New temporal benchmarks show LLMs struggle with outdated facts, and a structured knowledge-organization memory improves accuracy over ICL and RAG.
Language models are few-shot learners
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Question Answering under Temporal Conflict: Evaluating and Organizing Evolving Knowledge with LLMs
New temporal benchmarks show LLMs struggle with outdated facts, and a structured knowledge-organization memory improves accuracy over ICL and RAG.