UChicago Study: LLM Self-Reported Decision Factors Are Unreliable

1 reports archived1 independent sourcesupdated 10/24/2026, 17:30:05

Synthesis & Latest Updates

1 Sources Cross-Validated

A new University of Chicago paper reveals that LLMs' self-reported accounts of their decision-making factors are often inaccurate and fail to identify the true variables driving answer variations. The study demonstrates that comparing multiple generated answers to find disagreement points (cruxes) is more effective than querying the model directly or analyzing a single response, while being significantly cheaper in compute than local LLM judges. This finding offers critical guidance for auditing AI decisions in high-stakes scenarios.

LATEST/A new University of Chicago paper reveals that LLMs' self-reported accounts of their decision-making factors are often inaccurate and fail to identify the true variables driving answer variations. The study demonstrates that comparing multiple generated answers to find disagreement points (cruxes) is more effective than querying the model directly or analyzing a single response, while being significantly cheaper in compute than local LLM judges. This finding offers critical guidance for auditing AI decisions in high-stakes scenarios.

Hourly heat curve

2 fully observed hours

Fewer than 3 fully observed hours; trend not plotted yet.

Coverage timeline

Total 1 reports · Latest first

Rohan Paul (@rohanpaul_ai)T2Oct 24, 17:30·78 pts
UChicago Study: LLM Self-Reported Decision Factors Are Unreliable
- LLMs' self-reported accounts of key decision factors (cruxes) have extremely low accuracy, falling below random baseline levels.
- Sampling multiple answers and comparing their disagreements effectively identifies the core factual variables driving model outputs.
- Using rerankers to match answer discrepancies costs approximately 1/86th of the compute required by local LLM judges.
View Related Reports & Analysis Original Source