Leelenstein, a neural network chess engine built on Lc0, the top machine-learning chess project, lost to Stockfish in CCC 9.
In the CCC 9 blitz time control, the champion engine won over 17 other engines in a “gauntlet” format.
In CCC 10: Double Digits, an 18-engine competition held over four rounds, Stockfish will now attempt to keep its championship. CCC 10 is now active.
On the edge between blitz and rapid, the first three rounds of CCC 10 will be played at a time control of 10 minutes plus a three-second increment.
The 400 games in the two-engine CCC 10 finals will be divided into three blitz and quick time limits to decide the winner.
The neural network engines, under the direction of Lc0, will attempt to defeat the venerable Stockfish. In the first two phases, Lc0 and Leelenstein will compete, and other, less-tested engines will attempt to qualify. Without participating in stage one, the Allie neural network chess engine will enter the competition in stage two.
Thanks to a recent patch, Stockfish’s most recent version, which was released on August 5, boasts a significant rating increase. According to CCC organizers, the fish will surpass its previous self by an amazing 13 Elo points.
CCC planners state that Lc0 will also be competing at full strength in the event, operating net 42850, which should be the strongest network created in the most recent training run.
The secret Stoofvlees engine, an experimental effort combining many state-of-the-art artificial intelligence algorithms, is making a comeback to the Computer Chess Championship. The neural network that powers the Stoofvlees engine is trained using an oracle of grandmaster games combined with feature recognizers. The new chess machine followed up by incorporating the results into the Deep Sjeng chess engine. Written by the former Leela Go author “GCP,” Stoofvlees made it to the CCC 9 quarterfinals.
List of Engines:
- 1. Stockfish
- 2. Leelenstein
- 3. Lc0
- 4. Ethereal
- 5. Laser
- 6. Xiphos
- 7. Andscacs
- 8. Dark Queen Lc0
- 9. Rofchade
- 10. Rubichess
- 11. Stoofvlees
- 12. Winter NN
- 13. Lc0-C
Stage 1 (qualifiers) of CCC 10:
Three stages of round-robin
13 engines
468 games
Control time: 10+3.
First chapter: sure
Decision-making: 20-ply dead draw rule plus bullet mode
Stage 2 of CCC 10 (quarterfinals):
Four rounds of round-robin
12 engines
528 games
Control time: 10+3.
First chapter: sure
Decision-making: 20-ply dead draw rule plus bullet mode
Engines List:
- Stockfish (CCC 9 Automatic)
2. Automated Leelenstein
3. Lc0 (Automatic)
4. Automatic Allie
5. Komodo (Machine)
6. Automated Houdini
7. Automatic Komodo MC
8. Automatic Fire
- 9–12. The top four qualifiers from the earlier round
Stage 3 of CCC 10 (semifinals):
format: 12-times round-robin
engines: 6
Games: 360
Control time: 10+3.
First chapter: sure
Decision-making: 20-ply dead draw rule plus bullet mode
Engine list:
- 1–6. Previous stage’s top six finishers
Stage 4 (finals) of CCC 10:
Type of game: 200 at 3+2, 120 games at 10+3, 80 games at 25+5.
Engines: 2
400 games
Time limits: 3 + 2, 10 + 5, 25 + 5.
First chapter: sure
Decision-making: 20-ply dead draw rule plus bullet mode
Engine list:
- 1-2. The top two finishers from the previous stage
CCC 9: Finals Crosstable
This is a game that ought to be included in chess textbooks on how to handle gambits. Stockfish checkmate in an incredible 26 moves after taking Leelenstein’s early sacrifice on f7 and launching a fierce counterattack on the other side of the board. Both brutal and lovely.
Assigning the huge -2.5 evaluation, Stockfish believed it was winning the next game. That is until Leelenstein uncorked the neural-network-themed (and completely crazy) sacrifice 33. Bxh6 to destroy the conventional engine, a move Stockfish was completely unprepared for.
This game shows off the superior neural network engines’ deep, subtle, and inevitably enigmatic knowledge of the game, where we can’t know what these machines are thinking. We just know that it performs.
Express Your Viewpoint
In the Computer Chess Championship, can Stockfish hold onto its crown against an increasing number of neural network adversaries? Comment with your thoughts below, and don’t forget to cast your vote on the chess game.