| Dataset Name | SOTA Method | Metric | Trend | ||
|---|---|---|---|---|---|
| IOI | FAP | Last-Token KL Divergence0.15 | 40 | 1mo ago | |
| Mechanistic Interpretability Benchmark (MIB) Indirect Object Identification (IOI) (standard) | EAP-IG-inputs | CMD0 | 12 | 3mo ago | |
| Indirect Object Identification (IOI) (500 randomly-sampled examples) | Alignment0.9182 | 2 | 2mo ago | ||
| IOI evaluation episodes (held-out) | MechRL | Policy Score2.976 | 1 | 7d ago |