonshen@creator.patreon.com)

 Sun, Sep 6 at 4:04 p.m.

6 Sept 2026

 James Watson 是20世紀最重要的科學家之一。1953年，他與 Francis Crick 提出 DNA 雙螺旋結構，後來與 Maurice Wilkins 共同獲得諾貝爾獎。今天我們談基因工程、DNA鑑證，都離不開這場科學革命。想不到到了晚年，Watson 卻因為另一套關於基因的理論 / 想法，被自己曾經領導的科學界集體割席。

2007年，Watson 公開表示黑人與白人在智力測試上的平均差異可能與遺傳有關，引起巨大爭議，最終離開領導多年的 Cold Spring Harbor Laboratory。2019年，已經90歲的 Watson 在 PBS 紀錄片再次被問到這個問題，仍然表示自己的看法沒有改變。Cold Spring Harbor 隨即褫奪他所有榮譽職銜，直指他的說法「沒有科學根據」。

Watson 是不是納粹或者今天所說的 far right？都不是。但他的思想確實具有明顯優生學色彩：他相信人類很多重要特徵受到遺傳影響，也不抗拒利用遺傳科技選擇某些特徵、避免另一些特徵。

一個破解 DNA 結構的諾獎得主這樣說，是否特別可信？起碼主流科學界認為「不是」。Watson獲得諾貝爾獎的研究，是 DNA 的分子結構和遺傳資訊如何傳遞，幫助我們了解基因如何運作，卻沒有證明「黑人與白人的平均智力差異由基因造成」。後者是另一個完全不同、而且複雜得多的命題。

換句話說，今天科學界並不否認智力受到遺傳因素影響，但一項特徵具有遺傳性，不代表不同群體之間的平均差異就是遺傳造成。最簡單的例子是南北韓：兩邊人口遺傳背景非常接近，但今天南韓年輕人的平均身高明顯高於北韓人。不是南韓人在兩三代之間突然進化出「高佬基因」，而是營養、醫療和成長環境不同所致。

智力比身高更複雜。教育、營養、家庭環境、收入、文化、甚至測量方法，都可能影響結果。因此由「智力具有遺傳性」，跳到「兩個族群的平均 IQ 差異是基因造成」，中間就缺少了最重要的一步，也就是缺乏 correlation。因此，Watson 晚年的爭議很諷刺：他最偉大的發現告訴我們基因非常重要，但他後來希望論證「群體差異可以由基因解釋」，卻處處碰壁，乃至接近身敗名裂。

(待續)

（第二部份繼續於）https://www.patreon.com/c/simonshen/posts?filters%5Btag%5D=%E5%9C%8B%E9%9A%9B%E9%97%9C%E4%BF%82%E6%97%A5%E8%A9%95)

I agree with the viewpoint that intelligence is more complex than height. Education, nutrition, family environment, income, culture, and even measurement methods can all influence the development of intelligence. Therefore, jumping from "intelligence is heritable" to "the difference in average IQ between two ethnic groups is caused by genes" misses an important link: to prove the correlation.

Note:

1. Simon Shen (沈旭暉) is a Hong Kong born international relations scholar, political commentator, and public intellectual. He is particularly known for his work and commentary on China's foreign policy, U.S.–China relations, nationalism, Hong Kong, Taiwan, and international politics. He studied at Yale University and subsequently obtained a doctorate in politics/international relations from the University of Oxford. He has held academic and research positions at universities and institutions in Hong Kong and overseas. (ChatGPT)

2. James Watson (1928–2025) was an influential American molecular biologist and geneticist. In 1953, he co-discovered the double-helix structure of DNA alongside Francis Crick, drawing on pivotal X-ray crystallography data. This landmark breakthrough revolutionized modern biology by demonstrating how genetic information is stored and replicated. For this achievement, Watson shared the 1962 Nobel Prize in Physiology or Medicine. Later in his career, he led Cold Spring Harbor Laboratory and directed the Human Genome Project. (Google)

作者與出版商就Anthropic人工智能15億美元和解協議展開爭奪（2/2）

 Recently The New York Times reported the following:

Authors Wrangle with Publishers Over $1.5 Billion Anthropic A.I. Settlement (2/2)

The tech giant must pay $3,000 per pirated book that it used to train its chatbot. Many authors fear they could lose funds to others in the book business.

The NYT - By Neil Vigdor

Sept. 5, 2026

(continue)

Another author, Amy Lupold Bair, who has written guidebooks about blogging and how families can navigate the digital world, vented on social media over her potential diminishing share of the settlement.

“My publisher is fighting the default 50/50 payout split from the Anthropic class action settlement,” she wrote on Threads. “They only want me, the author whose entire work was stolen, to get 10%. So that’s a fun email to wake up to.”

Her post, which did not name the publisher, received more than 1,000 likes and started a lengthy discussion about the settlement process.

She declined to comment.

John Wiley & Sons Inc., which was listed as the publisher for two of her books on her website, “Raising Digital Families for Dummies” and “Blogging for Dummies,” did not address whether it was seeking a 90 percent cut.

In a statement on Friday, a representative for the company said that it was following the settlement terms when it filed claims for “all Wiley-published” works covered by the Anthropic agreement.

Allocations for educational titles were set by individual contracts, said the representative, which did not immediately say whether Wiley put its Dummies guidebook series in that category.

The Association of American Publishers, a trade group that represents most of the major publishing houses in the United States and has hundreds of members, did not immediately provide a comment on Friday.

In 2024, a group of authors filed a class-action lawsuit against Anthropic after learning that the company, seeking huge amounts of digital text needed to train A.I. technologies, had used their work without paying for it and stockpiled millions of pirated books.

Anthropic chose to settle the case after a judge paved the way for the case to go to trial, finding that the company’s stealing of books had given the plaintiffs cause.

At the same time, the judge said that the company’s training of its chatbot with books that it had legally acquired was a “fair use.”

Anthropic referred to a statement in May by Aparna Sridhar, the company’s deputy general counsel, as the settlement was going through final approval.

“We are pleased that more than 91 percent of authors and publishers covered by the settlement have claimed their share of the payment, and we’re looking forward to bringing this matter to a close,” the statement said.

JND Legal Administration, the settlement administrator, did not immediately respond to requests for comment on Thursday. Neither did a lawyer for the plaintiffs in the case.

Kristelia García, a professor at Georgetown Law who specializes in copyright law, said the dynamics of divvying up the Anthropic settlement funds were reminiscent of a 2007 federal lawsuit over the distribution of digital royalties for the music of the rapper Eminem.

Producers who discovered Eminem sued a subsidiary of the Universal Music Group and argued that they should receive a higher percentage of royalties from downloads at iTunes and other digital stores.

The two sides reached a settlement in 2012.

Not all book contracts have language about who is entitled to settlement payments in copyright infringement cases or revenues from emerging technologies, including A.I., García said.

“So I think this is just kind of the publishing industry having that sort of streaming moment where their contracts didn’t contemplate this,” she said.

Muddying the settlement process further, according to the Authors Guild, are more recent book ownership changes.

Rasenberger said an author contacted the guild this week with a dilemma: Rights to the books had reverted to the author this year.

“It depends on who owned the rights when the books were downloaded when the infringement happened, which is, you know, 2021 and ’22,” Rasenberger said.

In instances when ownership disputes remain unresolved, Rasenberger said, a court-appointed arbitrator will have the final say, but she added the guild “will fight tooth and nail for them if they are, in fact, being screwed over.”

Translation

作者與出版商就Anthropic人工智能15億美元和解協議展開爭奪（2/2）

這家科技巨頭必須為每本用於訓練其聊天機器人的盜版書籍支付3,000美元。許多作者擔心他們的資金可能會被其他出版業人士所瓜分

（繼續）

另一位作者 Amy Lupold Bair 撰寫了關於部落格寫作和家庭如何駕馭數位世界的指南，她在社交媒體上表達了對和解協議中可能減少的份額的不滿。

她在Threads上寫道:「我的出版商正在反對Anthropic集體訴訟和解協議中預設的50/50賠償分配方案」; 「他們只想讓我 - 一個作品全部被盜用的作者 - 拿到10%。所以，一早醒來就收到這樣的電郵，真係幾開心喎」。

她的貼文並未指明出版商，但獲得了超過1,000個贊，並引發了關於和解流程的長時間討論。

她拒絕置評。

約翰威立父子出版公司（John Wiley & Sons Inc.）是她網站上列出的兩本書的出版商，兩書分別是《養育數位家庭入門指南》（Raising Digital Families for Dummies）和《部落格入門指南》（Blogging for Dummies）。該公司並未回應是否要求90%的版稅。

該公司一位代表在週五的聲明中表示，他們已按照和解條款，就所有受Anthropic協議約束的 “威立(Wiley)出版” 作品提出索賠。

這位代表表示，教育類書籍的版稅分享比例由單獨的合約決定，但並未立即說明威立是否將其「入門指南」系列歸入此類別。

美國出版商協會（Association of American Publishers）是一個代表美國大多數主要出版社的行業組織，擁有數百名會員。該協會週五未立即對此事發表評論。

2024年，一群作家對 Anthropic 公司提起集體訴訟，因為他們發現該公司為了訓練人工智能技術，未經許可使用了他們的作品，並囤積了數百萬冊盜版書籍。

在法官裁定該公司盜版書籍的行為構成原告的訴訟理由，允許案件進入審判程序後，Anthropic公司選擇與原告達成和解。

同時，法官表示，該公司使用合法取得的書籍訓練聊天機器人屬於「合理使用」。

在和解協議最終獲批之際，Anthropic 公司援引了其副總法律顧問 Aparna Sridhar 於5月發表的聲明。

聲明中寫道：“我們很高興超過91%的受和解協議約束的作家和出版商已領取了他們應得的賠償金，我們期待著此事​​能夠圓滿結束。”

和解管理人 JND Legal Administration 週四未立即回應置評請求。本案原告律師也未回應。

喬治城大學法學院專攻版權法的 Kristelia García 教授表示，Anthropic 和解金的分配方式讓人想起2007年一起關於說唱歌手 Eminem 的電子音樂版稅分配的聯邦訴訟。

發掘 Eminem 的製作人起訴了環球音樂集團的子公司，並主張他們應該從iTunes和其他電子商店的下載中獲得更高的版稅比例。

雙方於2012年達成和解。

García 指出，並非所有圖書合同都包含有關在版權侵權案件中指出誰有權獲得和解金，或源自新興技術（包括人工智能）的收入的條款。

她說:「所以我認為這只是出版業在經歷串流媒體時代之後，合約中沒有考慮到這一點的一種體現」。

據美國作家協會稱，近期圖書版權所有權的變更進一步加劇了和解過程的複雜性。

Rasenberger 表示，本週有一位作家聯繫了協會带來了一個難題：今年圖書的版權又返回到了作者手中。

Rasenberger 說: 「這取決於侵權發生時 - 也就是2021年和2022年 - 圖書被下載時的版權歸屬」。

Rasenberger 表示，如果版權糾紛無法解決，法院指定的仲裁員將擁有最終裁決權，但她補充說，如果作家協會「確實受到了不公平對待，會將竭盡全力為他們爭取權益」。

              So, in 2024, a group of authors filed a class-action lawsuit against Anthropic after learning that the company, seeking huge amounts of digital text needed to train A.I. technologies, had used their work without paying for it and stockpiled millions of pirated books. Now, thousands of authors are waiting for their share of the largest copyright settlement in U.S. history from Anthropic. In instances when copyright ownership disputes remain unresolved, a court-appointed arbitrator will have the final say. Apparently, training the chatbot is a costly endeavor.