New MarCognity-AI study measures epistemic failure rates across 8 domains to identify knowledge gaps in large language models

MarCognity-AI project evaluates how frequently LLMs provide incorrect or uncertain responses across different knowledge domains

Study focuses on epistemic failures—instances where models lack reliable knowledge or confidence calibration

Research spans 8 distinct domains to provide comprehensive assessment of LLM reliability and knowledge boundaries

Findings available on GitHub as part of open-source AI research initiative