DeepSeek’s AI model has been recognized as China’s most impressive AI development by Google DeepMind CEO Demis Hassabis. However, he also noted that the model does not introduce any groundbreaking scientific advancements.
Speaking at a Google event in Paris ahead of the AI Action Summit, Hassabis acknowledged DeepSeek’s strong engineering but stated that it relies on existing AI techniques.
Last month, DeepSeek gained attention with claims that its model was trained at a fraction of the cost of other major AI systems using less advanced Nvidia chips. This announcement caused a stock sell-off and raised concerns about whether large tech firms are overspending on AI infrastructure.
Despite the excitement, experts have questioned DeepSeek’s cost claims, suggesting that the actual expenses might be higher than reported. Hassabis also compared DeepSeek’s AI to Google’s new Gemini 2.0 Flash models, asserting that Google’s models are more efficient.
While DeepSeek’s technology has geopolitical significance, Hassabis believes the media and industry hype surrounding it has been exaggerated.
The discussion around AI advancements also included predictions on the timeline for artificial general intelligence (AGI). Hassabis believes AGI, which refers to AI systems that match or surpass human cognitive abilities, could be as little as five years away.
He emphasized the need for society to prepare for the potential impacts, ensuring that AGI benefits everyone while mitigating risks. His views align with other industry leaders, including OpenAI CEO Sam Altman, who recently expressed confidence in achieving AGI.
However, concerns remain about the risks associated with AGI, including the possibility of losing control over highly intelligent AI systems. Prominent AI scientists, such as Max Tegmark and Yoshua Bengio, have voiced similar concerns. The ongoing debate highlights the challenges and responsibilities that come with advancing AI toward AGI-level intelligence.
