A toddler beats ChatGPT: Yann LeCun's world-model bet
SMRTR summary
A four-year-old child, it turns out, has absorbed roughly as much data as the largest AI language model ever trained. That's the provocative claim at the heart of Yann LeCun's new venture.
LeCun, one of AI's founding figures, spent 12 years as Meta's chief AI scientist before stepping away earlier this year to launch AMI, backed by a reported one billion dollar seed round. His argument is sharp: large language models are useful, but they're fundamentally limited. They predict text. They don't understand the physical world.
His alternative focuses on "world models," systems that learn how reality works the way children do, through observation, not labels.
He's also proposing something called Tapestry, a federated open model where countries train AI on their own data without ever sharing it, a direct answer to concerns about cultural and linguistic dominance by US and Chinese tech giants.
LeCun's bet is that Silicon Valley is trapped in a rut, and the next real AI breakthrough won't come from bigger chatbots.
SMRTR provides this summary for quick context. The original article belongs to Daily.dev.
Read the original article