“Mrs. Li, there’s no question your daughter is a bright girl, but I’m worried she’s not taking her future as seriously as she could. It’s never too early to start preparing for final exams, for example, so I often ask each student to share the books they’re reading with the class. Most cite textbooks, prep manuals, and selections from the school’s approved reading list. Fei-Fei’s answers this week concerned me, however, and—”
“My daughter has been an avid reader for her entire life,” my mother interjected, making no effort to conceal her contempt.
“Well, yes, of course. And uh, she certainly listed more books than anyone in the class—”
“So what’s the problem?”
The teacher sighed. It was clear this conversation wasn’t playing out as she’d expected.
“It’s what she’s reading. I mean, The Unbearable Lightness of Being? The Brontë sisters? And all these magazines she subscribes to. Marine life, fighter jets, something about UFOs ... the list goes on. She’s just not prioritizing literature that reflects the values and ideas of the curriculum.”
“Yeah? And?”
I sat beside my mother for the moment of silence that followed, doing my best to keep the glee coursing through my veins from reaching my face. The tension hung in the air for a moment or two longer, before the teacher leaned forward and made one last attempt, a new sternness in her voice.
“I’ll be frank, Mrs. Li. Your daughter may be bright, but there are many bright students in her class. That’s why intellect is only one ingredient in success. Another is the discipline to put aside one’s personal interests in favor of the studies that will prove most useful in the years ahead.”
I’m not sure if what my mother said next was meant to be a response. She looked down, speaking more softly than before. “Is this what Fei-Fei wants? Is this what I want for her?”
“What was that, Mrs. Li?” The teacher leaned in closer, clearly as confused as I was.
My mother sighed quietly, then looked back at the teacher as a determined expression returned to her face. The look would have to do. She was done throwing jabs. She stood up, thanked the teacher for her time, and gestured to me that we were leaving.
“I might have taught you too well, Fei-Fei,” she said with resignation as I tried to keep up with her pace on the walk out. “You don’t belong here any more than I did.”
“飛飛確實是個聰明的孩子,但我擔心她對自己的前途不夠上心。例如,現在得趕緊準備期末考,不能再拖了。我經常要求每一個學生跟班上同學分享他們正在閱讀的書。大多數學生讀的是課本、學習手冊,或是學校指定閱讀書目,並分享心得。然而,飛飛這個禮拜讀的東西讓我擔心……」
「我女兒從小就愛看書。」我母親插嘴道,毫不掩飾她的輕蔑。
「噢,當然。她讀的書肯定比班上任何一個同學都要來得多……」
「所以,有問題嗎?”
“老師嘆了一口氣。顯然,她沒想過會碰到這麼難纏的家長。
「問題出在她讀的書。她怎麽在讀《生命中不能承受之輕》?還有勃朗特姊妹的小說?還有,她訂的雜誌……都是海洋生物、戰鬥機、幽浮之類的。她應該先讀反映課程價值和思想的文學作品。」
「是嗎?然後呢?」
在接下來的靜默裡,我坐在母親身旁。我血液中流淌著喜悅,但我努力克制自己,不喜形於色。這種緊張對峙又持續了一、兩分鐘,接著老師傾身向前,做最後的努力。她的聲音多了一分嚴厲。
「我就老實跟您說吧,您的女兒也許很聰明,但她班上有很多聰明的學生。所以說,聰明才智只是成功的一個因素,另一個因素是紀律,必須放下個人興趣,專注於對未來最有用的學習。”
“我不知道我母親接下來說的話是不是在回應老師。她低著頭,聲音比之前更輕柔。「這是飛飛要的嗎?我希望她這麼做嗎?」
「您指的是?」老師靠過來,顯然和我一樣困惑。
我母親無聲地嘆了口氣,看著老師,臉上恢復堅定的表情。然而,她也只能這樣。該說的,她都說了。她站起來,對老師說聲謝謝,並向我示意我們得走了。
她快步走,我努力追上。「飛飛啊,也許我把你教得太好了,」她無奈地說,「你和我一樣,都不屬於這裡。”
For a Chinese student raised in the schools of Chengdu, my first days at Parsippany High School were an assault on the senses. The mood was manic and unsteady, and everything around me was brighter, faster, heavier, and noisier than the world I left behind. Nothing quite registered, no matter where I looked, as if the very nature of light and sound were somehow different here.
The colors alone were overwhelming. The clothes worn by students and teachers alike were more vibrant than anything I’d seen before, the palette ranging from earth tones to primaries to fluorescents, solid or broken by stripes and patterns, and adorned with lettering, illustrations, abstract designs, and logos. And they were accentuated with a blur of hats, sunglasses, earrings, purses, and branded backpacks, to say nothing of the makeup the girls wore—something I’d never once seen on teenagers.
The necessity of the backpacks became clear when I was given my new textbooks, which dwarfed their modestly sized paperback equivalents in China. Although most copies were scuffed and ragged along the edges, their quality shocked me; every class was accompanied by a heavily bound volume with vibrant cover art and hundreds upon hundreds of full-color pages. And their sheer weight seemed unreal.
Even more intense was the way everything moved. After a life spent in the ever-stationary seat of a Chinese student, the urgency with which the entire school seemed to spill from room to room was bewildering. My memories of China felt docile in comparison to the ritual that separated classes here, as loud bells unleashed even louder crowds that roared through halls like flash floods of teenage energy.
Finally, there were the people themselves. Rowdiness and irreverence seemed to be the norm among the kids. Even from behind a still-towering language barrier, I knew I’d never seen students talk to teachers the way Americans did. But what astonished me most was how the informalities appeared to cut both ways. Their dynamic was often adversarial, but jocular as well. Even warm. On an otherwise imposing first day, I instantly knew one thing: I would love American teachers.
“對一個在成都長大的中國學生來說,在帕西潘尼高中上學的第一天對我的感官造成劇烈衝擊。我覺得狂躁不安,周圍的一切都比我離開的那個世界要來得明亮、快速、沉重、喧鬧。這一切讓我眼花撩亂,彷彿光線和聲音的本質都不同了。
光是顏色就令人目不暇給。學生和老師都穿著我從來沒看過的衣服,色彩鮮豔,從大地色到原色到螢光色,有單色,也有條紋和圖案,有的還加上字母、插圖、抽象的設計和標誌。帽子、太陽眼鏡、耳環、皮包和名牌背包使他們的打扮更加亮眼,更不用說女孩的化妝──我從未在青少年身上看過這樣的妝容。
我一拿到教科書,就知道背包的必要了。相形之下,以前在中國使用的課本真是輕薄。儘管這裡的書封面破舊、邊緣磨損,但內容充實。每一門課都有一本厚重的教科書,封面圖案鮮豔奪目,全彩頁面多達好幾百頁,每一本都重得像巨大的磚頭。
更令人驚奇的是,這裡的學生總是跑來跑去。在中國上學,我們總是固定不動地坐在座位上,在美國,一到下課時間,整個學校的學生都從教室蜂擁而出,跑到下一間上課的教室。在中國,下課時間幾乎都是安安靜靜的,但在這裡,一旦鐘聲響起,青少年的精力有如瞬間爆發的山洪,哄鬧喧譁貫穿整個走廊。
最後,這裡的人也大不相同。這裡的孩子活潑吵鬧,不拘一格。即使我有語言障礙,也聽得出這裡的學生跟老師說話非常直接,這是我前所未見。“更讓我驚訝的是,老師和學生之間的互動似乎隨意、輕鬆,會互相詰問、開玩笑,甚至熱情洋溢。打從在美國上學的第一天,我就發覺:我喜歡這裡的老師。”
沒錯,上面的摘錄沒有涉及WordNet到ImageNet的迷人,沒有十九年到十八年到十五到五到一年可能讀完的博士學位劇情,而是完美的買櫝還珠的她的教育經歷對比。
原因很簡單,這本書,都該讀全本的,各取所需。
ImageNet“從將近10億張圖像篩選出1,500萬張,分布在22,000個類別之中,並由來自全球167個國家的48,000多名貢獻者標注完成”,教導機器學習的核心,是讓電腦從大數據自己找模式學習怎麼做,而不是依照人類的清晰指令;那麼,教學生學習呢?
真正的戲劇性是互聯網領域已經古代vs現代衹需要10年嗎?不止。戲劇的一直是反轉:
Modern AI was revealing itself to be a puzzle, and one whose pieces bore sharp edges.
“My daughter has been an avid reader for her entire life,” my mother interjected, making no effort to conceal her contempt.
“Well, yes, of course. And uh, she certainly listed more books than anyone in the class—”
“So what’s the problem?”
The teacher sighed. It was clear this conversation wasn’t playing out as she’d expected.
“It’s what she’s reading. I mean, The Unbearable Lightness of Being? The Brontë sisters? And all these magazines she subscribes to. Marine life, fighter jets, something about UFOs ... the list goes on. She’s just not prioritizing literature that reflects the values and ideas of the curriculum.”
“Yeah? And?”
I sat beside my mother for the moment of silence that followed, doing my best to keep the glee coursing through my veins from reaching my face. The tension hung in the air for a moment or two longer, before the teacher leaned forward and made one last attempt, a new sternness in her voice.
“I’ll be frank, Mrs. Li. Your daughter may be bright, but there are many bright students in her class. That’s why intellect is only one ingredient in success. Another is the discipline to put aside one’s personal interests in favor of the studies that will prove most useful in the years ahead.”
I’m not sure if what my mother said next was meant to be a response. She looked down, speaking more softly than before. “Is this what Fei-Fei wants? Is this what I want for her?”
“What was that, Mrs. Li?” The teacher leaned in closer, clearly as confused as I was.
My mother sighed quietly, then looked back at the teacher as a determined expression returned to her face. The look would have to do. She was done throwing jabs. She stood up, thanked the teacher for her time, and gestured to me that we were leaving.
“I might have taught you too well, Fei-Fei,” she said with resignation as I tried to keep up with her pace on the walk out. “You don’t belong here any more than I did.”
“飛飛確實是個聰明的孩子,但我擔心她對自己的前途不夠上心。例如,現在得趕緊準備期末考,不能再拖了。我經常要求每一個學生跟班上同學分享他們正在閱讀的書。大多數學生讀的是課本、學習手冊,或是學校指定閱讀書目,並分享心得。然而,飛飛這個禮拜讀的東西讓我擔心……」
「我女兒從小就愛看書。」我母親插嘴道,毫不掩飾她的輕蔑。
「噢,當然。她讀的書肯定比班上任何一個同學都要來得多……」
「所以,有問題嗎?”
“老師嘆了一口氣。顯然,她沒想過會碰到這麼難纏的家長。
「問題出在她讀的書。她怎麽在讀《生命中不能承受之輕》?還有勃朗特姊妹的小說?還有,她訂的雜誌……都是海洋生物、戰鬥機、幽浮之類的。她應該先讀反映課程價值和思想的文學作品。」
「是嗎?然後呢?」
在接下來的靜默裡,我坐在母親身旁。我血液中流淌著喜悅,但我努力克制自己,不喜形於色。這種緊張對峙又持續了一、兩分鐘,接著老師傾身向前,做最後的努力。她的聲音多了一分嚴厲。
「我就老實跟您說吧,您的女兒也許很聰明,但她班上有很多聰明的學生。所以說,聰明才智只是成功的一個因素,另一個因素是紀律,必須放下個人興趣,專注於對未來最有用的學習。”
“我不知道我母親接下來說的話是不是在回應老師。她低著頭,聲音比之前更輕柔。「這是飛飛要的嗎?我希望她這麼做嗎?」
「您指的是?」老師靠過來,顯然和我一樣困惑。
我母親無聲地嘆了口氣,看著老師,臉上恢復堅定的表情。然而,她也只能這樣。該說的,她都說了。她站起來,對老師說聲謝謝,並向我示意我們得走了。
她快步走,我努力追上。「飛飛啊,也許我把你教得太好了,」她無奈地說,「你和我一樣,都不屬於這裡。”
For a Chinese student raised in the schools of Chengdu, my first days at Parsippany High School were an assault on the senses. The mood was manic and unsteady, and everything around me was brighter, faster, heavier, and noisier than the world I left behind. Nothing quite registered, no matter where I looked, as if the very nature of light and sound were somehow different here.
The colors alone were overwhelming. The clothes worn by students and teachers alike were more vibrant than anything I’d seen before, the palette ranging from earth tones to primaries to fluorescents, solid or broken by stripes and patterns, and adorned with lettering, illustrations, abstract designs, and logos. And they were accentuated with a blur of hats, sunglasses, earrings, purses, and branded backpacks, to say nothing of the makeup the girls wore—something I’d never once seen on teenagers.
The necessity of the backpacks became clear when I was given my new textbooks, which dwarfed their modestly sized paperback equivalents in China. Although most copies were scuffed and ragged along the edges, their quality shocked me; every class was accompanied by a heavily bound volume with vibrant cover art and hundreds upon hundreds of full-color pages. And their sheer weight seemed unreal.
Even more intense was the way everything moved. After a life spent in the ever-stationary seat of a Chinese student, the urgency with which the entire school seemed to spill from room to room was bewildering. My memories of China felt docile in comparison to the ritual that separated classes here, as loud bells unleashed even louder crowds that roared through halls like flash floods of teenage energy.
Finally, there were the people themselves. Rowdiness and irreverence seemed to be the norm among the kids. Even from behind a still-towering language barrier, I knew I’d never seen students talk to teachers the way Americans did. But what astonished me most was how the informalities appeared to cut both ways. Their dynamic was often adversarial, but jocular as well. Even warm. On an otherwise imposing first day, I instantly knew one thing: I would love American teachers.
“對一個在成都長大的中國學生來說,在帕西潘尼高中上學的第一天對我的感官造成劇烈衝擊。我覺得狂躁不安,周圍的一切都比我離開的那個世界要來得明亮、快速、沉重、喧鬧。這一切讓我眼花撩亂,彷彿光線和聲音的本質都不同了。
光是顏色就令人目不暇給。學生和老師都穿著我從來沒看過的衣服,色彩鮮豔,從大地色到原色到螢光色,有單色,也有條紋和圖案,有的還加上字母、插圖、抽象的設計和標誌。帽子、太陽眼鏡、耳環、皮包和名牌背包使他們的打扮更加亮眼,更不用說女孩的化妝──我從未在青少年身上看過這樣的妝容。
我一拿到教科書,就知道背包的必要了。相形之下,以前在中國使用的課本真是輕薄。儘管這裡的書封面破舊、邊緣磨損,但內容充實。每一門課都有一本厚重的教科書,封面圖案鮮豔奪目,全彩頁面多達好幾百頁,每一本都重得像巨大的磚頭。
更令人驚奇的是,這裡的學生總是跑來跑去。在中國上學,我們總是固定不動地坐在座位上,在美國,一到下課時間,整個學校的學生都從教室蜂擁而出,跑到下一間上課的教室。在中國,下課時間幾乎都是安安靜靜的,但在這裡,一旦鐘聲響起,青少年的精力有如瞬間爆發的山洪,哄鬧喧譁貫穿整個走廊。
最後,這裡的人也大不相同。這裡的孩子活潑吵鬧,不拘一格。即使我有語言障礙,也聽得出這裡的學生跟老師說話非常直接,這是我前所未見。“更讓我驚訝的是,老師和學生之間的互動似乎隨意、輕鬆,會互相詰問、開玩笑,甚至熱情洋溢。打從在美國上學的第一天,我就發覺:我喜歡這裡的老師。”
沒錯,上面的摘錄沒有涉及WordNet到ImageNet的迷人,沒有十九年到十八年到十五到五到一年可能讀完的博士學位劇情,而是完美的買櫝還珠的她的教育經歷對比。
原因很簡單,這本書,都該讀全本的,各取所需。
ImageNet“從將近10億張圖像篩選出1,500萬張,分布在22,000個類別之中,並由來自全球167個國家的48,000多名貢獻者標注完成”,教導機器學習的核心,是讓電腦從大數據自己找模式學習怎麼做,而不是依照人類的清晰指令;那麼,教學生學習呢?
AlexNet呢?
In fact, AlexNet could catch all of those things—and many more— not just because it had been trained on ImageNet, but, crucially, because it remained faithful to the evolved spirit of biological vision. Rather than arbitrarily deciding in advance which features the network should look for, the authors allowed each of its hundreds of thousands of neurons to learn their own sensitivities gradually, exclusively from the training data, without manual intervention. Like a biological intelligence, AlexNet was a natural product of its environment.
“這一切、甚至還有更多更多的圖像,全都在AlexNet的掌握之中,而這不只因為它是用ImageNet訓練而成,更重要的是這套演算法忠實呈現了生物視覺的演化精神。這些作者沒有自己先入為主、預設這套神經網路演算法該去尋找什麼特徵,而是讓幾十萬個神經元完全從訓練資料集裡,自己去慢慢學習培養敏感度,人類不加以干預。AlexNet就像生物智能,是所屬環境的自然產物。” In fact, AlexNet could catch all of those things—and many more— not just because it had been trained on ImageNet, but, crucially, because it remained faithful to the evolved spirit of biological vision. Rather than arbitrarily deciding in advance which features the network should look for, the authors allowed each of its hundreds of thousands of neurons to learn their own sensitivities gradually, exclusively from the training data, without manual intervention. Like a biological intelligence, AlexNet was a natural product of its environment.
真正的戲劇性是互聯網領域已經古代vs現代衹需要10年嗎?不止。戲劇的一直是反轉:
overnight, a commitment often written off as misguided seemed downright prescient.
在李飛飛說全世界都應該感謝電子遊戲時,我們的教育在做什麼?禁手機呀。
But these were differences of degree, not of kind; at the level of theory, astonishingly little had changed. And yet AlexNet was performing like no other neural network in history.
How?
Part of the explanation was surely the hardware on which it ran. A defining flaw of neural networks—long considered fatal—was the difficulty of training them. Even the far smaller networks of bygone decades often proved impractical. Indeed, training a network like AlexNet with the world’s largest collection of images seemed incomprehensible. But technology had advanced significantly, especially when it came to cheap, high-performance computing hardware optimized for specific applications. Funnily enough, the world owed it all to the popularity of video games.
In yet another twist of fate, the style of number-crunching favored by neural networks is functionally similar to the kind used in rendering the graphics for video games—a multibillion-dollar industry that had been driving the advancement and commercialization of custom hardware since the 1990s, fueling the growth of megabrands like Nvidia, the company at the forefront of the field. By 2012, such hardware—specialized processors known as “graphics processing units,” or GPUs—had attained affordable, consumer-level status. For Hinton’s lab, that meant the silicon needed to bring AlexNet to life was no longer an investment requiring a government grant and construction permits. It was available off the shelf at Best Buy.
然而,這些都只是程度上的差異,而非本質上的差異;說到理論層面,差異小得驚人。話雖如此,AlexNet“表現的優異程度,卻遠遠超越歷史上其他神經網路。
這究竟是為什麼?
部分的原因肯定在於所使用的硬體。神經網路一直有個明顯、甚至長期以來被認為是致命的缺陷,就是訓練的難度實在太高。在過去幾十年間,即使是某些規模小得多的神經網路演算法,最後也都證明無法投入實用。說要用全世界最大的圖像集、訓練AlexNet這樣的神經網路,簡直就像是天方夜譚。然而此時技術已然有了長足的進步,特別是針對特定的應用,出現了成本實惠且性能優異的運算硬體。神奇的是,這全拜電玩流行所賜,全世界都該感謝電玩。
這可說是命運的另一個轉折,神經網路所需要的數字運算方式,功能上很類似電玩遊戲算繪(render,又譯「渲染」、「算圖」、「成像」等)圖像的方式,而電玩從1990年代以來已經發展成產值高達數十億美元的產業,推動了客製化硬體的進步與商業化,也帶動“輝達(Nvidia)這樣在產業最前端的超大型品牌成長。到了2012年,這種稱為「圖形處理器」(graphics processing units, GPU)的硬體已經來到了價格實惠、能做為日常消費用品的等級。在辛頓的實驗室看來,這代表即使沒有政府補助、無需申請施工許可,也已經能夠取得所需晶片,讓AlexNet從理論變成現實。只要到百思買(Best Buy)大賣場,架上直接就有得買!
一項過去常常被視為誤入歧途的努力,一夜之間成了眼光精準的先見之明。
太多我們不理解的,可能一直很精妙,最佳的例子,甚至我們都快忘了:
But these were differences of degree, not of kind; at the level of theory, astonishingly little had changed. And yet AlexNet was performing like no other neural network in history.
How?
Part of the explanation was surely the hardware on which it ran. A defining flaw of neural networks—long considered fatal—was the difficulty of training them. Even the far smaller networks of bygone decades often proved impractical. Indeed, training a network like AlexNet with the world’s largest collection of images seemed incomprehensible. But technology had advanced significantly, especially when it came to cheap, high-performance computing hardware optimized for specific applications. Funnily enough, the world owed it all to the popularity of video games.
In yet another twist of fate, the style of number-crunching favored by neural networks is functionally similar to the kind used in rendering the graphics for video games—a multibillion-dollar industry that had been driving the advancement and commercialization of custom hardware since the 1990s, fueling the growth of megabrands like Nvidia, the company at the forefront of the field. By 2012, such hardware—specialized processors known as “graphics processing units,” or GPUs—had attained affordable, consumer-level status. For Hinton’s lab, that meant the silicon needed to bring AlexNet to life was no longer an investment requiring a government grant and construction permits. It was available off the shelf at Best Buy.
然而,這些都只是程度上的差異,而非本質上的差異;說到理論層面,差異小得驚人。話雖如此,AlexNet“表現的優異程度,卻遠遠超越歷史上其他神經網路。
這究竟是為什麼?
部分的原因肯定在於所使用的硬體。神經網路一直有個明顯、甚至長期以來被認為是致命的缺陷,就是訓練的難度實在太高。在過去幾十年間,即使是某些規模小得多的神經網路演算法,最後也都證明無法投入實用。說要用全世界最大的圖像集、訓練AlexNet這樣的神經網路,簡直就像是天方夜譚。然而此時技術已然有了長足的進步,特別是針對特定的應用,出現了成本實惠且性能優異的運算硬體。神奇的是,這全拜電玩流行所賜,全世界都該感謝電玩。
這可說是命運的另一個轉折,神經網路所需要的數字運算方式,功能上很類似電玩遊戲算繪(render,又譯「渲染」、「算圖」、「成像」等)圖像的方式,而電玩從1990年代以來已經發展成產值高達數十億美元的產業,推動了客製化硬體的進步與商業化,也帶動“輝達(Nvidia)這樣在產業最前端的超大型品牌成長。到了2012年,這種稱為「圖形處理器」(graphics processing units, GPU)的硬體已經來到了價格實惠、能做為日常消費用品的等級。在辛頓的實驗室看來,這代表即使沒有政府補助、無需申請施工許可,也已經能夠取得所需晶片,讓AlexNet從理論變成現實。只要到百思買(Best Buy)大賣場,架上直接就有得買!
Modern AI was revealing itself to be a puzzle, and one whose pieces bore sharp edges.
“現代AI正展現出它就像一幅拼圖,每一片都有著尖銳的稜角。”
每一個學生,每一部手機,從盜火至今日盜出AI乃至AGI,哪一個沒有尖銳稜角?
禁止一個東西從來容易,想來人類第一次擁有火時,一定有部落是禁火的。
英文原版臺灣大陸兩個譯本,一併推給學生,要求:必讀!
Comments
Post a Comment