A debate is spreading across Silicon Valley: How far can scaling methods take the technology?
Google DeepMind CEO Demis Hassabis, fresh off the release of Gemini 3 to widespread acclaim, has made his position on the issue clear.
“We need to push the scale of our current system as much as possible, because at least that's going to be a key component of the final AGI system,” he said at Axios' AI+ Summit in San Francisco last week. “It could be the entire AGI system.”
AGI (artificial general intelligence) is a still-theoretical version of AI that reasons in the same way humans do. This is a goal that all major AI companies are racing to achieve, driving massive spending on infrastructure and talent.
The law of AI scaling suggests that the more data and computations given to an AI model, the smarter it becomes.
Hassabis said he thinks the industry will likely reach AGI through scaling alone, but that “one or two” other breakthroughs may be needed.
The problem with just scaling is that there is a limit to the data exposed, and adding compute requires building data centers, which is expensive and taxing on the environment.
Some AI watchers are concerned that AI companies behind major large-scale language models are starting to see diminishing returns on their massive investments in scaling.
Researchers like Meta's chief AI scientist Yann LeCun, who recently announced he was leaving to run his own startup, think the industry needs to consider a different approach.
“Most interesting problems are very badly scaled,” he told the National University of Singapore in April. “We cannot simply assume that more data and more computing means smarter AI.”
LeCun will task Meta with building a world model that is an alternative to large-scale language models that rely on collecting spatial data rather than language-based data.
“The startup's goal is to bring about the next big revolution in AI: systems that understand the physical world, have persistent memory, can reason, and can plan complex courses of action,” he said on LinkedIn in November.
