Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Preference Prediction on Composite Dataset (Arena-Expert-5K, HelpSteer3, HH-RLHF, UltraFeedback)

70.5Accuracy

PREMISE (+ pref repair)

62.764.72566.7568.775May 29, 2026
Updated 2d ago

Evaluation Results

MethodLinks
2026.05
70.5
2026.05
70.5
2026.05
70.5
2026.05
70.3
2026.05
70.2
2026.05
69.9
2026.05
69.9
2026.05
69.8
2026.05
69.7
2026.05
69.6
2026.05
69.5
2026.05
69.2
2026.05
69
2026.05
69
2026.05
68.8
2026.05
68.6
2026.05
68.4
2026.05
68.3
2026.05
68.2
2026.05
68.1
2026.05
68.1
2026.05
68
2026.05
67.9
2026.05
67.9
2026.05
67.5
2026.05
67.4
2026.05
67.4
2026.05
67.1
2026.05
67
2026.05
67
2026.05
66.4
2026.05
66.4
2026.05
66.2
2026.05
66.1
2026.05
66
2026.05
65.8
2026.05
65.3
2026.05
65.2
2026.05
64.8
2026.05
64
2026.05
63.6
2026.05
63