LeCun criticizes large language models (LLMs) like Meta’s LLaMA, OpenAI’s GPT-3, and Google’s Bard, stating that they are severely restricted by their reliance on text for learning, which he believes is not the primary method of human learning. He proposes an alternative architecture called "objective-driven AI", which learns about the physical world through sensors and training on video data, and is built to fulfill specific goals set by humans. Despite his skepticism, LeCun believes that machines will eventually surpass human intelligence, but not as soon as some predict.
Key takeaways:
- Yann LeCun, Meta’s chief AI scientist and a winner of the prestigious Turing Award, believes that artificial general intelligence (AGI) is a distant destination and current AI systems can't perform essential characteristics of human intelligence like reasoning, planning, persistent memory, and understanding the physical world.
- LeCun argues that large language models (LLMs) are not as intelligent as they seem due to their reliance on text for learning, which he believes is a superficial understanding of reality.
- LeCun proposes an alternative architecture called “objective-driven AI,” where AI systems are built to fulfil specific goals set by humans and learn about the physical world through sensors and training on video data.
- Despite his skepticism, LeCun believes that machines will eventually surpass human intelligence, but it will take a significant amount of time, contrary to predictions made by tech leaders like Elon Musk.