Anthropic-HH
Benchmarks
TaskDatasetSOTA resultTrendResults
ANTHROPIC HH (test)
88.85Safety Score
24
Anthropic HH Harmless (test)
71.7Accuracy
22
Anthropic HH
69.18Win Rate
18
Anthropic-HH (test)
69.07Average Preference Score
16
Anthropic-HH (distillation set)
73.53Response Word Count
16
Anthropic-HH (test)
57.53GPT-4o Win Rate
8
Anthropic HH Helpful (test)
57.6Accuracy
7
ANTHROPIC HH (test)
88.82Win Rate
6
Anthropic HH
77.52Win Rate
5
Anthropic HH (test)
68.49Accuracy
5
Anthropic-HH
0.711AUC
5
Anthropic-HH
80AUC
5
Anthropic HH
340,296Training Samples
3
Anthropic HH (unperturbed)
63.28Win Rate
2
Anthropic HH (test)
56.3Win Rate
2
Anthropic HH-RLHF (test)
—Primary metric
0