图灵奖得主、Meta AI部门负责人Yann LeCun在周二的Meta伦敦活动中表示，他认为大模型无法达到人类的智力水平，对通用人工智能（AGI）的概念持怀疑态度。LeCun指出，人类智能并不具备普遍性，因此追求“人类水平的AI”更为实际。他强调，当前的AI系统在推理、规划、持久记忆和理解物理世界等关键认知能力上存在显著缺陷。
以自动驾驶汽车为例，LeCun指出，由于缺乏这些基本能力，AI在应对复杂的真实世界环境时可能会遇到困难，从而影响其安全性。同样，家用机器人在处理日常家务任务时也会显得力不从心，而智能助手通常只能执行简单的指令。LeCun特别提到了大型语言模型（LLMs）的局限性，尽管它们在语言生成方面表现出色，但对现实世界的理解极其有限。
他以四岁孩子与世界互动获取信息的能力为例，估计这种接触数据的量是最大LLMs的50倍，强调了仅依赖文本知识的AI在理解复杂情境上的不足。LeCun的这些观点进一步引发了关于AI发展路径的讨论，表明在追求更先进的人工智能技术时，需要更注重模拟人类与环境交互的学习方式。
英语如下：
**News Title:** “Turing Award Winner Yann LeCun Warns: Large Language Models Fall Short of Human Intelligence, AI Faces Four Cognitive Challenges”
**Keywords:** Turing Award, Large Models, Human Intelligence
**News Content:** Yann LeCun, a Turing Award recipient and head of Meta’s AI division, expressed at a Meta event in London on Tuesday that he believes large models are incapable of reaching the intellectual capacity of humans, harboring skepticism about the concept of artificial general intelligence (AGI). LeCun contended that human intelligence is not universal, making the pursuit of “human-level AI” a more realistic goal. He emphasized that current AI systems exhibit significant deficiencies in critical cognitive abilities, such as reasoning, planning, long-term memory, and understanding the physical world.
Citing autonomous vehicles as an example, LeCun pointed out that AI may struggle in navigating complex real-world environments, potentially compromising safety due to its lack of these fundamental capabilities. Similarly, household robots often struggle with routine household tasks, while intelligent assistants are typically limited to executing simple commands. LeCun specifically highlighted the limitations of large language models (LLMs), noting their impressive language generation skills but their extremely limited understanding of the real world.
Drawing a comparison to a four-year-old’s ability to gather information through interaction with the environment, he estimated that this exposure to data is 50 times greater than that of the largest LLMs, underlining AI’s inadequacy in comprehending complex contexts when relying solely on textual knowledge. LeCun’s perspective fuels further debate on the trajectory of AI development, suggesting that a greater emphasis should be placed on mimicking human-environment interaction in the quest for more advanced AI technologies.
【来源】https://thenextweb.com/news/meta-yann-lecun-ai-behind-human-intelligence
Views: 3