Sure, #AI #LLMs are "next word predictors", but to do so optimally you need to model the entire process that produces those words. This involves not only language, but also human reasoning, physics, history, economics, politics, etc. In essence predicting the next word optimally, based on a data set generated by human brains, is the same as simulating the human #brain.

Taken from: https://www.reddit.com/r/MachineLearning/comments/12qe5hm/d_microsoft_research_paper_sparks_of_artificial/