chameleon-datasets-popqa-strategyqa

Summary

The knowledge conflict tests draw on two question sets with different shapes: one checks whether the model can correctly recall facts about specific entities (people, places, things) at scale, and the other requires chaining multiple facts together to reach a yes-or-no conclusion. Together they cover both simple factual lookup and multi-step reasoning, so the results show where the model's knowledge conflicts surface under different cognitive demands, not just one narrow skill.