Matches in SemOpenAlex for { <https://semopenalex.org/work/W4385571289> ?p ?o ?g. }
Showing items 1 to 69 of
69
with 100 items per page.
- W4385571289 abstract "Recent model editing techniques promise to mitigate the problem of memorizing false or outdated associations during LLM training. However, we show that these techniques can introduce large unwanted side effects which are not detected by existing specificity benchmarks. We extend the existing CounterFact benchmark to include a dynamic component and dub our benchmark CounterFact+. Additionally, we extend the metrics used for measuring specificity by a principled KL divergence-based metric. We use this improved benchmark to evaluate recent model editing techniques and find that they suffer from low specificity. Our findings highlight the need for improved specificity benchmarks that identify and prevent unwanted side effects." @default.
- W4385571289 created "2023-08-05" @default.
- W4385571289 creator A5024922900 @default.
- W4385571289 creator A5030546839 @default.
- W4385571289 creator A5041251477 @default.
- W4385571289 creator A5050436152 @default.
- W4385571289 creator A5076400003 @default.
- W4385571289 date "2023-01-01" @default.
- W4385571289 modified "2023-09-24" @default.
- W4385571289 title "Detecting Edit Failures In Large Language Models: An Improved Specificity Benchmark" @default.
- W4385571289 doi "https://doi.org/10.18653/v1/2023.findings-acl.733" @default.
- W4385571289 hasPublicationYear "2023" @default.
- W4385571289 type Work @default.
- W4385571289 citedByCount "0" @default.
- W4385571289 crossrefType "proceedings-article" @default.
- W4385571289 hasAuthorship W4385571289A5024922900 @default.
- W4385571289 hasAuthorship W4385571289A5030546839 @default.
- W4385571289 hasAuthorship W4385571289A5041251477 @default.
- W4385571289 hasAuthorship W4385571289A5050436152 @default.
- W4385571289 hasAuthorship W4385571289A5076400003 @default.
- W4385571289 hasBestOaLocation W43855712891 @default.
- W4385571289 hasConcept C119857082 @default.
- W4385571289 hasConcept C121332964 @default.
- W4385571289 hasConcept C124101348 @default.
- W4385571289 hasConcept C127413603 @default.
- W4385571289 hasConcept C13280743 @default.
- W4385571289 hasConcept C138885662 @default.
- W4385571289 hasConcept C154945302 @default.
- W4385571289 hasConcept C168167062 @default.
- W4385571289 hasConcept C176217482 @default.
- W4385571289 hasConcept C185798385 @default.
- W4385571289 hasConcept C205649164 @default.
- W4385571289 hasConcept C207390915 @default.
- W4385571289 hasConcept C21547014 @default.
- W4385571289 hasConcept C41008148 @default.
- W4385571289 hasConcept C41895202 @default.
- W4385571289 hasConcept C97355855 @default.
- W4385571289 hasConceptScore W4385571289C119857082 @default.
- W4385571289 hasConceptScore W4385571289C121332964 @default.
- W4385571289 hasConceptScore W4385571289C124101348 @default.
- W4385571289 hasConceptScore W4385571289C127413603 @default.
- W4385571289 hasConceptScore W4385571289C13280743 @default.
- W4385571289 hasConceptScore W4385571289C138885662 @default.
- W4385571289 hasConceptScore W4385571289C154945302 @default.
- W4385571289 hasConceptScore W4385571289C168167062 @default.
- W4385571289 hasConceptScore W4385571289C176217482 @default.
- W4385571289 hasConceptScore W4385571289C185798385 @default.
- W4385571289 hasConceptScore W4385571289C205649164 @default.
- W4385571289 hasConceptScore W4385571289C207390915 @default.
- W4385571289 hasConceptScore W4385571289C21547014 @default.
- W4385571289 hasConceptScore W4385571289C41008148 @default.
- W4385571289 hasConceptScore W4385571289C41895202 @default.
- W4385571289 hasConceptScore W4385571289C97355855 @default.
- W4385571289 hasLocation W43855712891 @default.
- W4385571289 hasOpenAccess W4385571289 @default.
- W4385571289 hasPrimaryLocation W43855712891 @default.
- W4385571289 hasRelatedWork W112744582 @default.
- W4385571289 hasRelatedWork W1485630101 @default.
- W4385571289 hasRelatedWork W2118451750 @default.
- W4385571289 hasRelatedWork W2379533788 @default.
- W4385571289 hasRelatedWork W2498017833 @default.
- W4385571289 hasRelatedWork W2961085424 @default.
- W4385571289 hasRelatedWork W2966858528 @default.
- W4385571289 hasRelatedWork W3192902148 @default.
- W4385571289 hasRelatedWork W4306674287 @default.
- W4385571289 hasRelatedWork W4224009465 @default.
- W4385571289 isParatext "false" @default.
- W4385571289 isRetracted "false" @default.
- W4385571289 workType "article" @default.