webhat on Nostr: when I open a zip file in a text editor, one can easily say the actual content of ...

when I open a zip file in a text editor, one can easily say the actual content of whatever was compressed isn't there. It doesn't mean there are no files in the zip file, it just means that I need to use the correct tool to turn the zipped representation of the contents into the contents

It's true that the big array of floatingpoints from LLMs can't easily be converted into an exact representative, rather it produces, at best, a poor facsimile of the data on which it was trained. Unlike with the zip example, LLM training is good at compressing the data, and poor at deflating it