This research introduces a mathematical framework to measure similarity between large language models, helping detect unauthorized copying and protect intellectual property.

• Uses perplexity curves and Menger curvature to create a distinctive signature for each model
• Provides a quantifiable metric to determine if one LLM is derived from another
• Demonstrates effectiveness across various model architectures and sizes
• Addresses growing concerns about AI model theft and copyright infringement

For security professionals, this method offers a practical solution to verify model originality and protect valuable AI assets in an increasingly competitive landscape.

A Perplexity and Menger Curvature-Based Approach for Similarity Evaluation of Large Language Models