Unverified50% confidenceBenchmarkExact time
Among 42 effective corruption samples, 16 were 'corrupt-and-overtrust' cases, in which the model accepted erroneous reasoning it should not have.
1
Sources
50%
Confidence
Long-term
Relevance
10/2/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedOnly differential contamination—where some models are contaminated far more than others—distorts rankings, and this situation is rare in practice.66% similarUnverified20 samples are insufficient to support a reliable multi-class model, and forcing subdivision will introduce noise and overfitting.66% similarUnverified19.4% (6,251 articles) of duplicate document contamination exists in the naive test set split.65% similarUnverifiedThe research team analyzed item-level response data from 47 publicly released models and 74 models fine-tuned with known contamination doses.65% similarUnverifiedOut of a total of 188 model × benchmark combinations, only 3 cases exhibit differential contamination cross-validated by two reference points.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/966692API
curl https://kongchang.com/api/v1/knowledge/claims/966692MCP
get_claim(id=966692)