Patterns Over Principles: The Fragility of Inductive Reasoning in LLMs under Noisy Observations

Chunyang Li, Weiqi Wang, Tianshi Zheng +1

Inductive reasoning, a cornerstone of human cognition, enables generalization from limited data but hasn't yet been fully achieved by large language models (LLMs). While modern LLM…