Google DeepMind CEO Demis Hassabis has praised China’s AI model, DeepSeek’s, as “the best work” coming out of the country, but also stated that there is no actual scientific advance behind its claims. Despite the hype surrounding DeepSeek’s announcement, which claimed to have trained an AI model at a fraction of the cost of leading tech firms and on less-advanced Nvidia chips, Hassabis says it is “using known techniques” in AI.
Hassabis’ comments come ahead of the AI Action Summit in Paris, where he emphasized that DeepSeek’s model shows good engineering skills but not a significant breakthrough. He also noted that his own company’s Gemini 2.0 Flash models are more efficient than DeepSeek’s.
The debate over when artificial general intelligence (AGI) will arrive has been ongoing for years. Hassabis believes the industry is on the path towards AGI, which he describes as a system exhibiting human-like cognitive capabilities, and that it could be just five years away.
However, experts have questioned the cost of development for DeepSeek’s models, suggesting they may be higher than claimed. The AI world is still grappling with concerns over potential risks associated with AGI, including loss of control over created systems.
Despite these challenges, Hassabis urges society to prepare for the benefits and mitigate the risks of AGI, emphasizing the need for a balanced approach to harnessing its power.
Source: https://www.cnbc.com/2025/02/09/deepseeks-ai-model-the-best-work-out-of-china-google-deepmind-ceo.html