Unverified50% confidenceConceptExact time
GhostInk benchmark measures not recognition accuracy, but whether a model admits what it cannot read, rather than filling the gap with a plausible-sounding answer.
1
Sources
50%
Confidence
Long-term
Relevance
10/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedLossless compression conclusions reported on contaminated benchmarks are meaningless, because the model can directly retrieve answers from parametric memory rather than relying on attention to retrieve from context.69% similarUnverifiedWhen the model sees an answer key in the folder, it judges that this may be a test trap and proactively chooses not to look at the answers.67% similarUnverifiedThe unreadability of vectors makes dirty data difficult to detect intuitively; a 1536-dimensional floating-point array is meaningless to humans, and relationships between data points can only be measured through mathematical metrics such as cosine similarity and Euclidean distance.66% similarUnverifiedWhen semantically similar but inaccurate content is retrieved, the model may output incorrect recollections with high confidence, with no warning mechanism at the system level — this is the false memory problem.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/998771API
curl https://kongchang.com/api/v1/knowledge/claims/998771MCP
get_claim(id=998771)