Contrastive Attribution in the Wild: An Interpretability Analysis of LLM Failures on Realistic Benchmarks Paper • 2604.17761 • Published Apr 20 • 4
