OpenAI联合创始人卡帕西：AI智能体距“真正有用”尚需十年

Core Insights - OpenAI co-founder Andrej Karpathy expresses that despite rapid advancements in AI technology, intelligent agents are still significantly lacking in functionality and practicality, predicting it will take another decade to address core issues, contrasting with some investors' expectations for 2025 to be the "year of intelligent agents" [1][2]. Group 1: Current Limitations of AI Agents - Current AI agents exhibit multiple shortcomings, including insufficient intelligence and multimodal capabilities, inability to independently perform complex tasks like computer operations, lack of continuous learning and memory retention, and limited cognitive abilities that fail to meet flexible demands in real-world applications [2]. - Karpathy critiques the current pace of AI development, stating that the speed of tool creation in the AI field far exceeds the improvement of AI capabilities, warning against the vision of fully automated systems that eliminate human involvement, which could lead to negative consequences [2]. Group 2: Ideal Development Path for AI - Karpathy emphasizes the importance of "human-machine collaboration" as a core value for the future of AI, envisioning AI as a collaborative partner rather than a replacement, capable of assisting in programming tasks and engaging in meaningful communication with humans [2]. - Despite a more cautious timeline for AI development compared to the optimistic views in the San Francisco AI community, Karpathy maintains a relatively optimistic outlook on AI's value, distinguishing his perspective from outright pessimism [3].