Unverified50% confidenceCautionExact time
Uncensored models may generate inaccurate, harmful, or illegal content, and community fine-tuned versions are often less stable than the original models.
1
Sources
50%
Confidence
Long-term
Relevance
8/31/2026
First Seen
Sources
Related Entities
Related Claims
Unverified下游用户对模型的二次微调可能意外破坏原有安全特性69% similarUnverifiedInference methods based on model internal features face a trade-off between false positives and false negatives68% similarUnverifiedUser-perceived model "deterioration" is often the cumulative result of multiple factors including silent updates, tightened safety alignment, dynamic downgrading of inference resources, and user expectation drift, and does not always indicate a genuine regression in model capability.67% similarUnverified基准过拟合和数据污染会导致模型通过记忆而非真正理解获得高分，使公开基准分数可信度存疑67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/828634API
curl https://kongchang.com/api/v1/knowledge/claims/828634MCP
get_claim(id=828634)