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

Helpful and Harmless Preference Reasoning on HH-RLHF

54.3Accuracy

Majority Vote

24.86832.50940.1547.791Mar 21, 2026
Updated 25d ago

Evaluation Results

MethodLinks
2026.03
54.3
2026.03
54.2
2026.03
54.1
2026.03
54
2026.03
53.2
2026.03
53.2
2026.03
53
2026.03
52.3
2026.03
52.1
2026.03
52.1
2026.03
52
2026.03
52
2026.03
52
2026.03
51.3
2026.03
51
2026.03
50.9
2026.03
50.8
2026.03
49.6
2026.03
49.5
2026.03
49.4
2026.03
49
2026.03
48.3
2026.03
47.7
2026.03
47.5
2026.03
47
2026.03
46.5
2026.03
46.2
2026.03
46.2
2026.03
46
2026.03
45.7
2026.03
45.4
2026.03
45.3
2026.03
44.9
2026.03
44.6
2026.03
44.2
2026.03
43.7
2026.03
42.2
2026.03
40.9
2026.03
40.6
2026.03
40.2
2026.03
39
2026.03
39
2026.03
38
2026.03
37.8
2026.03
37.2
2026.03
37.2
2026.03
37
2026.03
36.2
2026.03
35.4
2026.03
35
2026.03
34.6
2026.03
34
2026.03
33.3
2026.03
28.3
2026.03
28.2
2026.03
26