ResearchBenchmarksPersonalized news headline generation on LaMP-4 (test)Follow40Alignment Win (%)DPS33.7635.383738.62Apr 24, 2026Evaluation ResultsMethodMethodLinksAlignment Win (%)RelevanceFluencyStyleAlignmentFactualityDPSEvaluator=HumanEvaluator=Human2026.0440—————CADEvaluator=HumanEvaluator=Human2026.0434—————CADEvaluator=GPT-5.2Evaluator=GPT-5.22026.04—3.974.513.623.634.08DPSEvaluator=GPT-5.2Evaluator=GPT-5.22026.04—4.454.833.913.934.21