Benchmark datasets need protection against training data leakage

Original: LLM Benchmark Datasets Should Be Contamination-Resistant

Source: arxiv.org ↗