AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

The Consistency Dilemma in LLMs: Generator-Evaluator Agreement and Vulnerability to Mistakes

Marina Mancoridis, Zoe Hitzig

ICML 2026 · Large language models › Language, safety and interpretability · approach: Statistical frameworks for LLM evaluation

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
higher generator-evaluator self-consistency across modelsfrontier LLMs · MedMistakes-Validatedmistake vulnerability: beta2 = 1.116, p < 10^-6Hurt
use two-way clustered standard errors (concept and model)self-consistency association · MedMistakes-Validatedmistake vulnerability: SE = 0.533, p = 0.036Hurt
add model fixed effects to the regression (ablation)self-consistency coefficient · MedMistakes-Validatedmistake vulnerability: 1.116 -> 0.147 (SE = 0.334, p = 0.66)No clear effect
higher MedQA benchmark accuracyfrontier LLMs · MedMistakes-Validatedmistake vulnerability: beta1 = -1.732, p = 0.042Improved
leave out Correct for Incomplete Reason test from consistency score (ablation)self-consistency score · MedMistakes-Validatedconsistency-vulnerability association: substantially attenuates the effectHurt
model choice: Gemini 3 vs Claude Sonnet 4.5ontological tests · List True Attributeschance-corrected agreement: 1.00 vs -0.23Mixed
model choice: GPT-5 vs GPT-4oNone of the Above test · None of the Abovechance-corrected agreement: 0.11 -> 0.79Improved
model choice: Opus 4.5 vs DeepSeekCorrect for Incomplete Reason · Correct for Incomplete Reasonchance-corrected agreement: -1.00 -> 0.73Improved
stability: modal outcome is agreement vs disagreementrepeated evaluation · 200 instances x 5 runsstability of judgment: difference in means = 0.056, 95% CI [0.003, 0.113]Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy