Open-ended scientific discovery on iNatDisco-50K
0.6667RecallDiscoPER
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DiscoPERType=LLM, H=H_open_code, Reflect=True, Prior Info=None2026.07 | 0.6667 | 74.2 | |
| DiscoPER w/o ReflectType=LLM, H=H_open_code, Reflect=False, Prior Info=None2026.07 | 0.5 | 66.6 | |
| ExperiGenType=LLM, H=H_guided_code, Reflect=True, Prior Info=Partial2026.07 | 0.25 | 67.8 | |
| HeurekaBenchType=LLM, H=H_guided_code, Reflect=False, Prior Info=Partial2026.07 | 0.1667 | 64.7 | |
| LLM+NOTEARSType=Causal, H=H_edge, Reflect=False, Prior Info=None2026.07 | 0.0833 | — | |
| GPT-4 BFSType=Causal, H=H_edge, Reflect=False, Prior Info=None2026.07 | 0.0833 | — | |
| LLM+PCType=Causal, H=H_edge, Reflect=False, Prior Info=None2026.07 | 0 | — |