In a surprise announcement, Yann LeCun, the chief scientist of Meta, has said that generative AI, the technology behind ChatGPT, is already at a dead end. Instead, LeCun is promising new artificial intelligence that resembles human rationality.
"Today AI and machine learning really sucks," LeCun said at a Meta launch event in Paris. "Humans have common sense, machines don't."
LeCun's comments come as a shock to many in the AI community, who have been quick to point out that ChatGPT and other generative AI models have achieved impressive results in recent years. For example, ChatGPT has been used to create realistic-looking fake news articles, write poetry, and even generate code.
However, LeCun argues that generative AI models are still fundamentally limited. They can only generate text or code that they have already seen before. They cannot understand the meaning of what they are generating, and they cannot reason about abstract concepts.
In contrast, LeCun's new AI project, called Joint Embedding Predictive Architecture (JEPA), is designed to mimic human rationality. JEPA will be able to understand the meaning of text and code, and it will be able to reason about abstract concepts.
LeCun is confident that JEPA will revolutionize the field of AI. He predicts that in a few years, generative AI models will be obsolete, and JEPA will be the new standard for AI.
It remains to be seen whether LeCun is right. However, his comments have certainly sparked a debate about the future of AI.
No comments:
Post a Comment