Unverified50% confidenceOpinionExact time
If a demo has no mislog, then it has not measured.
1
Sources
50%
Confidence
Long-term
Relevance
10/14/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedMultiplying the unlearning loss of the forget set by 0.0 causes the forget items to produce no gradients, meaning the model performs no real unlearning updates.65% similarUnverifiedGhostInk benchmark measures not recognition accuracy, but whether a model admits what it cannot read, rather than filling the gap with a plausible-sounding answer.63% similarUnverifiedZero retention does not equal zero learning—the fact that data is not stored does not mean that patterns and information extracted from it will not influence the model in some form63% similarUnverified线性探测器可能无法捕捉非线性编码的信息，因此内部探测的'未检测到'可能是探测方法本身的局限而非表征真的缺失61% similarUnverifiedWhen a certain delta happens to be empty tool-call data, if the parsing logic handles it improperly, it may incorrectly judge that the entire reasoning stream has ended and thus lose subsequent content.60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1024727API
curl https://kongchang.com/api/v1/knowledge/claims/1024727MCP
get_claim(id=1024727)