MMKC-Bench provides a human-verified benchmark of multimodal knowledge conflicts and shows that current LMMs prefer internal parametric knowledge over external evidence.
A survey on rag meeting llms: Towards retrieval-augmented large language models
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
unclear 1representative citing papers
citing papers explorer
-
Benchmarking Multimodal Knowledge Conflict for Large Multimodal Models
MMKC-Bench provides a human-verified benchmark of multimodal knowledge conflicts and shows that current LMMs prefer internal parametric knowledge over external evidence.