Unverified50% confidenceFactExact time
For attacks that rely on gradient estimation (such as finite-difference-type methods), GCD's confusion direction causes the gradient estimate obtained from each query to have a systematic bias, and the attacker needs more queries to converge to a usable perturbation direction.
1
Sources
50%
Confidence
Long-term
Relevance
10/14/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedGCD's confusion effect is amplified as the proportion of defended queries rises.79% similarUnverifiedWhen the guidance signal deviates from the model's own sampling distribution, the policy gradient estimate incurs bias.71% similarUnverifiedOnce a benchmark is made public and repeatedly attacked, targeted overfitting may occur—that is, learning the distribution patterns of the problems without necessarily corresponding to underlying generalized intelligence.70% similarUnverifiedWriting to more positions in parallel speeds up inference but also increases the risk of deviating from the true distribution, constituting a trade-off between speed and fidelity.69% similarUnverifiedBefore multi-path retrieval results enter reranking, they typically need to undergo deduplication and score normalization; otherwise, inconsistent score scales across different retrieval paths will introduce bias.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1023780API
curl https://kongchang.com/api/v1/knowledge/claims/1023780MCP
get_claim(id=1023780)