诺奖得主惊人预测：4年推出广义相对论，就是AGI，做完人类580亿年任务

Core Viewpoint - Demis Hassabis, head of Google DeepMind, has redefined AGI (Artificial General Intelligence) with a rigorous standard known as the "Einstein Test," which assesses an AI's ability to independently derive theories like general relativity from a limited knowledge base [1][3][5]. Group 1: Definition and Implications of AGI - The "Einstein Test" emphasizes the originality and scientific discovery capability of AI rather than just its knowledge base [3][5]. - There is a consensus among industry leaders that AGI is approaching, with Hassabis shortening his timeline to potentially within five years [9][11]. - Sam Altman, CEO of OpenAI, predicts AGI could be achieved by 2028, suggesting that current students will enter a world with AGI upon graduation [11][15]. Group 2: Diverging Perspectives on AGI - Elon Musk argues that Hassabis's definition aligns more with "superintelligence" rather than AGI, as it sets a bar that surpasses human capabilities [6][7]. - Different leaders have varying definitions of AGI, with Musk's being more accessible, while others like Yann LeCun express skepticism about current AI architectures achieving true AGI [29][30]. Group 3: Acceleration of AI Development - Recent evaluations indicate that advanced AI models are experiencing exponential growth in their ability to complete complex tasks, doubling in task length approximately every four months [31]. - Predictions suggest that by 2041, AI could theoretically accomplish tasks that would take humans 58 billion years, far exceeding the current age of the universe [33]. - The rapid development of AI is creating a sense of urgency and anxiety within the industry, as leaders acknowledge that the world is not prepared for the impending changes [37][39].