QQJ: Quantifying Qualitative Judgment for Scalable and Human-Aligned Evaluation of Generative AI

In three linesQQJ is an evaluation framework for generative AI combining expert-designed multi-dimensional rubrics and LLM evaluator calibration on small high-quality annotation sets. Tested on text and image generation, QQJ shows stronger alignment with human judgment than traditional automatic metrics and unconstrained LLM-based evaluators.

Summary generated by Claude — human-verified