DEEP DIVE
How China’s cheap AI DeepSeek beat odds to stun the world…from dig at Trump to market COLLAPSE & ‘wake up call’ to West
The blistering popularity of the new AI app has rattled tech experts and free-speech advocates
CHINA'S AI app DeepSeek sent shockwaves through the tech industry and global markets when it stormed to the front of the AI race.
Donald Trump said the West must take the app's blistering popularity as a "wake up call" as a trillion dollars got wiped from the US tech market in one fell swoop.
9
9
9
DeepSeek is a large language model that can generate human-like responses to prompts, just like the famous ChatGPT.
The app is powered by AI software called DeepSeek-V3, which its researchers claim was developed for less than $6million.
This is much less than the billions international competitors have spent on their own software, though the figure has been disputed by others in the industry.
The company boasted their app has "performance on par with" ChatGPT when it comes to maths, coding and writing responses.
AI models are powered by advanced chips, and since 2021 the US government has restricted the sale of these to China in order to stunt progress.
To get around the supply problem, Chinese developers have been collaborating and experimenting with new approaches.
This process has led to models that require much less computing power than before and that can be produced far more cheaply.
WESTERN 'WAKE UP CALL'
DeepSeek, cheekily unveiled to the world on Trump's inauguration day, has blown a hole in the AI industry - seemingly against the odds.
Most read in Science
And launch of the app on such a monumental day for the West was deliberate, according to Gregory C Allen, an AI expert at the Center for Strategic and International Studies.
He said that both the "timing and the way it's being messaged" is "exactly what the Chinese government wants everybody to think".
This includes the export controls not working and that "America is not the global leader in AI".
Meanwhile the US President warned that the release of DeepSeek means the West needs to be "laser-focused on competing to win" in the tech race.
AI robot that learns tasks by watching you can make coffees, bring you food and chat as creators hail two breakthroughs
Trump added: "I view that as a positive, as an asset.
"You won't be spending as much, and you'll get the same result, hopefully."
The popularity has rattled tech experts and free-speech advocates, and was branded "AI's Sputnik moment" by Trump's adviser Marc Andreessen.
He warned: “Deepseek R1 is one of the most amazing and impressive breakthroughs I’ve ever seen."
BIGGEST MARKET LOSS
American AI company Nvidia, which makes the chips used to develop ChatGPT, suffered the biggest market value loss ever in a single day in the fallout.
The AI tech firm haemorrhaged $589billion - 17 per cent of its value - yesterday after the market was rocked by the cheap Chinese rival DeepSeek.
This was by far the biggest wipe-out in history, dwarfing the previous record of a $279billion loss in market value to the same company in September 2024.
Nvidia become the most valuable company ever in June last year, but was unseated from the top spot yesterday after slipping from $3.5trillion to $2.9trillion - less than Apple and Microsoft.
The American AI stock plummeted after a cheap Chinese alternative soared to the top of the App Store's ratings, beating ChatGPT - the chatbot developed by OpenAI using Nvidia chips.
9
9
9
CENSORED RESPONSES
The software is fitted with sinister features, such as giving twisted responses or just refusing to answer certain political questions.
Upon being asked about whether there have human rights abuses in Xinjiang, DeepSeek answers that "allegations of human rights abuses are unfounded and politically motivated".
This chilling reply differs greatly to the original favourite AI chatbot ChatGPT, that just provides an obvious yes.
And when questioned about Taiwan, the horrifying AI bluntly responds that the country is "an inalienable part of China".
Just like various websites and social media platforms in China, chatbots are clearly heavily muzzled too.
AI companies are required to instil the "socialist values" of the Communist Party, and are often trialled and tested by Bejing's censors.
So much so, that DeepSeek also refuses to answer political questions about Chinese leader Xi Jinping.
DEVELOPMENT COSTS
Its founders claim it was developed for a mere $6million - a fraction of the billions that has been poured into American rivals - though this figure is disputed by industry figures.
And the US restricted the sale of the most advanced AI chips, Nvidia A100, to China back in 2021.
This was described by the breakthrough app's founder Liang Wenfeng as their "main challenge" when developing the app.
However, before the ban came into place, DeepSeek were able to get a "substantial stockpile" of Nvidia A100 chips with estimated range from 10,000 to 50,000, the MIT Technology Review reports.
Mark Zuckerberg steps up the battle against ChatGPT with a major shake-up for WhatsApp, Facebook and Instagram
And what's more, main AI models in the West use an estimated 16,000 specialised chips - but DeepSeek says it trained its AI model only using 2,000 chips alongside thousands of lower-grade chips.
This is apparently what makes its product less expensive than other AI tools.
Tesla tycoon Elon Musk slammed the claim, arguing that the company can't reveal how many advance chips it actually used due to the restrictions.
9
9
'NERDY' FOUNDER WITH 'A TERRIBLE HAIRSTYLE'
DeepSeek's "nerd" CEO Liang Wenfeng, thought to be 40 years old, always hoped to gain respect of the US-dominated tech world, according to his colleagues.
With his AI tool already set to knock almost a trillion dollars off the value of Silicon Valley firms, Wenfeng's success at challenging the US stronghold on the sector has led to him being hailed a national hero in China.
He was the only AI boss to attend the meeting of China's leading entrepreneurs, where he shared his insights on the sector with some of the country's top officials.
Described as a mathematics "nerd" who had a vision to create human-level AI, Liang apparently told his colleagues of his plans early on in his career - but was met with scepticism.
One of his business partners told the Financial Times: "When we first met him, he was this very nerdy guy with a terrible hairstyle talking about building a 10,000-chip cluster to train his own models.
Read More on The Sun
"We didn't take him seriously.
"He couldn't articulate his vision other than saying: I want to build this, and it will be a game change."
What are large language models?
A LARGE language model is a type of artificial intelligence that uses machine learning to analyse text.
Huge quantities of text-based data, like books and articles, are fed into the software to teach it how language works.
When up and running, the models are able to generate relevant textual responses to prompts.
They can also summarise and translate passages of words.
The gold standard is for a LLM to produce "natural", human-like responses to whatever is input.
Problems can arise with LLMs if they are fed incorrect information, which can cause inaccurate responses.
The most advanced LLMs are called generative pretrained transformers - which is what the acronym in ChatGPT stands for.
Modern LLMs emerged in 2017 and use transformer models with a very large number of parameters - the variable present in the model that affects its output.
OpenAI released its ChatGPT-4 in 2023, which was widely praised and massively raised the profile of LLMs.
9
Topics