Can large language models generalize analogy solving like children can? [pdf]

2 pointsposted 8 hours ago
by Gys

1 Comments

Gys

8 hours ago

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.