Theory explains why language models fail on unfamiliar reasoning tasks

Original: A Measure-Theoretic Analysis of Reasoning: Structural Generalization and Approximation Limits

Source: arxiv.org ↗