Zero-source LLM Hallucination Detection with Human-like Criteria Probing

In three linesHCPD, a hallucination detection method without access to model internals or external references. An LLM agent adaptively decomposes judgment into weighted, interpretable criteria aligned via weak supervision on semantic consistency. Code released.

Summary generated by Claude — human-verified