Unverified50% confidenceBenchmarkExact time
The approach using frozen visual features yielded a ROC-AUC of only 0.482, meaning the model essentially learned nothing.
1
Sources
50%
Confidence
Long-term
Relevance
10/9/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedRoc作为0.x版本仍不建议用于生产环境62% similarUnverifiedPR-AUC 相比 ROC-AUC 不会被大量负样本的正确分类所稀释，更适合稀有事件检测场景62% similarUnverifiedThe model in the testing sandbox was not equipped with all the safeguards available to deployed models, such as classifiers capable of flagging suspicious behavior.56% similarUnverifiedThe original report only cited the statement of a single executive and did not disclose the specific code name, development stage, or specific manifestations of the instruction non-compliance of the abandoned model52% similarUnverifiedAUX's public endpoints use only security test data, do not perform real external validation, and have no production-environment SLA guarantees.52% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/990930API
curl https://kongchang.com/api/v1/knowledge/claims/990930MCP
get_claim(id=990930)