[KongchangAI]
Unverified50% confidenceReleaseExact time

The researchers proposed a single-stage, attack-free proxy metric for evaluating the robustness of unlearning that can capture the parametric knowledge level inaccessible to output-level metrics.

1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/4/2026
First Seen
Valid until: 2/2/2027

Sources

Related Entities

Related Claims

Cite This Claim

Stable URI
https://kongchang.com/claim/1134672
API
curl https://kongchang.com/api/v1/knowledge/claims/1134672
MCP
get_claim(id=1134672)