The field of artificial intelligence (AI) has undergone significant transformations in recent years, raising new questions about the future of technology. In an interview, Levente Szabados, Associate Professor at the Frankfurt School and Co-Founder of Neuron Solutions, shared his insights on the latest advancements and their potential implications.
Scalability and the Data Dilemma
Until recently, the prevailing belief in AI model development was that larger models deliver superior performance. However, this paradigm is now being challenged. The training of OpenAI’s ChatGPT-5 revealed that increasing data volume and computational power did not yield the expected improvements. Meanwhile, China’s DeepSeek introduced a smaller, more cost-effective model that outperforms larger models in certain areas.
The core issue lies in both the quality and availability of data. Existing models have already utilized a significant portion of readily available textual data from the internet, making it increasingly difficult to source new, high-quality training material. This challenge is prompting developers to explore alternative solutions, such as generating synthetic data or optimizing the use of existing datasets.
The Role of Synthetic Data
Synthetic data refers to artificially generated datasets designed to replicate the characteristics of real-world data. These datasets are created through computational models rather than direct observation or measurement. A notable example is DeepSeek, which trained its model using explanations generated by ChatGPT—a compelling demonstration of synthetic data’s potential in AI development.
Overcoming Knowledge Limitations with New Learning Methods
One of AI’s key limitations stems from its reliance on textual information, as it lacks direct interaction with the physical world. This constraint is akin to trying to understand birdsong through a correspondence course. NVIDIA’s Cosmos world model aims to address this by learning from video and visual data, establishing a more direct connection to real-world experiences.
Future advancements in AI are expected to prioritize logical reasoning, strategic thinking, and planning capabilities. The ultimate challenge is not merely storing facts but equipping AI with adaptive strategies to solve new, previously unseen problems.
Europe’s Opportunity in the AI Race
While the United States has historically led AI innovation, Europe has primarily focused on regulation. However, recent developments suggest that Europe may be poised to catch up. The French government’s recent €109 billion investment in AI signals a commitment to enhancing competitiveness.
Europe’s late entry into the race may prove advantageous, as it allows for the adoption of cutting-edge technologies while bypassing earlier development stages—an effect known as the “innovation paradox.” DeepSeek, for example, leveraged advancements introduced by ChatGPT within just six months.
The European AI ecosystem is also bolstered by the presence of key industry players. French company Mistral has developed a high-quality AI model, while Hugging Face has established a widely supported open-source infrastructure within the developer community.
Conclusion
The evolving landscape of AI development calls for a shift in focus—from sheer model size to efficiency, data quality, and innovative learning methodologies. Europe has the potential to emerge as a major player in this transformative era, provided that it fosters innovation and research. Future AI systems are expected to more closely resemble human intelligence, demonstrating enhanced problem-solving capabilities and a deeper understanding of the world.
