Under controlled conflicting specifications, LLMs systematically prefer formal and naturalized-formal task descriptions over pure natural language and input–output examples, with domain-specific exceptions.
Whose Facts Win? LLM Source Preferences under Knowledge Conflicts
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
When Specifications Conflict: A Symmetry-Based Framework for Measuring LLM Preferences
Under controlled conflicting specifications, LLMs systematically prefer formal and naturalized-formal task descriptions over pure natural language and input–output examples, with domain-specific exceptions.