[2L4-GS-1] Fundamental AI, theory, algorithm:

[2L4-GS-1-01] Theory of Emergence in LLM

〇Tomohiro Isshiki¹ (1. Financial Technology System Inc.)

Keywords:neural network, learning, theory, Emergence

LLM is one of the central themes in AI in recent years. In LLM, when the increase in training data and computational resources exceeds a certain threshold, a phenomenon is observed in which unexpected new capabilities are acquired. This phenomenon is called emergence. This emergence goes beyond conventional technical understanding, and many researchers are attempting to elucidate its mechanism.
Therefore, this study also aims to elucidate the mechanism of emergence, and mathematically discusses the mechanism by which this emergence occurs.
To be more specific, LLM is a neural network and has the properties of a neuronal manifold. Therefore, in this study, we use a mathematical theory that can be applied to neuronal manifolds to show that emergence can occur under the above conditions.