Physics of generative AI’s atom: Repetition, bias, and beyond

We derive a first-principles physics theory of the individual “atom” at the heart of generative AI such as ChatGPT: the basic attention head. The theory shows how, when, and why its output can become repetitive or suddenly switch to potentially harmful content. In situations where a small subset of attention heads dominates generative AI’s output or multi-layer effects average out, such undesirable microscale behavior will emerge at the macroscale to threaten generative AI’s safety in medical, legal, and business settings. The theory also quantifies the impact of bias from training and fine-tuning. The theory’s 2-spin form suggests why generative AI such as ChatGPT can work so well but hints that a generalized 3-spin attention might be even better. The theory’s similarity to spin bath physics means existing physics expertise could be harnessed to help generative AI become more trustworthy and resilient to manipulation.

• Publication year

  2026

• Venue

  AIP Advances

• Publication date

  2026-03-01

• Fields of study

  Not labeled

• Identifiers
  DOI 10.1063/5.0296911
• External record

  Open on Semantic Scholar

• Source metadata

  Semantic Scholar

• No claims are published for this paper.

• No concepts are published for this paper.

• Phase Transitions in Large Language Models and the O(N) Model

  2025cited by this paper
• Training large language models on narrow tasks can lead to broad misalignment

  2025cited by this paper
• Basic Attention Head as a Building Block toward Understanding Transformer-based Generative AI

  2025cited by this paper
• Multispecies Cohesion: Humans, Machinery, AI, and Beyond.

  2024cited by this paper
• Talking Heads: Understanding Inter-layer Communication in Transformer Language Models

  2024cited by this paper
• Progress measures for grokking via mechanistic interpretability

  2023cited by this paper
• Mapping of attention mechanisms to a generalized Potts model

  2023cited by this paper
• Energy Transformer

  2023cited by this paper
• Transformer Variational Wave Functions for Frustrated Quantum Spin Systems.

  2022cited by this paper
• Emergent Abilities of Large Language Models

  2022cited by this paper
• A Theoretical Analysis of the Repetition Problem in Text Generation

  2020cited by this paper
• DeeBERT: Dynamic Early Exiting for Accelerating BERT Inference

  2020cited by this paper
• BERT Loses Patience: Fast and Robust Inference with Early Exit

  2020cited by this paper
• The Curious Case of Neural Text Degeneration

  2019cited by this paper
• Attention in Natural Language Processing

  2019cited by this paper
• DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter

  2019cited by this paper
• Reducing Transformer Depth on Demand with Structured Dropout

  2019cited by this paper
• Are Sixteen Heads Really Better than One?

  2019cited by this paper
• Analyzing Multi-Head Self-Attention: Specialized Heads Do the Heavy Lifting, the Rest Can Be Pruned

  2019cited by this paper
• Attention is All you Need

  2017cited by this paper
• Neural Machine Translation by Jointly Learning to Align and Translate

  2014cited by this paper

• No citing papers are available for this paper.