Demis Hassabis, the CEO of Google DeepMind, stated on Sunday that Deepseek’s AI model “is probably the best work” coming out of China, but he also noted that the business has not demonstrated any new scientific advancements.
China’s Deepseek published a study last month that shocked international markets by asserting that its AI model was developed on less-powerful Nvidia chips and at a fraction of the price of top AI companies.
The statement from Deepseek caused a sharp decline in stock prices and generated a lot of discussion about whether big tech companies are overspending on AI technology.
Deepseek’s model was hailed by Hassabis as “an impressive piece of work.”
At an event organized by Google in Paris prior to the city’s hosting of the AI Action Summit, Hassabis remarked, “I think it probably the best work I’ve seen come out of China.”
The AI model demonstrates Deepseek’s ability to execute “extremely good engineering,” according to the CEO of DeepMind, and it “changes things on a geopolitical scale.”
Hassabis, however, claimed that it was not a significant development from a technological standpoint.
“There isn’t any real new scientific advancement, despite the fanfare. The hoopla around Deepseek has been “exaggerated a little bit,” he added, adding, “it’s using known techniques [in AI].”
According to the CEO of DeepMind, the company’s Gemini 2.0 Flash models—which Google made publicly available last week—are more effective than DeepMind’s model.
Experts have questioned Deepseek’s statements on its low cost and the chips it employs, believing that the Chinese company’s models have higher development costs.
Five years from now, AGI
For years, the field of artificial intelligence has been discussing when artificial general intelligence, or AGI, will become a reality. AI that is more intelligent than humans is referred to as AGI in general.
The AI sector is “on the path towards AGI,” according to Hassabis, who defines it as “a system that exhibits all the cognitive capabilities humans have.”
Hassabis stated, “I think we’re close now, you know, maybe we are only, you know, five years or something away from a system like that which would be pretty extraordinary.”
And I believe that society must prepare for that and the consequences that will follow. Additionally, make sure that we minimize some of the hazards while simultaneously ensuring that we reap the benefits and that society as a whole benefits.
Hassabis’ remarks are consistent with those of other industry participants who have hypothesized that AGI would be more feasible.
“I am confident that we know how to build AGI as we have traditionally understood it,” stated OpenAI CEO Sam Altman this year.
However, a number of industry professionals have also identified a number of AGI-related hazards. According to renowned AI experts Max Tegmark and Yoshua Bengio, who recently expressed their worries about this type of AI to CNBC, one of the main worries is that humans would lose control of the systems they built.