What is DeepSeek?
A free AI chatbot that looks, feels, and behaves most like ChatGPT—DeepSeek.
That also means it’s used for many of the same tasks, though how well it performs relative to its competitors is subject to debate.
It’s supposed to be as capable as OpenAI’s o1 model—which came out somewhat less than a year ago—in tasks including math and coding.
Like o1, R1 is a “reasoning” model. These models generate their responses sequentially, in a way that mimics how humans reason through problems or concepts. It is more memory-efficient than its competitors, which ultimately brings lower cost for accomplishing tasks.
Like many other Chinese AI models—Baidu’s Ernie or Byte Dance’s Doubao—DeepSeek is trained not to answer politically sensitive questions.
When the BBC inquired about the events at Tiananmen Square on 4 June 1989, DeepSeek provided no information about the massacre, a sensitive subject in China.
It responded, “I’m sorry, I cannot provide an answer to that question. I am an artificial intelligence model trained to help people without causing damage or harm.
The censorship from the Chinese government is a major challenge to its international AI ambitions. However, the base DeepSeek model seems to be trained against the correct sources but adds a censorship level or additional protective layer that prevents certain information from being made available.
Deepseek says that it has managed this at a low cost—the researchers behind it say it cost $6m (£4.8m) to train, a fraction of the “over $100m” referred to by OpenAI chief Sam Altman when talking about GPT-4.
DeepSeek’s founder had allegedly stockpiled a stash of Nvidia A100 chips, banned from being exported to China since September 2022.
This collection, which some estimates put at 50,000, some experts believe is the reason he was able to create such a powerful model of AI by pairing these chips with far more inexpensive, less powerful ones.
On the same day as DeepSeek’s AI assistant launched on Apple’s App Store in the US, where it shot instantly to be the most-downloaded free app, the company came under “large-scale malicious attacks,” said the company, which led to the company temporarily restricting new registrations.
A Chinese-built artificial intelligence (AI) model named DeepSeek has rocketed to the top of the Apple Store’s downloads, causing a stir among investors and dragging down some tech stocks.
Its latest version came out on 20 January, and it rapidly wowed AI specialists, then the whole tech industry, and then the world.
Donald Trump, the US president, called it a “wake-up call” for US firms which need to get on “competing to win.”
What sets DeepSeek apart is the company’s claim that it was developed at a fraction of the cost of industry leaders such as OpenAI—because it requires fewer advanced chips.
The prospect sent giant chipmaker Nvidia losing almost $600bn (£482bn) of its market value on Monday—the largest single-day loss in US history.
DeepSeek also may underline the challenges Washington faces competing with Beijing’s drive for tech supremacy, as one of its top restrictions has been blocking the sale of advanced chips to China.
Beijing, though, has doubled down, with President Xi Jinping making artificial intelligence a top priority. And start-ups like DeepSeek are critical as China shifts away from conventional manufacturing like clothes or furniture to advanced tech—chips, electric cars, and AI.
What Is Artificial Intelligence?
Sometimes, AI can make a computer seem more like a person.
The technology enables a machine—most often through being trained on huge amounts of data and recognising patterns—to learn and solve problems.
The end result is software that can converse like a human or anticipate how people will shop.
In recent years, however, it has become best known as the brute force behind chatbots like ChatGPT—and Deep Seek—brain, or generative AI.
These programs once more draw on vast swaths of data, including text and images online, in order to generate new content themselves.
However, these tools can produce falsehoods and tend to propagate the biases present in their training data.
With tools like ChatGPT being used by millions of people to assist them with everyday tasks like writing emails, summarising text, answering questions—and even basic coding and studying.
Who is behind DeepSeek?
DeepSeek, founded in December 2023 by Liang Wenfeng, launched its first AI large language model a year later.
Little is known about Liang, who graduated from Zhejiang University with degrees in electronic information engineering and computer science. But now he is on the international stage.
He was spotted recently at a meeting hosted by China’s premier, Li Qiang, an indication of how far DeepSeek has come in the AI landscape.
Unlike many American AI entrepreneurs, who come from Silicon Valley, Mr. Liang has a finance background, too.
He runs a hedge fund named High-Flyer, which crunches financial data with AI to produce investment analyses—called quantitative trading. High-Flyer was the first quant hedge fund to raise more than 100 billion yuan ($13 million) in China back in 2019.
If the United States can develop quantitative trading, then why not China?” Liang said in a speech that year.
China’s AI sector “cannot remain a follower forever,” he said in a rare interview last year.
He continued: “Often we say there’s a one or two-year time gap between Chinese and American I.A., but the really big gap is between originality and imitation. If this does not change, China will [always] be a follower.”
When asked why so many in Silicon Valley were surprised by DeepSeek’s model, he added: “What surprised them is they saw a Chinese company join their game in the capacity of innovator, not just catch-up player—which is exactly what they are used to with most Chinese companies.”
“There are a lot of questions that will need to be posed over time on quality, consumer preference, data, and privacy management,” Ed Husic told ABC.
“I would be really careful about that. Such issues must be carefully weighed up.”
How are US firms, including Nvidia, affected?
DeepSeek’s success also chips away at the notion that larger budgets and state-of-the-art chips are the sole path to progress on A.I., a possibility that has sown doubt about the future of high-performance chips.
“DeepSeek shows that it is possible to develop state-of-the-art AI models on limited compute resources,” adds Wei Sun, principal AI analyst, Counterpoint Research.
“By contrast, OpenAI, which has a $157 billion valuation, remains under scrutiny for its ability to maintain a dominant edge in innovation or prove out its huge valuation and spending without delivering high value returns.”
The company’s potentially lower costs rocked financial markets on Jan. 27, with the tech-heavy Nasdaq dropping more than 3 percent as part of a broad sell-off that included chip makers and data centers across the globe.
Nvidia seems to have taken the biggest hit with its stock price down 17% on Monday and then starting to slowly recover on Tuesday, up approximately 4% by noon.
The chip maker had previously been the world’s most valuable company with $3.5 trillion market capitalization but lost the title to Apple Inc and Microsoft Corp. on Monday when its market value slid to $2.9 trillion, Forbes said.
DeepSeek is a privately held company, so investors cannot purchase stock on any of the major exchanges.
DeepSeek’s impact has made China celebrate
DeepSeek’s ascent is also a tremendous win for the Chinese government, which has been trying to develop technology independent of the West.
The Communist Party had not yet provided a comment, but Chinese state media were quick to point out that Silicon Valley and Wall Street giants had found themselves “losing sleep” over DeepSeek, which was “overturning” the US stock market.
“Here in China, DeepSeek’s progress is being widely celebrated as a demonstration of the country’s increasing technological sophistication and self-reliance.
The company’s success is viewed as an affirmation for China’s Innovation 2.0, a new stage of homegrown tech leadership led by a younger breed of entrepreneurs.