Articles
'Neural Nakamura' Analyzes Top Computer Chess Games
Hikaru Nakamura. | Photo: Maria Emelianova / Chess.com.

'Neural Nakamura' Analyzes Top Computer Chess Games

Pete
| 22 | Amazing Games

Hikaru Nakamura, fresh off his 2019 U.S. chess championship, is back streaming chess on his popular channel.

Today, Nakamura reviews the best games played by neural-network chess engines, helping you understand why these mysterious machine-learning engines like Lc0 and AlphaZero chose the winning moves in some unbelievable games.

Watch Hikaru Nakamura on Twitch.tv/GMHikaru, and follow some of the games yourself with our list below. 

The top 10 neural-network computer chess games:

1. The match that started it all.

We start with an incredible sample game from a "shock announcement" by Google in December 2017 that made headlines around the world as AlphaZero defeated Stockfish in a private match. 

This is the first look most chess fans ever had at the human-like play of a neural-network engine.


2. AlphaZero's return, destroying Stockfish 8.
From the new set of games released December 2018—Alphazero proves that its earlier sample games were no fluke.


3. Lc0 debuts at Computer Chess Championship.
The open-source machine-learning project crushes Crafty with an early rook sacrifice. Lc0 wows fans with intuitive and creative play. 



4. Lc0 "trolls" Nirvana with underpromotion in brilliant attack
.

In what became something of a trademark for the lifelike engine, Lc0 loves to underpromote.

Analysis by IM Daniel Rensch. 



5. Lc0 shocks Komodo with Nxh7.

Lc0's shocking 14. Nxh7! was not foreseen by Komodo (or, on analysis, by Stockfish) and showcases the creative, flexible, and intuitive style that has generated so much fan support for the neural-network engine.



6. Lc0 uses the "fawn pawn."

Lc0 continued to show its ingenuity and strength by quickly destroying Nemorino from the black side of the board. Lc0 checkmated its opponent in just 45 moves after advancing a pawn all the way to pester the enemy king, in a style similar to AlphaZero that may be a hallmark of neural-network engines.


7. Lc0 and the endgame beyond human comprehension.

This endgame from Lc0 might be beyond the comprehension of both humans and traditional chess engines. The machine-learning engine Lc0 squeezes a win out of the depths of its neural network. 


8. Antifish stuns Stockfish's evaluation.

Antifish, an Lc0 variant neural network specifically designed to beat Stockfish, wins a game where the traditional engine never saw what went wrong. Before neural network engines came along, Stockfish losing a self-proclaimed advantage and losing the game simply never happened.


9. Lc0 "adopts" Lc0-Dev.

Lc0 showed its younger version who's boss by winning a ton of material in this crushing victory.


10. Lc0 makes history, wins first Computer Chess Championship. 

Lc0 does heavy damage to Stockfish on the h-file in a game that could be in a textbook for humans on how to defeat the Pirc Defense. The endgame shows the substantial power of a queen and three connected passed pawns as the mighty fish could only delay its own demise.

More from Pete
Can You Solve Our 2024 Holiday Puzzler?

Can You Solve Our 2024 Holiday Puzzler?

Can You Solve Our 2023 Holiday Puzzler?

Can You Solve Our 2023 Holiday Puzzler?