Unverified50% confidenceFactExact time
LEAPS adopts a data-contamination-resistant design, with datasets built entirely anew to avoid inflated scores due to models' pretraining memory.
1
Sources
50%
Confidence
Long-term
Relevance
10/20/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedBenchmark data contamination refers to the phenomenon where, after a test set is made public, its questions may appear in the pre-training corpora of subsequent models, causing scores to reflect memorization rather than genuine reasoning ability.74% similarUnverifiedOut-of-sample design can substantially eliminate the interference of memorization retrieval on evaluation by using contemporary works created after the model's training cutoff date or from controlled sources.72% similarVerifiedThe true difficulty of model fine-tuning lies in the construction and cleaning of high-quality training data, preventing catastrophic forgetting, and inference deployment after quantization.71% similarUnverified差分隐私训练、去重预处理和机器遗忘等缓解记忆化的技术方案均会在不同程度上牺牲模型性能70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1074007API
curl https://kongchang.com/api/v1/knowledge/claims/1074007MCP
get_claim(id=1074007)