Online LLM Picker Reduces Model Selection Annotation Costs by 71%

The Online LLM Picker framework uses active learning to select the most informative prompts for annotation in streaming data environments. Testing across 130 models and 10 datasets shows the method reduces annotation costs by up to 71.67% while maintaining high accuracy in identifying optimal models.