LLMs don't have these structures, so their intelligence is different. Even if you trained an LLM with a trillion params, it won't reach our intelligence unless it matches our structures abilities. (And the more params you add, the more trade offs you get, different conversation).
Now, once you have these structures, you have AGI. (Just like if you have a brain, you can perform all human cognitive tasks). However, how you connect these structures will be the difference between an Einstein AI and a Picasso AI. It doesn't mean some AIs will be less intelligent or more (as long as they have the same structures like each brain does (except for neurodegenerative diseases)), but rather will optimize the usage of those structures already there, just like our human brains.
If there are more structures you can add that makes intelligence and consciousness enhanced, then you arrive at superintelligence.
Otherwise, thinking of g as making human brains or AI more or less intelligent will not lead you to solving education for humans nor finding AGI.