Cross-Model Behavioral Measurement via Thermodynamic Peclet Number: Breaking Scoring Circularity with Public Benchmark Data
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Maps 27 large language models from public benchmark scores (TruthfulQA, MMLU, HellaSwag, ARC-Challenge, Arena Elo, MT-Bench, sycophancy rates) to the Void Framework's three-dimensional behavioral space and composite Peclet number. Tests whether Pe predicts cognitive performance beyond what any single benchmark captures. Partial correlations controlling for TruthfulQA show Pe significantly predicts MMLU, HellaSwag, and ARC-Challenge (all p
Fields of Study & Subjects
Topic Modeling
Ethics and Social Impacts of AI
Natural Language Processing Techniques
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 29 March 2026
- Language
- EN
- Document type
- Preprint