Leela Chess Zero vs Stockfish 9

Hi. Because I'm not very good at chess yet, I have installed Arena to practice on. Which hopefully will pay off soon. :) I came across Leela Chess Zero today. So I added its engine and have it play against Stockfish. So far Stockfish wins every time. I know that the purpose of LCS is to learn to play better and better. Which is utterly fascinating to me. And I'd like to know exactly how that works. How exactly does it learn? I have a general understanding, but I really want to learn about the details. Which leads me to two questions to those of you who are knowledgeable on this: 1. How does it learn specifically? 2. What can be predicted about the end result? Can it become the strongest engine or is that not possible yet with current technology? Thanks in advance.

 1. https://medium.com/applied-data-science/alphago-zero-explained-in-one-diagram-365f5abf67e0

2. Neural network  has very limited search capacity, current best NN A0 would be about 300 elo weaker than SF on your common desktop. A0 won SF because of x800 times hardware advantage.

https://www.chess.com/forum/view/general/leela-zero-a-neural-network-engine-similar-to-alpha-zero?page=6

https://www.chess.com/forum/view/general/leela-zero-stockfish-without-human-knowledge-like-alpha-0?page=3

This shows A0 practically stalled at 3300 rating,

Thanks. So far LCZ keeps losing from SF 9, but I'm curious to see if that changes in the future.

Pretty unlikely to win Stockfish on normal desktop PC within 1-2 years.  ( 2000 rated leela vs 3500 rated stockfish.)

SF has been tuned manually for 700 million games or 1024 CPU years. Leela gets 2 million games only. ( dont know CPU year for leela, may be 50 CPU year , I guess)

are you using CPU or GPU?