An 11M-parameter model writes a story in this tab, badly. A 100M-parameter model reads it back in my voice. One of them fits in a GIF; the other does not.
The writer is 2.31 MB and runs here. The reader is 131 MB and does not, because asking you to download 131 MB to hear a story about a key would be rude. It costs fifty-seven times more to say the words than to think of them, which is the whole point of this page. the model and my voice.