Hosted by Darius, creator of chessengeria.eu, May 2022.
„(…) within search there are still many interesting heuristics to be tried”.
Dear Readers, it is my honor and pleasure to present to you an interview with Jay Honnold, who is the author of Berserk - one of the most powerful chess engines.
On our blog, there is a page entirely dedicated to the Berserk chess engine.
If you want to find out more (download, games, etc.), I strongly invite you to visit it.
Jay, please introduce yourself to our Readers.
Jay Honnold, source: GitHub
Hi, my name is Jay. I'm a software developer in Phoenix, AZ and I'm the author of the Berserk Chess Engine.
What made you interested in creating and developing a chess engine ?
I've played chess since I was a child. Programming started as a hobby and eventually became a career. At some point I decided it would be fun to mix my "hobbies" together.
Tell us about the history of Berserk.
Berserk's first code was written in Dec 2020. Originally I wrote it in Java as that is the language I have most experience in. I realized that this was simply too slow of a language (or at least my implementation was) and I decided to move to C. From here Berserk followed a pretty standard path of development and most recently has made the switch to NNUE.
What is your main work to be done with Berserk right now ?
I'm heavily focused on deep networks. Berserk has used a shallow architecture since NNUE was implemented and I'm looking too improve on this architecture by adding additional layers. At this time, the speed hit has not balanced the improved networks.
What motivates you to develop the Berserk engine ?
I simply enjoy working on Berserk. It's a fun project and is something I'm happy to put time and investment into.
What innovations may we expect in future versions of Berserk ?
Only the future knows.
How did you achieve that Berserk can see moves invisible to other engines ?
This is purely luck :).
When can we expect Berserk 9 ?
I would say to expect a stronger network as I don't plan on releasing until then. Apart from that, I have fine tuned Berserk's search and continue to work through small patches there as well.
Do you collaborate with anyone while working on your chess engine ?
Berserk would not be the engine it is without the people I've collaborated with. Early on, Martin (Cheng's author) was a huge help for me. Later, Terje (author of Weiss), Connor (author of Seer), Finn and Kim (authors of Koivisto), Andrew (author of Ethereal), Aryan (author of Bit Genie), Pali (author of Black Marlin), Vizvez (contributor on SF) as well as many others.
Do you see any weaknesses in Berserk, or is there something you would like to significantly improve or strengthen ?
Berserk has a relatively weak network. It's small and fast, but has room for improvement. That has been my main focus of development.
Is the playing power of a chess engine the most important feature of a modern chess engine today ?
That is up to the author. I develop Berserk to be strong, so that is most important to me. Other authors may want their engine to focus on tactical play or position play.
Berserk is capable of winning against any other chess engine. It does well in competition with other super-strong chess engines. What can Berserk bring to make it more interesting to the average chess player?
I'm unsure what Berserk's speciality is. It has great attacking prowess, but in many games I enjoy watching it grind an endgame. I think this is what can interest an average chess player. A game where everything looks equal, but Berserk can take it to a win.
Some claim that modern chess engines using NNUE play in an "artificial style" devoid of the human element of the traditional position evaluation function. Do you think there is some truth in this ?
I think engines are so much stronger than humans, even without NNUE, that all chess engines at a high level simply play without human element. Berserk 4.5.1 had no network, but still played artificially.
What do you think will have a greater impact on chess engines in the future, the continued development of NNUE (Efficiently Updatable Neural Networks) or the development of MCTS (Monte-Carlo Tree Search) ? Or maybe something else ?
NNUE. I think within search there are still many interesting heuristics to be tried, but none as drastic as MCTS.
Do you have a favorite chess engine / engines ?
I like all chess engines that are open source and contribute to the growth of this community as a whole. If I had to choose a favorite, I think Seer's development of its original networks really make that engine unique and interesting.
Do you have any other hobby besides computer chess ?
I like to write code in general, doesn't have to be with computers. I'v also started to play OTB chess again, and that has been fun. Separate from chess and programming, I like to run and backpack. While networks train, I go outside and enjoy some nature.
Favorite Games ?
https://lichess.org/study/QhDbEdk7 (Game 11 vs Fire)
Link to Berserk game viewer page (added by Darius): Click here
Jay, on behalf of Chessegeria Readers and myself, thank you so much for taking the time to do this interview. All the best in further development of Berserk !
Further publication and reproduction of this interview is permitted only with the prior permission of Darius the creator of chessengeria.eu
The condition is that the content is preserved in its entirety and that a link to chessengeria.eu is included.