DeepSeek: China's AI progress as a wake-up call for Europe

Kaiserslautern (dpa) - The success of the Chinese AI start-up DeepSeek should encourage Europe to be more innovative, according to the German Research Center for Artificial Intelligence (DFKI). "There are many funding programs in the EU," said Paul Lukowicz, Scientific Director and Head of DFKI's Embedded Intelligence research department. "But for someone to take a billion euros in hand and say: Now we are building a European model. That just doesn't seem to be happening. And I think that's a great pity," Lukowicz told the German Press Agency.

For Peter Liggesmeyer, Director of the Fraunhofer Institute for Experimental Software Engineering (IESE) in Kaiserslautern, the development from China has the potential to change the research landscape. "DeepSeek's language model really does seem to be breaking new ground in terms of achieving high performance with moderate use of resources."

Energy efficiency a major topic

Research results on energy-efficient AI will become important in the future in order to be able to use large language models in practice, said Liggesmeyer. The current solution of implementing large language models with ever-increasing hardware performance is the exact opposite of "green IT" and contradicts greater sustainability.

For Liggesmeyer, the appearance of DeepSeek's language model came as quite a surprise. Lukowicz from DFKI, which is also based in Kaiserslautern, said that China had probably looked at the tools currently on the market, such as the US-based ChatGPT. "And they have not only caught up with DeepSeek, but have also slightly overtaken it." However, hysteria surrounding the new AI model would be exaggerated. "It's not something from another planet. China has simply caught up and is now at the forefront."

Aspect of political influence

Lukowicz does not expect a broader debate about data security, as was the case with the video app TikTok. "There will be discussions, there always are on this topic. Data is processed almost everywhere these days. You should be aware of this when using such systems." However, he does not consider the issue to be of primary importance. "The aspect that I find critical is the political influence. When I read that DeepSeek doesn't answer critical questions about China properly, that's much more of a problem." It is to be expected that certain contents of DeepSeek's language model should be used with caution, Liggesmeyer continued. "However, it seems to me that the different approach underlying the model is much more interesting than the content of the model."