Reasoning test expamples (IMAGE) PNAS Nexus Caption Manipulating content within fixed logical structures. In each of the author’s three datasets, they instantiate different versions of the logical problems. Different versions of a problem offer the same logical structures and tasks but instantiated with different entities or relationships between those entities. The relationships in a task may either be consistent with, or violate real-world semantic relationships, or may be nonsense, without semantic content. In general, humans and models reason more accurately about belief-consistent or realistic situations or rules than belief-violating or arbitrary ones. Credit Lampinen et al Usage Restrictions None License Original content Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of news releases posted to EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system.