标题: 「外刊开麦」deepseek余震,中国AI在世界舞台上抢占美国光环 [打印本页] 作者: ZqUIC 时间: 昨天 01:24 标题: 「外刊开麦」deepseek余震,中国AI在世界舞台上抢占美国光环 Deepseek aftershocks
How China is quietly upstaging America来自：The Economist
原文
While America tech giants are spending megabucks to learn the secrets of their rivals‘'proprietary artificial-intelligence (AI) models, in China a different battle is under way. It is what Andrew Ng， a Stanford University-based AI boffin, recently called the “Darwinian life-or-death struggle” among builders of China's more open large language models (LLMs). Their competitive zeal should be a wake-up call for the West.
In January DeepSeek, a Chinese start-up, rocked global stockmarkets by making available free of charge an advanced AI model it had developed on a shoestring. Since then Chinese models from Alibaba, a tech giant, and others have quietly continued to gain traction abroad. When entrepreneurs walk into the offices of Andreessen Horowitz, a big American venture-capital firm, the odds these days are that their startups are using AI models made in China. “I'd say 80% chance [they are] using a Chinese open-source model, says Martin Casado.
Strictly speaking, China specialises in open-weight models. Unlike open-source software, for which the source code is shared publicly for anyone to modify, most non-proprietary LLMS provide only the numerical parameters (or “weights”) they have learned during training, and not the source code or underlying data. But call them what you will, on a variety of intelligence tests Chinese models released this year have outperformed their similarly open American peers, such as those from Meta, a social-media giant. Moreover, their capabilities are closing in on the best proprietary models.
OpenAI, maker of ChatGPT, illustrates the pressure this is creating. In the mid-2010s, it popularised the more open approach to AI (hence its name), but in order to make money and prevent misuse of increasingly powerful AI， it switched to selling only proprietary LLMs in 2020. Recently, though, it has seen an uptick in its customers use of open-weight models, including those from China, and wants to get in on the action. This month it released its first open-weight language model since 2019,called gpt-oss.
The use of the lower case is telling. The model is relatively small. In the same week OpenAI unveiled the long-awaited-and underwhelming-GPT-5, its latest proprietary model. Such timing made OpenAI's embrace of openness look half-hearted. That may prove true of other American companies' efforts, too. Ali Farhadi of the Allen Institute for AI, a Seattle-based non-profit organization, says that while Chinese firms go all-in, releasing their best models openly, American ones keep the “shiny new thing” proprietary. “As hard as it is for us all to swallow, I think we're behind [on open weights] now,”he says.
Even Meta reinforces that idea. It was widely celebrated in the open-source world for making Llama open and widely available. But Mark Zuckerberg, its boss, is now focused on building so-called super-intelligence. In the future, his company will be more cautious about what it chooses to make open, he has said.
From a business perspective, how much does this matter? After all, the revenues generated by American proprietary models are far greater than those produced by the Chinese open-weight ones. The valuations of the former-up to $500bn in the case of OpenAI-dwarf those of the latter；Alibaba's entire market capitalisation is only $285bn. It is easier to make money from proprietary models, and the proceeds can be poured back into innovation.
Yet open source is not just for the also-rans. Percy Liang, co-founder of Together AI, a platform for open-weight LLMs, says the models spur different forms of adoption than proprietary technology. They can be more easily adapted by companies, governments and researchers to the “nooks and crannies” of individual use cases, and help users run their AI tools on premises rather than relying on the cloud. Money can still be made from ancillary services, including support with customisation.
In other words, while American labs are betting big on the fortunes to be made by pushing the frontiers of intelligence, their open-weight Chinese rivals are more focused on encouraging adoption of AI. If they succeed, the DeepSeek shock may be just the beginning.
Key Takeaways
(, 下载次数: 0)
上传
点击文件名下载附件
on a shoestring
If you do something on a shoestring, you do it with a very small amount of money以极大批资金
The film was made on a shoestring.
这部电影是个小制造。
traction
the fact of an idea, product, etc. becoming popular or being accepted（观点或产品等）变得盛行（或被接受）
In our digital age, it takes less time for new words and phrases to gain traction than it did in the past.
在当前数码时代，新词或词组盛行起来要过去快得多。
close in
to gradually get nearer to someone, usually in order to attack them（通常指为了防御）渐渐围下去，包围，逼近
The advancing soldiers closed in on the town.
行进的兵士一步步包围了镇子。
underwhelming
not causing someone to feel any excitement or admiration未留下深入印象的;未惹起宏大反响的
The food was good but unfortunately we found the rest of the experience distinctly underwhelming.
食物很好，但很遗憾，我们觉得其他的体验分明不尽人意。
half-hearted
showing no enthusiasm or interest不热情的;兴味不大的
He made a half-hearted attempt to clear up the rubbish.
他不大情愿地去打扫渣滓。
proceeds
the amount of money received from a particular event or activity or when something is sold（从事某种活动或变卖财物的）支出，收益
The proceeds of today's festival will go to several local charities.
明天活动的支出将捐给当地的几家慈善机构。
also-ran
someone in a competition who is unlikely to do well or who has failed（比赛、竞争中的）失败者，落选者
nook
a small space that is hidden or partly sheltered角落;隐蔽处;幽静处
a cosy/sheltered/quiet nook
温馨的／不受风雨侵袭的／僻静的角落
cranny
a small, narrow opening in something solid裂痕，缝隙
There were small plants growing in every nook and cranny of the wall.
墙上每道缝隙里都长着低矮的植物。
ancillary
providing support or help辅助的，补充的；附属的，附加的
ancillary staff/workers
勤杂人员／工人
(, 下载次数: 0)
上传
点击文件名下载附件
That may prove true of other American companies' efforts, too. 其他美国公司的努力能够证明也是如此。（熟读原句，对照中文复述英文；再试试运用加粗的词语对任何一样事物发表观点 ）
(, 下载次数: 0)
上传
点击文件名下载附件
In other words, while American labs are betting big on the fortunes to be made by pushing the frontiers of intelligence, their open-weight Chinese rivals are more focused on encouraging adoption of AI. 换句话说，当美国实验室大肆押注于拓展人工智能前沿可创造的财富时，其开放权重的中国竞争者则更专注推进人工智能的运用。
What is one key difference between the American and Chinese approaches to AI development?模板1（日常交流）
直接回答 Well, the key difference is their focus.
简单举例 American companies, like OpenAI, are mostly creating closed, proprietary AI to make money. But Chinese firms are sharing their open-weight models more freely to encourage wider adoption.
总结This openness is helping Chinese AI become very competitive quickly.
模板2（雅思口语/其他口语考试）
主题句 The fundamental distinction lies in their core philosophy and business models.
阐述+举例 On one hand, American labs, such as OpenAI and Meta, are investing heavily in proprietary AI systems. Their primary goal is to monetize these cutting-edge models by keeping them under tight control.
On the other hand, Chinese companies are fiercely competing in the open-weight domain. They are releasing their powerful models, like those from DeepSeek and Alibaba, often for free. For instance, the article mentions how the Chinese startup DeepSeek rocked the market by offering a advanced model developed on a shoestring budget.
总结 Consequently, this strategy is accelerating global adoption of Chinese AI and is seen as a serious wake-up call for the West, potentially challenging American dominance in the long run.
欢迎留言讨论，或者后台音频回答!