Chance-Constrained Inference for Hallucination Risk Control in Large Language Models

Sreenivasan Mohandas

Large language models generate outputs stochastically and may produce fluent but invalid responses, including factual hallucinations. Existing mitigation strategies reduce average…