GM Larry Kaufman Interview: 'New Repertoire For Black And White'

GM Larry Kaufman's book Kaufman's New Repertoire for Black and White is now available in Europe in all formats, and in the United States as an e-book. It will soon be sold worldwide in all formats, including as an e-book from Forward Chess.

The book serves as a follow-up to the excellently received The Kaufman Repertoire for Black and White, but it is not a second edition. Kaufman has completely revised his repertoire from the White side and is now recommending 1.e4 instead of 1.d4—the most impactful of many changes.

Kaufman's perspective is unique among chess authors. Kaufman is a graduate of MIT who has been involved in computer chess since the 1960s. Today, most chess players will be familiar with his work on first Rybka and now the Komodo chess engine.

Kaufman first became an IM in 1980 and a GM in 2008 when he won the World Senior Championship. He is currently a vital part of the Chess.com and Komodo team.

Chess.com: How did the idea for this book come about, and what did you find compelling enough about the idea to warrant the commitment a book requires?

GM Larry Kaufman: My first complete repertoire for both colors was written in 2003 (The Chess Advantage in Black and White, Random House), and was partly done as a way to force myself to work out a complete repertoire that was both sound and compact enough to fit in one book, so that I could hope to remember most of it for my own tournament games. I reasoned that many others would also like such a book. The 2012 book was a completely new one for New in Chess, but with largely similar motivation, with the difference that now the engine I had been developing, Komodo, would play a major role.

The 2019 book was largely motivated by the MCTS [Monte Carlo Tree Search, a method for computer engines to choose which moves to investigate] revolution, because MCTS engines favor lines that are apt to work in over-the-board play against humans, not just ones intended for correspondence play vs. engines. The reason is that normal engines assume "perfect" play by the opponent, while MCTS merely assumes "good" play. My hope was that the analysis would be very different from books that use traditional engines. Of course there are good and bad moves in chess, so there is considerable overlap, but I expect people will see substantial differences from existing theory.

Why did you think the time was right to release the new repertoire?

Because Lc0 [a machine-learning chess engine that uses neural networks and MCTS] and Komodo MCTS had reached high-enough level to make an MCTS-based book logical. Also because the improvement in both hardware and software in seven years made the earlier book rather obsolete.

Who do you think will most benefit from this book?

The actual analysis is at a very high level, so is suitable for strong players, even for grandmasters. However the explanations of the moves are aimed at average amateur players, maybe in the 1200 to 1800 range. So I hope that players of a wide range of abilities can learn from this book, although they will learn different things. I suppose that players in the 1600 to 2000 range might get the most benefit, as they are not too strong to need the explanations but strong enough to utilize the advantages they may get from the lines.

Chess players today often suggest that opening theory is draining chess of its excitement. Do you agree?

Yes, I do agree, even though saying so won't help sales of the book! But if you are going to play competitive chess with reasonably strong players, whether over the board or online, you will get better results playing good openings that you know than playing poor ones or good ones that you don't know. I am a big advocate of reforms in chess, whether that means balloted openings, playing Fischer Random, Armageddon playoffs, special anti-draw rules, or whatever. But this book is written for chess as it is played currently.

How do you think opening preparation is different today from when you first became an IM and GM?

Well, that's two completely different questions, because I became an IM in 1980, before computers were of any use to chess players, but a GM in 2008, when I got my title by winning the World Senior Championship with the help of massive computer preparation both before the event and for every game. In 1980, preparation meant carrying around a couple books and reviewing the lines in the books that you wanted to play. Two different worlds!

Which major openings are you no longer recommending in the new repertoire, and which ones are you recommending now?

For White, it was a total switch from 1.d4 to 1.e4, motivated by some positive developments for 1.e4 and some negative ones for 1.d4. For Black, the main change is that the Breyer defense to the Spanish is now my backup line, with the Marshall becoming my main one, along with a chapter on the Moller that I suggest might be better for correspondence play than for over the board. No major change vs. 1.d4, although the games and analysis are heavily revised.

In analyzing the opening, in which aspects was Komodo superior, and in which aspects was Lc0 superior?

Lc0 was generally superior on my computer, because it has a very powerful, expensive GPU with 3,000 cores; Komodo just uses six of the eight CPUs. Lc0 has some weak points though, which Komodo patches up: Lc0 is rather blind to perpetual checks, relatively weak in evaluating many endgames, and lacks specialized chess knowledge that applies in infrequent situations. Also I would say that Komodo is generally superior when it is necessary to see a long, precise series of moves to justify the initial move choice. Lc0 will usually find good moves quicker with my GPU, but will be slow to admit that it is wrong.

Is Lc0 as strong as AlphaZero now?

If they both ran on the same or comparable hardware, I think they are close in strength. But even Chess.com doesn't have hardware like Google. Probably a five or 10-minute Lc0 think on my computer is comparable to a one-minute think of AlphaZero on Google's hardware.

Given your expertise with engines, have you any interest in competing in the International Correspondence Chess Federation?

The percentage of draws between top ICCF players is in the mid or high 90s, I'm told. Playing a game with such a draw percentage doesn't interest me.

How do you evaluate the future of Chess960 (Fischer Random) as a means of circumventing opening theory?

I am a big advocate of Chess960 (Fischer Random), especially after seeing the semifinals and finals of the World Fischer Random Chess Championship. I actually won the only U.S. open championship of the game ever held (I believe), about a decade ago. I don't find it interesting to watch two super-GMs reproduce 20+ moves of computer analysis in a broadcast standard chess game, but I try not to miss a single game of 960 live between the top players, knowing that they are thinking for themselves after just a couple moves. I only regret that there are hardly any live, over-the-board events that most players can join. I would love to see 960 take an equal footing with standard chess before I become too old to play.

Is the French Defense as bad as IM Danny Rensch says it is?

Normal chess engines think it's more or less as good as anything, but statistics and the neural-network engines rate it as inferior to the "big three" (1...e5, Sicilian, Caro.) I think that you need to know a lot to prove that the French is inferior, but at super-GM or correspondence level, it is just the fourth-best defense. But I owe my GM title and World Senior Championship to the French!

Is Bobby Fischer right that 1.e4 is "best by test"?

Yes, I agree with him on this (as well as on 960 and on the use of increment in chess, although I have a better claim to being the inventor of increment chess than he does!). But 1.d4 and 1.Nf3 are not far behind.

Which question do you wish you had been asked about the new repertoire?

"How can I write a technical opening book full of the latest games and novelties played and analyzed in 2019, when I'm old enough to have had a chess teacher (Harold M. Phillips) who played against Steinitz in 1894?"

As an active partner in Komodo, I have the technical knowledge needed to make best use of computers. I can afford the best practical hardware. Also I probably understand better than any other GM where the displayed scores are coming from, at least in the case of Komodo, and so I can explain the +.53 eval to the reader. I'm still an active tournament player, the oldest active American GM, about to turn 72. My parents lived to an average age of 100, so I'm pretty young still for an old man!