Trilemma Dataset
The Trilemma of Truth in Large Language Models
홈페이지 · 논문 1편
The Trilemma of Truth is a multiclass probing dataset for evaluating the veracity-tracking mechanism of large language models.
It comprises three configurations— city_locations (statements about where cities are), med_indications (drug–condition assignments), and word_definitions - each annotated as true, false, or neither.
Designed to plug straight into sAwMIL and various veracity probes.