What AI says about HalluDetect

HalluDetect is a project that detects hallucinations in large language model (LLM) outputs using a token-probability based approach. It provides implementations to train and evaluate classifiers (logistic regression and MLP) on features derived from LLM-generated text, using an auxiliary LLM (LLM-E) to compute token probabilities. It includes datasets (HaluEval and related corpora) and documentation linking to a paper with theory and results, enabling evaluation across different tasks.