Neural networks are like brains, duh!

This was an awkward realization, because the clue was right there in the name all along.

I struggled to understand how LLMs store information in them but not in a way traditional databases do by storing the actual data.

Traditional databases are like a library with a really good librarian that knows where any book in it is kept and can give them to you, but the librarian wouldn't answer any questions about the books contents or even read the books.

LLMs on the other hand are like a brain. A really fast reader that read every single book in the library. With a good memory as well, they can quote some of the books, summarize their chapters and develop ideas based on those books. But the actual books aren't stored in the brain.

And they are also like someone from Buenos Aires, Argentina. If you ever met a porteño you know they will never admit they don't know something. Instead they will make up stuff whenever they don't remember it or never read about.

Where's porteños do it because of pride, LLMs do it because they don't know that they don't know. There's no distinction between quoting a book they actually read making up a quote from a book that doesn't exist, it's feels like the same thing to them.