Correlation with Human Majority Vote on Novelty-Bench Human Majority Vote benchmark leaderboard · SOTA2 Research

Correlation with Human Majority Vote on Novelty-Bench Human Majority Vote

0.88Spearman Correlation (All)GPT-4o

Evaluation Results

Method
GPT-4o2025.09	0.88	1	1	1	0.83	0.64	0.83	0.85
Claude-3-Sonnet2025.09	0.88	1	1	1	0.83	0.74	0.71	0.85
Gemini-1.5-Flash2025.09	0.83	1	1	1	1	0.74	0.41	0.85
Embedding Diversity2025.09	0.8	0.43	0.82	0.59	0.66	0.77	0.66	0.77
Vocabulary Diversity2025.09	0.64	0.25	0.51	0.44	0.46	0.7	0.54	0.7
Novelty-Bench Functional Diversity2025.09	0.59	0.37	0.13	0.45	0.69	0.59	0.62	0.73
Compression Diversity2025.09	0.26	0.06	0.62	0.45	0.62	0.73	0.39	0.49