ResearchBenchmarksDistractor Generation on Human Evaluation Set (test)Follow4.14RelevanceGPT-32.4762.9083.343.772Apr 19, 2026Evaluation ResultsMethodMethodLinksRelevanceDifficultyFluencyGPT-3Selection Strategy=k-NNSelection Strategy=k-NN2026.044.143.514.78GPT-3Selection Strategy=COT...Selection Strategy=COT-k-NN2026.043.743.154.65Ground-truth2026.043.733.284.58GPT-3Selection Strategy=clu...Selection Strategy=clustering2026.043.452.934.56TinyLlama2026.043.342.974.47T5Method=beamMethod=beam2026.0432.513.9T5Method=contrastMethod=contrast2026.042.542.183.58