At the recent ‘All In Summit’, Google DeepMind CEO Demis Hassabis provided a rational assessment of the current state of development in the field of artificial intelligence. He openly questioned the AI systems on the market that claim to have reached ‘PhD-level intelligence’, asserting that their actual capabilities are far from this standard.
Hassabis emphasized that true general intelligence should manifest as comprehensive abilities across multiple fields and tasks, meaning that it can achieve PhD-level performance in various disciplines, which current AI systems clearly cannot meet. He cited OpenAI CEO Sam Altman’s earlier mention of GPT-5 as an example, pointing out that while such models excel in specific tasks, there remains a significant gap in their overall capabilities compared to ‘PhD-level intelligence’.
‘Existing systems still make errors on simple tasks, and it is even harder for them to achieve PhD-level understanding and judgment in complex situations,’ Hassabis stated frankly at the conference. He also pointed out that a key limitation of current AI technology is the lack of continuous learning capabilities, which prevents it from self-improvement and capability enhancement through the accumulation of experience, as humans do.
Regarding the realization of systems with true PhD-level comprehensive intelligence, Hassabis provided a relatively conservative time expectation, suggesting that it will still take 5 to 10 years. This assessment undoubtedly cools the enthusiasm of companies eager to claim they have created ‘super AI’ and pours a bucket of rational cold water on the current AI hype.
Hassabis’ views have sparked a rethinking of the current state of AI development in the industry. Several experts have expressed agreement, stating that although AI technology has made significant breakthroughs in specific areas, it is still far from achieving true general intelligence. The industry is calling for a rational attitude, urging against excessive promotion to prevent the public from developing unrealistic expectations and misunderstandings about AI capabilities. (Suky)返回搜狐，查看更多
平台声明：该文观点仅代表作者本人，搜狐号系信息发布平台，搜狐仅提供信息存储空间服务。