Large Language Models feel the direction of time
Researchers have found that AI large language models, like GPT-4, are better at predicting what comes next than what came before in a sentence. This “Arrow of Time” effect could reshape our understanding of the structure of natural language, and the way these models understand it.