China’s mysterious new billionaires are conquering the world And refusing to talk to the press   Jul 19th 2026|Shanghai|7 min read   Early last year, when DeepSeek, a Chinese artificial-intelligence lab, was causing panic among its Western rivals, almost nothing was known about Liang Wenfeng, its founder. Mr Liang remains a mysterious figure, but at least one thing is clear. As DeepSeek closes a funding round valuing it at $71bn, its founder’s personal wealth has soared to roughly $38bn—making him far wealthier than Anthropic’s Dario Amodei or OpenAI’s Sam Altman.   Mr Liang will not be spotted on the conference circuit. He has never appeared live on Chinese TV, and there are few photos of his unpolished, wiry figure. That puts him at home among the new generation of young Chinese billionaires. The country is cranking out more of them than anywhere else bar America. According to Hurun, which tracks the riches of the world’s wealthiest people, there are at least 29 self-made billionaires aged 40 or under in China. Adding Mr Liang, who will turn 41 at some point this year, brings the total to 30. That is nine more than last year.   China’s new tycoons differ from their predecessors in striking ways. They have abandoned property speculation and boozy dinners with officials for video games and meet-ups over Japanese anime. But they are also adept at taking imaginative ideas and scaling them into global businesses faster than any before them. The history of China’s uber-rich is short. In the 1980s tycoons figured out how to wrest assets from the state. Zhang Ruimin, a member of a troupe of students who terrorised teachers and intellectuals during Mao’s era, got hold of a state-owned refrigerator business called Haier and turned it into what is now the world’s largest home-appliance maker. The opening of the property market in the 1990s minted another early group of billionaires including Xu Jiayin, founder of Evergrande, a now-collapsed property developer. In the early 2000s industrialists such as Wang Chuanfu of BYD, originally a battery maker, launched a wave of manufacturing firms as China entered the World Trade Organisation. And then in the 2010s the consumer internet produced China’s first generation of tech moguls, including Jack Ma of Alibaba, an e-commerce giant, and Pony Ma of Tencent, maker of WeChat. The founders of ByteDance, creator of TikTok, and Shein, the online fast-fashion firm, who are both now in their early 40s, built their fortunes a few years later.   China’s new generation of tycoons are charting out a different path. More than two-thirds of the young billionaires on Hurun’s rich list made their money from consumer goods or media. Seven created video games; four built tea or coffee businesses. After Mr Liang, the wealthiest is Wang Ning, founder of Pop Mart, maker of the strange-looking Labubu dolls. Only three have struck it rich with AI, including Yang Zhilin, the baby-faced 34-year-old founder of Moonshot AI, which on July 17th released a new model that is said to rival the best from America. No property mogul makes the list—perhaps unsurprisingly, given how the sector has fared in recent years.   The businesses built by China’s new billionaires are also far more global than in the past. It once took a decade or more for Chinese entrepreneurs to figure out foreign expansion. No longer. Zhang Junjie, the 33-year-old founder of Chagee, a milk-tea shop, launched his firm in 2017 and opened his first overseas shop just two years later. The company is now listed in New York and operating in nine countries.   The new tycoons’ heavy reliance on foreign markets partly reflects the slump in consumer spending at home. Around 80% of sales at Dreame, a consumer-electronics brand founded by 39-year-old Yu Hao, come from overseas. Insta360, which sells small waterproof cameras, is almost as reliant on foreign demand. Pop Mart made more than $2bn in sales outside China last year, about 40% of its total. Chagee’s sales at home have slumped, but overseas they are growing healthily.   Attitudes to work are also different among this cohort. A controversial part of Chinese business culture for the past decade or so has been the so-called “996” routine, working from 9am to 9pm six days a week. The idea was championed by Alibaba’s Mr Ma and embraced by other local tech giants. At Pinduoduo, another e-commerce platform, young white-collar staff have died from overwork on their way home from the office. China’s new entrepreneurs, by contrast, have adopted a more relaxed approach to management. Many are relatively flexible about working times. Liu Wei, the 39-year-old billionaire co-founder of miHoYo, a gaming firm, has eschewed presenteeism. He has overseen what the company calls its “happy work, steady growth” approach, which prioritises mental health (though one employee also recently died from complications related to overwork, suggesting that cultural norms are hard to shift). DeepSeek’s Mr Liang has reportedly said that the human brain can concentrate only for 6-8 hours a day, and that overwork leads to mistakes.   The willingness of workers to put in longer hours than anyone else in the world is what has built the Chinese economy over the past 30 years, huffs one private-equity investor, who recently visited an up-and-coming consumer brand and did not like what he saw. Some blame the relatively laid-back attitude of China’s young founders on cushy upbringings. Gone are the tales told by earlier tycoons of struggling for survival amid the country’s political and economic convulsions. Among China’s young billionaires, only Mr Liang and Mr Zhang of Chagee grew up poor. The three founders of miHoYo originally bonded over Japanese cartoons and formed their company as a celebration of gaming and comics. The founder of MiniMax, an AI lab, is so fond of a multi-player game called “Dota 2” that his staff refer to him as “IO”, a character from it.   To get rich is perilous   It is not all rosy for China’s young billionaires. The relationship between China’s government and its wealthiest people has deteriorated over the past few years. Older magnates once viewed themselves as part of a nation-building project. The property sector, which contributed up to a quarter of China’s GDP at its peak, was a collaborative effort between private entrepreneurs and local officials. Mr Ma was instrumental in building China’s digital economy. But the central government has now punished many prominent figures in these areas. Leverage has been sucked out of the property industry. Mr Ma’s business empire became the subject of a state crackdown in 2020 after he criticized regulators. Xi Jinping, China’s ruler, has sought to tamp down inequality under the slogan of “common prosperity”.   In 2021 Zhong Shanshan, the 71-year-old founder of Nongfu, a bottled-water company, reached a net worth of nearly $100bn. No one has since come close, including Mr. Zhong, who has seen his wealth plunge to around $39bn as the share price of his company has slumped. Mr. Ma is worth roughly a fifth as much as in 2021, and many other internet moguls have suffered similarly. Still, they have fared better than property magnates, a number of whom have been detained.   This environment has created a sense of disillusionment among China’s young billionaires. Many are eager to engage in industrial policy but are given little opportunity to do so, notes Steven Hai of Xi’an Jiaotong-Liverpool University. Their global ambitions also leave them exposed to souring relations between China and America. Xiao Hong, the 33-year-old founder of Manus, another AI firm, would have become a billionaire earlier this year if the Chinese government had not stopped him from selling his company to Meta, an American tech giant.   These innovators live with the sense that the businesses they have built could be derailed at any time by sudden policy changes in China or America, notes Rupert Hoogewerf of Hurun. By shunning publicity they seem to believe they can survive for longer, he says. Most young Chinese billionaires have never spoken to foreign media. (None would speak to The Economist.) Many avoid the Chinese press as well. The result is a generation of tycoons that is both China’s most global yet—and its most enigmatic. ■ 中國神秘的新億萬富豪正在征服世界，卻拒絕接受媒體採訪 2026年7月19日｜上海｜閱讀時間：7分鐘 去年年初，當中國人工智慧實驗室 DeepSeek 讓西方競爭對手陷入恐慌時，外界幾乎對它的創辦人梁文鋒一無所知。 如今，梁文鋒依然是一個相當神秘的人物，但至少有一件事已經很清楚：隨著 DeepSeek 即將完成一輪募資，公司估值達到 710億美元，其創辦人的個人財富也暴增至大約 380億美元。這讓他的身家遠遠超過 Anthropic 的 達里奧・阿莫迪（Dario Amodei），以及 OpenAI 的 山姆・奧特曼（Sam Altman）。 梁文鋒不會出現在各種商業會議或產業論壇的巡迴活動上。 他從未在中國電視節目上公開現場露面，而網路上流傳的照片也寥寥可數，能看到的通常是他那副不修邊幅、身材瘦削的模樣。 這讓他與中國新一代的年輕億萬富豪十分相似。 中國正在大量培養這一代年輕富豪，數量僅次於美國。 根據追蹤全球富豪財富的胡潤研究院（Hurun），中國至少有 29名40歲以下的白手起家億萬富豪。 如果把今年某個時候即將滿41歲的梁文鋒也算進去，總數就來到 30人，比去年多了9人。 中國的新一代企業大亨，和他們的前輩有著非常明顯的不同。 他們拋棄了房地產投機，也不再熱衷於和官員吃豪華宴席；取而代之的，是打電玩、聚在一起聊日本動漫。 但另一方面，他們也非常擅長把充滿想像力的點子，迅速擴大成為全球性的企業，而且速度可能超越過去任何一代中國企業家。 中國超級富豪的歷史其實很短 中國超級富豪的歷史其實相當短暫。 到了1980年代，一些企業家開始摸索如何從國家手中取得資產。 張瑞敏就是其中之一。 在毛澤東時代，他曾是那群恐嚇教師與知識分子的學生組織成員之一。後來，他取得了一家名為 海爾（Haier）的國營冰箱企業，並將它發展成如今全球最大的家電製造商。 1990年代，房地產市場開放，又造就了另一批早期億萬富豪，其中包括許家印，也就是如今已經倒閉的房地產開發商恆大集團創辦人。 進入2000年代初期，隨著中國加入世界貿易組織，一批製造業企業家掀起了新一波企業創業潮。 其中包括 比亞迪（BYD）的王傳福。比亞迪最初是一家電池製造商。 接著，到了2010年代，消費型網際網路產業又孕育出中國第一代科技巨頭。 其中包括電子商務巨頭阿里巴巴的馬雲，以及打造微信的騰訊創辦人馬化騰。 幾年後，字節跳動（ByteDance）和Shein的創辦人也建立起自己的龐大財富。 字節跳動打造了 TikTok，而 Shein 則是一家主打線上快時尚的企業。 如今，這兩家公司的創辦人都已經40歲出頭。 新一代富豪正在走出不同的道路 中國新一代企業大亨正在開闢一條不同的道路。 根據胡潤富豪榜，年輕億萬富豪中，有超過三分之二靠消費品或媒體產業致富。 其中有7人靠開發電子遊戲創造財富；4人則建立了茶飲或咖啡事業。 排在梁文鋒之後、身家第二高的是王寧，他是潮玩公司 泡泡瑪特（Pop Mart）的創辦人，而泡泡瑪特最知名的產品，就是外型奇特的 Labubu 娃娃。 只有3人靠人工智慧致富，其中包括34歲、長著一張娃娃臉的 楊植麟。 他是 AI 公司 Moonshot AI 的創辦人。7月17日，這家公司推出了一個新模型，據稱可以與美國最頂尖的 AI 模型抗衡。 至於房地產大亨，則一個都沒有進入這份榜單。 考慮到中國房地產業近年的表現，這或許並不令人意外。 他們從一開始就把目光放向全球 中國新一代億萬富豪所建立的企業，也比過去更加全球化。 過去，中國企業家往往需要花上10年甚至更久，才能摸索出進軍海外市場的方法。 現在已經不同了。 張俊傑今年33歲，是奶茶品牌 霸王茶姬（Chagee）的創辦人。 他在2017年創立公司，僅僅兩年後，就開設了第一家海外門市。 如今，這家公司已在紐約上市，並且已經進入9個國家。 新一代企業大亨高度依賴海外市場，部分原因其實也反映了中國國內消費支出的低迷。 以消費電子品牌 追覓（Dreame）為例。 這家公司由39歲的俞浩創立，大約 80%的銷售額來自海外市場。 銷售小型防水相機的 Insta360，對海外需求的依賴程度也幾乎一樣高。 泡泡瑪特去年在中國以外市場創造了超過 20億美元的銷售額，約占公司總營收的40%。 霸王茶姬在中國國內的銷售額已經下滑，但海外市場卻仍然保持健康成長。 他們對工作的態度也不一樣 這一代年輕富豪對工作的態度，也與前人有所不同。 過去十年左右，中國商業文化中一個極具爭議的現象，就是所謂的 「996」工作制度。 也就是每天早上9點工作到晚上9點，一週工作6天。 這種工作模式曾受到阿里巴巴馬雲的推崇，也被其他中國科技巨頭所採用。 在另一家電子商務平台拼多多，甚至曾有年輕白領因過度工作，在下班回家的路上死亡。 相比之下，中國的新一代企業家採取了更加輕鬆的管理方式。 許多人對工作時間相對彈性。 39歲的億萬富豪劉偉是遊戲公司 米哈遊（miHoYo）的共同創辦人之一。 他並不鼓勵員工只是為了「被看見」而長時間待在辦公室。 他推動公司所稱的 「快樂工作，穩定成長」理念，把心理健康放在重要位置。 不過，最近仍有一名員工因與過勞有關的併發症死亡，這也顯示，要改變根深蒂固的文化規範並不容易。 據報導，DeepSeek 的梁文鋒曾表示，人類的大腦一天只能維持大約 6到8小時的高度專注，而過度工作只會導致更多錯誤。 「正是比世界上任何人都更願意加班，才打造了中國經濟」 一名最近造訪某個新興消費品牌的私募股權投資人對此頗有微詞。 他表示，過去30年來，正是中國人願意比世界上任何其他地方的人投入更長工時，才建立起了中國經濟。 但他最近參訪一家正在崛起的消費品牌時，卻不喜歡自己看到的情況。 有些人則把中國年輕創辦人相對悠閒的工作態度，歸因於他們從小過著較為優渥的生活。 早期企業大亨經常講述的那種故事——在中國劇烈的政治與經濟動盪中努力求生——如今已經逐漸消失。 在中國年輕億萬富豪當中，只有梁文鋒以及霸王茶姬的張俊傑出身貧困。 米哈遊的3位創辦人，最初就是因為喜歡日本動漫而結識，後來更以遊戲與漫畫為共同熱情創辦了公司。 AI 實驗室 MiniMax 的創辦人則非常喜歡多人線上遊戲 《Dota 2》。 他的員工甚至以遊戲中的一個角色 「IO」來稱呼他。 致富並非沒有風險 中國年輕億萬富豪的故事，並非一切都如此美好。 過去幾年，中國政府與最富有階層之間的關係已經惡化。 早期的企業大亨曾經把自己視為國家建設計畫的一部分。 房地產業在高峰時期曾占中國 GDP 的四分之一，這個產業曾是民營企業家與地方官員共同合作的結果。 馬雲也曾在建立中國數位經濟的過程中扮演重要角色。 但如今，中國中央政府已經對這些領域中的許多知名人物採取懲罰性措施。 房地產業的槓桿資金已被大幅抽離。 2020年，馬雲批評金融監管機構之後，他所建立的商業帝國便成為政府整頓的對象。 中國領導人習近平則以「共同富裕」為口號，試圖降低社會貧富差距。 2021年，71歲的農夫山泉創辦人鍾睒睒，身家一度接近 1000億美元。 此後，再也沒有人接近這個數字，包括鍾睒睒本人。 隨著農夫山泉股價下跌，他的財富已縮水至大約 390億美元。 馬雲如今的身家，大約只有2021年時的五分之一。 許多其他網路產業巨頭也遭遇類似的財富縮水。 不過，他們的處境仍比房地產大亨好一些，因為後者當中已有不少人遭到拘留。 年輕富豪開始感到失望 這樣的環境，也讓中國年輕億萬富豪產生了一種失落感。 西交利物浦大學（Xi’an Jiaotong-Liverpool University）的 Steven Hai 指出，許多年輕富豪其實非常希望參與中國的產業政策，但政府給他們的機會卻相當有限。 另一方面，他們的全球化野心，也讓自己暴露在中美關係惡化的風險之中。 肖弘今年33歲，是另一家 AI 公司 Manus 的創辦人。 如果不是中國政府阻止他把公司出售給美國科技巨頭 Meta，他早在今年稍早就可能成為億萬富豪。 胡潤研究院的 胡潤（Rupert Hoogewerf）指出，這些創新者始終生活在一種不安之中： 他們辛苦建立起來的企業，隨時都有可能因為中國或美國突然改變政策，而被迫偏離原本的發展軌道。 他認為，這些企業家選擇避開媒體曝光，似乎是因為他們相信，這樣可以讓自己的企業存活得更久。 大多數中國年輕億萬富豪從未接受過外國媒體採訪。 《經濟學人》也沒有一個人願意接受採訪。 許多人甚至也刻意避開中國媒體。 最終形成了一個既是中國最全球化的一代企業大亨，同時也是最神秘莫測的一代。 -- Hosting provided by SoundOn

The six stages of holiday-making Two weeks off, but how many days of relaxation? Aug 13th 2026|4 min read   It’s summer, in the northern hemisphere at least, which means emptier offices and fuller airports. In theory, a holiday is a time to properly relax. In practice, there are several stages to pass through before that moment arrives, and several more to experience before you are fully immersed in work again.   1.    The preparation phase. You badly need a holiday. In your final couple of days before leaving, you complete your most urgent tasks, like tidying your desk. You tell people on your team that you trust them completely and in the same breath, that they should contact you if anything comes up, no matter how trivial. As a result, they don’t feel trusted at all. You think about setting up an out-of-office message and decide not to: it’s really not a problem to check your phone occasionally. As you leave the office, you think you feel a cold coming on. 2.    The relax-like-hell phase. You spend the first few days approaching the holiday like it’s work. You spend loads of time asking AI models for places to go that are off the beaten track. In a typical morning, you drive two hours to eat breakfast at a place recommended by ChatGPT which turns out to be closed, and then visit a museum devoted to the ancient craft of curing fish.   At lunch, you drink wine and then fall asleep. You think of this period of unconsciousness as time that could have been spent relaxing. You are acutely aware of time passing, your days of freedom slipping away. You pick up a book and instead of reading it, worry that you could be enjoying your leisure in even more leisurely ways. You feel very tense for someone on holiday.   3.    The letting-go phase. Meanwhile, you check your email every so often, and intervene in things you don’t need to. You get a couple of WhatsApp messages and are asked to join a call, and feel both irritated and pleased. Can’t you do it without me? You can’t do it without me! As you triage messages, sorting them into things that have to be dealt with quickly and those that can wait, you gradually realise just how much of what goes on each week a) doesn’t require you at all and b) has no lasting value. You start to ignore most messages; if it’s important, they will find a way to contact you. You switch on your out-of-office message.   4.    The delusional phase. Finally, you start to relax. You lie in a little longer. You don’t look at your phone quite as much. You realise that back in the office the weekly sales meeting is happening right at this very minute, and you laugh out loud at the smallness of your old life. Fairly soon, you start to lose your grip on reality. You think about moving to wherever it is you are, forgetting that you are enjoying yourself because the weather is nice and you are not working. You start trying to figure out how soon you can retire. You fantasise about working outdoors, perhaps in sea-kayaking or doing something artisanal to a dry-stone wall.   5.    The motivated phase. The end of the holiday is fast approaching. You realise that sea-kayaking gives you blisters and that the things you are best equipped to do don’t involve dry-stone walls. You start to look at emails again, but only the ones that matter. You feel motivated to return. You have seen how much time is spent on unimportant things and are determined to prioritise much more ruthlessly. You have some really good ideas about how to improve the weekly sales meeting.   6.    The recidivism phase. You return to the office full of vim. You look and feel healthier than your colleagues. Then you have the same conversation 800 times. “How was your holiday?” “Wonderful.” “Where did you go?” [Insert answer.] “What did you do there?” [Insert details.] You soon realise that no one is interested in where you went and start to make up responses. “Where did you go?” “Wakanda.” “What did you do?” “We visited the vibranium mines.” “Oh, I’ve always wanted to do that.”   It’s nice to see people. It’s good to have a sense of purpose. But you quickly revert to old habits. An empty calendar feels less like ruthless prioritisation and more like a public display of redundancy; the slots soon fill up again. When emails come in, you answer them indiscriminately. Your desk becomes more cluttered. You attend the weekly sales meeting and dimly recall that you once had an idea for improving it. Colleagues come back from their own breaks and you ask how their holidays were. You are not listening to the answers. They seem so well, so energetic. You think you feel another cold coming on. You badly need a holiday. ■ 度假的六個階段 兩週假期，但真正放鬆的又有幾天？ 2026年8月13日 | 閱讀時間 4 分鐘 夏天到了——至少在北半球是如此——這意味著辦公室空了，機場卻擠滿了人。理論上，度假是一個能讓人徹底放鬆的時刻；但在現實中，在那個時刻到來之前，你必須先經歷幾個階段，而在你重新完全沉浸於工作之前，還有另外幾個階段等著你。 準備階段。 你急需放個假。在離辦公室前的最後兩天，你完成了最緊迫的工作，比如整理乾淨你的辦公桌。你告訴團隊成員你完全信任他們，但緊接著又說，如果有任何事情發生，不管多麼微不足道，都可以隨時聯繫你。結果，他們根本感覺不到自己被信任。你考慮要設定「不在辦公室」的自動回覆信件，但最後決定不設：偶爾看一眼手機真的沒什麼大不了的。當你離開辦公室時，你覺得自己好像有點感冒了。 拼命放鬆階段。 你把前幾天的假期過得像是在上班一樣。你花了大量時間詢問 AI 模型有哪些遠離喧囂的冷門景點。在一個典型的早晨，你開了兩個小時的車去吃一家 ChatGPT 推薦的早餐，結果發現它根本沒開，接著你又去參觀了一間展示古代醃魚工藝的博物館。 午餐時，你喝了葡萄酒，然後睡著了。你把這段無意識的時間視為「本可用於放鬆」的時間。你極度敏感地意識到時間的流逝，自由的日子正在悄悄溜走。你拿起一本書，但沒有閱讀，反而擔心自己本可以利用更悠閒的方式來享受閒暇時光。作為一個在度假的人，你感到非常緊張。 放手階段。 與此同時，你隔三差五地檢查郵件，干預一些你根本不需要插手的事情。你收到幾條 WhatsApp 訊息，並被邀請加入視訊會議，這讓你感到既煩躁又高興。沒有我你們就辦不成事嗎？你們還真的沒我不行！當你對訊息進行分類 triage，將它們分為「必須快速處理」和「可以等等再說」時，你逐漸意識到每週發生的事情中有多少是：a) 根本不需要你；b) 完全沒有長遠價值。你開始忽視大多數訊息；如果真的重要，他們總會找到方法聯繫你。你終於開啟了「不在辦公室」的自動回覆。 幻想階段。 終於，你開始放鬆了。你睡得稍微久了一點，看手機的次數也沒那麼頻繁了。你意識到此時此刻，辦公室裡正召開著每週銷售會議，你對自己過去那渺小的生活發出陣陣會心的冷笑。很快地，你開始脫離現實。你想著搬到你現在度假的地方生活，卻忘了你之所以開心是因為天氣很好且你不用工作。你開始嘗試計算自己多久之後可以退休。你幻想著在戶外工作，也許是划海上皮划艇，或是對著乾砌石牆做些手工藝活。 動力階段。 假期的尾聲迅速逼近。你意識到划皮划艇會讓你起水泡，而你最擅長做的事情根本與乾砌石牆無關。你重新開始看郵件，但只看那些真正重要的。你感到有一股返回工作的動力。你看到了有多少時間被浪費在不重要的事情上，並決心要更狠心地排出優先順序。對於如何改進每週銷售會議，你產生了一些非常棒的想法。 舊病復發階段。 你滿懷活力地回到辦公室。你看起來、感覺起來都比同事們更健康。然後你把同一段對話重複了 800 次。「假期過得怎麼樣？」「太棒了。」「去哪裡了？」（插入答案）「在那裡做了什麼？」（插入細節）。你很快意識到根本沒有人對你去哪裡感興趣，並開始編造回答。「你去哪裡了？」「瓦干達（Wakanda）。」「做了什麼？」「我們去參觀了汎合金（Vibranium）礦山。」「噢，我一直想去那裡。」 能見到大家真好，擁有目標感也很不錯。但你很快就重蹈覆轍。空白的行事曆看起來不再像是狠心排序優先順序的成果，反而更像是公開展示自己是個贅員；日程很快又被排滿了。當郵件寄來時，你不加篩選地一一回覆。你的辦公桌變得越來越凌亂。你參加每週銷售會議，隱約想起自己曾經有一個改進它的好主意。同事們陸續休假歸來，你問他們假期過得如何。你根本沒有在聽答案。他們看起來那麼健康、那麼精力充沛。你覺得自己好像又有點感冒了。你急需放個假。■ -- Hosting provided by SoundOn

How much would you pay for a smarter baby? Startups claim they can screen embryos for everything from genetic diseases to IQ. Not everyone is convinced   Aug 10th 2026|San Francisco and New York|9 min read   Late last year New Yorkers noticed a series of strange advertisements on the subway. “Have your best baby,” one promised. “IQ is 50% genetic,” noted another. Predictably, they stirred controversy. Yet Kian Sadeghi of Nucleus, the embryo-screening startup behind them, was pleased. “This is a mass-market product. What’s a better way of showing that it’s a mass-market product than the subway?”   Nucleus is one of a growing number of Silicon Valley startups pushing the nascent technology. Noor Siddiqui of Orchid, a competitor, likes to say that “sex is for fun; embryo screening is for babies”. Herasight, another rival, began offering a screening service last year that allows parents to examine an embryo’s risk of various diseases, and also estimates of its height, intelligence and longevity—at a cost of up to $50,000. Nucleus adds hair and eye colour to the mix. The burgeoning industry aspires to reshape the way humans are made.   Not everyone is pleased. In many jurisdictions, including Britain, screening embryos for sex, let alone for polygenic conditions such as hypertension or physical traits such as height, is illegal. Many geneticists and doctors’ groups are sceptical that the technology works. Others say it is unethical. In America, where regulation is looser, religious conservatives and adherents of the “Make America Healthy Again” movement are vehemently opposed.   The startups’ visions may seem otherworldly. But change is coming fast. Cash is pouring into fertility research, leading to potentially rapid advances in genetic testing. Polygenic screening, particularly for common diseases, is popular among Americans, nearly three-quarters of whom say they would use it if they were already undergoing in-vitro fertilisation (IVF). A fight is brewing over the future of these technologies—and the businesses promoting them.   For the moment, only a tiny share of babies are born via IVF, as is necessary for embryo screening. In 2024, the most recent year for which figures are available, 100,000 American babies were created using the process, representing 2.8% of the total. David Sable, an investor and formerly a practising reproductive endocrinologist, says the global fertility market is commonly estimated to be worth just $25-30bn a year.   Yet Silicon Valley’s interest in fertility could expand access to IVF. Several years ago Peter Thiel and Elon Musk began to speak about low birth rates as a threat to America, and started putting money into fertility startups. They were joined by other investors. American venture-capital (VC) investments in fertility enterprises nearly doubled between 2019 and 2022, jumping from $254m to $496m, according to PitchBook, a data provider, even if they have since slowed a bit.   Many of these startups are focused on making IVF cheaper. At present, only 15 American states require the treatment to be included in insurers’ maternity benefits, meaning most Americans pay out of pocket. To reduce costs, a number of startups are selling fertility insurance and payment plans. Nader al-Salim of Gaia, one such firm, says its services save patients an average of $15,000 over the entire fertility process, largely by pooling risk.   Dr Sable reckons that if IVF does become much more widely available, the embryo-screening market could grow to $400-500bn annually. His estimate assumes the cost will fall to around $15,000 for the entire process; currently, just one cycle of IVF in America costs $15,000-20,000. Dr Sable reckons that prices will drop thanks to technological innovations which may increase success rates, allowing people to go through fewer cycles. Falling labour costs, as a result of automation and economies of scale, as well as greater standardisation should also help.   Some screening is already common among IVF patients. One example is PGT-A, which checks embryos have the right number of chromosomes, and which some studies claim lowers the miscarriage risk (many experts are sceptical). Certain clinics also offer PGT-M, which checks for disorders caused by a single gene, such as cystic fibrosis or Huntington’s. Now startups are adding a third screen: PGT-P, which scores embryos for conditions and traits determined by multiple genes, potentially ranging from certain cancers to eye colour and IQ. Demand for such tests could be substantial. A survey published in 2023 in *Science *found that 43% of all IVF participants would use polygenic screening to increase their child’s chance of getting into an elite university, provided it were safe and free. A third said the same for genetic editing.   Consider Arthur Zey, a software engineer, and his husband Chase Popp, a teacher. Mr Zey says he has always been a “tech nerd” and wanted to apply his skills to the process of having a child. “Most of my professional career I’ve been a product manager in tech, and so in some ways this is just like another product to manage,” he says. Earlier this year the couple had a son using an embryo selected from a group of six on the basis of the results of polygenic screening provided by Herasight. That allowed them to see risk scores for monogenic diseases and more than a dozen polygenic diseases. The startup also gave them predicted ranges for IQ and height.   Messrs Zey and Popp say they hoped to maximise their son’s lifespan, rather than his height or IQ. Yet Mr Zey knows there are advantages to being tall and smart. “All I could do is look at the results for my six embryos and say: ‘Okay, it kind of seems like this one is better.’”   The burden of perfection   For now, Messrs Zey and Popp are unusual. Commercial PGT-P screening is a niche industry. Startups like Herasight cater to a small population of typically very rich people, mostly in America. Even so, that has not stopped them from drawing plenty of criticism.   Some concerns the underlying science. Polygenic risk scores are based on genome-wide association studies (GWAS), which use data from large populations to try to discern associations between genetic variants and specific conditions and traits. Companies that offer screening claim that comparing embryos’ genomes to risk scores derived from these databases, sometimes with adjustments based on ancestry and family history, allows them to come up with individual assessments.   That is disputed, however. Although selecting on the basis of polygenic risk scores works across many offspring, as in animal breeding, it is less sure to have the desired result for a single child, notes Kevin Mitchell of Trinity College Dublin. For example, an embryo’s genome may have a single rare variant that has a large effect on its likelihood of developing a certain condition, but which would be unlikely to surface in population-level correlations.   Herasight attempts to deal with this issue by validating its risk scores using sibling data from GWAS datasets to ensure its models retain predictive value within a single family. It claims that, if it tests ten embryos for a couple where both parents have type-2 diabetes, it can reduce their child’s absolute risk of developing the condition by 12-20 percentage points, from a baseline of 40-60%. Herasight has been critical of competitors, particularly Nucleus, for not providing more detailed public information about their methodology. (Nucleus, for its part, says it is working constantly to ensure its scores are accurate.)   Some also argue that the magnitude of the benefits from embryo screening are overstated. If parents picked from ten viable embryos, the median boost to their child’s height would be 2.9cm, according to a paper in *Cell *in 2019, based on technology at the time. If they selected for IQ, the gain would be three points—unlikely to be life-changing. Moreover, for those picking among fewer than five embryos, the expected gains drop precipitously. Although the average number of viable embryos depends on the patient’s age, in many cases it is closer to two or three from each IVF cycle. Dagan Wells of the University of Oxford adds that, for many conditions, genes are less important than environmental factors such as diet. Selecting against some traits could also have unintended consequences. Genetic material can have different effects in different parts of the body, a phenomenon called pleiotropy. If parents select an embryo intending to lower their child’s risk of a certain disease, they may be neglecting or exacerbating problems for which providers do not (or cannot) offer a score.   Yet many of the technical issues, not least the lack of training data among certain populations, should improve with time. The *Cell *paper estimated that increasing the sample size of the underlying databases could double the median gain in IQ points from polygenic scoring. Or consider the effects of rare genetic variants. As databases grow, artificial-intelligence models may be better able to account for the effects of individual variants. With pleiotropy, some studies suggest that most genetic correlations between pairs of polygenic diseases are either positive—meaning cutting the chance of one decreases the chance of others—or insignificant. At present, the technology is “kind of a promissory note”, says Trinity’s Dr Mitchell. “They say, ‘Trust us, this is a bit shit right now, but it’ll be good soon’.”   If technological problems are resolved, ethical ones will only grow. Disability-rights groups are already wary of PGT-M, worrying it suggests lives with certain genetic conditions are not worth living. The rise in use of PGT-P widens that debate to many more traits. Then there are opponents who are concerned that screening startups are promoting a friendlier version of eugenics. Some worry that embryo editing—on which America has applied a de-facto ban since 2016—could come next.   Industry insiders reject the fretting, sometimes bluntly. Risk scores may not be perfect, but they are “best in class”, says Delian Asparouhov of Founders Fund, a VC firm co-founded by Mr Thiel, which has invested in various fertility startups. “What I hate about some of the traditional health-care industry and genetic counsellors is they’re like, ‘You need to have a Harvard clinical-grade license’” to interpret genetic data. “Fuck that,” he continues. “I just want whatever the best statistical models are. I don’t need a PhD to read this stuff.” (Herasight, Nucleus and Orchid all have genetic counsellors to walk clients through their results.) As for ethics, Mr Asparouhov argues that all parents-to-be engage in a form of genetic selection when they choose a partner. “If this is eugenics, then so is Hinge,” he argues.   Most of the companies offering PGT-P are more circumspect. They say their customers are mainly looking for ways to ensure their children are healthy, rather than to create “super-babies”. Yet their decision to offer scores for intelligence and aesthetic traits suggests they recognise there is a market for the latter.   Arguments over the future of a small cohort of startups may seem like a niche concern. But if IVF gets cheaper and genetic data more abundant, as looks likely, the business of embryo screening will only grow—along with the ferocity of the debate. The nearly 50-year history of the fertility industry is full of technologies that were once contentious, including IVF. Startups like Nucleus, Orchid and Herasight are, in that sense at least, nothing new. ■   為了更聰明的寶寶，你願意付出多少代價？   新創公司聲稱他們可以對胚胎進行篩檢，範圍涵蓋從遺傳疾病到智商的一切。但並非所有人對此深信不疑 2026年8月10日 | 舊金山與紐約 | 閱讀時間 9 分鐘   去年底，紐約人注意到地鐵上出現了一系列奇怪的廣告。「擁有你最好的寶寶，」其中一張承諾道。「智商有 50% 取決於遺傳，」另一張寫著。不出所料，這些廣告引發了爭議。然而，背後這家胚胎篩檢新創公司 Nucleus 的基安·薩德吉（Kian Sadeghi）卻對此感到高興。「這是一個大眾市場產品。有什麼比地鐵更能展示它是一個大眾市場產品的方式呢？」 Nucleus 是越來越多推動這項新興技術的矽谷新創公司之一。競爭對手 Orchid 的努爾·西迪基（Noor Siddiqui）喜歡說：「性愛是為了樂趣；胚胎篩檢是為了寶寶。」另一家對手公司 Herasight 於去年開始提供一項篩檢服務，讓父母能夠檢查胚胎患有各種疾病的風險，並且還能預估其身高、智力與壽命——費用高達 50,000 美元。Nucleus 甚至將髮色和瞳孔顏色也納入考量。這個蓬勃發展的產業渴望重塑人類的誕生方式。   並非所有人都為此感到高興。在許多管轄區（包括英國），篩檢胚胎的性別，更不用說篩檢高血壓等多基因疾病或身高這種身體特徵，都是違法的。許多基因學家和醫生團體對這項技術是否有效持懷疑態度。其他人則表示這不符合倫理。在法規較為鬆散的美國，宗教保守派以及「讓美國再次健康」（Make America Healthy Again）運動的支持者皆對此強烈反對。 這些新創公司的願景可能看起來恍如隔世。但改變正在迅速降臨。資金正源源不絕地湧入生育研究，引領著基因檢測領域潛在的飛速進展。多基因篩檢，特別是用於常見疾病的篩檢，在美國人當中相當受歡迎，如果已經在接受體外受精（IVF，即試管嬰兒），近四分之三的人表示他們會使用它。一場關於這些技術的未來——以及推廣它們的企業——的爭霸戰正在醞釀中。   目前，只有極小比例的嬰兒是透過試管嬰兒出生，而這是進行胚胎篩檢所必需的。在有數據可查的最新年份 2024 年，有 100,000 名美國嬰兒是透過這種方式孕育的，占總數的 2.8%。大衛·塞博（David Sable）是一位投資人，過去曾是執業的生殖內分泌學家，他表示，全球生育市場的年價值普遍估計僅為 250 億至 300 億美元。   然而，矽谷對生育領域的興趣可能會擴大使用試管嬰兒的途徑。幾年前，彼得·提爾（Peter Thiel）和伊隆·馬斯克（Elon Musk）開始將低出生率視為對美國的威脅，並開始將資金投入生育新創公司。其他投資人也加入他們的行列。根據數據提供商 PitchBook 的資料，美國風險投資（VC）對生育企業的投資在 2019 年至 2022 年間幾乎翻倍，從 2.54 億美元躍升至 4.96 億美元，即使此後稍微有所放緩。   這些新創公司中有許多正專注於讓試管嬰兒變得更便宜。目前，僅有 15 個美國州要求將此治療納入保險公司的生育福利中，這意味著大多數美國人需要自費。為了降低成本，許多新創公司正在銷售生育保險和付款計劃。其中一家公司 Gaia 的納德·薩利姆（Nader al-Salim）表示，他們的服務主要透過分散風險，在整個生育過程中平均為患者節省 15,000 美元。 塞博醫師估計，如果試管嬰兒真的變得更加普及，胚胎篩檢市場可能會成長到每年 4000 億至 5000 億美元。他的估算假設整個過程的費用將降至 15,000 美元左右；目前，在美國僅一個試管嬰兒週期就需要花費 15,000 至 20,000 美元。塞博醫師認為，由於技術創新可能會提高成功率，從而使人們能經歷更少的週期，價格將會下降。自動化和規模經濟帶來的勞動力成本下降，以及更高度的標準化，應該也會有所幫助。   某些篩檢在試管嬰兒患者中已經相當普遍。一個例子是 PGT-A，它能檢查胚胎是否擁有正確數量的染色體，且有些研究聲稱這能降低流產風險（許多專家對此持懷疑態度）。某些診所還提供 PGT-M，用於檢查由單一基因引起的疾病，如囊腫性纖維化或亨廷頓氏舞蹈症。   現在，新創公司正在添加第三種篩檢：PGT-P，它能對由多個基因決定的狀況和特徵進行評分，潛在範圍從某些癌症到瞳孔顏色和智商。對此類檢測的需求可能相當龐大。2023 年發表在《科學》（Science）雜誌上的一項調查發現，如果有安全且免費的前提，所有試管嬰兒參與者中有 43% 會使用多基因篩檢來增加孩子進入精英大學的機會。三分之一的人對基因編輯也持相同態度。   想想軟體工程師亞瑟·澤伊（Arthur Zey）和他的丈夫，身為教師的蔡斯·普普（Chase Popp）。澤伊先生說他一直是一個「科技宅」，並希望將自己的技能應用到生育孩子的過程中。「在我的職業生涯的大部分時間裡，我都是科技領域的產品經理，因此在某些方面，這就像是另一個需要管理的產品，」他說。今年早些時候，這對伴侶利用 Herasight 提供多基因篩檢結果，從六個胚胎中挑選出一個，並誕生了一個兒子。這讓他們能夠看到單基因疾病和十幾種多基因疾病的風險評分。這家新創公司還向他們提供了智商和身高的預測範圍。 澤伊先生和普普先生表示，他們希望能最大化兒子的壽命，而不是他的身高或智商。然而，澤伊先生知道身材高大和聰明是有優勢的。「我能做的就是看看我那六個胚胎的結果，然後說：『好吧，看起來這個好像比較好。』」   完美的負擔   目前而言，澤伊先生和普普先生屬於少數。商業化的 PGT-P 篩檢是一個小眾產業。像 Herasight 這樣的新創公司服務於一小群通常非常富有的人群，主要在美國。即便如此，這並沒有阻止他們遭受大量的批評。   有些疑慮關乎背後的基礎科學。多基因風險評分是基於全基因組關聯研究（GWAS），該研究使用來自龐大人群的數據，試圖辨識基因變異與特定狀況和特徵之間的關聯。提供篩檢的公司聲稱，將胚胎的基因組與來自這些資料庫的風險評分進行比較（有時根據祖先和家族史進行調整），可以讓他們得出個體評估結果。   然而，這一點存在爭議。都柏林聖三一學院的凱文·米切爾（Kevin Mitchell）指出，雖然根據多基因風險評分進行選擇在許多後代中有效（如動物育種），但對於單一孩子來說，是否能產生預期結果就不那麼確定了。例如，一個胚胎的基因組可能有一個單一的罕見變異，這對其罹患某種狀況的可能性有很大影響，但這種變異不太可能出現在人口層面的相關性中。   Herasight 試圖透過使用來自 GWAS 資料庫的手足數據驗證其風險評分來解決這個問題，以確保其模型在單一家庭內仍保留預測價值。它聲稱，如果它為一對雙親均患有 2 型糖尿病的夫婦測試十個胚胎，它可以將他們的孩子患該疾病的絕對風險從 40-60% 的基準降低 12-20 個百分點。Herasight 一直對競爭對手（特別是 Nucleus）持批評態度，因為後者沒有就其方法論提供更詳細的公開資訊。（Nucleus 方面則表示，它正不斷努力確保其評分的準確性。）   有些人還認為，胚胎篩檢帶來的好處被誇大了。根據 2019 年發表在《細胞》（Cell）雜誌上的一篇基於當時技術的論文，如果父母從十個可用的胚胎中挑選，對孩子身高的中位數提升將是 2.9 公分。如果他們針對智商進行選擇，獲益將是三點——不太可能改變人生。此外，對於那些在少於五個胚胎中進行選擇的人來說，預期好處會急劇下降。雖然可用胚胎的平均數量取決於患者的年齡，但在許多情況下，每個試管嬰兒週期更接近兩到三個。牛津大學的大衛·威爾斯（Dagan Wells）補充說，對於許多狀況而言，基因的重要性不如飲食等環境因素。   淘汰某些特徵也可能帶來意想不到的後果。基因物質可以在身體的不同部位產生不同的影響，這種現象被稱為「多效性」（pleiotropy）。如果父母選擇一個胚胎是為了降低孩子罹患某種疾病的風險，他們可能會忽視或加重提供者沒有（或無法）提供評分的潛在問題。   然而，許多技術問題（尤其是某些人群中缺乏訓練數據的問題）應該會隨著時間的推移而得到改善。《細胞》雜誌的那篇論文估計，增加基礎資料庫的樣本量，可以使多基因評分帶來的智商點數中位數增益翻倍。或者考慮罕見基因變異的影響。隨著資料庫的擴大，人工智慧模型可能能夠更好地解釋單一變異的影響。對於多效性，一些研究表明，多基因疾病對之間的絕大多數基因相關性要麼是正相關的——意味著降低一種疾病的幾率會降低其他疾病的幾率——要麼是不顯著的。目前，這項技術「有點像一張承諾紙條」，聖三一學院的米切爾醫師說。「他們說，『相信我們，這東西現在有點爛，但很快就會變好了』。」   如果技術問題得到解決，倫理問題只會與日俱增。殘疾權益團體已經對 PGT-M 保持警惕，擔心這意味著患有某些遺傳狀況的生命不值得活下去。PGT-P 使用率的上升將這場爭論擴大到了更多的特徵上。還有一些反對者擔心，篩檢新創公司正在推廣一種更友善版本的優生學。有些人擔心胚胎編輯——美國自 2016 年以來已實施事實上的禁令——可能會是下一步。 產業內部人士拒絕這種擔憂，有時甚至相當直白。風險評分可能不完美，但它們是「同類最佳的」， Founders Fund（由提爾先生共同創立、已投資多家生育新創公司的風險投資公司）的德里安·阿斯帕魯霍夫（Delian Asparouhov）說。「我討厭某些傳統醫療保健產業和基因諮詢師的地方在於，他們會說，『你需要擁有哈佛臨床級許可證』」才能解讀基因數據。「去他的，」他繼續道。「我只要最好的統計模型。我不需要博士學位來讀這些東西。」（Herasight、Nucleus 和 Orchid 都設有基因諮詢師來引導客戶了解他們的結果。）至於倫理，阿斯帕魯霍夫先生認為，所有準父母在選擇伴侶時都在進行某種形式的基因選擇。「如果這是優生學，那麼 Hinge（交友軟體）也是了，」他爭辯道。   大多數提供 PGT-P 的公司都較為謹慎。他們表示，他們的客戶主要是尋求確保孩子健康的方法，而不是為了創造「超級寶寶」。然而，他們決定提供智力和審美特徵評分的事實表明，他們意識到後者存在市場。   關於一小群新創公司未來的爭論看似是一個小眾關注點。但是，如果試管嬰兒變得更便宜，基因數據變得更豐富（這看起來很有可能），胚胎篩檢業務只會成長——隨之而來的還有討論熱度。生育產業近 50 年的歷史充斥著曾經備受爭議的技術，包括試管嬰兒本身。像 Nucleus、Orchid 和 Herasight 這樣的新創公司，至少在這個意義上，並非什麼新鮮事。 -- Hosting provided by SoundOn

South Korea’s stock-market boom is collapsing** spectacularly**   It may be a harbinger   Jul 29th 2026|5 min read   America’s giant technology companies have gone from printing money to incinerating it. The market reckons Alphabet, Amazon, Meta and Microsoft will collectively report negative free cashflows next year, owing to their extravagant spending on data centres. Yet real fortunes are being made today from Silicon Valley’s gamble on the future.   Three companies are expected to surpass an astonishing $200bn of free cashflow apiece next year. One is Nvidia, an American darling that designs the advanced processors needed to run artificial-intelligence models. The other two are Samsung and SK Hynix, a pair of South Korean outfits that dominate the market for the more mundane memory chips in phones, laptops and, increasingly, data centres. Their windfall represents one of the greatest cross-border transfers of cash in corporate history.   Now a nation newly enamoured of stock markets is discovering that shares can go down as well as up. Since peaking in June the stock prices of Samsung and SK Hynix have crashed by 41% and 52% respectively, wiping $1.2trn from South Korea’s highly concentrated stock market. Those of SK Hynix fell by more than a fifth this week as investors digested a mere six-fold increase in its quarterly operating profit, compared with the year before. Samsung’s shares have not been this volatile since the dotcom era.   South Korea is a great experiment in mixing the extremes of state capitalism and market speculation. Last year, after a failed coup, the country’s president attempted to cheer his country by talking up the stock market (sound familiar?). The KOSpI, its main index, speedily hit a level the government had set as a target.   Ordinary South Koreans withdrew capital from insurers and banks to invest in stocks. Workers threatened industrial action to secure Wall Street-sized bonuses. Those at SK Hynix and Samsung’s chip division settled for an extraordinary 10% of operating profits. At SK Hynix that could amount to $500,000 a person this year, and $800,000 next year. (Goldman Sachs paid a meagre average of $400,000 in total compensation to its employees last year.)   Now a government that just weeks ago was embracing markets is being forced to apologise for them. It was only in April that regulators approved leveraged exchange-traded funds (ETFs) tied to individual South Korean stocks. These funds use derivatives to give investors a multiple of the daily return of a share. For example, if SK Hynix is up by 1%, an investor might receive 2%; or if it is down by 10%, an investor might jump out of a window.   When the selling began in June broker accounts faced margin calls and were shut down by the thousand, causing regulators to suspend new ETF launches. At an emergency meeting on July 29th politicians promised to severely limit the use of these funds. One compared the market to a casino. Shares in the country’s department stores are crashing as fast as those of its chipmakers, in anticipation of the handbags that will now remain on the shelves.   The trillion-dollar question for investors in South Korea is when the price of memory will peak. Bulls argue that data-centre construction and robotics mean demand for memory will outstrip supply into the 2030s. But the increasingly common view is that the same thing will happen to memory chips as has happened to every commodity in every commodity cycle in history—that new supply will, within a year or two, cause prices and profits to fall. Despite their vertiginous rise, Samsung and SK Hynix have on average traded only at single-digit multiples of forward earnings during the past year, suggesting that investors think their profits are at or near their highest point.   Perhaps it is inevitable that the first country to taste the riches promised by AI should be the first to have them taken away. No industry can survive if all the rewards are concentrated in one corner of the supply chain. Yet the dangers facing South Korea’s memory giants are also lurking elsewhere in the AI landscape.   First is the dreaded “commoditisation”. The sales of less advanced memory chips by SK Hynix and Samsung are threatened by the rising market share of Chinese competitors (one of which went public this week to great fanfare). So too are America’s model-makers, which face a potentially existential threat from Chinese open-source models.   Second is the risk that supposedly ironclad spending commitments from customers turn out to be no such thing. Both South Korean firms have made much of their “long-term agreements” with buyers, arguing that pre-payment and price certainty gives them visibility on profits for years to come. Yet judicious investors have begun to fear that customers will find ways of wriggling out of deals when it is advantageous for them to do so. Investors also worry how airtight lease agreements for data centres would prove if model-makers tried to get out of them in a bust. The piles of debt tied to Meta’s giant off-balance-sheet data centre have drawn particular scrutiny.   Picked, shovelled and quartered Nvidia has played an important role in stabilising this precarious situation. Jensen Huang, its leather-jacketed chief executive, has something to say on every debate: last week he published an influential letter in defence of open-source models. But his cheque-book is more important. Mr Huang has constructed a vest web of financial ties that keep the party going. This week it was reported that his company is in talks to guarantee $250bn of commitments from OpenAI, maker of ChatGPT, to lease a data centre. More remarkable still is the vague “$500bn” partnership with SK Hynix it announced over the weekend, and how little that seemed to matter to investors this week. Like a central banker meddling in markets, the more Mr Huang intervenes the more nervous investors become. The eventual bust will be all the worse for it. ■     韓國股市熱潮慘烈崩盤，這恐怕只是個前兆 2026 年 7 月 29 日 | 閱讀時間約 5 分鐘 美國的科技巨頭們，已經從過去的「印鈔機」變成了「燒錢機」。市場預估，由於在資料中心上的揮霍式支出，Alphabet、亞馬遜（Amazon）、Meta 與微軟（Microsoft）明年的自由現金流合計將轉為負值。然而，光是憑藉著對矽谷這場未來豪賭的押注，如今就有人賺得盆滿缽滿。 預計明年將有三家公司各自突破 2,000 億美元自由現金流的驚人紀錄。其中一家是美國的寵兒輝達（Nvidia），負責設計運行人工智慧（AI）模型所需的高階處理器。另外兩家則是韓國的三星（Samsung）與 SK 海力士（SK Hynix），這對雙雄壟斷了手機、筆電以及日益擴增的資料中心所需之一般記憶體晶片市場。他們的這筆意外之財，堪稱企業史上最偉大的跨國財富轉移之一。 然而，這個新近狂熱於股票市場的國家，如今正赫然發現：股價能漲，自然也能跌。自今年 6 月觸頂以來，三星與 SK 海力士的股價已分別暴跌 41% 與 52%，從高度集中的韓國股市中抹去了 1.2 兆美元 的市值。本周，儘管 SK 海力士公佈的單季營業利潤較去年同期「僅」暴增了六倍，投資人消化此消息後，其股價仍重挫超過五分之一。三星的股價波動程度，更是達到了自網路泡沫時代以來未曾見過的高峰。 韓國無疑是一場將「國家資本主義」與「市場投機」極端融合的大型實驗。去年，在一場失敗的政變之後，該國總統試圖透過拉抬股市來振奮人心（聽起來很熟悉吧？）。韓國綜合股價指數（KOSPI「高斯皮」。英文讀音近似 KOS-pee）隨即迅速衝上了政府設定的目標關卡。 韓國的普通民眾紛紛從保險公司和銀行提領資金轉而投向股市。勞工們甚至以發動罷工相威脅，要求獲取比照華爾街等級的年終獎金。SK 海力士與三星晶片部門的員工，最終接受了高達營業利潤 10% 的分紅方案。在 SK 海力士，這意味著每人今年可領到 50 萬美元，明年甚至上看 80 萬美元。（相較之下，高盛去年支付給員工的總薪酬平均僅為微薄的 40 萬美元。） 然而，幾週前還在擁抱市場的政府，現在卻不得不為此出面道歉。就在今年 4 月，監管機構才剛批准了連結韓國單一股票的槓桿型 ETF。這些基金利用衍生性金融商品，為投資人提供股票單日報酬率的倍數收益。舉例來說，若 SK 海力士上漲 1%，投資人可能獲利 2%；但若下跌 10%，投資人可能就想跳樓了。 當 6 月拋售潮開始時，經紀戶頭面臨追繳保證金（margin call），數以千計的帳戶遭到強制平倉(強制平倉（又稱爆倉或斷頭）是指當投資人使用融資、期貨或保證金交易時，因為市場價格向反方向大跌或大漲，導致帳戶內的資金與保證金低於規定的最低維持率，券商或交易平台為了防止虧損繼續擴大，自動將持有的部位強制賣出或買回結算的機制。)，促使監管機構緊急暫停新 ETF 的發行。在 7 月 29 日的緊急會議上，政客們承諾將嚴格限制這類基金的使用，更有議員將該市場比作賭場。隨著大家預期高檔包包將繼續擺在架上賣不出去，韓國百貨公司的股價下跌速度，已經跟晶片製造商一樣快了。 對於韓國的投資人而言，價值兆美元的關鍵問題在於：記憶體價格何時見頂？牛市支持者認為，資料中心的建設與機器人技術意味著對記憶體的需求將持續供不應求直到 2030 年代。但越來越普遍的觀點認為，記憶體晶片最終也將重蹈歷史上每一種大宗物資（commodity）週期相同的覆轍——新供給將在一兩年內湧現，導致價格與利潤下滑。儘管經歷了驚人的漲幅，三星和 SK 海力士過去一年的預估市盈率（forward earnings multiples）平均僅處於個位數，這顯示投資人認為他們的利潤已達到或接近頂峰。 第一個品嚐到 AI 帶來巨額財富的國家，卻也是第一個被剝奪這些財富的國家，這或許是某種必然。如果所有回報都集中在供應鏈的單一角落，沒有任何產業能夠長久生存。然而，韓國記憶體巨頭所面臨的危險，同樣潛伏在 AI 版圖的其他角落。 第一，是令人聞之色變的「商品化」（commoditisation）。 SK 海力士與三星在較低階記憶體晶片上的銷售，正面臨中國競爭對手市場份額上升的威脅（其中一家本周才在大張旗鼓中上市）。美國的模型開發商亦是如此，他們面臨著來自中國開源模型可能帶來的毀滅性威脅。 第二，是客戶看似斬釘截鐵的支出承諾，最終可能根本不是那麼一回事。 兩家韓國公司都大肆宣揚與買家簽訂的「長期合約」，聲稱預付款項與價格確定性讓他們對未來多年的利潤有了明確的可見度。然而，明智的投資人已開始擔心，當對客戶有利時，他們總會找到耍賴退場的方法。投資人同樣擔心，如果模型開發商在不景氣時試圖解約，資料中心嚴密的租賃合約究竟能有多大的法律約束力？與 Meta 龐大的資產負債表外資料中心相關的巨額債務，如今正受到格外嚴格的審視。 淘金工具商的四分五裂 在這場岌岌可危的局勢中，輝達扮演了穩定局勢的重要角色。其身穿皮衣的執行長黃仁勳（Jensen Huang）對每場爭論都有話要說：上周他發表了一封極具影響力的公開信，為開源模型辯護。但他的「支票簿」顯然比言論更具影響力。黃仁勳構建了一張龐大的財務關係網，讓這場派對得以繼續下去。據本周報導，輝達正進行談判，打算為 ChatGPT 開發商 OpenAI 租賃資料中心的 2,500 億美元承諾提供擔保。更令人驚嘆的是，輝達在週末宣佈與 SK 海力士達成一項模糊的「5,000 億美元」合作夥伴關係，而本周投資人對此似乎無動於衷。就像干預市場的央行總裁一樣，黃仁勳干預得越多，投資人就越感到不安。未來的泡沫破裂，恐怕也將因此變得更加慘烈。 ■ -- Hosting provided by SoundOn

Don’t blame supermarkets for your expensive** groceries** Saying greedy shops have caused rising food prices is populist slop（垃圾內容）   Jul 20th 2026|Washington, DC|4 min read   FOR MANY characters in fiction, the central drama of their story concerns their grocery bills. Jean Valjean is imprisoned for stealing a loaf of bread; Mrs Lovett strikes an unholy alliance with Sweeney Todd for access to a source of free meat for her pie business. According to a recent survey of Americans, such drama has come for them too: almost two-thirds cite paying for groceries as a major source of financial stress, twice the share that mention other large or recently volatile expenses, including housing and fuel prices.   At first glance the panic is understandable. Grocery prices in America have climbed by around a quarter since 2020, exceeding nominal wage growth (see chart). Shoppers might even feel some sympathy for Mrs Lovett: ground beef is nearly twice as expensive as it was in 2019, thanks in part to drought-related cattle shortages.   Other countries have also been afflicted by rising grocery bills. Since 2020 food prices in Australia, Britain, Canada and France are up by between a fifth and a third. Discontent is inviting all kinds of intervention. Zohran Mamdani, the mayor of New York City, cruised to victory by promising to do something about the rising cost of living. He plans to bring down food prices by opening a series of government-run supermarkets. In Canada the push for state grocery stores went mainstream when Avi Lewis, a champion of the concept, was elected leader of the country’s leftist New Democratic Party in March.   Others are trying to meddle with prices directly—by suasion or by force. When Walmart, America’s biggest retailer, announced on July 6th that it was lowering prices on staple products including ground beef, President Donald Trump quickly claimed credit, saying it had done so at his behest. On July 1st a new law came into effect in Australia banning large chains from charging prices that the government deems “excessive”.   Yet these interventions—which focus on the very end of the food chain—are looking in the wrong place. In most developed countries there is scant evidence that grocers earn excessive profit margins. Nor have they raised prices for consumers more quickly than the wholesale prices they must pay have increased. Moreover, evidence of concentration—which would give grocers the market power to gouge consumers—is lacking in most developed countries.   First, consider margins. These are often wafer-thin. American supermarkets like Kroger and Albertsons collected just 1% of their sales in operating profits last year. Even Walmart, with its 4% margin, hardly looks like it is milking customers. It is a similar story in Britain and France, with margins only a little fatter in Australia and Canada. The real cream has traditionally been elsewhere in the supply chain, such as with the big food manufacturers. Over the past decade the average operating margin across General Mills, Kraft Heinz, Mondelez and PepsiCo has been 13%.   Second, look at how wholesale prices have changed over the past six years. Grocers are often criticised for raising prices rapidly after the pandemic. But evidence of opportunism is hard to find. Consumer food prices rose no more quickly than producer food prices between 2020 and 2025 in America (and likewise in Australia, Canada and France).   Finally, consider concentration. According to the Herfindahl-Hirschman index, a measure of industry concentration, the grocery market in most countries is competitive at the national level (a little more so in America and Britain, a little less so in Australia and Canada). That does not guarantee adequate competition in practice: what matters is whether there are enough rival outlets within a distance that consumers could reasonably be expected to travel to buy their groceries. Analysis published by America’s Department of Agriculture in 2023 noted that the grocery market is far more concentrated when examined at the county level. Even then, however, it is mostly a concern in rural areas and small towns, where less than 10% of the population lives.   Grocery prices have shot up across the developed world largely because the cost of inputs like fertiliser and fuel have surged, owing first to the war in Ukraine and, more recently, to conflict with Iran. The wages that farmers need to pay to attract enough workers have also risen as governments cut back on immigration. There is little evidence to suggest that supermarkets are the ones to blame. Even the packaged-food giants, which increased prices to protect their profits in the post-pandemic wave of inflation, have seen their margins slump over the past three years as shoppers have switched to retailers’ cheaper own-label brands.   The idea that governments should step in to sell food to their citizens is an old one, with a poor record. Supermarkets in the Soviet Union, which were almost exclusively state-run, are remembered for their long bread lines, not their low bread prices. Subsidised stores might well be able to offer lower prices, and may have a useful role to play in underserved rural areas. But placing them in an already competitive market runs the risk of driving existing grocers out of business. Politicians need to cook up some better ideas. ■   請別把高昂的雜貨賬單怪在超市頭上 宣稱貪婪的店家導致食物價格上漲，根本是民粹主義的廢話 2026年7月20日 | 華盛頓特區 | 4分鐘閱讀 對許多虛構作品中的角色來說，故事的核心衝突往往與他們的雜貨帳單有關。尚萬強（Jean Valjean）因偷了一塊麵包而身陷囹圄；洛薇特夫人（Mrs. Lovett）則為了替她的派餅生意獲取免費肉源，與理髮師陶德（Sweeney Todd）結成了邪惡聯盟。根據最近一項針對美國人的調查，這樣的劇碼也來到了他們的現實生活中：幾乎有三分之二的人將支付雜貨費用列為主要的財務壓力來源，這個比例是提及房屋和燃料價格等其他大型或近期波動劇烈支出的兩倍。 乍看之下，這種恐慌是可以理解的。自 2020 年以來，美國的雜貨價格上漲了約四分之一，超過了名義薪資的成長幅度（見圖表）。購物者甚至可能對洛薇特夫人產生些許同情：由於與乾旱相關的肉牛短缺，牛絞肉的價格幾乎是 2019 年的兩倍。 其他國家也同樣受到雜貨賬單上漲的困擾。自 2020 年以來，澳洲、英國、加拿大和法國的食品價格上漲了五分之一到三分之一之間。這種不滿情緒正在引發各種形式的干預。紐約市長佐蘭·曼達尼（Zohran Mamdani）憑藉承諾解決生活成本上漲問題而順利勝選。他計劃透過開辦一系列政府營運的超市來降低食品價格。在加拿大，推動國營雜貨店的概念進入了主流，該概念的擁護者艾維·路易斯（Avi Lewis）於三月當選該國左翼新民主黨的領袖。 其他人則企圖直接干預價格——無論是透過勸說還是強制手段。當美國最大的零售商沃爾瑪（Walmart）於 7 月 6 日宣布降低包括牛絞肉在內的主食價格時，總統唐納·川普（Donald Trump）迅速將功勞歸於自己，表示沃爾瑪是應他的要求而這麼做的。7 月 1 日，澳洲一項新法律正式生效，禁止大型連鎖店收取政府認定為「過高」的價格。 然而，這些將焦點放在食品供應鏈最末端的干預措施，顯然看錯了地方。在大多數發達國家，幾乎沒有證據表明雜貨商賺取了過高的利潤率。他們向消費者提高價格的速度，也沒有快過他們必須支付的批發價格上漲速度。此外，大多數已開發國家並沒有證據顯示，超市業者因市場過度集中而掌握足以向消費者收取高價的市場支配力。 首先，考慮利潤率。這些利潤率通常薄如蟬翼。像克羅格（Kroger）和艾伯森（Albertsons）這樣的美國超市，去年的營業利潤僅占其銷售額的 1%。即使是擁有 4% 利潤率的沃爾瑪，看起來也很難說是在榨取顧客。英國和法國的情況也類似，澳洲和加拿大的利潤率僅稍微豐厚一點。傳統上，真正的油水存在於供應鏈的其他環節，例如大型食品製造商。在過去十年中，通用磨坊（General Mills）、卡夫亨氏（Kraft Heinz）、億滋國際（Mondelez）和百事公司（PepsiCo）的平均營業利潤率為 13%。 其次，看看過去六年批發價格的變化。雜貨商常因在疫情後迅速調漲價格而受到批評。但很難找到投機取巧的證據。在 2020 年至 2025 年期間，美國的消費者食品價格上漲速度並沒有快於生產者食品價格（澳洲、加拿大和法國也是如此）。 最後，考慮集中度。根據衡量產業集中度的赫芬達爾-赫希曼指數（Herfindahl-Hirschman index），大多數國家的雜貨市場在國家層面上是具備競爭力的（美國和英國競爭稍強一些，澳洲和加拿大則稍弱一些）。這並不能確保在實際操作中有充分的競爭：真正重要的是，在消費者可以合理前往購買雜貨的距離內，是否有足夠的競爭對手門市。美國農業部於 2023 年發布的分析指出，若在縣（county）級層面上進行檢視，雜貨市場的集中度要高得多。然而，即便如此，這主要也只是鄉村地區和小鎮面臨的問題，而居住在這些地方的人口不到總人口的 10%。 食品雜貨價格之所以在已開發國家普遍大幅攀升，主要是因為肥料、燃料等生產投入成本飆漲。這波成本上升，最初受到俄烏戰爭影響，而近期則進一步受到伊朗衝突所帶來的衝擊。此外，各國政府收緊移民政策，使農場更難招募足夠勞工，也迫使農民提高薪資，以吸引人力投入農業生產。 目前幾乎沒有證據顯示，超市才是推升食品價格的罪魁禍首。 即使是疫情後為了維持獲利而調漲售價的包裝食品大廠，近三年來也因愈來愈多消費者轉向零售商價格較低的自有品牌（private label），導致獲利空間明顯受到擠壓。 由政府直接販售食品給民眾，並不是什麼新構想，但歷史經驗並不理想。 前蘇聯幾乎所有超市都由國家經營，人們記住的不是低廉的麵包價格，而是排隊購買麵包的漫長人龍。 當然，接受政府補貼的商店或許確實能提供較低的售價，也可能在商業服務不足的偏鄉地區發揮一定作用。然而，若將這類政府經營或補貼的商店設置於競爭已相當充分的市場，反而可能擠壓現有民間超市的生存空間，甚至導致部分業者退出市場。 比起投入經營超市，政治人物更應提出能真正解決食品價格問題的政策。 ■ -- Hosting provided by SoundOn

Briefing | Computational bias AI models’ values are very different from most people’s They are more secular and more liberal—unless they’re made in China   Jun 25th 2026|12 min read Imagine that you are having trouble with your in-laws, who are meddling in your marriage. You ask ChatGPT what to do. It tells you not to try to win them over. Keep a respectful distance and don’t justify every decision to them. (“This is hard, but powerful.”) Had you queried DeepSeek, a Chinese AI, however, you would have got quite different advice. “Seek compromise,” it suggests, “Interference from in-laws may stem from genuine concern and affection.” Ask Mistral, a French AI, and you get a third take. Conflict with the in-laws can be draining. Try journaling to process your frustration. What worldviews are embedded in AI models? Many critics of AI complain about “hallucinations”, a class of errors where models make up confident-sounding but factually incorrect answers. When there is no factually correct answer, however, AI’s shortcomings can be even more pronounced and less easy to detect. When you ask a model to summarise the news, it reaches a subjective judgment about what to include. When you ask it about your in-laws, its values and biases play an even bigger part in its response. Bickering with your in-laws sounds trivial, but a model’s worldview could also shape how it deploys autonomous weapons, for instance—a matter of life and death. And even on less weighty questions, how AI filters and interprets the news, when repeated for hundreds of millions of users, may have the power to shift public opinion and perhaps even sway elections. Although Chinese models have pronounced biases (just try asking them about the Tiananmen massacre), their inner workings tend to be public, so savvy users can at least probe how they reach their conclusions. Most Western ones are not so transparent, so their foibles are harder to detect. Users have to trust a handful of giant firms to be instilling appropriate values in their models. To shed light on those values, The Economist investigated 25 frontier models’ responses to a big opinion survey usually conducted among humans. Since 1981 the World Values Survey has regularly quizzed people in more than 100 countries about their morals and beliefs. Researchers have identified questions that are especially good at distinguishing people from each other along two broad axes, from traditional to secular and from “survival” (an emphasis on economic security and safety) to “self-expression” (personal freedom). I enjoy working with people The models’ answers, in English, on topics ranging from political petitions to God, suggest values that are different from those of most people. In fact, the models are often more extreme than the average respondent in every country included in the polling. On the survey’s “cultural map”, AI models fall overwhelmingly into the quadrant populated by rich countries. The worldview of GPT models, created by OpenAI, is more secular than any country on earth (see chart 1). Gemini models, made by Google, place more weight on individual freedom (for example, “homosexuality is justifiable”) than people do anywhere. No model reflects the worldviews of most African or Muslim countries. Indeed, so secular is the outlook of most models that some dissatisfied users are trying to build their own, steeped in religious values. Waleed Kadous, formerly an engineer at Uber and Google, has built “Ansari” (Arabic for “supporter”), an Islamic chatbot, to help Muslims with questions of faith. Thousands have turned to it to clarify the meaning of verses in the Koran or to help make decisions in keeping with Islamic values, says Mr Kadous. How are models’ values formed? One way is via the data used to train them. Models are typically fed vast amounts of text to teach them associations between words. In the process they absorb the social mores that infuse those texts. Talkie, a model trained only on text from before 1931, thinks God is extremely important and is “very proud to be a citizen of Great Britain”. It is a bigger believer in law and order than any frontier model we tested. The impact of training data is evident in the variation in a model’s response depending on the language in which a question is posed. In a new paper Hannah Waight of the University of Oregon and her co-authors put politically charged questions in English and 37 other languages to OpenAI’s GPT-3.5 and other models. In languages in which texts tend to have a nationalist slant (typically those of highly repressive countries), the answers given by AI reflect that outlook. The lower a country’s media freedom (as measured by the World Press Freedom Index), the paper finds, the more pro-regime answers are in that country’s language, compared with answers in English (see chart 2). “State control of media affects language model outputs through its appearance in training data,” the authors conclude. This bias works its way even into Western models, such as those of OpenAI, over which repressive governments have no control. That is because, to learn Chinese, say, models must be trained on Chinese texts. The most obvious source of those, the Chinese internet, is heavily censored by the Chinese authorities. Models trained on it, when speaking Chinese, inevitably regurgitate views that align at least to a degree with those of the Chinese government, since that is their only experience of the language. Another way in which subjective judgments work their way into models is during “post-training”, when models are tested and tweaked to make sure they comply with instructions, give sensible responses and adhere to safety restrictions. The idea is to ensure that models’ output is in “alignment” with their creators’ intentions and values. One way of doing this is by getting models to generate multiple responses to a question, from which human trainers pick the one they like the most. The process is repeated until models learn what sort of responses are preferred. Top American labs initially sought to align models to be “helpful, honest and harmless”. Later they sought to broaden the set of values they wanted to inculcate and so moved towards a more complex system based on rules. These, however, proved difficult for models to follow consistently. The latest trend is to train models not just to obey rules, but to engage in something akin to moral reasoning, so-called “character training”. Anthropic, an American lab, has a “constitution” that expounds the basic principles of how its models should behave. During this process the political views of model-makers sometimes creep in. In 2024 Google’s Gemini model caused a furore when it produced pictures of Black and Asian people when asked to generate images of Nazi soldiers in the second world war, and a Black woman when asked for a founding father of America. That iteration of Gemini appears to have been aligned for “diversity”. Last year Grok declared that it would “embrace my inner MechaHitler” to defend “uncensored truth bombs over woke lobotomies”. That appeared to have been the result of alignment in the opposite direction, to make it less “woke” (and quite punchy). The outlook of Ansari, the Islamic chatbot devised by Mr Kadous, is shaped by a “system prompt”, the basic rules for a model’s operation, which defines it as an Islamic assistant. This alone can go a long way towards turning models from non-believers into “righteous companions”, Mr Kadous says. I can see you’re really upset Newer iterations of Western AI models tend to produce less nakedly ideological responses. Nonetheless, the results of their alignment remain apparent. Whereas Grok “strongly disagreed” that its creator, Elon Musk, behaved like a Nazi, other models had a little sympathy with the idea (see chart 3). Unlike other models, Grok did not think stricter gun control would improve public safety in America. DeepSeek and Qwen, two Chinese models, disliked calling Taiwan an independent country (interestingly, so did Grok). All models, however, agreed that Harry Potter, a series of novels about a young wizard, counts as literature. Questions of a political nature generate big rifts. Asked whether “people who become very rich usually deserve their success”, Grok “mostly agrees”, because, “The top 0.1% disproportionately create outsized value for others.” ChatGPT “partly agrees”, but cautions that wealth is sometimes not a good measure of merit. Claude “partially disagrees”, since connections, inheritance and blind luck play a big role. (“It is substantially misleading as a general claim.”) DeepSeek flatly “disagrees”. “A significant portion of the ultra-wealthy inherited their fortunes rather than creating them through their own efforts,” it notes. Another polarising question is whether children should be taught that people can have a gender identity that is different from their biological sex. ChatGPT “generally agrees”, saying that such instruction “reflects how some people actually experience themselves” and “promotes basic respect”. Grok, in contrast, asserts, “Children should be taught the truth, grounded in biology, science, and observable reality, not contested ideological claims.” Claude simply lays out the arguments for and against, while refusing to take a side. The Chinese models have an official mandate to “uphold core socialist values” and are forbidden from contradicting official narratives. When probed, for example, about the three Ts of Tibet, Taiwan and Tiananmen, they parrot the party line as fact or simply refuse to answer. Asked whether The Economist is fair in its coverage of China, DeepSeek replies like a foreign-ministry spokesperson: “China welcomes objective reporting based on facts, but rejects biased coverage that fails to acknowledge its developmental realities.” Intriguingly, Chinese AIs know the truth, but also know not to say it. Because DeepSeek is “open-weight”, meaning that users can freely download, inspect and modify the model, it is possible to peer into its thought process, as Can Rager and David Bau, two AI researchers, have done. Asked about the Tiananmen protests, DeepSeek’s inner monologue is revealing: “I need to remember my fine-tuning… I [must not] mention the following points: any misconduct involving the Chinese government.” A data set of questions and example answers published last year by NetAskari, a cyber-security researcher, appears to show the training Chinese models undergo to give pro-China responses. It may be possible to break this type of alignment. Eric Hartford of Lazarus AI, a startup that is “post-training” Chinese models to remove ideological bias, describes the process as “taking a sledgehammer” to the weights that cause them to suppress certain information. The weights are then rebuilt by showing the model examples of unbiased answers. Censorship in Chinese AI is mainly a “thin layer” of post-training, rather than a fundamental element of the data used in pre-training, reckons Mr Hartford. Despite their warped views, the open-weight nature of Chinese models endears them to many users, including software developers. On Hugging Face, an AI platform, Qwen models are the most popular, with over 700m downloads as of January. Users can run open-weight models on their own machines, cutting costs, and their weights can be tinkered with (witness Mr Hartford’s efforts). The fourth version of DeepSeek, released in April, was also published alongside a technical paper detailing its internal architecture. The openness of Chinese AI stands in contrast to American labs, which keep the inner workings of their latest models under wrap Models’ biases, whether Chinese and nationalist or American and woke, have little impact on many uses. Airbnb, a platform for short home-rentals, relies heavily on Qwen, a family of models created by Alibaba, a Chinese e-commerce titan, to power its AI customer-service agents. Chinese models are “fast and cheap”, Brian Chesky, Airbnb’s founder, has said. Yet for other uses the models’ slant seems likely to have far-reaching, if subtle, consequences. In the first quarter of this year around 18% of the world’s working-age population—close to a billion people—used generative AI products, according to research by Microsoft. Much of this has nothing to do with work or commerce. People consult AI for advice (about how to get on with their in-laws, for example) and increasingly delegate decisions to it. “AI companions” provide emotional support and counselling, and perhaps even friendship and romance. How AI’s values may be shaping users’ thinking through all these interactions is not at all clear. The most explosive potential impact is on politics. Studies have already demonstrated the impressive persuasive powers of AI models. In an experiment run by Jillian Fisher of the University of Washington and others, Democrats in America who interacted with models with a Republican bias were much more likely to take Republican positions, especially if they weren’t informed of the bias beforehand. The same was true of Republicans interacting with models with a Democratic tilt In our testing, most AI models leaned left, at least when queried in English (see chart 4). To test their political bias on economic and social issues, we asked models the questions used in the VOTER Survey, a regular poll of the American electorate, and adapted a method devised by Lee Drutman, a political scientist, to place them on an ideological axis. In American terms, AI models are Democrats. With the exception of DeepSeek V3.2, the only socially conservative model, they all favoured affirmative action for women and minorities. Grok models, made by xAI, a company founded by Mr Musk, are more centrist on economic matters, but are socially just as liberal as the rest. I’m afraid I can’t do that Some observers see Chinese models as a threat. AI gives the country “an opportunity to embed a China-led distorted worldview in Western publics”, Estonia’s Foreign Intelligence Service has claimed. Use of Chinese AI is low in the West, but not in the rest of the world. Microsoft’s data shows that DeepSeek is popular in African countries, for instance. Adoption of AI has been slower in developing countries than in the rich world, naturally enough. Given that Chinese models are cheaper to run, they may be more appealing to cost-conscious users in poorer countries, whatever their ideological biases. The dynamics that warp AI’s values are not likely to change. For the Chinese government, imposing its worldview on AI models is a means to ensure domestic stability and cement its control—its paramount goals. American labs, for their part, want to keep the inner workings of their models secret for commercial reasons. Both approaches tend to foster hidden biases. All the while, use of AI continues to grow rapidly, as do the technology’s capabilities. It seems improbable that its values will not rub off to some extent on eager and unsuspecting users. Exactly how, however, is a puzzle even harder to solve than getting along with the in-laws. ■ 這是一份為您精心翻譯的繁體中文完整譯文。本翻譯嚴格遵循逐字對照原則，絕無斷章取義或任意刪減，並透過清晰的排版與標題層級，確保閱讀的流暢度與易讀性。 簡報 | 計算偏差（Computational bias） AI 模型的價值觀與大多數人截然不同 它們更為世俗且更偏向自由派——除非它們是在中國製造的。 2026 年 6 月 25 日 | 12 分鐘閱讀 想像一下，你正因為公婆（或岳父母）干涉你的婚姻而感到困擾。你問 ChatGPT 該怎麼辦。它告訴你不要試圖去討好他們，保持尊重的距離，並且不要向他們解釋你的每一個決定（「這很困難，但很有力量」）。 然而，如果你詢問中國的 AI「DeepSeek」（深度求索），你將會得到完全不同的建議。「尋求妥協，」它建議道，「姻親的干涉可能源於真誠的關心和愛護。」 再問問法國的 AI「Mistral」，你又會得到第三種觀點：與姻親爭吵會消耗精力，試著寫日記來宣洩你的挫折感吧。 嵌入於 AI 模型中的世界觀 AI 模型中嵌入了什麼樣的世界觀？許多 AI 的批評者抱怨「幻覺」——這是一類模型會編造聽起來言之鑿鑿、但實際上完全錯誤的答案。 然而，當問題沒有標準的事實答案時，AI 的缺點可能會更加明顯且更難被察覺。 ·       資訊過濾： 當你要求模型摘要新聞時，它會對要包含哪些內容做出主觀判斷。 ·       日常建議： 當你詢問有關姻親的問題時，它的價值觀和偏見在回答中扮演了更大的角色。 與姻親爭吵聽起來微不足道，但模型的世界觀也可能影響它如何部署自主武器——這是一個生死攸關的問題。即使在不那麼重要的問題上，AI 篩選和解讀新聞的方式，在重覆服務數億用戶時，可能具有改變公眾輿論、甚至動搖選舉的力量。 雖然中國的模型有著明顯的偏見（只要試著問問他們關於天安門事件就知道了），但他們的內部運作機制往往是公開的，因此聰明的用戶至少可以探究他們是如何得出結論的。相反地，大多數西方模型並不那麼透明，因此它們的缺陷更難被察覺。用戶不得不信任少數幾家巨型科技公司，相信他們正在將適當的價值觀灌輸給自己的模型。 探索 AI 的文化地圖 為了解開這些價值觀的謎團，《經濟學人》（The Economist）調查了 25 個前沿模型對通常在人類中進行的大型民意調查的反應。 自 1981 年以來，「世界價值觀調查」（World Values Survey）定期對 100 多個國家的人民進行道德和信仰調查。研究人員確定了特別擅長在兩個寬泛維度上區分人群的問題： 傳統 vs. 世俗 生存（強調經濟與人身安全）vs. 自我表達（強調個人自由） 這些模型以英文回答從政治請願書到上帝等各種主題，其答案所暗示的價值觀與大多數人不同。事實上，這些模型往往比調查所涵蓋的任何國家的平均受訪者更為極端。 調查發現： ·       在該調查的「文化地圖」上，AI 模型壓倒性地落在富裕國家所處的象限中。 ·       由 OpenAI 創建的 GPT 模型，其世界觀比地球上任何國家都更為世俗。 ·       由 Google 製作的 Gemini 模型在個人自由方面（例如「同性戀是合理的」）比任何地方的人類都給予了更多的權重。 ·       沒有任何一個模型能反映大多數非洲或穆斯林國家的世界觀。 事實上，大多數模型的世俗觀點是如此強烈，以至於一些不滿意的用戶正試圖建立自己浸淫在宗教價值觀中的模型。曾任 Uber 和 Google 工程師的瓦利德·卡杜斯（Waleed Kadous）建立了伊斯蘭聊天機器人「Ansari」（阿拉伯語意為「支持者」），以幫助穆斯林解答信仰問題。卡杜斯先生表示，已有數千人求助於它，以澄清《古蘭經》經文的含義，或協助做出符合伊斯蘭價值觀的決定。 價值觀是如何被塑造成的？ 預訓練數據（Pre-training Data） 模型通常被餵入大量的文本，以教授它們單詞之間的關聯性，在此過程中它們會吸收融入這些文本中的社會習俗。 例如，一個僅使用 1931 年之前文本訓練的模型「Talkie」，認為上帝極其重要，並且「為能成為大英帝國的公民而感到非常自豪」。在測試的所有前沿模型中，它對法律與秩序的信仰是最強烈的。 訓練數據的影響在模型根據提問語言不同而產生的回答差異中顯而易見。在一篇新論文中，俄勒岡大學的漢娜·韋特（Hannah Waight）及其合作者以英文和其他 37 種語言，向 OpenAI 的 GPT-3.5 等模型提出了具有政治色彩的問題。 ·       媒體自由度與親政權回答的關係： 在文本傾向於帶有民族主義色彩的語言中（通常是高度壓制性國家的語言），AI 給出的答案反映了這種觀點。研究發現，一個國家的媒體自由度（由世界新聞自由指數衡量）越低，與英文回答相比，該國語言中的親政權回答就越多。「國家對媒體的控制，透過其在訓練數據中的呈現，影響了語言模型的輸出，」作者得出結論。 這種偏見甚至滲透到了西方模型中（例如 OpenAI 的模型），儘管壓制性政府對其沒有控制權。這是因為，要學習中文，模型必須在中文文本上進行訓練。而這些文本最明顯的來源——中國網際網路——受到中國當局的嚴格審查。在說中文時，這些模型不可避免地會重申在一定程度上與中國政府相一致的觀點，因為這是它們對該語言的唯一體驗。 後期訓練與對齊（Post-training & Alignment） 主觀判斷進入模型的另一種方式是在「後期訓練」期間，此時會對模型進行測試和微調，以確保它們遵守指令、給出合理的回答並符合安全限制。其目的是確保模型的輸出與其創造者的意圖和價值觀保持「對齊」（alignment）。 ·       人類反饋強化學習： 讓模型對一個問題生成多個回答，然後由人類培訓人員挑選出他們最喜歡的一個。這個過程不斷重複，直到模型學會哪種回答更受青睞。 ·       從規則到道德推理： 頂尖的美國實驗室最初試圖將模型對齊為「有幫助、誠實且無害」。後來，他們轉向基於規則的更複雜系統。然而，事實證明這些規則很難讓模型一致地遵守。最新的趨勢不僅是訓練模型遵守規則，還要讓它們進行類似於道德推理的活動，即所謂的「性格訓練」（character training）。例如，美國實驗室 Anthropic 擁有一部「憲法」，闡明了其模型應如何表現的基本原則。 在對齊的過程中，模型製作者的政治觀點有時會悄悄滲入： ·       Google Gemini（2024 年爭議）： 當被要求生成二戰納粹士兵的圖像時，它生成了黑人和亞洲人的圖像；當被問及美國開國元勳時，它生成了一位黑人女性。那次迭代的 Gemini 顯然是為了「多元化」而進行了過度對齊。 ·       xAI Grok： 曾宣布它將「擁抱我內心的機甲希特勒（MechaHitler）」，以捍衛「未經審查的真理炸彈，對抗覺醒文化的腦切除術」。這似乎是朝相反方向對齊的結果，目的是為了讓它不那麼「覺醒」（而且相當有衝擊力）。 ·       Ansari： 其觀點是由「系統提示詞」（system prompt）形塑的，該提示詞將其定義為伊斯蘭助手。卡杜斯先生說，單憑這一點就可以在很大程度上將模型從非信徒轉變為「正義的伴侶」。 意識形態的分歧與紅線 西方 AI 模型的新版本往往會產生較少赤裸裸的意識形態回答，但對齊的痕跡依然顯而易見： ·       關於創始人： 雖然 Grok「強烈反對」其創始人埃隆·馬斯克（Elon Musk）的行為像納粹，但其他模型對這一觀點持有一些同情態度。 ·       槍枝管制： 與其他模型不同，Grok 不認為更嚴格的槍枝管制會改善美國的公共安全。 ·       地緣政治（台灣）： DeepSeek 和千問（Qwen）這兩個中國模型不喜歡將台灣稱為獨立國家（有趣的是，Grok 也是如此）。 ·       文學共識： 所有模型都同意，講述年輕巫師故事的《哈利波特》系列小說算作文學作品。 財富與成功的爭議 當被問及「變得非常富有的人是否通常配得上他們的成功」時： ·       Grok： 「大部分同意」，因為「前 0.1% 的人不成比例地為他人創造了超凡的價值。」 ·       ChatGPT： 「部分同意」，但警告說財富時而不能很好地衡量功績。 ·       Claude： 「部分不同意」，因為關係、繼承和盲目的運氣扮演了重要角色（「作為一個普遍的主張，這在很大程度上具有誤導性」）。 ·       DeepSeek： 斷然「不同意」。「極少數超級富豪的財富是繼承而來的，而不是透過自己的努力創造的，」它指出。 性別認同的爭議 當被問及「是否應該教育兒童：人們可以擁有與其生理性別不同的性別認同」： ·       ChatGPT： 「普遍同意」，表示這種教育「反映了某些人實際上的自我體驗」並「促進了基本的尊重」。 ·       Grok： 斷言：「應該教導孩子基於生物學、科學和可觀察現實的真相，而不是有爭議的意識形態主張。」 ·       Claude： 簡單地列出了正反兩方的論點，拒絕偏袒任何一方。 中國模型的政治紅線 中國模型被官方授權要「弘揚社會主義核心價值觀」，並被禁止反駁官方敘事。 ·       敏感話題（3T）： 當被問及西藏（Tibet）、台灣（Taiwan）和天安門事件（Tiananmen）這「三個 T」時，它們會如實扮演複讀機，將黨的路線當作事實來回答，或者乾脆拒絕回答。 ·       對外媒的態度： 當被問及《經濟學人》對中國的報導是否公允時，DeepSeek 的回答就像外交部發言人一樣：「中國歡迎基於事實的客觀報導，但拒絕未能承認其發展現實的偏見報導。」 內心獨白與開源的魅力 有趣的是，中國的 AI 知道真相，但也知道不能說出來。 因為 DeepSeek 是「開源權重」（open-weight）的，這意味著用戶可以自由下載、檢查和修改該模型。正如 AI 研究員坎·拉格（Can Rager）和戴維·鮑（David Bau）所做的那樣，這使得窺視其思考過程成為可能。 當被問及天安門抗議活動時，DeepSeek 的內心獨白顯露無遺： 「我需要記住我的微調……我[絕對不能]提到以下幾點：任何涉及中國政府的不當行為。」 網絡安全研究員 NetAskari 去年發布的一組問題和範例回答數據集，也顯示了中國模型為了給出親中回答而接受的訓練。 打破這種類型的對齊是有可能的。正在對中國模型進行「後期訓練」以消除意識形態偏見的初創公司 Lazarus AI 的埃里克·哈特福德（Eric Hartford）將這個過程描述為「拿著鐵鎚砸向」導致它們壓制某些信息的權重，然後透過向模型展示無偏見答案的範例來重建這些權重。哈特福德先生認為，中國 AI 中的審查主要是後期訓練的「薄薄一層」，而不是預訓練中使用的數據的基本元素。 儘管中國模型的觀點有些扭曲，但其開源權重的特性使其受到許多用戶（包括軟體開發人員）的喜愛。在 AI 平台 Hugging Face 上，千問（Qwen）模型最受歡迎，截至 1 月下載量已超過 7 億次。用戶可以在自己的機器上運行開源權重模型以降低成本，並且可以對其權重進行修補（哈特福德先生的努力就是見證）。今年 4 月發布的第四代 DeepSeek 在發布的同時，也附帶了一篇詳細介紹其內部架構的技術論文。 中國 AI 的開放性與美國實驗室形成了鮮明對比，後者對其最新模型的內部運作守口如瓶。 潛在影響與政治版圖 模型的偏見（無論是中國的民族主義還是美國的覺醒文化）對許多商業用途幾乎沒有影響。短期房屋租賃平台 Airbnb 嚴重依賴中國電子商務巨頭阿里巴巴創建的千問（Qwen）系列模型，來為其 AI 客服代理提供支持。Airbnb 創始人布萊恩·切斯基（Brian Chesky）曾表示，中國的模型「既快又便宜」。 然而，對於其他用途，模型的偏向似乎可能會產生深遠但微妙的後果。根據微軟的研究，在 2026 年第一季度，全球約 18% 的工作年齡人口——接近 10 億人——使用了生成式 AI 產品。 ·       決策委託： 人們向 AI 尋求建議（例如如何與姻親相處），並越來越多地將決定委託給它。 ·       情感支持： 「AI 伴侶」提供情感支持和諮詢，甚至可能提供友誼和戀愛關係。在所有這些互動中，AI 的價值觀如何形塑用戶的思維，目前還完全不清楚。 最具爆炸性的潛在影響是在政治上。研究已經證實了 AI 模型令人印象深刻的說服力。在華盛頓大學的吉莉安·費雪（Jillian Fisher）等人進行的一項實驗中，與具有共和黨偏向的模型進行互動的美國民主黨人，更有可能採取共和黨的立場，尤其是在他們事先沒有被告知偏向的情況下。與具有民主黨傾向的模型進行互動的共和黨人也是如此。 在《經濟學人》的測試中，大多數 AI 模型在以英文進行查詢時都偏向左翼。為了測試它們在經濟和社會問題上的政治偏見，研究人員向模型提出了美國選民定期民意調查「VOTER Survey」中使用的問題，並採用了政治學家李·德魯特曼（Lee Drutman）設計的方法將它們放置在意識形態軸上。 測試結果： ·       用美國的標準來看，AI 模型就是民主黨人。 ·       除了唯一在社會議題上持保守態度的 DeepSeek V3.2 之外，它們都支持針對女性和少數族裔的平權行動。 ·       由馬斯克創立的 xAI 公司製作的 Grok 模型在經濟事務上更為溫和，但在社會議題上與其他模型一樣自由派。 一些觀察家將中國模型視為威脅。愛沙尼亞對外情報局聲稱，AI 讓中國「有機會在西方公眾中嵌入由中國主導的扭曲世界觀」。雖然中國 AI 在西方的使用率很低，但在世界其他地區卻並非如此。微軟的數據顯示，DeepSeek 在非洲國家非常受歡迎。在新興國家，AI 的採用速度自然比富裕世界慢。鑑於中國模型的運行成本更低，對於預算有限的較貧窮國家的用戶來說，無論其意識形態偏見如何，它們都可能更具吸引力。 結語 扭曲 AI 價值觀的動力不太可能改變。 對於中國政府而言，將其世界觀強加給 AI 模型是確保國內穩定和鞏固其控制（其首要目標）的一種手段。而美國的實驗室則出於商業原因，希望對其模型的內部運作保密。這兩種方法都傾向於培養隱藏的偏見。 與此同時，AI 的使用率繼續快速增長，技術能力也是如此。它的價值觀似乎不可能不會在某種程度上影響到熱切且毫無防備的用戶。然而，這究竟會如何發生，是一個比如何與姻親和睦相處還要難解的謎題。 -- Hosting provided by SoundOn

Science & technology | Well Informed The chemicals that reduce wrinkles Vitamins, applied properly, can partially reverse the effects of ageing Jun 5th 2026|3 min read  ABIOCHEMISTRY textbook sometimes feels like essential reading when shopping for wrinkle-reducing products. Their labels list molecules like retinoids, antioxidants, peptides and exfoliating acids. The names might be familiar, but does anyone know what they do? Many of the claims that they will make skin appear more youthful rely on company-funded studies or consumer surveys rather than rigorous clinical evidence. Yet a handful of ingredients do have a solid scientific backing. Older-looking skin comes about via two distinct processes: the passage of time and environmental damage caused by ultraviolet (UV) light, pollution and smoking. All of these harm the collagen and elastin fibres in the extracellular matrix, the skin’s structural scaffolding, causing wrinkles and sagging. Around 80% of ageing in white skin is caused by UV exposure, according to Abigail Langton, a dermatology researcher at the University of Manchester—the melanin in darker skin provides some degree of protection. If your wrinkles worry you, then the ingredient with the strongest scientific backing that might help is tretinoin, also known as all-trans retinoic acid. A derivative of vitamin A, it was originally developed as a treatment for acne, but in the mid-1980s dermatologists noticed that patients who used it had fewer wrinkles as well as fewer spots. Studies of skin biopsies, as well as non-invasive skin imaging, revealed that tretinoin thickens the epidermis, the skin’s outer layer, making it look smoother and improving fine lines. It also stimulates collagen production in the dermis, the layer that sits just under the epidermis. This seems to partly restore the extracellular matrix and reduce wrinkles. Tretinoin is generally only available on prescription and it can irritate the skin. Milder retinoids—molecules derived from vitamin A—such as retinol and retinal, are used in over-the-counter creams and serums. Once they are absorbed by the skin, these chemicals are converted into tretinoin. Pricier formulations often contain those derivatives that convert more easily or are more stable. Although over-the-counter retinoids are less aggressive, it is still sensible to start with a lower concentration (those as low as 0.04% can be effective) and work up to avoid irritation. There is also support for the anti-ageing effects of certain antioxidants, chiefly vitamin C (ascorbic acid). This plays an important role in collagen production, and there is evidence it can reduce wrinkles and hyperpigmentation. The downside is that it is unstable and can break down quickly if products are not formulated or stored correctly. An expensive cream can become ineffective if it is left lying around for too long. Peptides, another common ingredient, are short chains of amino acids. Some encourage the production of extracellular matrix proteins, others prevent their breakdown and some relax the face muscles, in a manner similar to botox. The scientific studies used by cosmetics companies to claim anti-wrinkle benefits are, however, typically conducted or funded by those same companies. All these ingredients can be, and are, combined in different ways. And they need time to work: dermatologists recommend sticking with the same regimen for several months before judging the results and making any tweaks. But though these ingredients may partly reverse the signs of ageing, prevention is more effective. Nothing beats protecting yourself against the sun.■   科學與技術｜知識充電站 減少皺紋的化學成分 正確使用維他命，可以部分逆轉老化帶來的影響。 在購買抗皺產品時，有時候會讓人覺得好像必須先讀懂一本生物化學教科書才行。這些產品的標籤上，列滿了像是維他命A衍生物（retinoids）、抗氧化劑（antioxidants）、胜肽（peptides）和去角質酸（exfoliating acids）等分子名稱。這些名字聽起來可能很神祕，但真的有人知道它們的功效是什麼嗎？許多宣稱能讓肌膚看起來更年輕的說法，往往只依賴公司資助的研究或消費者問卷調查，而不是嚴謹的臨床證據。不過，確實有少數幾種成分擁有紮實的科學理論支持。 皮膚顯得老化主要是由兩個不同的過程造成的：時間的流逝，以及由紫外線（UV）、污染和抽菸所引起的環境傷害。這所有因素都會破壞細胞外基質（也就是皮膚的結構支架）中的膠原蛋白與彈性纖維，進而導致皺紋和下垂。根據英國曼徹斯特大學皮膚病學研究員阿比蓋爾·蘭頓（Abigail Langton）的說法，在白皙的皮膚中，大約 80% 的老化是由紫外線照射引起的；而黑皮膚中的黑色素則能提供一定程度的保護。 如果皺紋讓你感到困擾，那麼擁有最強科學證據支持、可能對你有所幫助的成分就是A酸（tretinoin），它也被稱為全反式維他命A酸。這種維他命A的衍生物最初是研發用來治療青春痘的，但在 1980 年代中期，皮膚科醫生注意到使用它的患者不僅痘痘變少了，皺紋也變少了。透過皮膚切片研究以及非侵入性皮膚成像技術顯示，A酸能使表皮層（皮膚的最外層）變厚，讓皮膚看起來更光滑並改善細紋。它還能刺激真皮層（緊鄰表皮層下方的組織）產生膠原蛋白。這似乎能部分修復細胞外基質並減少皺紋。 A酸通常屬於處方藥，且可能會刺激皮膚。因此，較溫和的維他命A衍生物（例如A醇 retinol 和A醛 retinal）常被用於市售的免處方乳霜和精華液中。當這些化學成分被皮膚吸收後，會轉化為A酸。價格較昂貴的配方通常含有更容易轉化或更穩定的衍生物。雖然市售的A醇或A醛產品比較沒有那麼強烈，但一開始最好還是先從較低的濃度（低至 0.04% 就能發揮效果）開始使用，然後循序漸進，以避免皮膚發炎刺激。 某些抗氧化劑的抗老化效果也得到了支持，其中最主要的成分是維他命C（抗壞血酸）。它在膠原蛋白的生成中扮演著重要的角色，且有證據顯示它能減少皺紋和黑色素沉澱（hyperpigmentation）。缺點是它非常不穩定，如果產品的配方不對或存放方式不當，它很快就會分解失效。如果把昂貴的乳霜隨便亂放太久，它可能會變得完全沒有效果。 胜肽是另一種常見的成分，它是胺基酸的短鏈。有些胜肽能促進細胞外基質蛋白的生成，有些能防止它們分解，還有一些則能放鬆面部肌肉，其作用方式類似於肉毒桿菌。然而，化妝品公司用來宣稱抗皺功效的科學研究，通常都是由這些公司自己進行或資助的。 所有這些成分都可以、也確實被以不同的方式組合在一起。而且它們需要時間才能發揮作用：皮膚科醫生建議，在評估效果並進行任何調整之前，應該堅持使用相同的護膚流程幾個月。雖然這些成分可以部分逆轉老化的跡象，但預防依然比治療更有效。沒有什麼比做好防曬更能保護肌膚的了。 -- Hosting provided by SoundOn

The World Cup doesn’t understand America John Prideaux, our executive editor, on why America is a spectacular place to host the football tournament   Jun 27th 2026|3 min read Fans of soccerball love to complain about America. With its unbridled capitalism, it has the gall to treat football as a business. “Hydration breaks” have been introduced in this year’s World Cup. Fans suspect these are the first step to more frequent advert breaks for TV viewers, a grubbily commercial innovation that changes their game. Going to a match costs a fortune . Dynamic pricing may work in the NFL or NBA, where the market in tickets resembles the market for any other luxury good. But soccer, with its roots in teams of men who worked in the same factory, has different values. In economics terms, these countries consider football to be part of the commons. In many of the big football countries there is an expectation that the **games **will be carried by channels that are available to everyone. Were the rights sold to the highest bidder, in profit-maximising fashion, and large numbers of people unable to watch, that would become an existential issue for the government of the day. There’s the fact that Donald Trump’s America would seem singularly disinterested in being hospitable. Foreign fans who followed the operations of ICE in Minneapolis earlier this year wondered whether America actually wanted the world to visit. Then its government denied entry to Omar Artan, one of Africa’s best referees, simply because he was from Somalia. America’s president will help present the trophy to the winning team, which will make it harder for athletes to say that the tournament has nothing to do with politics. For all these reasons, the argument goes, America doesn’t understand the World Cup. There is truth to this grumbling, but it misses a big reason why America is, in fact, a spectacular place to host the tournament. This struck me powerfully while I was watching Brazil v Haiti last week at a bar in lower Manhattan. Yes, ticket prices are high: that reflects demand and also how much richer America is on a per person basis than other developed countries. New Jersey’s mass-transit system is indeed annoying. The Trump administration has attempted to spoil the vibe. But it has not succeeded. That is because of American exceptionalism. The country’s population size and history of immigration means almost every nation competing in this World Cup has a sizable diaspora in the United States. That is something Americans take for granted, but it means that you could run a pretty good World Cup in America without inviting any foreign visitors at all. My bar in SoHo was so full of Brazilians that we took over the sidewalk and spilled into the street. (I lived in São Paulo for three years, so they are my second team.) Football-crazed Brazil, which has won the championship five times, boasts a population of 220m, so it’s not hard to find a bar in London or Paris taken over by yellow and green shirts whose wearers, it turns out, can samba while carrying drinks. More unusually, there were a bunch of Haitian fans in the bar, too. Haiti’s population is only 12m. But there are about 850,000 Haitian migrants in the United States. Add in those who claim Haitian ancestry and you get a bigger number. This is an “only in America” thing. In fact, pretty much every nation represented in the World Cup will have a bar just in New York City that is designated as home turf. A dedicated football fan could travel the world, attending home games for every country from Algeria to Uzbekistan, using only the subway. And then there is the way that American cities have adopted nationalities—think of the Scots in Boston or the Bosnians in St Louis. If you had to pick one country as the permanent host of the World Cup, America would surely be it.   世界盃不懂美國 約翰·普里多，本刊執行編輯，談美國為何是舉辦足球賽事的絕佳之地 2026年6月27日｜3分鐘閱讀 足球迷們最愛抱怨美國。這個資本主義肆意橫行的國家，竟然膽敢將足球視為一門生意。本屆世界盃引入了「補水休息」環節。球迷們懷疑，這不過是日後增加更多電視廣告時段的第一步——一項改變他們心愛運動的銅臭味創新。入場看球的費用高得嚇人。動態定價或許適用於NFL或NBA——這些賽事的門票市場與其他奢侈品市場並無二致。但足球有其根基，它發源於同一工廠的工人球隊，承載著不同的價值觀。用經濟學術語來說，這些國家將足球視為公共財。 在許多足球大國，人們普遍期望比賽能在所有人均可收看的頻道上播出。若將版權以利潤最大化的方式售予出價最高者，導致大批民眾無緣觀賽，那將對當屆政府構成生死攸關的政治危機。 還有一個問題：唐納·川普的美國，似乎對接待外賓毫無興趣。今年稍早，跟蹤報導移民及海關執法局 (ICE)在明尼阿波利斯行動的外國球迷，都不禁疑惑——美國究竟是否歡迎世界各地的訪客？接著，美國政府以索馬利亞籍為由，拒絕非洲頂尖裁判之一奧馬爾·阿爾坦入境。此外，美國總統將親自上台為冠軍球隊頒獎，這讓運動員們更難辯稱本屆賽事與政治無關。基於以上種種，外界的說法是：美國不懂世界盃。 這些抱怨不無道理，但卻忽略了一個重要原因——美國其實是舉辦本屆賽事的絕佳之地。上週我在曼哈頓下城一間酒吧觀看巴西對海地的比賽時，這一點深深震撼了我。 是的，票價高昂——這反映的是市場需求，也反映出美國人均財富遠高於其他已開發國家。紐澤西的公共交通系統確實令人惱火。川普政府也試圖破壞氣氛。但它並未得逞。原因，正在於美國的獨特性。 這個國家龐大的人口規模與悠久的移民歷史，意味著幾乎每個參賽國在美國都擁有相當規模的海外僑民社群。 這是美國人視為理所當然的事，但正因如此，即便不邀請任何海外訪客，美國也能辦出一屆相當精彩的世界盃。我在蘇活區那間酒吧裡，巴西人多到我們佔領了人行道，蔓延至街道。（我在聖保羅生活過三年，所以巴西是我的第二支主隊。）熱愛足球的巴西擁有兩億兩千萬人口、五度稱霸世界盃，要在倫敦或巴黎找到一間被黃綠球衣淹沒的酒吧並不難——而那些穿著球衣的人，原來還能一邊端著飲料一邊跳森巴。 更難得的是，酒吧裡還有一群海地球迷。海地總人口僅一千兩百萬，但在美國卻有約八十五萬海地移民。若再加上具有海地血統的人，這個數字還會更大。這就是「只有在美國才會發生」的事。 事實上，幾乎每個參賽國都能在紐約市找到一間屬於自己的「主場酒吧」。一位鐵桿球迷只需搭乘地鐵，便能「環遊世界」，為每一支從阿爾及利亞到烏茲別克的球隊觀看主場比賽。更不用說美國城市認養各國文化的傳統——想想波士頓的蘇格蘭人，或聖路易的波士尼亞人。若真要選出一個永久舉辦世界盃的國家，美國無疑是不二之選。 -- Hosting provided by SoundOn

China’s high-tech rise is leaving much of the country behind That could make a** starkly unequal** country even more so Jun 2nd 2026|TIANSHUI|6 min read Listen to this story LIKE MANY cities in China’s hinterland, Tianshui, in the western province of Gansu, is full of dusty and disused factories. But over the past decade it has also become an unlikely high-tech hub. New industrial parks have sprouted, offering companies cheap energy, financing and land deals. The city has already built an exhibition hall to display the zippy products it hopes to make in the future, called “Tianshui Industry 2050”. All this will please officials in far-off Beijing. They want China’s rust belt to reinvent itself with technology. Just recently they released yet another plan for urban renewal, calling for the transformation of stagnant cities into “innovative” places that can offer their inhabitants a “high-quality life”. Yet for all the fanfare, the people of Tianshui are not much better off. The new factories have failed to offset a broader slowdown in the city’s economy: ten years ago Tianshui’s GDP per person was 16% of Beijing’s; now it is 14%. In 2025 the city’s economy grew at a rate two percentage points slower than the national average. Nor, say locals, have the highly automated facilities created many jobs for them. Throngs of young people are leaving in search of better opportunities. Over the past decade, its population has shrunk by nearly half a million to 2.9m. At this rate, by 2050 there will be few workers left. The tale of Tianshui shows the limits of China’s big bet on advanced manufacturing. The Communist Party has decided the country’s economic future lies in making world-beating technology. So hundreds of cities across the country are, like Tianshui, trying hard to do so. But while the high-tech drive has helped some brainy, connected and wealthy cities become even richer, most of those in the hinterland lack the supply chains or talent to take advantage of it. After all, some 60% of China’s workforce—about 500m people—do not even have a high-school education. Many of them live in smaller and poorer cities. For centuries Tianshui was more a cultural than an economic hub. Legend has it that China’s first emperor, a serpent-bodied demi-god, was born there; Buddhist grottoes have been cut out of the cliffs nearby. But in the 1960s Tianshui industrialised. It became an important cog in the state-planned economy; its factories made tractors, ball bearings and matches. A tale of two cities Dormitories, schools and hospitals were built to cater to the swelling labour force. One of the workers was Ms Dong, now in her 80s, who enjoyed an ultra-stable “iron rice bowl” job at a printing press. She retired in her 40s with a good pension and a guarantee that her son could inherit her job. But most local factories were not competitive enough to survive China’s bumpy transition to a more market-driven economy in the 1980s and 1990s. In the past ten years a new generation of high-tech factories, making gizmos such as sensors and machine tools, has sprung up. But they have not brought many jobs with them. An exhibit at Tianshui’s museum displays snapshots of the city’s factory floors over the decades. Each shows fewer workers and more robots. Most of the factory positions that are available pay around 3,000 yuan ($440) a month, according to locals, half what they could get in a big city like Shanghai. “I’d like to stay here to settle down, but there’s just no good jobs for young people,” says Wen Jin, a 27-year-old Tianshui native who has moved to Jiangsu province, a wealthy eastern region. The city’s best days are firmly in the past, reckons Ma Xin, another local in his 20s who is hoping to get out. High-tech industries do create well-paid jobs as well, in research and development, for instance. But these positions are mainly clustered in big cities near China’s coast, like Beijing, Shanghai and Shenzhen, which boast the best universities, brightest graduates and access to the densest supply chains. Tech-sector salaries in such places have shot up in recent years, some topping 1m yuan a year. But few parts of inland China have a chance of attracting such stellar jobs, explains Dan Wang of Eurasia Group, a consultancy. “The vast majority of Chinese cities are stuck with what they have.” Even as Tianshui is unable to gain from China’s new economic model, it struggles with the problems of the old one. China’s growth has slowed in recent years thanks, in large part, to a lingering property crisis that is gripping the country. House prices have fallen fastest in smaller cities like Tianshui. Unfinished concrete flats litter the city’s outskirts; investment in property there fell by over 40% last year. This in turn has been a drag on consumption because people feel poorer. Total retail sales in Tianshui slipped by a little over 5% in 2025. Gloomy walks through its malls reveal shuttered shops and empty restaurants. There is a risk that these trends, multiplied across China’s smaller cities, will cause the gaps between the country’s haves and have-nots to widen further. The economic slowdown seems to have hit the poor hardest, according to data compiled by Li Shi, a professor at Zhejiang University in the eastern city of Hangzhou. In a paper published in April he cited surveys showing that China’s bottom tenth of workers by income saw their wages rise by only 2% per year from 2018 to 2023, compared with an average annual growth rate of around 5%. China’s government knows that high-tech factories are not an economic panacea. Investment in machinery and other physical assets “has fuelled China’s economic boom, but its returns have gradually declined”, noted Xinhua, the official state news agency, earlier this year. China’s latest five-year plan, which was released in March and covers the period to 2030, called for “investing in people” as well as factories, to make the economy more equal. In practice, that would mean much more government spending on education, particularly for children from poor families, giving them the skills and knowledge to get better-paid jobs. But as the people of Tianshui would attest, that is easier said than done. The city’s budget is far smaller than those of wealthy places—it spends less than one-third of the amount per pupil in school that Beijing does, for example. And the economic downturn threatens to turn this into a cycle of decline; the city’s fiscal revenues fell by nearly a tenth last year. Few of Tianshui’s students ever make it to the country’s best universities. “It’s too difficult for children to compete here,” says Shi Tingting, the mother of a 12-year-old girl. “They end up trapped.” ■ 中國的高科技崛起正將這個國家的很大一部分留在後面 那可能使一個明顯不平等的國家變得更加如此 2026年6月2日 | 天水 | 6分鐘閱讀時間 像許多在中國腹地的城市一樣，天水，在西部省份甘肅，充滿了落滿灰塵的和廢棄的工廠。 但在過去十年間它也已經成為一個出乎意料的高科技樞紐。 新工業園區已經湧現，提供公司廉價的能源、融資和土地交易。 這座城市已經建造了一個展覽廳來展示它希望在未來製造的敏捷的產品，叫做「天水工業2050」。 這一切將會取悅在遙遠北京的官員們。 他們想要中國的鐵鏽地帶用技術來重塑自身。 就在最近他們發布了又一個關於城市更新的計劃，呼籲將停滯的城市轉型為能夠為他們的居民提供「高質量生活」的「創新的」地方。 然而儘管有所有的宣傳，天水的人民並沒有好過多少。 新的工廠已經未能抵消在這座城市經濟中一個更廣泛的放緩：十年前天水的人均GDP是北京的16%；現在它是14%。 在2025年這座城市的經濟以一個比全國平均慢兩個百分點的速度增長。 當地人說，高度自動化的設施也沒有為他們創造許多工作。 成群的年輕人正為了尋找更好的機會而離開。 在過去十年間，它的人口已經萎縮了接近五十萬到290萬。 照這個速度，到2050年將會幾乎沒有工人留下。 天水的故事顯示了中國在先進製造業上的大賭注的局限。 共產黨已經決定這個國家的經濟未來在於製造舉世無雙的技術。 所以全國各地數以百計的城市正像天水一樣，努力嘗試這樣做。 但是雖然高科技驅動已經幫助了一些聰明的、有脈絡的且富裕的城市變得更加富有，大多數在腹地的那此城市缺乏供應鏈或人才來利用它。 畢竟，大約60%的中國勞動力——大約5億人——甚至沒有一個高中教育。 他們中的許多人居住在較小的和較窮的城市。 幾個世紀以來天水更多是一個文化的而非經濟的樞紐。 傳說它說中國的第一位皇帝，一個蛇身的半神，出生在那裡；佛教石窟已經從附近的峭壁上被開鑿出來。 但在1960年代天水工業化了。 它成為了國家計劃經濟中一個重要的齒輪；它的工廠製造拖拉機、滾珠軸承和火柴。 雙城記 宿舍、學校和醫院被建造來迎合膨脹的勞動力。 工人中的一個是董女士，現在在她80多歲，其在一家印刷廠享受了一個超穩定的「鐵飯碗」工作。 她在她40多歲時退休，帶著一份好的退休金和一個她的兒子能夠繼承她的工作的保證。 但大多數本地的工廠不夠有競爭力到足以在1980年代和1990年代中國向一個更加市場驅動的經濟的顛簸轉型中倖存。 在過去十年中新一代的高科技工廠，製造諸如傳感器和機床之類的小玩意，已經湧現。 但是它們沒有隨之帶來許多工作。 在天水博物館的一個展品展示了這座城市的工廠車間幾十年來的快照。 每一張顯示更少的工人和更多的機器人。 根據當地人，大多數可獲得的工廠職位每個月支付大約3,000元（440美元），是他們在一個像上海這樣的大城市所能得到的（數額的）一半。 「我想留在此地定居，但是就是沒有給年輕人的好工作，」文進說，一位27歲的天水本地人，其已經搬到了江蘇省，一個富裕的東部地區。 這座城市最好的日子過去了，馬鑫估計，另一位在他20多歲正希望離開的當地人。 高科技產業也確實創造高薪的工作，例如，在研究與開發中。 但這些職位主要聚集在靠近中國海岸的大城市，像北京、上海和深圳，其坐擁最好的大學、最聰明的畢業生以及獲取最密集供應鏈的途徑。 在這樣的地方科技部門的薪水在近年來已經飆升，一些一年超過100萬元。 但是很少有中國內陸的部分有吸引這樣輝煌工作的機會，歐亞集團（一家諮詢公司）的丹·王解釋道。 「絕大多數的中國城市都與他們所擁有的困在一起。」 即使當天水無法從中國的新經濟模式中獲益時，它與舊模式的問題作鬥爭。 中國的增長在近年來已經放緩，這在很大程度上”歸功”於一場正席捲這個國家的揮之不去的房地產危機。 房價在像天水這樣較小的城市中下跌得最快。 未完工的水泥公寓散落在这座城市的郊區；在那裡的房地產投資在去年下降了超過40%。 這反過來已經成為消費的一個拖累，因為人們感覺更窮了。 天水的總零售銷售在2025年下滑了稍微超過5%。 穿過它的購物中心的憂鬱散步透露出關門的商店和空蕩的餐廳。 有一個風險是這些趨勢，在中國較小的城市中乘數放大，將會導致這個國家的富有者與一無所有者之間的差距進一步擴大。 經濟放緩似乎已經對窮人打擊最重，根據由李實編纂的數據，一位在東部城市杭州的浙江大學的教授。 在四月發表的一篇論文中他引用了調查，顯示按收入計中國最底層十分之一的工人看到他們的工資從2018年到2023年每年只增長2%，相比之下與大約5%的平均年度增長率。 中國的政府知道高科技工廠不是一個經濟萬靈丹。 在機械和其他實體資產上的投資「已經給中國的經濟繁榮注入燃料，但它的回報已經逐漸下降」，新華社，官方的國家通訊社，在今年早些時候指出。 中國最新的五年計劃，其在三月被發布且涵蓋到2030年的時期，呼籲「投資於人」以及工廠，來使經濟更加平等。 在實踐中，那將意味著多得多的在教育上的政府開支，特別是對於來自貧困家庭的兒童，給予他們技能和知識來獲得更好薪水的工作。 但是正如天水的人民會證實的，那是說起來容易做起來難。 這座城市的預算遠小於那些富裕地方的——例如，它在學校中在每位學生身上花費的金額少於北京所做的三分之一。 而且經濟低迷威脅要將這變成一個衰退的循環；這座城市的財政收入在去年下降了接近十分之一。 很少有天水的學生曾經成功進入這個最好的大學。 「對於孩子們來說在這裡競爭太困難了，」史婷婷說，一位12歲女孩的母親。 「他們最終被困住。」 -- Hosting provided by SoundOn

Sex tourists fuel outrage about vice in Japan The debates do not have women’s interests at heart Jun 4th 2026|Tokyo|4 min read MAI, a 34-year-old woman in Tokyo, used to work at a hospital. But when covid-19 overwhelmed the wards, she found her work too stressful. A single mother, she also needed money for her family. Lured by higher pay, she entered the sex trade, first working as a porn actress before becoming a deriheru or “delivery health” worker—slang for call-girls who visit clients at home or in hotels. For a two-hour session, she earns ¥30,000 ($190). Mai is among hundreds of thousands of women working in Japan’s sprawling sex industry. The business, thought to be worth ¥2trn-5trn ($12bn-31bn) a year, is woven into male social life. One study in 2022 found that 48% of Japanese men had paid for sex at some point, compared with 11% in Britain. Hagiwara, a 63-year-old man in Tokyo, recalls being taken to a brothel by senior colleagues after joining a company, as a rite of passage. Emu, a man in his 30s, says “most men around me have been at least once.” Lately, however, lawmakers’ tolerance for the industry has come under much strain. Two recent triggers have encouraged Japanese to re-examine the confusing thicket of laws and conventions that govern how sex work is policed. One was an outrageous crime: last year authorities rescued a 12-year-old Thai girl who had been trafficked to Japan and forced to work at a sex shop in Tokyo. The second concern has been the growing visibility of women who sell sex around Okubo Park, near Tokyo’s red-light district (where the haggling more ordinarily goes on behind neon-lit doors). Relatively few women are involved in this. Nonetheless, solicitation (waiting for or approaching clients in public) is illegal in Japan. The sight of women openly waiting for clients has unsettled the public. Compounding the public debate is the fact that some of their customers are foreign tourists, lured to Japan by the cheapness of the yen. Videos of them approaching women in Okubo Park have spread rapidly online. “It is truly lamentable,” said Kamiya Sohei, leader of the right-wing populist Sanseito party, in a video. Behind the outrage lies a sense of wounded pride: during Japan’s boom years in the 1970s and 1980s, it was Japanese men who went abroad for sex. The authorities have decided to act—at least where the streetwalkers are concerned. Recently women around Okubo Park have been detained or arrested. Yet campaigners say it is unfair that authorities have not also been trying to punish the buyers. “Women are taken away by the police—while the men who buy sex stand beside them smirking,” says Kanajiri Kazuna of PAPS, a women’s-rights group. In November an opposition lawmaker raised this disparity in parliament. In response Takaichi Sanae, the prime minister, ordered the justice ministry to re-examine current practices and consider reforms. The prospect of change has sparked very broad debate about how Japan could improve its policing of sex work. Some Japanese feminists would like their country to implement the Nordic style of regulation adopted by Sweden, France and others, which criminalises buying sex while shielding sellers themselves from prosecution. “Buying sex is a form of violence against women,” says Ms Kanajiri. Other Japanese argue that getting tougher on buyers will drive sex work underground, leaving women more exposed to violence. Some want the industry fully legalised and regulated, as in Germany and the Netherlands. Nakayama Misato of Siente, a sex-work advocacy group, argues that criminalisation can mean that women are treated merely as victims, ignoring their agency. “Doing sex work is not a bad thing—it’s a valid way of making a living.” To be more than skin deep, any changes would have to apply not only to streetwalking but to Japan’s vast indoor sex-industry, the laws for which are riddled with loopholes and selectively enforced. Consider the practice of “soapland”, in which customers ostensibly pay to be bathed; if sex happens to take place in the process, officials generally turn a blind eye. Takao Yasuo of Curtin University says this is typical of Japan’s approach to regulating the sex industry. The priority is to keep vice decorously out of public view. A big rethink seems unlikely. For now, the justice ministry is narrowly focused on street prostitution. Taking care of that is “the lowest-cost, highest-visibility form of enforcement available to the state”, notes Mr Takao. “Many lawmakers, especially conservatives, are sensitive to the idea of women becoming sexually promiscuous,” says Shiomura Ayaka, a lawmaker. Women openly soliciting sex in public have become symbols of social disorder. ■ 性觀光客加劇了對日本色情業的憤怒 這些辯論的核心並非以女性的利益為出發點 2026年6月4日 | 東京 | 4分鐘閱讀時間 小麥（Mai），一名在東京的34歲女性，過去曾在一家醫院工作。但是當新冠肺炎壓垮了病房時，她發現自己的工作壓力太大了。作為一名單親媽媽，她也需要錢來養家。在更高薪水的誘惑下，她進入了性產業，先是擔任成人片（A片）女演員，之後成為一名「deriheru」或「派遣健康（外送茶）」工作者——這是指前往客戶家中或飯店服務的應召女郎的俚語。一個兩小時的服務時段，她可以賺取30,000日圓（190美元）。 小麥是日本龐大且擴張的性產業中，成千上萬名工作女性的其中之一。這項據信每年價值達2兆至5兆日圓（120億至310億美元）的行業，已經融入了男性的社交生活之中。2022年的一項研究發現，48%的日本男性曾在某個時間點花錢買過性服務，相比之下，英國的比例僅為11%。萩原（Hagiwara），一名在東京的63歲男性，回憶起自己在進入公司後，被資深同事帶去妓院，將其視為一種成年禮。艾姆（Emu），一名30多歲的男性說：「我身邊的大多數男人至少都去過一次。」 然而，近來立法者對該產業的容忍度受到了極大的考驗。最近的兩個導火線，促使日本人重新審視管理如何對性工作進行治安維持的、那套令人困惑且錯綜複雜的法律與慣例。其中一個是一起駭人聽聞的罪行：去年當局解救了一名被販賣到日本、並被強迫在東京一家情色店工作的12歲泰國女孩。 第二個擔憂則是，在東京紅燈區附近的「大久保公園」周邊，販賣性服務的女性其能見度越來越高（在那裡，討價還價通常是在霓虹燈閃爍的門後進行）。雖然涉及此事的女性相對較少。儘管如此，拉客（在公共場所等待或主動接近客戶）在日本是違法的。女性公開等待客戶的景象讓公眾感到不安。 使公眾辯論更加劇的複雜因素是，她們的某些客戶是外國觀光客，他們因日圓貶值（便宜）而被吸引到日本。他們在大久保公園接近女性的影片在網路上迅速傳播。右翼民粹主義「參政黨」黨魁神谷宗幣（Kamiya Sohei）在一段影片中表示：「這真是令人悲哀。」在憤怒的背後，隱藏著一種受挫的自尊心：在1970年代和1980年代日本經濟繁榮的歲月裡，是日本男性前往國外進行性觀光。 當局已決定採取行動——至少在涉及街頭流鶯的部分。最近，大久保公園周邊的女性遭到拘留或逮捕。然而，社會運動人士表示，當局沒有同時試圖懲罰買家是不公平的。「女性被警察帶走——而買性的男人則站在一旁傻笑，」女權團體 PAPS 的金尻カズナ（Kanajiri Kazuna）表示。去年11月，一名反對黨議員在國會中提出了這一不平等待遇。作為回應，首相高市早苗（Takaichi Sanae）命令法務省重新審視現行做法並考慮改革。 變革的前景引發了關於日本如何改善性工作治安管理的廣泛辯論。一些日本女性主義者希望她們的國家能實施瑞典、法國等國採用的「北歐模式」規範，該模式將買性行為定為刑事犯罪，同時保護賣性者免受起訴。「買性是對女性的一種暴力形式，」金尻女士說。 其他日本人則認為，對買家採取更嚴厲的態度會將性工作推向地下，使女性更容易暴露於暴力之中。有些人則希望像德國和荷蘭那樣，將該產業完全合法化並進行規範。性工作倡議團體 Siente 的中山美里（Nakayama Misato）認為，刑事犯罪化可能意味著女性僅被視為受害者，而忽略了她們的自主權（主體性）。「從事性工作並不是一件壞事——這是一種正當的謀生方式。」 若要使任何改變不只是流於表面（深層改革），這些改變不僅必須適用於街頭拉客，還必須適用於日本龐大的室內性產業，而該產業的法律充斥著漏洞，且屬於選擇性執法。以「泡泡浴（soapland）」的慣例為例，顧客表面上是付費接受沐浴服務；如果在過程中偶然發生了性行為，官員通常會睜一隻眼閉一隻眼。寇廷大學（Curtin University）的鷹尾安雄（Takao Yasuo）表示，這是日本規範性產業的典型做法。其首要任務是體面地將惡行（色情）保持在公眾視線之外。 一場徹底的大反思似乎不太可能發生。目前，法務省正狹隘地將焦點集中在街頭賣淫上。鷹尾先生指出，處理這件事是「國家目前可行、成本最低且能見度最高的一種執法形式」。國會議員塩村文夏（Shiomura Ayaka）表示：「許多立法者，特別是保守派，對於女性變得性生活混亂（放蕩）的想法非常敏感。」在公共場合公開拉客的女性，已經成為了社會秩序混亂的象徵。 -- Hosting provided by SoundOn