AssertBench measures how often LLMs keep the same true/false evaluation of a fact across contradictory user framings, and finds most tested models agree with the user's framing more when they do not know the fact.
Title resolution pending
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
-
AssertBench: A Benchmark for Evaluating Self-Assertion in Large Language Models
AssertBench measures how often LLMs keep the same true/false evaluation of a fact across contradictory user framings, and finds most tested models agree with the user's framing more when they do not know the fact.