SEP 3, 2026 · PREPRINT
Legibility is Not Interpretability: Comparing Judged and Actual Importance in Chain-Of-Thought Reasoning
arXiv
This is an empirical study of LLM reasoning traces using a novel operationalization of step importance, but it is exploratory and methodological rather than addressing a settled clinical or therapeutic question, and the findings raise cautions rather than resolve them.
Reported
Primary comparisonLLM judges outperform prevalence baseline b…
Fine-tuning effectstrong improvement for incorrect responses;…