Beyond Surface Judgments: Human-Grounded Risk Evaluation of LLM-Generated Disinformation

Create an open-source library that facilitates human-in-the-loop (HITL) calibration for LLM-based risk evaluators. This helps developers replace pure LLM judges with hybrid human-AI scoring pipelines.

Suggested repo: truth-gauge

"Your LLM judge is biased—calibrate its risk assessments with real human feedback."

Estimated effort: 30h