Can large language models generalize analogy solving like children can? [pdf]arxiv.org 2 pointsGys8 months ago1 commentSaveHideCopy link On HNComments−GysOP8moFrom the abstract:Children and adults easily generalized their knowledge to unfamiliar domains, whereas LLMs did not. This key difference between human and AI performance is evidence that these LLMs still struggle with robust human-like analogical transfer.
Comments
From the abstract:
Children and adults easily generalized their knowledge to unfamiliar domains, whereas LLMs did not. This key difference between human and AI performance is evidence that these LLMs still struggle with robust human-like analogical transfer.