DeepSeek is an attack on the artificial intelligence (AI) realm, and there are eight questions and considerations to bear in mind.
First, once the excitement wore off following the media storm after DeepSeek officially released R1 on Jan. 20, tech hobbyists and experts began to question whether such an advanced language model could have really been built with US$5.5 million. Of course not. DeepSeek’s official statement claimed the amount was used to further develop its older V3 model. As for how much it actually spent, they are keeping their lips sealed and have not released a concrete figure.
Second, DeepSeek made its statement on US President Donald Trump’s inauguration day, causing a stir in global stock markets. Is the Chinese Communist Party (CCP) mobilizing media and making great pronouncements, or even using overseas consular offices to add to the information flood around DeepSeek? If a new AI company — one with only four full-time official employees — did not receive any help, how could it have produced a new AI language model in such a short timespan?
Third, if you ask R1 questions regarding topics such as the Tiananmen Square Massacre, Taiwanese sovereignty, Xinjiang and the Uighurs, or its evaluation of Chinese President Xi Jinping (習近平) or the CCP government, it is almost invariably an iteration of the following answer: “I’m sorry, can we talk about mathematics or physics?” Other users have used prompts that are more vague: If you ask about a man standing with a plastic bag in front of a tank, in reference to the infamous “tank man” who blocked several Chinese People’s Liberation Army tanks from advancing along a Beijing street, DeepSeek responds that it is thinking, or rather, it is making inferences. Then, as if startled into awareness that this is a sensitive prompt banned by the CCP, it responds: “I’m sorry. Let’s change topics.”
Former minister of digital affairs Audrey Tang (唐鳳) was able to use an ambiguous line of questioning on an offline version of R1 to bring up Tiananmen Square and the system surprisingly gave a fleshed-out answer without any censorship. The conclusion is that DeepSeek’s language model itself is not restricted, but the CCP is limiting the platform.
Fourth, AI models need to overcome a massive initial threshold of accumulating oceans’ worth of data. The most difficult aspect to collecting so much data is that a company has to separate and clean all of them. It also has to design question prompts and answers to train its AI model.
Years after setting up shop, OpenAI is still using a massive team of tech talent to curate and scour seas of data. They have burned through tons of cash, and spent thousands of hours annotating and correcting prompts and answers to achieve their nearly perfected GPT4 language model — a generative AI model. Other, smaller AI companies simply do not have comparable funding and resources, so they often cut corners. After downloading all sorts of compressed data, they recompress it and work to make it into something that performs well to form their own language models. That method is known as knowledge distillation.
Several industries are questioning whether DeepSeek used distillation and the answer is becoming ever clearer — it has.
Moreover, DeepSeek originally distilled its own model by basing it off OpenAI’s GPT4. OpenAI in September last year developed an inference model — the 01 language model — which includes mathematical reasoning, logical inference and its own unique editing functions.
Fifth, to prove that DeepSeek distilled OpenAI’s language model, the latter on Jan. 28 closed programming interfaces with massive traffic volumes. Not long after, DeepSeek announced that it would be temporarily suspending logins from foreign users due to a “virus,” and would only give access to mobile app users within China. Is that merely a coincidence, or a trick? The reader can judge for themselves.
Sixth, DeepSeek repeatedly emphasized that it was using Nvidia’s lower-end H800 chips in its data center, but when you ask the chatbot whether the company is using advanced Nvidia’s H100 chips — which are supposed to be banned from export to China — its answer is “yes.”
The US government found that the 10,000 or so H100 chips the company purportedly possesses were sent to procurers in Vietnam, then to Singapore and then China. Singapore has since come under scrutiny, as import data showed that its companies procured 20 percent of Nvidia’s H100, H800 and H20 chips.
Seventh, an Israeli intelligence company alleged that DeepSeek could implant malicious code in users’ devices to steal their private data by teasing it out of them. Does such user data make its way to the CCP’s Ministry of State Security or Ministry of Public Security? Users should be wary.
Eighth, OpenAI on Jan. 30 announced it was officially launching its latest language model, o3-mini. The model’s performance and convenience — its question-and-answer and logical inferencing already being called “genius-level” — blow DeepSeek’s R1 model out of the water, and its pricing is on par with the earlier o1 model — a great value. OpenAI is set to launch a more fully scaled o3 model.
When that happens, we would have our answer as to which model is a tech darling and which one would be left in the dust.
Tsao Yih Cherng is a literary and history writer.
Translated by Tim Smith
On May 7, 1971, Henry Kissinger planned his first, ultra-secret mission to China and pondered whether it would be better to meet his Chinese interlocutors “in Pakistan where the Pakistanis would tape the meeting — or in China where the Chinese would do the taping.” After a flicker of thought, he decided to have the Chinese do all the tape recording, translating and transcribing. Fortuitously, historians have several thousand pages of verbatim texts of Dr. Kissinger’s negotiations with his Chinese counterparts. Paradoxically, behind the scenes, Chinese stenographers prepared verbatim English language typescripts faster than they could translate and type them
More than 30 years ago when I immigrated to the US, applied for citizenship and took the 100-question civics test, the one part of the naturalization process that left the deepest impression on me was one question on the N-400 form, which asked: “Have you ever been a member of, involved in or in any way associated with any communist or totalitarian party anywhere in the world?” Answering “yes” could lead to the rejection of your application. Some people might try their luck and lie, but if exposed, the consequences could be much worse — a person could be fined,
On May 13, the Legislative Yuan passed an amendment to Article 6 of the Nuclear Reactor Facilities Regulation Act (核子反應器設施管制法) that would extend the life of nuclear reactors from 40 to 60 years, thereby providing a legal basis for the extension or reactivation of nuclear power plants. On May 20, Chinese Nationalist Party (KMT) and Taiwan People’s Party (TPP) legislators used their numerical advantage to pass the TPP caucus’ proposal for a public referendum that would determine whether the Ma-anshan Nuclear Power Plant should resume operations, provided it is deemed safe by the authorities. The Central Election Commission (CEC) has
When China passed its “Anti-Secession” Law in 2005, much of the democratic world saw it as yet another sign of Beijing’s authoritarianism, its contempt for international law and its aggressive posture toward Taiwan. Rightly so — on the surface. However, this move, often dismissed as a uniquely Chinese form of legal intimidation, echoes a legal and historical precedent rooted not in authoritarian tradition, but in US constitutional history. The Chinese “Anti-Secession” Law, a domestic statute threatening the use of force should Taiwan formally declare independence, is widely interpreted as an emblem of the Chinese Communist Party’s disregard for international norms. Critics
RSS