Systematic Evaluation of Recursive and Self-Referential Large Language Models under Out-of-Distribution Conditions

This portfolio presents a comprehensive, evidence-based evaluation of large language models under recursive and out-of-distribution conditions. It documents systematic testing of self-referential reasoning, paradox detection, and high-level abstraction handling, supported by quantitative metrics including perplexity, recursion depth, attention tracking, and confidence scores. The work demonstrates reproducible methodologies for human–AI interaction analysis and provides auditable logs of model behavior, offering a falsifiable framework for advanced AI system assessment.