Deepseek aftershocks
How China is quietly upstaging America去自：The Economist
本文
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
on a shoestring
If you do something on a shoestring, you do it with a very small amount of money以极多量资本
The film was made on a shoestring.
那部影戏是个小制作。
traction
the fact of an idea, product, etc. becoming popular or being accepted（概念或者产物等）变患上流行（或者被承受）
In our digital age, it takes less time for new words and phrases to gain traction than it did in the past.
正在目前数码时期，新词汇或者词汇组流行起去要已往快很多。
close in
to gradually get nearer to someone, usually in order to attack them（凡是指为了防备）垂垂围上来，包抄，迫近
The advancing soldiers closed in on the town.
前进的战士一步步包抄了镇子。
underwhelming
not causing someone to feel any excitement or admiration已留住深化影像的;已引起弘大反应的
The food was good but unfortunately we found the rest of the experience distinctly underwhelming.
食品很佳，但是很遗恨，咱们以为其余的体会清楚没有尽人意。
half-hearted
showing no enthusiasm or interest没有热忱的;兴致没有年夜的
He made a half-hearted attempt to clear up the rubbish.
他没有年夜甘愿天来清扫残余。
proceeds
the amount of money received from a particular event or activity or when something is sold（处置某种举动或者变买财物的）收入，支益
The proceeds of today's festival will go to several local charities.
来日诰日举动的收入将捐给本地的多少野慈悲机构。
also-ran
someone in a competition who is unlikely to do well or who has failed（角逐、合作中的）失利者，落第者
nook
a small space that is hidden or partly sheltered边际;荫蔽处;寂静处
a cosy/sheltered/quiet nook
舒适的／没有受风雨侵害的／荒僻冷僻的边际
cranny
a small, narrow opening in something solid裂缝，漏洞
There were small plants growing in every nook and cranny of the wall.
墙上每一讲漏洞里皆少着高低的动物。
ancillary
providing support or help帮助的，弥补的；从属的，附带的
ancillary staff/workers
勤纯职员／工人
That may prove true of other American companies' efforts, too. 其余好国公司的勤奋可以证实也是云云。（生读本句，比较华文复述英文；再尝尝使用减细的词汇语对于所有一致实物揭晓概念 ）
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. 换句话道，当好国尝试室鼎力大举押注于拓展野生智能前沿可缔造的财产时，其盛开权沉的华夏合作者则更专一促进野生智能的使用。
What is one key difference between the American and Chinese approaches to AI development?模板1（一样平常交换）
间接答复 Well, the key difference is their focus.
简朴举例 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.
归纳This openness is helping Chinese AI become very competitive quickly.
模板2（俗思白话/其余白话测验）
中心句 The fundamental distinction lies in their core philosophy and business models.
论述+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.
归纳 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.
欢送留行会商，大概背景音频答复!
面打存眷尔的公家号👇👇 |
提升卡
置顶卡
沉默卡
喧嚣卡
变色卡
千斤顶
照妖镜
广告
