The user interface of DeepSeek. (CFP Photo)
At the AI Action Summit in Paris, DeepSeek, a Chinese company that develops an open-source large language model (LLM), grabbed the attention of political leaders such as Emmanuel Macron, CEOs of AI firms worldwide, as well as notable computer scientists and Nobel economic laureate Daron Acemoglu.
Founded in 2023 by Liang Wenfeng, a 39-year-old programmer born in Guangdong's city of Zhanjiang, DeepSeek soon gained worldwide attention at the beginning of 2025 owing to its LLMs and low cost.
Since the release of the R1 model, however, some Western pundits and newspapers have alarmed that DeepSeek heralds a "Sputnik moment" of AI, implying a Cold War mentality, as the former Soviet Union launched the first artificial satellite to orbit Earth in October 1957, which triggered panic in the US.
In a recent exclusive interview, Dr. Patrick Glauner, Professor of AI at the Deggendorf Institute of Technology in Germany, noted that DeepSeek is "a big win for open source models" and called for less regulation within European countries and worldwide.
The user interface of DeepSeek. (CFP Photo)
A big win for open source
Using the model that circulated in the news two weeks ago, Prof. Glauner conducted some tests, such as generating Chinese New Year wishes for his friends. He felt that the messages generated for that use case worked quite well.
"It's a big win for open source," Prof. Glauner confessed, "they built on top of open source and use the state-of-the-art models."
According to him, DeepSeek embraces a great opportunity for all, and countries can advance together by building on each other's knowledge. Knowledge does not get locked in companies due to open source, but is rather available to the whole world.
Compared with ChatGPT and Mistral, its American and European counterpart, Prof. Glauner categorized DeepSeek and Mistral as one group and ChatGPT as another due to the degree of open source.
He detailed that ChatGPT from OpenAI was initially founded as a company to build open source. Currently, they release little insight about their models, nonetheless blurring their working mechanisms.
Mistral has lots of open source and progress, similar to DeepSeek. They are much more accessible to the public. Arthur Mensch, co-founder of Mistral in France, noted in an interview with POLITICO that DeepSeek is a player that is very similar to Mistral, as the two are about the same age despite the former having better funding.
DeepSeek, ChatGPT, Claude and Gemini. (CFP Photo)
Less regulation worldwide
From a sporadically downloaded app to a sensational product, DeepSeek ranks first in downloads on the Apple App Store free list in both China and the US in late January.
Data published by WISEAPP·RETAIL on February 4 unveiled that it ranked as the second most used generative AI application by South Koreans in the fourth week of last month, with 1.21 million users.
Given its waxing global popularity, countries such as the US and South Korea are embarking on banning DeepSeek from being utilized in their governments.
Against the status quo, Prof. Glauner underscored that all the banning moves would be in vain and voiced the need for fewer regulations within Europe and worldwide.
He stated that everything is available to everyone through the Internet nowadays, and blocking is impossible.
"It's anyway largely open source. You can download it, look into how it works, and critically assess it. So I don't really see that risk from DeepSeek," he added.
Additionally, the AI practitioner grumbled that Europe is also grappling with AI over-regulation.
Provided that the European Union implements the world's most stringent AI regulation—General Data Protection Regulation (GDPR) — Prof. Glauner cried out that Europe needs less regulation and more action.
"We need more innovators and fewer regulators. Otherwise, we become even more dependent on other countries," he said.
French President Emmanuel Macron speaks during the Artificial Intelligence Action Summit in Paris on Feb 10, 2025. (CFP)
An arduous but promising future
Upon its new R1 model in two months for under $6 million, DeepSeek now has a valuation ranging from US$1 billion to US$150 billion. The startup won the favor of Prof. Glauner, who also admitted that there is an onerous road ahead in the AI domain.
He noted that DeepSeek offers a novel approach to the development of LLMs. "I've always been very skeptical of building larger models and spending more energy and training them on more data. Because eventually, this is not sustainable. We as humans also learn from very little data."
"This is now a start. They'll also be working on models that generate images, audio, or videos soon. It's not just text. If they add more features, more models will ultimately follow," he said.
He observed that AI is a very competitive, dynamic, and volatile environment. AI firms need to keep working hard as others will try to catch up and vie for the most advanced models and applications.
"There is a good model. Maybe one week later, there is a better model, and thousands of models get released each week on platforms such as Hugging Face," he warned. "There is no moat; you need resources: data, computing capabilities, and smart people. Any idea that there are companies years ahead of others and that no one can start fresh and catch up is just not true."
This comment was made by Christopher Manning, Director of Stanford AI Laboratory, after DeepSeek's roaring debut in late January. As the firm's name suggests, you must seek deep ahead on the evolving AI road, indeed.
Reporter | Zhang Ruijun
Editor | Yuan Zixiang, James, Shen He
