A four-year-old has seen more data than an LLM: AI pioneer Yann LeCun

Yann LeCun, Vice-President and Chief AI Scientist at Meta

AI pioneer Yann LeCun said a four-year-old child has probably seen more data than a large language model (LLM), calling for the need to build AI systems that can understand the world, reason, and plan.

“Robot intelligence doesn't come anywhere close to what cats and dogs do. Domestic robots still can't accomplish what a ten-year-old can do. The world model needs to be trained from sensory inputs," LeCun said during a lecture at a packed Music Academy in Chennai.

He further quipped, “If you are interested in human-level AI, don't work on LLMs. Abandon generative models if you are in favour of human-level intelligence. Abandon probabilistic models in favour of energy-based models.”

LeCun, who is a Vice-President and Chief AI Scientist at Meta, which owns Facebook, Instagram, and WhatsApp, was in Chennai to deliver the Indian Institute of Technology Madras’ (IIT Madras) "Prof. Subra Suresh Institute Lecture Series" on October 22, 2024.

LeCun, also a faculty at New York University, gave an address on the topic of ‘How Could Machines Reach Human-Level Intelligence?’.

He was awarded, along with Geoffrey Hinton and Yoshua Bengio, the 2018 Turing Award, often referred to as the ‘Nobel Prize in Computer Science’, for his work on Deep Learning. Together, the three of them are often referred to as the Godfathers of AI.

LeCun noted that India has a large community of AI enthusiasts, and predicted that all human interactions with the digital world will soon be mediated by AI assistants.

“We shouldn't be threatened by machines who are smarter than us. We need systems that resemble human intelligence or advanced machine intelligence (AMI). Human intelligence is not general at all,” he said preferring to use the term AMI rather than Artificial General Intelligence (AGI).

LeCun envisions a world where intelligent assistants can help us in our daily lives. Smart glasses that can communicate through voice, vision, and display are expected to play a vital role.

“We need to solve for large-scale world model training. We need a lot of data from India. We need a way to do distributed training without copying the data. Open-source AI platforms are necessary. Open research should not be regulated out of existence,” he said.

Referring to the recent wins by AI leaders in Nobel Physics and Chemistry, he quipped, “Nobel committee was under some pressure to reward deep learning."