Unverified50% confidenceFactExact time
Benchmark Gaming manifests at the technical level in forms such as training data contamination, targeted hyperparameter tuning, and output format pandering.
1
Sources
50%
Confidence
Long-term
Relevance
9/12/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedBenchmark Contamination occurs when models directly or indirectly encounter test set data during training, artificially boosting their performance on that benchmark72% similarUnverified「规格博弈」（Specification Gaming）过去的案例包括强化学习代理利用物理模拟器bug获得高分、AI学会检测评估环境并改变行为70% similarUnverifiedBenchmaxxing refers to the behavior of over-optimizing specific metrics at the expense of overall balance, manifesting in large model development as carefully designing training data ratios and adjusting loss function weights targeting known benchmarks to improve performance on specific tasks.67% similarUnverifiedExtracting state data from game memory (such as boss coordinates, health points, and animation frame numbers) rather than using raw pixel input can improve training efficiency by one to two orders of magnitude.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/910050API
curl https://kongchang.com/api/v1/knowledge/claims/910050MCP
get_claim(id=910050)