JTS webinar: UG researchers discuss the 'irrational’ response to DeepSeek

The launch of the Chinese AI app DeepSeek caused quite a stir in the world of artificial intelligence (AI). Stocks of technology companies such as NVIDIA and ASML temporarily plunged and several public institutions banned the chatbot. The Jantina Tammes School of Digital Society, Technology and AI hosted an online meeting on DeepSeek, where several UG researchers shared their views on the new entrant into the AI arena.

Text: Jelle Posthuma

The gathering was kicked off by Lambert Schomaker, professor emeritus of AI at the University of Groningen. He shared a post on LinkedIn shortly after the news about DeepSeek, in which he expressed his surprise about the uproar. He argued that the markets showed an ‘irrational’ response. While the development and claimed cost-effectiveness of DeepSeek are quite praiseworthy, they are certainly not a revolution, Schomaker stressed.

Dark jungle

The Chinese software company cleverly and efficiently used existing technology, Schomaker explained. He used the analogy of exploring the jungle. With machetes, several AI players had already made their way through the dark jungle. DeepSeek, as a latecomer, walked the beaten path and unsurprisingly found its way more easily. Why does this scare the financial world so much, the professor openly wondered. He also pointed to Jevons' paradox, describing how more efficiency often in fact leads to more consumption and usage.

According to Schomaker, financial markets (and journalists) have too simplistic a view of artificial intelligence. The professor argued that scientists need to become more involved in the debate. Several ‘AI experts’ who speak out about artificial intelligence in the media lack fundamental knowledge about the basics, leading to a distorted view, he concluded.  

Unreliable and immense energy use

Bart Verheij, professor of AI & Argumentation, likewise called DeepSeek as a ‘normal step’ in technological progress during the online meeting. According to him, DeepSeek shows once again that it is feasible to build Large Language Models similar to ChatGPT. As such, it has shown some technological innovation, efficiency, and lower costs, although the Chinese company's claims are difficult to verify, the professor stressed.

In his presentation, Verheij pointed out the ‘real’ issues of Large Language Models, like ChatGPT and DeepSeek. As of today, these models are still unreliable and therefore far too risky for serious tasks. It is often overlooked that when ChatGPT or DeepSeek claim they can perform a task, there is no guarantee they are correct. He also mentioned the immense energy use of the models: plans have even been made for small-scale nuclear power plants to power them.

After a few questions, Tommaso Caselli, Assistant Professor and theme coordinator ‘AI and Language’ at the Jantina Tammes School and Dirk Pleiter, professor of Innovative Computer Architectures at CogniGron, gave their reflections on the DeepSeek development. According to them, the Chinese tool show that more can be done with less, although some caution is needed. The DeepSeek team's open sharing of source code, results, and technical insights supports efforts to develop large-scale models in other location, Pleiter mentioned.

Universities

This also provides opportunities for Europe to develop new AI innovations. Verheij referred to the European Union's AI Factories, through which member states want to develop new generative AI models. He mentioned Groningen as an excellent candidate for such a Factory, offering smart people and a well-functioning artificial intelligence ecosystem.

In the online meeting, all researchers stressed that universities have an important role to play in driving these innovations forward. Scientists at the UG are working on more energy-efficient hardware and responsible AI. DeepSeek demonstrates the potential of current technology, but this development is by no means in its final phase.