Prior conversational context, especially from a different knowledge domain, can sharply reduce LLM multiple-choice accuracy, and repeating the task near the query mitigates the drop.
Beyond the chat: Executable and verifiable text-editing with llms
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
other 1
citation-polarity summary
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1roles
other 1polarities
unclear 1representative citing papers
citing papers explorer
-
Evaluating the Sensitivity of LLMs to Prior Context
Prior conversational context, especially from a different knowledge domain, can sharply reduce LLM multiple-choice accuracy, and repeating the task near the query mitigates the drop.