Anthropic HH-RLHF
Benchmarks
TaskDatasetSOTA resultTrendResults
Anthropic hh-rlhf
3.36Gold Reward
14
Anthropic-hh-rlhf (test)
5.83LLM-as-a-Judge Helpful Score
12
Anthropic/hh-rlhf HH-helpful core250
0.292Delta RM
6
Anthropic HH-RLHF
82.5Preference Coverage
6
Anthropic HH-RLHF 2022 (test)
62Win Rate
4
Anthropic HH-RLHF+VI Preference (test)
64Overall Accuracy
3