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

Preference Alignment on UltraFeedback

83Win Rate

EFT

44.5254.5164.574.49Sep 30, 2025Nov 5, 2025Dec 12, 2025Jan 18, 2026Feb 23, 2026Apr 1, 2026May 8, 2026
Updated 1mo ago

Evaluation Results

MethodLinks
2025.09
830.5
2026.05
81-
2026.05
79-
2025.09
770.58
2026.05
75.33-
2025.09
72-0.39
2025.09
701.04
2025.10
67-3.12
2025.09
65-0.03
2025.09
651.22
2025.10
64-3.75
2025.10
63-3.59
2025.09
610.18
2026.05
58.33-
2025.10
58-4.42
2025.10
56-9.25
2025.09
560.3
2026.05
53.67-
2025.10
52-9.51
2025.10
52-9.7
2025.10
51-9.87
2025.10
50-5.47
2025.10
50-10.02
2026.05
46-
2025.09
-0.51
2025.09
-1.59