The funny thing about LLM is, you can build one from scratch and yet you still won't understand how it works. You would understand what kind of matrix multiplications the neural network performs (in fact that's not that hard. An OS is orders of magnitudes more complex), but you would still have no idea why it does what it does.