Mechanistic Interpretability Benchmark
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
Mechanistic Interpretability Benchmark (MIB) Indirect Object Identification (IOI) (standard)
0CMD
12
Mechanistic Interpretability Benchmark (MIB) MCQA (standard)
0.04CMD
9
Mechanistic Interpretability Benchmark (MIB)
83IOI
9