Radioactive Feedback: Tracing Models Trained by Label-Only LLM Supervision
Conventional distillation transfers capability through teacher-generated demonstrations. LLM-as-a-Judge knowledge distillation offers a distinct and increasingly practical alternative. In the label-only variant studied here, the student generates every response and improves from scalar scores alone. Prior work shows th...
Yue-Jia Li, Junheng Li, Hao Wu et al.
· 0 citations