English

Position: Understanding LLMs Requires More Than Statistical Generalization

Machine Learning 2024-06-18 v3 Machine Learning

Abstract

The last decade has seen blossoming research in deep learning theory attempting to answer, "Why does deep learning generalize?" A powerful shift in perspective precipitated this progress: the study of overparametrized models in the interpolation regime. In this paper, we argue that another perspective shift is due, since some of the desirable qualities of LLMs are not a consequence of good statistical generalization and require a separate theoretical explanation. Our core argument relies on the observation that AR probabilistic models are inherently non-identifiable: models zero or near-zero KL divergence apart -- thus, equivalent test loss -- can exhibit markedly different behaviors. We support our position with mathematical examples and empirical observations, illustrating why non-identifiability has practical relevance through three case studies: (1) the non-identifiability of zero-shot rule extrapolation; (2) the approximate non-identifiability of in-context learning; and (3) the non-identifiability of fine-tunability. We review promising research directions focusing on LLM-relevant generalization measures, transferability, and inductive biases.

Keywords

Cite

@article{arxiv.2405.01964,
  title  = {Position: Understanding LLMs Requires More Than Statistical Generalization},
  author = {Patrik Reizinger and Szilvia Ujváry and Anna Mészáros and Anna Kerekes and Wieland Brendel and Ferenc Huszár},
  journal= {arXiv preprint arXiv:2405.01964},
  year   = {2024}
}

Comments

Accepted as a position paper at ICML2024, Code: https://github.com/rpatrik96/llm-non-identifiability

R2 v1 2026-06-28T16:15:19.446Z