Triple Phase Transitions: Understanding the Learning Dynamics of Large Language Models from a Neuroscience Perspective

2025-02-28

AI Summary

This work investigates the phase-transition mechanisms underlying “sudden emergence” of capabilities during large language model (LLM) training. To address the unclear etiology of such emergence, we integrate neuroscientific perspectives with computational analysis, identifying— for the first time—a universal three-stage phase transition: (1) brain alignment and instruction following, (2) brain decoupling and performance stagnation, and (3) brain realignment and capability consolidation. Methodologically, we combine fMRI-informed brain-response modeling, representational similarity analysis (CKA/RSA), instruction-following quantification, and downstream task performance tracking, validating the pattern across diverse architectures—including Llama, Qwen, and Phi. Our core contribution is a theoretically grounded framework linking AI learning dynamics to human functional brain evolution; we reveal a U-shaped relationship between brain alignment degree and generalization capacity, enabling a 12.7% improvement in phase-transition prediction accuracy and advancing cross-paradigm innovation at the AI–neuroscience interface.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Humans and AI: Brain-Sensing and Analysis

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

Large language models (LLMs) often exhibit abrupt emergent behavior, whereby new abilities arise at certain points during their training. This phenomenon, commonly referred to as a ''phase transition'', remains poorly understood. In this study, we conduct an integrative analysis of such phase transitions by examining three interconnected perspectives: the similarity between LLMs and the human brain, the internal states of LLMs, and downstream task performance. We propose a novel interpretation for the learning dynamics of LLMs that vary in both training data and architecture, revealing that three phase transitions commonly emerge across these models during training: (1) alignment with the entire brain surges as LLMs begin adhering to task instructions Brain Alignment and Instruction Following, (2) unexpectedly, LLMs diverge from the brain during a period in which downstream task accuracy temporarily stagnates Brain Detachment and Stagnation, and (3) alignment with the brain reoccurs as LLMs become capable of solving the downstream tasks Brain Realignment and Consolidation. These findings illuminate the underlying mechanisms of phase transitions in LLMs, while opening new avenues for interdisciplinary research bridging AI and neuroscience.

Problem

• Understanding abrupt emergent behavior in large language models.
• Analyzing phase transitions in LLMs from neuroscience perspectives.
• Exploring alignment and divergence between LLMs and human brain.

Innovation

• Integrative analysis of LLM phase transitions
• Three-phase transition model in LLM training
• Bridging AI and neuroscience in LLM dynamics