Physics in Next-token Prediction
This work addresses the fundamental mechanisms of intelligence emergence and energy efficiency in large language models for researchers and practitioners in AI.
The paper tackles the problem of understanding the physics behind next-token prediction in auto-regressive models by identifying laws of information conservation and energy consumption, resulting in the formulation of the First and Second Laws of Information Capacity and demonstrating consistency with existing scaling laws.
We discovered the underlying physics in Next-token Prediction (NTP). We identified the law of information conservation within NTP and proposed the First Law of Information Capacity (IC-1), demonstrating that the essence of intelligence emergence in auto-regressive models is fundamentally a process of information transfer. We also introduced Landauer's Principle into NTP, formulating the Second Law of Information Capacity (IC-2), which establishes the relationship between auto-regressive model training and energy consumption. Additionally, we presented several corollaries, which hold practical significance for production practices. Finally, we demonstrate the consistency between the Law of Information Capacity and the Scaling Law for Neural Language Models, the Knowledge Capacity Scaling Laws, and the Scaling Laws for Precision.