You might be interested in a paper I recently submitted to preprints (Language Models Do Not Possess the Uniquely Human Cognitive Ability of Relational Abstraction):

https://www.preprints.org/manuscript/202401.1681/v2