Filenews 28 January 2025
Investors are fleeing U.S. artificial intelligence stocks amid surprise at a new, cheaper, but at the same time effective alternative Chinese technology. But what is and how does the DeepSeek app work?
What is DeepSeek?
DeepSeek is a Hangzhou-based Chinese artificial intelligence (AI) company that emerged a few years ago from a university startup. Its stated goal is to create an artificial general intelligence – a term for human-level intelligence that no tech company has yet achieved. It's not there yet, but that may be one reason why the computer scientists at DeepSeek have taken a different approach to building their AI model, making it look many times cheaper to operate than its competitors in the US.
Another reason why he seems to have taken the low-cost approach could be the fact that Chinese computer scientists have long had to go beyond limits on the number of computer chips available to them as a result of U.S. government restrictions.
Why hadn't we heard it before?
The company has long quietly impressed the world of artificial intelligence with its technical innovations, including a cost-to-performance ratio several times lower than that of models made by Meta (Llama) and OpenAI (Chat GPT). It hasn't made as much noise about the potential of its innovations as Silicon Valley companies. The latter have been circulating product announcements for months, as they become increasingly urgent to finally generate returns from multibillion-dollar investments. We should expect to see more of DeepSeek's cheerful blue whale logo though, as more and more people around the world download it to experiment.
What is the R1 model that the whole world is talking about?
It's the DeepSeek AI model that people are most excited about at the moment, as it claims to have performance comparable to OpenAI's o1 model, which was rolled out to Chat GPT users in December. On Monday it was the most popular free app downloaded on Apple's app store in the UK and other parts of the world.
But there are many artificial intelligence models from OpenAI, Google, Meta, and other companies. Where is the big issue?
This model uses a different kind of internal architecture that requires less memory usage, thus significantly reducing the computational cost of each search or interaction with the chatbot-like system. It has been praised by researchers for its ability to tackle complex reasoning tasks, particularly in mathematics and coding, and appears to produce results comparable to competitors, with a fraction of the computing power they need. DeepSeek said it took two months and less than $6 million to develop the model, though some observers warn that the amount has likely been deliberately underestimated. However, it's much smaller than the billions Silicon Valley tech companies spend on AI development anyway, and it's less expensive to operate.
Who owns it?
One key figure is Liang Wenfeng, who once ran a Chinese hedge fund that now funds DeepSeek. In a rare interview, he said: "For many years, Chinese companies have become accustomed to others bringing technological innovation, while we focused on monetizing applications – but this is not inevitable. In this wave, our starting point is not to seize the opportunity for quick profit, but rather to reach our technological limits and drive the development of the entire ecosystem... We believe that as the economy grows, China should gradually become a contributing factor."
Why did U.S. tech stocks fall?
Hundreds of billions of dollars disappeared from the market capitalization of major tech stocks after news spread about the performance of the DeepSeek chatbot over the weekend. The timeline was significant, as in recent days U.S. tech companies had committed hundreds of billions more to investment in artificial intelligence — much of which will go toward building the computing infrastructure and energy sources needed, widely believed, to achieve the AI goal. DeepSeek's performance seems to challenge, at least in the foreground, this narrative.
What's the concern about Nvidia?
Nvidia is one of the companies that has gained the most from the explosion of artificial intelligence. From a graphics card manufacturer for video games, it became the dominant chipmaker in the voracious hungry artificial intelligence industry. He was compared to a moderate pickaxe and bucket dealer in 19th-century California who happened to be on the spot when the gold rush occurred and thus became a huge supplier to the world's richest industry. Tech companies looking obliquely at DeepSeek are probably wondering if they should now buy so many of Nvidia's tools. Its market capitalisation fell by 600 billion dollars on Monday.
What doesn't DeepSeek do?
It hasn't reached artificial general intelligence, the limit at which artificial intelligence is starting to approach and that OpenAI and other giants in Silicon Valley are pursuing. Sam Altman, CEO of OpenAI, warned that the breakthrough is unlikely to be imminent. But it does what others can do, at a fraction of the cost.
Is the emergence of DeepSeek good news?
One possibility is that advanced AI capabilities can now be achieved without the vast amount of computing power, microchips, energy and cooling water that were thought necessary. As with all technological discoveries, time will be the judge of what will become reality.
Capital.gr
