A new evaluation framework shows that even the best tested LLM only reliably adjusts response complexity in the intended direction 46% of the time across 98 scientific queries.
Blinded by Generated Contexts: How Language Models Merge Generated and Retrieved Contexts When Knowledge Conflicts?
3 Pith papers cite this work. Polarity classification is still indexing.
3
Pith papers citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CL 3verdicts
UNVERDICTED 3roles
background 1polarities
background 1representative citing papers
QREAM rewrites documents to question-focused style using iterative ICL and distilled FT models, boosting RAG performance by up to 8% relative improvement.
A survey on LLM-as-a-Judge that reviews reliability strategies, proposes evaluation methods, and introduces a novel benchmark for assessing such systems.
citing papers explorer
-
Explain Like I'm 5 or Whatever I Choose: Evaluating the Interactive Potential of Language Model Responses
A new evaluation framework shows that even the best tested LLM only reliably adjusts response complexity in the intended direction 46% of the time across 98 scientific queries.
-
Align Documents to Questions: Question-Oriented Document Rewriting for Retrieval-Augmented Generation
QREAM rewrites documents to question-focused style using iterative ICL and distilled FT models, boosting RAG performance by up to 8% relative improvement.
-
A Survey on LLM-as-a-Judge
A survey on LLM-as-a-Judge that reviews reliability strategies, proposes evaluation methods, and introduces a novel benchmark for assessing such systems.