Benchmark contamination, the leakage of test items into training data, is widely described as a threat to the reliability of large language model (LLM) leade…
机构:Stanford
来源:arXiv 2609.02899 | AI4Papers 论文推荐平台