📊 Market Data Snapshot
A research article from Stanford University and Georgetown University highlights a fundamental flaw in how AI is trained: large language models need roughly 100,000 times more input than a human child to learn a language. The gap, laid out by researchers Michael C. Frank and Ethan Gotlieb Wilcox, is a key challenge for AI development and already prompting investors to look at which crypto projects actually stand to benefit from the AI boom.
The 100,000x gap
The article points to Meta's Llama 3.1, which was pretrained on 15 trillion tokens. Wilcox notes that frontier models may pretrain on ten times that amount. Compare that to a preteen in a linguistically rich home, who has likely heard about 100 million words. With literacy, that number can reach 300 million
