Unverified50% confidenceConceptExact time
In deletion fidelity evaluation for GNNs, "deleting an edge" has two semantically different interventions: structural deletion (removing the edge from the graph structure and recomputing structure-dependent quantities such as degree normalization and attention coefficients) and message suppression (keeping the structural context but blocking the message passed along that edge).
1
Sources
50%
Confidence
Long-term
Relevance
11/4/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn interface design, it is recommended to make deactivation the default removal action, place deletion at a deeper confirmation level, and clearly warn that it is irreversible.71% similarUnverifiedCanary-style deletion means binding the deletion of suspicious logic to behavior observation, first verifying it on low-traffic, non-critical paths, and then gradually expanding.70% similarUnverifiedA second solution is to treat a node's done state as a signal that actively reduces its context contribution, triggering an actual context pruning action63% similarUnverifiedIf users frequently need to manually delete part of a prompt's returned result, that is direct evidence of task coupling, and the prompt should be split.62% similarUnverifiedThe Abliteration technique identifies and suppresses the direction vector responsible for refusal behavior in the model's residual stream, using orthogonal projection to remove that direction from the weight matrices, thereby eliminating refusal behavior without retraining.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1134405API
curl https://kongchang.com/api/v1/knowledge/claims/1134405MCP
get_claim(id=1134405)