paper_type.json (354B)
1 { 2 "paper_type": "empirical", 3 "reason": "The paper proposes RLTHF, a human-AI hybrid framework, and validates it through quantitative experiments on standard benchmarks (HH-RLHF, TL;DR) with ablation studies and comparisons to baselines, demonstrating that the method achieves comparable accuracy with significantly reduced human annotation effort." 4 }