A metaphor to compare LLMs with AGI

The difference between large language models and future artificial general intelligence is like the difference between a square and a cube.
A cube is infinitely more complex than a square: You could say a cube is made up of infintely many squares, stacked on top of each other along a higher dimension.
But a cube only needs 8 points with 3 coordinates each to define it, versus 4 points with 2 coordinates each for a square.
So a cube is really only three times as complex as a square.

A true AI, capable of thought and feeling, will also have to be several times larger and more complex than even the most sophisticated LLM. But even if it's a thousand times larger, that won't really matter. Just a multiplication of required resources.
The hard part will be the exponential difficulty in training it.