Can you beat a machine that dont search anything!!

This Neural Network has been handicapped with no search, ( only 1 node).

Previously, people think that machines outplay human because they can search million of positions per second search . However, this Leela can beat 99% of human without searching anything. 

Note: Neural Network of this Leela is 4x bigger  than Alpha Zero.

My games against her so far in 6+3 TC.

I am 2000+ blitz in here and 2200+ blitz in that site.

I understand machine learning can make many of its moves brilliant no calculation necessary, but I don't understand how it wont blunder tactics if it does no calculation.

Of course she blunder tactics. In game 1 , both players blundered .

I blundered a piece on move 16. dxc5?? (If leela played 16. ...... Na4, I lose a piece)

The worse case with 1 node is that she doesnt know the history and doesnt know the perpetual. Here is a game vs 1100 player.

How does she knows tactics? She play with her knowledge, which is a kind of pattern recognition associated with statistics. During her training the she was allowed to search 800 positions or 1600 positions per move.  She decide 1 node with the highest chance of winning . e.g If she saw king opposition in endgame has the highest chance of winning, she will move king opposition. In opening and middle game, she learn patterns with winning chance. e.g She is allowed to play all possible 20 first moves, and she learns which move has the highest chance of winning. (eg. 1. e4 or d4 or a3  in first move, And then After 1.e4 e5,  2. Nf3 or 2. d4 and so on)

In the terms of a standard engine, a search and calculation is essentially the same thing.

In the example post, Leela isn't calculating or searching, just looking at the possible moves at one-ply and comparing those positions to it's learned positions, chosing the position that has the highest winning percentage from it's network of positions.

Yes, search= calculate

You can handicap Stockfish without search, which is like depth 1.    Leela 1 node approx 2200 elo in blitz.  You can play depth 1 stockfish in many GUI, e.g Fritz, Arena , and it is approximately  1000 elo in blitz TC.

If you wanna see the intrinsic evaluation value of stockfish, you can paste any FEN into this website.

https://hxim.github.io/Stockfish-Evaluation-Guide/

For example, in this position, Stockfish evaluate as  + 1.31 , without search. However, actual working Stockfish will search depth 20 , 30 etc and it will show the evaluation of the end position of the principal variation .  Athough Stockfish think white is much better without search, after searching depth 30, 40 etc, she will realize that white mobility and space benefit is very minimal like  +0.15 etc.

1st time, I won her in 6+3 TC.

https://lichess.org/Zl0vWTaK/white

That doesn't compute. (muhahaha!)