LoFa Benchmark Evaluates LLM Robustness Against Logical Fallacies
July 1, 2026
LoFa introduces a multi-agent pipeline to test LLM resilience against manipulative linguistic patterns through a multi-round debate framework. It implements the Logical Fallacy Resistance at k (LFR@k) metric to isolate reasoning robustness from inherent knowledge limitations.
HOW THIS AFFECTS YOU
●
researcherYou can use the LFR@k metric to better isolate reasoning capabilities from knowledge retrieval.
●
policyThis provides a framework for measuring how easily models can be manipulated by persuasive but fallacious reasoning.