What Does This Mean?
[ SLANG ][ AI ][ NOUN ]
When a model's training data accidentally (or not-so-accidentally) includes the actual questions and answers from a benchmark it's later evaluated on, inflating its score without reflecting genuine improved capability.
REAL-WORLD EXAMPLE
"Turns out that shockingly high score was just benchmark contamination, the test questions were sitting in the training set."
LORE & ORIGIN
Named directly by AI researchers auditing suspiciously high benchmark results, borrowing 'contamination' from the same concept in data science where test data leaks into a training set.
FIRST SEEN
2023
PEAK ERA (EDITORIAL)
2023-2026 (Current Era)
CURRENT STATUS
Mainstream Slang
Related To:Eval GamingEval Awareness
AURA IMPACT ⓘ
+15 AURA
TREND STATUS
→ STEADY
EDITORIAL ESTIMATE
CULTURE CATEGORY
[ SLANG ][ AI ][ NOUN ]
RELATED TERMS
MENTIONS OVER TIME
DOCUMENTED SINCE
2023
NOW
Radar observation data not yet available for this term.
RELATED SEARCHES
>what is benchmark contamination?>benchmark contamination meaning?>benchmark contamination examples?>benchmark contamination vs eval gaming?>eval awareness meaning?
RELATED TOPICS
TOP PLATFORMS
Not enough Radar data yet to show a platform breakdown.
WHEN DID YOU FIRST HEAR THIS?
CLICK AN OPTION BELOW TO CAST YOUR VOTE.
[ VOTE TO REVEAL COMMUNITY RESULTS ]