Does GPT-4 really understand what we’re saying?

David Krakauer is the president of Sante Fe Institute, a complex science institute that explores the evolution on intelligence and stupidity.

GPT-4 understands what we say?

Answer: Yes and no. We surveyed AI researchers in my new paper, co-authored with Melanie Mitchell, on the notion that large pretrained language models like GPT-4 can understand language. It’s unclear whether we agree on the definition of understanding when they claim that these models can understand us or not. Shannon made it clear when he invented information theory that he was only interested in communication and not meaning. You can have two messages equally informative with one message having a lot of meaning, but the other one has none.

Source:
https://nautil.us/does-gpt-4-really-understand-what-were-saying-291034/