• Daniyar Kylyzhov

Fat Fritz 2, a commercial computer chess system, was convicted of plagiarizing an open-source engine


A few days ago, chess developer ChessBase unveiled "the new number 1 chess engine" Fat Fritz 2. According to the description on the site, it is "on a new neural network trained by Albert Silver with Fat Fritz."



It turned out that Fat Fritz 2 is, in fact, a search for Stockfish on a new neural network and with minimal changes in the code.


In July 2018, Silver secretly used a free open source Leela Chess Zero chess engine with a custom neural network that he trained to compete in a TCEC computer chess tournament under the name DeusX.


The network was trained using scripts from the Leela project, and the network architecture followed the original engine. There is nothing difficult in training such a network; Leela project members do this regularly.


Silver later grossly downplayed the Leela project in an interview, claiming that he did in a few months what other engine writers had taken decades to do. Moreover, Silver called himself the author of the new engine.


The following year, an updated version of DeusX, called Fat Fritz, was released as part of the € 79.90 ChessBase Fritz package. Again, the Leela engine was used with no major functional changes, apart from some default parameter values.


Fat Fritz was promoted as an innovative engine. The description of the product began with the words: "This is a semi-secret development, a clone of AlphaZero, created in the last nine months." True, one of the articles contained a brief mention of the fact that Fat Fritz still uses Leela "as a basis." In fact, this article is about an "inspiring" talk by a DeepMind employee to ChessBase programmers, giving the false impression that ChessBase played a significant role in the development of Fat Fritz's code. DeepMind has developed its own machine learning algorithm AlphaZero, which has managed to master not only ordinary chess but also Japanese shogi chess and the game of go. In 2018, he beat Stockfish in a 1,000-game match.


In the ChessBase articles, Fat Fritz was described in a way that implied superiority over Stockfish and Leela. However, when comparing with Stockfish, for some reason an outdated version was used, although it was known that the new one was much more powerful. Likewise, compared to the Leela, the most powerful engine configuration was not used.


In 2020, Stockfish began supporting NNUE, fast neural networks that can run on a CPU. This feature has significantly improved Stockfish's position, returning it to its status as the strongest chess engine in existence.


And now Fat Fritz 2 is out, which ChessBase sells for € 99.90. The engine copies Stockfish instead of Leela. The product description states that FF2 "learns from the accuracy of the legendary Stockfish search." The engine is advertised as the most powerful at the moment, but comparative characteristics are presented only with the old version of Stockfish, and not the version used in FF2 itself.



Independent results show that the current versions of Stockfish that the engine is based on are in fact superior to FF2.


ChessBase published an interview with Silver. The text describes him as the "inventor" of Fat Fritz 2 and claims that he started the project "almost entirely from scratch."

Recent Posts

FYBERUS

  • Facebook
  • Twitter
  • Instagram

Copyright © 2020 Fyberus WebSite. A Fyberus. All rights reserved. Reproduction in whole or in part without permission is prohibited.