We know that some humans can use information to solve harder problems, while others can not do it. We say the first kind built understanding about the informati

We know that some humans can use information to solve harder problems, while others can not do it. We say the first kind built understanding about the information. You can easily test for this by just throwing hard problems at them that they haven't seen before, Google does that in their interviews for example (but many who do similar just takes problems people definitely have seen). So doing similar kinds of testing on ChatGPT we see that it doesn't understand the information. It can repeat a lot of stuff more or less verbatim, but it cannot apply most of the knowledge it can repeat, so it doesn't understand any of it. The only thing ChatGPT really understands are relationships between words that has been repeated in texts it has seen, that is some level of understanding but that has no relationship to what ChatGPT says about the words. When you ask ChatGPT to describe something it repeats a description it has seen of the word, when you ask it to do something with a word then it fully ignores that description and instead look at the relationships it has seen the word have to other words. So its understanding and its knowledge are completely separate things, so you can say that it doesn't understand anything it knows.