Google DeepMind Pushes AI to Its Limits
Demis Hassabis, CEO of Google DeepMind, affirmed that the future of artificial general intelligence (AGI) depends on pushing current AI systems to their limits. Speaking at the Axios AI+ Summit in San Francisco, Hassabis emphasized that the principle of “scaling to the limit” is the most direct path to achieving systems capable of thinking, planning, and learning with human-like abilities.
He stated that increasing data volume and computing power is not merely a technological expansion, but may constitute the backbone, or even the core component, of the ultimate AGI system.
Hassabis explained that the launch of the Gemini 3 model is just one step in a long process based on what are known as the “laws of scaling.” These laws indicate that doubling data and computing power directly impacts the quality of AI. DeepMind believes this strategy represents the solid foundation upon which companies are building their race to achieve artificial general intelligence, a goal that is driving billions of dollars toward developing giant data centers, increasing processing capacities, and accelerating innovation in AI architecture.
But this view isn't universally shared within the AI community. Yann Lecon, a leading scientist in the field and former head of AI at Meta, argues that scaling alone isn't enough. He cautions against the assumption that more data necessarily equates to higher intelligence. In a lecture at the National University of Singapore, he explained that some of the most complex problems "deteriorate in solvability as scale increases," and that solutions may not come from expanded language models, but rather from new approaches based on "world-understanding" models that build a realistic perception grounded in space, physics, and long-term memory.
Lecon, who is working on a new project in this direction, asserts that the future of AI may require a completely different leap from the current paradigm. The ability to understand the real world, not just text, will be the decisive factor in the next AI revolution. DeepMind, however, believes that continuing to scale to its fullest extent is the essential path, and that even if it isn't the ultimate solution, it will be an integral part of any true AI system.
Between these two trends, the sector finds itself facing a fundamental question: Should we continue to inflate current models to the maximum extent, or is it time to rethink the scientific foundations on which intelligence systems are built? Despite the escalating controversy, DeepMind seems to still be sticking to the option of expansion, ready to push artificial intelligence capabilities to limits that the industry has never reached before.