Unverified50% confidenceFactExact time
Evaluating knowledge editing typically focuses on three dimensions: edit success rate, generality and locality; sequential editing scenarios additionally examine portability and long-term stability.
1
Sources
50%
Confidence
Long-term
Relevance
10/20/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe authors split state reconciliation evaluation into three core dimensions: target update accuracy (whether the model made changes justified by the conversation), retention accuracy (whether it left unrelated fields untouched), and incidental mutation (how often it changed things that should have remained unchanged).75% similarUnverifiedEditing effects are mutually dependent: the impact of modifying one token depends on which other tokens are edited simultaneously, so editing sets must be considered rather than evaluating individual positions in isolation.75% similarUnverifiedThe research proposes dynamically routing different editing strategies based on a task's locality, rather than making a fixed choice between diff or full-file rewrite.69% similarUnverifiedDiffusion models exhibit quality degradation during multi-round iterative editing, and the fundamental solution typically involves explicit tracking of editing history or switching to a non-destructive editing strategy.69% similarUnverifiedA single set of corrections based on Fisher-Rao geometry can simultaneously improve multiple tasks including guided control, knowledge editing, feature attribution, dictionary learning, and fine-tuning.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1073998API
curl https://kongchang.com/api/v1/knowledge/claims/1073998MCP
get_claim(id=1073998)