DO LANGUAGE MODELS REALLY UNDERSTAND THE WORLD?

Summary from the TLDR AI newsletter

Researchers looked into how well pretrained language models understand the basic building blocks and rules of the world, like 'a dog is an animal' or 'an apple is a fruit.' They found that these programs are decent at remembering some of this foundational knowledge, but aren't perfect at reasoning with it.