Camilo Chacón Sartori

Score inflation in emotional-support chatbot evaluations can mask critical safety failures, with model-specific performance diverging by up to six points on key metrics.

Algorithmic expertise can now be explicitly represented, learned, and transferred as executable knowledge graphs, unlocking zero-shot generalization across domains.

LLMs can be confidently wrong about *why* they succeed, and accurately explain failures they can't fix, revealing a fundamental disconnect between explanation and competence.