Inverse Scaling in Chain-of-Thought Faithfulness

A study of whether CoT reasoning faithfulness decreases as model scale increases, tested across 11 open-weight Llama and Qwen models.

Research Project Details

Add detailed information about your research project here.

Abstract

Brief summary of the research.

Methodology

Describe your approach.

Results

Across all experiments, LLMs produced plausible yet inaccurate estimates, showing that they rely on heuristics rather than actual computation.