- Published on
DeepSeek颠覆硅谷的AI认知
- Authors
DeepSeek颠覆硅谷的AI认知
一项人工智能突破在股市掀起冲击波、令硅谷巨头感到恐慌、引发了关于美国技术主导地位终结的热议。 
双语精读
Why DeepSeek Could Change What Silicon Valley Believes About AI
为什么DeepSeek可能改变硅谷关于AI的认知
The artificial intelligence breakthrough that is sending shock waves through stock markets, spooking Silicon Valley giants, and generating breathless takes about the end of America's technological dominance arrived with an unassuming, wonky title: "Incentivizing Reasoning Capability in LLMs via Reinforcement Learning."
一项人工智能突破在股市掀起冲击波、令硅谷巨头感到恐慌、引发了关于美国技术主导地位终结的热议，这项突破却以一个低调、学究气的标题出现：“通过强化学习激励大语言模型的推理能力”。
The 22-page paper, released last week by a scrappy Chinese AI start-up called DeepSeek, didn't immediately set off alarm bells. It took a few days for researchers to digest the paper's claims, and the implications of what it described.
这份22页的论文由一家结构松散的、名为DeepSeek的中国AI初创公司在上周发布，当时论文并没有立即敲响警钟。研究人员花了几天时间来消化论文的观点，以及所描述内容的可能影响。
The company had created a new AI model called DeepSeek-R1, built by a team of researchers who claimed to have used a modest number of second-rate AI chips to match the performance of leading American AI Models at a fraction of the cost.
该公司创造了一个名为DeepSeek-R1的AI新模型，构建模型的研究团队声称，他们用数量不多的二流AI芯片、以极低的成本就达到了堪与美国一流AI公司相媲美的性能。
DeepSeek said it had done this by using clever engineering to substitute for raw computing horsepower. And it had done it in China, a country many experts thought was in a distant second place in the global AI race.
DeepSeek表示，它是通过用巧妙的工程技术来替代原始算力而做到的。而且它是在中国做到了这一点，许多专家认为中国在全球AI竞赛中处于远远落后的第二位。
Some industry watchers initially reacted to DeepSeek's breakthrough with disbelief. Surely, they thought, DeepSeek had cheated to achieve R1's results, or fudged their numbers to make their model look more impressive than it was. Maybe R1 was actually just a clever re-skinning of American AI models that didn't represent much in the way of real progress.
一些行业观察家最初对DeepSeek的突破表示怀疑。他们认为，DeepSeek为了达到R1的结果肯定作了弊，或者捏造了数据，让模型看起来比实际更厉害。或许R1实际上只是对美国AI模型进行了巧妙的换皮包装，并不代表真正取得了多少进步。
Eventually, as more people dug into the details of DeepSeek-R1 - which, unlike most leading AI models, was released as open-source software, allowing outsiders to examine its inner workings more closely - their skepticism morphed into worry.
最终，随着越来越多的人深入研究DeepSeek-R1的细节（与大多数领先的AI模型不同，它是作为开源软件发布的，让外部人员能更仔细地审视其内部工作原理），他们的怀疑变成了担忧。
Based on conversations I've had with industry insiders, and a week's worth of experts poking around and testing the paper's findings for themselves, it appears to be throwing into question several major assumptions the American tech industry has been making.
根据我与业内人士的交谈，以及一周以来专家们的探索和对论文结果的亲自测试，这个模型似乎让美国科技行业一直做出的几个主要假设受到了质疑。
The first is the assumption that in order to build cutting-edge AI models, you need to spend huge amounts of money on powerful chips and data centers. It's hard to overstate how foundational this dogma has become.
第一个假设是，为了构建最尖端的AI模型，你需要在强大的芯片和数据中心上花费巨额资金。这个教条的根深蒂固怎么夸大都不为过。
Companies like Microsoft, Meta and Google have already spent tens of billions of dollars building out the infrastructure they thought was needed to build and run next-generation AI models.
微软、Meta、谷歌之类的公司已经花费了数百亿美元来建造他们认为构建和运行下一代AI模型所需的基础设施。
But DeepSeek's breakthrough on cost challenges the "bigger is better" narrative that has driven the AI arms race in recent years by showing that relatively small models, when trained properly, can match or exceed the performance of much bigger models.
但是DeepSeek在成本方面的突破挑战了近年来推动AI军备竞赛的“越大越好”的叙事，它让人们看到，如果训练得当，相对较小的模型也可以达到或超过大模型的性能。
词汇预习
- source [高考]
美[sɔːrs] | 英[sɔːs]
n. 来源；发源地；原始资料
v. 从...获得 - relatively [高考]
美[ˈrelətɪvli] | 英[ˈrelətɪvli]
adv. 相对地；比较地 - impressive [高考]
美[ɪmˈpresɪv] | 英[ɪmˈpresɪv]
adj. 给人深刻印象的；令人钦佩的 - inner [高考]
美[ˈɪnər] | 英[ˈɪnə(r)]
adj. 内心的；内部的；里面的
n. 里面 - technological [高考]
美[ˌteknəˈlɑːdʒɪkl] | 英[ˌteknəˈlɒdʒɪkl]
adj. 科技的；工艺的 - disbelief [高考]
美[ˌdɪsbɪˈliːf] | 英[ˌdɪsbɪˈliːf]
n. 不相信；怀疑 - initially [高考]
美[ɪˈnɪʃəli] | 英[ɪˈnɪʃəli]
adv. 最初 - exceed [高考]
美[ɪkˈsiːd] | 英[ɪkˈsiːd]
vt. 超过；超出
vi. 领先 - infrastructure [高考]
美[ˈɪnfrəstrʌktʃər] | 英[ˈɪnfrəstrʌktʃə(r)]
n. 基础；基础设施 - fraction [高考]
美[ˈfrækʃn] | 英[ˈfrækʃn]
n. 小部分；少许；分数；片段 - data [高考]
美[ˈdeɪtə] | 英[ˈdeɪtə]
n. 数据；资料 - rate [高考]
美[reɪt] | 英[reɪt]
n. 比率；速度；价格；费用；等级
v. 认为；估价；定等级；值得；怒斥；责骂 - generation [高考]
美[ˌdʒenəˈreɪʃn] | 英[ˌdʒenəˈreɪʃn]
n. 一代（人）；一代人的时间；（产品发展的）代；产生 - unlike [高考]
美[ˌʌnˈlaɪk] | 英[ˌʌnˈlaɪk]
adj. 不同的，不相似的
prep. 不像，与…不同；与 ... 平时不同 - intelligence [高考]
美[ɪnˈtelɪdʒəns] | 英[ɪnˈtelɪdʒəns]
n. 智力；智慧；才智；（尤指关于敌国的）情报 - artificial [高考]
美[ˌɑːrtɪˈfɪʃl] | 英[ˌɑːtɪˈfɪʃl]
adj. 人造的；虚伪的；武断的 - software [高考]
美[ˈsɔːftwer] | 英[ˈsɒftweə(r)]
n. 软件 - distant [高考]
美[ˈdɪstənt] | 英[ˈdɪstənt]
adj. 疏远的；遥远的；远房的；陌生的；远行的 - via [高考]
美[ˈviːə] | 英[ˈvaɪə]
prep. 经由；通过 - assumption [高考]
美[əˈsʌmpʃn] | 英[əˈsʌmpʃn]
n. 假定；设想；担任(职责等)；假装 - raw [高考]
美[rɔː] | 英[rɔː]
adj. 未加工的；生的；原始的；湿冷的
n. 触到某人的痛处 - substitute [高考]
美[ˈsʌbstɪtuːt] | 英[ˈsʌbstɪtjuːt]
n. 代用品；代替者；替补队员
vt. 代替；用 ... 代替；代以 - digest [高考]
美[daɪˈdʒest , ˈdaɪdʒest] | 英[daɪˈdʒest , ˈdaɪdʒest]
n. 摘要
vt. 消化；理解
vi. 消化 - breakthrough [高考]
美[ˈbreɪkθruː] | 英[ˈbreɪkθruː]
n. 突破，突破性进展；关键问题的解决 - breathless [高考]
美[ˈbreθləs] | 英[ˈbreθləs]
adj. 喘不过气来的 - in order [高考]
美[ɪn 'ɔːrdər] | 英[ɪn 'ɔːdə]
整齐；秩序井然；按顺序 - horsepower [四级]
美[ˈhɔːrspaʊər] | 英[ˈhɔːspaʊə(r)]
n. 马力(功率单位) - narrative [四级]
美[ˈnærətɪv] | 英[ˈnærətɪv]
n. 故事；叙述
adj. 叙述的；叙述体的 - tech [四级]
美[tek] | 英[tek]
n. 技术学校；技术；技术人员 - overstate [四级]
美[ˌoʊvərˈsteɪt] | 英[ˌəʊvəˈsteɪt]
v. 夸大，对…言过其实，夸大的叙述 - in the way [四级]
美[ɪn ði we] | 英[in ðə wei]
挡道，妨碍某人；[英]在场，在眼前，在附近 - in the way of [四级]
美[ɪn ðə weɪ əv] | 英[ɪn ðə weɪ əv]
在 ... 方面； 妨碍 - a fraction of [四级]
美[ə ˈfrækʃn əv] | 英[ə ˈfrækʃn ɒv]
一小部分, 少许 - engineering [四级]
美[ˌendʒɪˈnɪrɪŋ] | 英[ˌendʒɪˈnɪərɪŋ]
n. 工程学；工程 - leading [四级]
美[ˈliːdɪŋ , ˈledɪŋ] | 英[ˈliːdɪŋ , ˈledɪŋ]
adj. 领导的；指导的；主要的；在前的 - dominance [四级]
美[ˈdɑmənəns] | 英['dɒmɪnəns]
n. 支配；控制；统治；优势；【生】优势度，显性 - substitute for [六级]
美[ˈsʌbstɪˌtut fɔr] | 英[ˈsʌbstitju:t fɔ:]
用…代替， （使）代替;替换;抵换 - dogma [六级]
美[ˈdɔːɡmə] | 英[ˈdɒɡmə]
n. 教条；教义；信条 - skepticism [考研]
美[ˈskeptɪsɪzəm] | 英[ˈskeptɪsɪzəm]
n. 怀疑论；怀疑态度；怀疑主义
重点讲解
Why DeepSeek Could Change What Silicon Valley Believes About AI
为什么DeepSeek可能改变硅谷关于AI的认知
take n. 看法；观点
【例】What's your take on the plan?
你对这项计划有什么意见？
【近义词】opinion/view n. 观点
unassuming adj. 不炫耀的；低调谦逊的
【例】an unassuming house 一栋朴实的房子
【近义词】modest adj. 谦逊的；低调的
【拓展】assume v. 假设；摆姿态；假装
【例】to assume an air of superiority
摆出高人一等的样子
The artificial intelligence breakthrough that is sending shock waves through stock markets, spooking Silicon Valley giants, and generating breathless takes about the end of America’s technological dominance arrived with an unassuming, wonky title: “Incentivizing Reasoning Capability in LLMs via Reinforcement Learning.”
一项人工智能突破在股市掀起冲击波、令硅谷巨头感到恐慌、引发了关于美国技术主导地位终结的热议，这项突破却以一个低调、学究气的标题出现：“通过强化学习激励大语言模型的推理能力”。
take n. 看法；观点
【例】What's your take on the plan?
你对这项计划有什么意见？
【近义词】opinion/view n. 观点
unassuming adj. 不炫耀的；低调谦逊的
【例】an unassuming house 一栋朴实的房子
【近义词】modest adj. 谦逊的；低调的
【拓展】assume v. 假设；摆姿态；假装
【例】to assume an air of superiority
摆出高人一等的样子
The 22-page paper, released last week by a scrappy Chinese AI start-up called DeepSeek, didn’t immediately set off alarm bells. It took a few days for researchers to digest the paper’s claims, and the implications of what it described.
这份22页的论文由一家结构松散的、名为DeepSeek的中国AI初创公司在上周发布，当时论文并没有立即敲响警钟。研究人员花了几天时间来消化论文的观点，以及所描述内容的可能影响。
set off
1. 使爆炸；使响起
【例】to set off firecrackers 点燃鞭炮
2. 引起；激发
【例】to set off a wave of protest
激起一阵抗议
implication n. 可能的影响；牵连
【例】The development of the site will have implications for the surrounding countryside.
这个地点的开发将会影响到周围的乡村。
【近义词】influence/impact n. 影响
The company had created a new AI model called DeepSeek-R1, built by a team of researchers who claimed to have used a modest number of second-rate AI chips to match the performance of leading American AI Models at a fraction of the cost.
该公司创造了一个名为DeepSeek-R1的AI新模型，构建模型的研究团队声称，他们用数量不多的二流AI芯片、以极低的成本就达到了堪与美国一流AI公司相媲美的性能。
a fraction of 一小部分；一点儿
【例】He hesitated for a fraction of a second and then agreed.
他略微犹豫了一下，然后同意了。
substitute for 代替；取代
【例】Nothing can substitute for a devoted friend.
挚友是无可替代的。
【近义词】replace v. 代替；取代
【搭配】substitute A for B 用A代替B
substitute B with A 用A代替B
DeepSeek said it had done this by using clever engineering to substitute for raw computing horsepower. And it had done it in China, a country many experts thought was in a distant second place in the global AI race.
DeepSeek表示，它是通过用巧妙的工程技术来替代原始算力而做到的。而且它是在中国做到了这一点，许多专家认为中国在全球AI竞赛中处于远远落后的第二位。
a fraction of 一小部分；一点儿
【例】He hesitated for a fraction of a second and then agreed.
他略微犹豫了一下，然后同意了。
substitute for 代替；取代
【例】Nothing can substitute for a devoted friend.
挚友是无可替代的。
【近义词】replace v. 代替；取代
【搭配】substitute A for B 用A代替B
substitute B with A 用A代替B
Some industry watchers initially reacted to DeepSeek’s breakthrough with disbelief. Surely, they thought, DeepSeek had cheated to achieve R1’s results, or fudged their numbers to make their model look more impressive than it was. Maybe R1 was actually just a clever re-skinning of American AI models that didn’t represent much in the way of real progress.
一些行业观察家最初对DeepSeek的突破表示怀疑。他们认为，DeepSeek为了达到R1的结果肯定作了弊，或者捏造了数据，让模型看起来比实际更厉害。或许R1实际上只是对美国AI模型进行了巧妙的换皮包装，并不代表真正取得了多少进步。
initially adv. 起初；最初
【例】Initially, the system worked well.
开始时系统运转良好。
【近义词】originally adv. 原本；起初
at first 一开始；起初
Eventually, as more people dug into the details of DeepSeek-R1 — which, unlike most leading AI models, was released as open-source software, allowing outsiders to examine its inner workings more closely — their skepticism morphed into worry.
最终，随着越来越多的人深入研究DeepSeek-R1的细节（与大多数领先的AI模型不同，它是作为开源软件发布的，让外部人员能更仔细地审视其内部工作原理），他们的怀疑变成了担忧。
dig (deep) into 探究；细查
【例】You'll need to dig deep into the records to find the figures you want.
你必须仔细查阅档案才能找到你需要的数据。
【拓展】dig v. 挖掘
morph v. 变形；变化
【例】The robot morphed into a car.
机器人变成了一辆车。
【近义词】transform v. 变形；变化
Based on conversations I’ve had with industry insiders, and a week’s worth of experts poking around and testing the paper’s findings for themselves, it appears to be throwing into question several major assumptions the American tech industry has been making.
根据我与业内人士的交谈，以及一周以来专家们的探索和对论文结果的亲自测试，这个模型似乎让美国科技行业一直做出的几个主要假设受到了质疑。
The first is the assumption that in order to build cutting-edge AI models, you need to spend huge amounts of money on powerful chips and data centers. It’s hard to overstate how foundational this dogma has become.
第一个假设是，为了构建最尖端的AI模型，你需要在强大的芯片和数据中心上花费巨额资金。这个教条的根深蒂固怎么夸大都不为过。
overstate v. 夸大；夸张
【例】The seriousness of the crime cannot be overstated.
这一罪行的严重性怎么说也不为过。
【近义词】exaggerate v. 夸张
【反义词】understate v. 轻描淡写
Companies like Microsoft, Meta and Google have already spent tens of billions of dollars building out the infrastructure they thought was needed to build and run next-generation AI models.
微软、Meta、谷歌之类的公司已经花费了数百亿美元来建造他们认为构建和运行下一代AI模型所需的基础设施。
overstate v. 夸大；夸张
【例】The seriousness of the crime cannot be overstated.
这一罪行的严重性怎么说也不为过。
【近义词】exaggerate v. 夸张
【反义词】understate v. 轻描淡写
But DeepSeek’s breakthrough on cost challenges the “bigger is better” narrative that has driven the AI arms race in recent years by showing that relatively small models, when trained properly, can match or exceed the performance of much bigger models.
但是DeepSeek在成本方面的突破挑战了近年来推动AI军备竞赛的“越大越好”的叙事，它让人们看到，如果训练得当，相对较小的模型也可以达到或超过大模型的性能。
properly adv. 正确地；适当地
【例】You acted properly in approaching me first.
你先来找我，做得很对。
【反义词】improperly adv. 不恰当地
【近义词】appropriately adv. 恰当地；适当地
课后练习
What can be learned about the AI breakthrough described in the DeepSeek paper?
A. The breakthrough has already ended America’s technological dominance.
B. The breakthrough had a significant impact on the technological industry.
C. The breakthrough was celebrated as a major achievement in Silicon Valley.
D. The breakthrough caught widespread attention due to the strange title.What is special about the DeepSeek-R1 model?
A. It was developed by a large international team.
B. It was built using a large number of high-end AI chips.
C. It has achieved impressive performance at a low cost.
D. It has improved AI chips through clever engineering.What was American experts' eventual reaction towards DeepSeek-R1?
A. They didn’t believe in the breakthrough.
B. They were eager to try and use the model.
C. They were puzzled by its internal workings.
D. They worried about losing to it in the competition.What assumption is being questioned following DeepSeek's breakthrough?
A. That AI models need to be open-source to be successful.
B. That building cutting-edge AI models requires huge spending.
C. That only American companies can develop powerful AI models.
D. That the quality of data used in training is key for AI technology.
