Meta’s AI Chief Thinks That AI Super Intelligence Is Not Coming Anytime Soon
by Shahnawaz Alam News 04 December 2023
Meta’s chief scientist and deep learning pioneer, Yann LeCun, expressed his belief that achieving true sentience in current AI systems is decades away. He emphasized the need for common sense to elevate AI capabilities beyond creative text summarization. This perspective contrasts with Nvidia CEO Jensen Huang’s assertion that AI will be “fairly competitive” with humans within five years, highlighting differing opinions within the AI community.
LeCun, speaking at an event commemorating Meta’s 10-year anniversary of its Fundamental AI Research team, suggested that Huang’s stance may be influenced by the competitive nature of the AI industry, referring to it as an “AI war” where Nvidia supplies the necessary hardware.
Regarding artificial general intelligence (AGI), LeCun noted that more GPUs would be required for its development. He acknowledged the ongoing pursuit of AGI by organizations like OpenAI, indicating a continued demand for Nvidia’s computer chips.
LeCun questioned the effectiveness of text-based training, deeming text a “poor source of information.” He highlighted the limitations of training models on vast amounts of text and emphasized the vast amount of time it would take a human to read equivalent material.
In response to these challenges, Meta’s AI executives, including LeCun, are actively researching how transformer models, like those used in ChatGPT, can be adapted for diverse data types such as audio, image, and video. The aim is to uncover hidden correlations among these data types, potentially enabling more advanced AI capabilities.
Meta’s research extends to practical applications, such as software designed to enhance tennis skills using augmented reality. This demo showcased the integration of 3D visual data, text, and audio to provide real-time guidance during tennis play.
The development of multimodal AI systems, capable of processing various data types, is considered the next frontier. However, this progress comes with significant costs. As companies like Meta and Google parent Alphabet invest in advanced AI models, Nvidia stands to gain a competitive edge, particularly if no alternative competitors emerge.