At a Meta farewell event in Paris, Meta’s lead researcher, Yann LeCun, made a remarkable statement, Meta declares that generative AI, generative simulated intelligence, has reached a dead end. Instead, he pledged the development of novel artificial intelligence models with rationality similar to that of humans.
LeCun expressed his displeasure with the current status of AI and machine learning, pointing out that robots lack common sense in contrast to humans. He was shown Meta’s most recent AI endeavour, the Joint Embedding Predictive Architecture (JEPA), which was image-based. This ambitious project aims to go beyond the capabilities of generative AI by allowing robots to conceptualize abstract concepts rather than just repeating information from the internet.
According to LeCun
According to LeCun, Meta plans to prioritize joint embedding predictive architecture, rendering generative models obsolete. He confidently said, “My prediction is that generative large language models will not be used anymore in a few years; we will have something superior to supplant them.” LeCun, well-known for his expertise in AI, has been critical of promoting generative artificial intelligence models like ChatGPT and the image-based Dall-E since they were introduced the year before.
LeCun acknowledges that the excitement and zeal around generative AI have created a distorted perception of its capabilities. Imprint Zuckerberg, president of Meta, announced on Facebook that the JEPA instrument would be open source, allowing researchers to investigate other options.
The goal is to develop AI closely resembling how people perceive the environment. As Zuckerberg said, “We want models that recognize the world & make predictions.” This study is another step in that direction.
Unlike its competitors, Meta has adopted a more careful approach to handling ChatGPT-style artificial intelligence for its social media platforms, Facebook and Instagram. Although Meta has incorporated generative AI into its products, Microsoft and Google have garnered more media attention. Additionally, Meta has introduced open-source AI models that are less computationally intensive than ChatGPT’s technology.
LeCun’s statement represents a significant shift in the AI field since Meta is driving the creation of new models of AI that seek to mimic human reason. The JEPA project is a ground-breaking step toward enhancing machine intelligence beyond the constraints of generative AI.