Triangular Vector Arbitration (TVA) and Multimodel Semantic Triangulation (MMST): A New Framework and Computational Protocol for Analizing Divergence Between Language Models

Key Points

• The study aims to introduce frameworks for assessing and analyzing divergence between language models at the embedding level.
• Introduction of Triangular Vector Arbitration (TVA) and Multimodel Semantic Triangulation (MMST) frameworks.
• Development of two new metrics: Divergence Index (ID) and Semantic Dispersion (SD).
• Computational protocols for language model evaluation.
• TVA and MMST provide novel approaches to measure language model divergence.
• The Divergence Index (ID) quantifies divergence effectively.
• Semantic Dispersion (SD) facilitates understanding of model variability.