> If you remove all of the seeded RNG during inference (meaning; no random seeds, no temps, just weights/tokenizer), you can actually create an equation that deterministically gives you the same string of text every time.

That answers the "what", but not the "why" nor the "how exactly", with the latter being crucial to any claim that we understand how these things actually work.

If we actually did understand that, we wouldn't need to throw terabytes of data on them to train them - we'd just derive that very equation directly. Or, at the very least, we would know how to do so in principle. But we don't.