Initially shows how using Hypernym's core technology to fingerprint LLM models can detect what model class has been used to process a result with a unique, orthogonal benchmark.,
For Agentic tool use: Proves that "Distractors" hamper tool using LLMs even if the LLMs themselves recognize it.,
This project whitepaper:, suggests using Hypernym's technology to detect these high-variability "Distractor".