Stockfish 14 (stockfishchess.org)

Article summary

Stockfish 14, a new version of the open-source chess engine, has been released, featuring significant improvements in evaluation and search capabilities. The engine's neural network has been updated with a new architecture and trained on a large dataset, including positions evaluated by Leela Chess Zero. This has resulted in a substantial increase in playing strength, with Stockfish 14 being at least 400 Elo ahead of Stockfish 7. The new version also includes some search refinements, minor bug fixes, and additional improvements.

Main themes

• Stockfish 14 release
• chess engine development
• neural networks
• open-source collaboration
• deep learning
• chess GUIs

What commenters say

• The release of Stockfish 14 is seen as a challenge to Alpha Zero, with some commentators suggesting a rematch between the two engines.
• The use of a large dataset to train the neural network is a key factor in Stockfish 14's improved performance, but the size of the dataset does not necessarily imply a large neural network.
• The difference in architecture and search algorithms between Stockfish and Leela Chess Zero is a significant factor in their relative strengths and weaknesses.
• Some commentators argue that the distinction between traditional chess engine methods and deep learning is becoming increasingly blurred, while others see Stockfish 14 as a demonstration of the power of open-source and open-data approaches.
• The command-line interface of Stockfish is designed for interoperability with chess GUIs, but can be used manually with some effort.
• The playing strength of Stockfish 14 is not just a matter of Elo points, but also of its ability to find complicated ideas and build long-term advantages.
• The collaboration between the Stockfish and Leela Chess Zero teams is seen as a positive development, demonstrating the benefits of open-source and open-data collaboration in the field of chess engines.