Illusory Pattern Perception Drives Spurious Inference in Large Language Models

Key points

• This paper investigates whether Large Language Models (LLMs) exhibit such perceptual tendencies, which can lead to systematic errors in downstream applications.1
• To our knowledge, this work presents the first systematic study of illusory pattern perception in LLMs, adapting classic psychological paradigms to three tasks with direct empirical comparison to human behaviors.1
• We find that LLMs frequently exhibit stronger illusory pattern perception than humans.1
• To uncover the mechanism behind these behaviors, we develop a feature interpretability framework based on Sparse Autoencoders (SAEs) to analyze internal representations.1

Sources (1)

• [1]Illusory Pattern Perception Drives Spurious Inference in Large Language Models
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 6, 05:37 AM

  Illusory pattern perception is a well-documented human cognitive tendency to infer meaningful relationships in data that is actually random.

  This paper investigates whether Large Language Models (LLMs) exhibit such perceptual tendencies, which can lead to systematic errors in downstream applications.

Extractive summary: sentences quoted from the sources.