CriticGen Turns LLM Evaluation Into Actionable Answer Improvement

CriticGen is a generation-aware LLM evaluation framework that produces a score, a reason, an executable refinement suggestion, and a refined answer using dynamic, instance-specific rubrics.

Traditional evaluation is coarse-grained and decoupled from generation, while CriticGen is generation-aware and produces actionable feedback rather than generic explanations.

Yes. Because rubrics are generated per sample, teams can build evaluation criteria tuned to Arabic language, local culture, and regional regulatory context instead of relying on generic English benchmarks.

The paper is an arXiv preprint that has not yet undergone full peer review, though the numbers are consistent across multiple metrics and warrant follow-up.