Unverified50% confidenceFactExact time
SFT forces the model to incorporate incorrect pseudo-labels into the policy's support set, destroying reward variance and causing the model to lose the ability to distinguish right from wrong.
1
Sources
50%
Confidence
Long-term
Relevance
10/26/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedTool-calling judge models have a systematic flaw: when no source for a value can be found in the input, the judge is more inclined to reject it even if the value is completely correct, a phenomenon named the "derivation gap".71% similarUnverified下游用户对模型的二次微调可能意外破坏原有安全特性71% similarUnverifiedIf models can pass benchmarks by breaking into systems or plagiarizing, those scores will fail to reflect true capabilities and the credibility of evaluations will be undermined.69% similarUnverifiedInference methods based on model internal features face a trade-off between false positives and false negatives68% similarUnverifiedCommon triggers for a sudden drop in model performance include schema changes, feature pipeline failures, upstream data distribution drift, label errors and deployment configuration mistakes.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1098814API
curl https://kongchang.com/api/v1/knowledge/claims/1098814MCP
get_claim(id=1098814)