3/20/2022 – Jon Speelman worked as commentator at this year’s edition of the long-standing Varsity Match between Oxford and Cambridge Universities. Looking at the games alongside computer-chess expert Matthew Sadler prompted him to reflect on the value of using engines for chess improvement and preparation. | Pictured: The Oxford team - Victor Vasiesiu, Tom O'Gorman, Daniel Karim Abbas, Daniel Sutton, 0Dominic Miller, Filip Mihov, AkShaya Kalaiyalahan, Max French. | Photo: John Saunders
Winning starts with what you know
The new version 18 offers completely new possibilities for chess training and analysis: playing style analysis, search for strategic themes, access to 6 billion Lichess games, player preparation by matching Lichess games, download Chess.com games with built-in API, built-in cloud engine and much more.
Winning starts with what you know The new version 18 offers completely new possibilities for chess training and analysis: playing style analysis, search for strategic themes, access to 6 billion Lichess games, player preparation by matching Lichess games, download Chess.com games with built-in API, built-in cloud engine and much more.
[Note that Jon Speelman also looks at the content of the article in video format, here embedded at the end of the article.]
Last Saturday, March 12th, I was at the RAC’s clubhouse (Royal Auromobile Club) in London’s Pall Mall for the annual Varsity match between Oxford and Cambridge Universities.
First played in 1873, this is the world’s oldest chess contest and was for years reported on in the pages of the famous Russian chess magazine “64”. When I played for Oxford from 1975-7, Cambridge were in the ascendant and we lost all three matches: personally, I lost to Michael Stean and drew twice with Jonathan Mestel. These things swing over time, and at the moment it’s very close. Cambridge started as the Elo favourites, but after an endgame save in the last game to finish, Oxford ran out the winners by the narrowest possible margin of 4½-3½, with the overall score now 60-58 to Cambridge with 22 draws.
The 1921 Oxford team | Find more info at BritBase, John Saunders’ excellent games archive
The match has been at the RAC now for nearly half a century, with a dinner afterwards, and in recent years internet coverage and commentary on site. This year’s commentator was Mathew Sadler and for some of the afternoon I acted as sous-commentator, chatting with Matthew about the games.
At one stage I mentioned that I normally use Houdini as my analysis engine, but Matthew [pictured], who of course is immensely knowledgable about computer chess and has written extensively on Alpha Zero, told me that the latest version of Stockfish is much stronger. I therefore decided to switch to it as my default analysis engine in ChessBase, but I’m now wondering (and of course this can be changed with the click of a mouse) whether I was right.
The question of course is how to use the analysis and assessments produced. Most computer engines (Alpha Zero and its daughter Leela are different) are giant bean counters which produce a “maximin”, maximizing the minimum score they get against the opponent's supposedly best play. Depending on the accuracy of the analysis and the size of the beans, the scores will vary, and while Houdini with its rating, I dunno, of 2700 or 2800 tends to bumble around with assessments quite close to zero, Stockfish thunders its pronouncements giving assessments like +/- 2.5 in positions which look to my human eye to be fairly but not entirely clear; and going up/down to +/- 6 or more when even my human eye can see that it “ought” to be winning.
Pavel Eljanov explains in depth what Gyula Breyer already saw in 1911 and what became an opening choice of the likes of Kasparov, Kramnik, Anand or Carlsen. The Breyer Variation, which is characterised by the knight retreat to b8.
The certainty is wondrous but rather unsettling. When I was a kid, I no doubt made the mistake of trying to play the best moves. Nowadays, of course, I know better, and while I will stop and indeed try to work out the best solution in an obviously utterly critical position, most of the time I poddle along choosing decent moves without worrying too much about whether there are better ones. To do this, I’ve created a story for myself that I can quickly select goodish moves in reasonable positions (of course it’s much harder if you’re under heavy pressure). But gazing into the “face of God”, I have to be careful not to be blinded and to undermine this essential fiction.
So I’m still thinking about what to do. Perhaps with enough time available I should use both, analysing both with St Houdini and the deity Stockfish. Certainly when I’m streaming I try much of the time to use my own carbon-based resources and sometimes dip into a fairly hobbled version of Stockfish which isn’t too scary. But occasionally, when I want to know “the truth” I turn to My Lord Sesse (the Norwegian-based fusion of Stockfish and ridiculously powerful hardware).
One point I should make in general is not to take too much notice of computer assessments, even if they are right. They are extremely relevant to the world’s top players when they are doing opening preparation, but for the rest of us they are just a tool. In particular, I’ve noticed that when people check their games after playing online, there are some engines which dish out ??s like confetti. Of course people do play some terrible moves, especially at blitz, but ?? should mean a move that loses a piece or maybe even a rook — or at a higher level makes a complete mess of the position. It shouldn’t mean that the assessment has dropped drastically without in human terms affecting the result.
One reason I go to the Varsity match is to help choose the Best Game and Brilliancy Prize — often with Ray Keene, in this case with Matthew. Both receive works by the artist Barry Martin and, in this case, since the Brilliancy Prize was shared, both players got prints.
Cambridge team: back, left to right: Miroslav Macko, Matthew Wadsworth, Imogen Camp, Harry Grieve. Front, left to right: Jan Petr, Declan Shafi (captain), Ognjen Stefanovic, Koby Kalavannan. | Photo: John Saunders
For the best game, we decided on the board 1 win by Oxford, and I’ve annotated it, out of interest, using both engines. I’ve given them a fairly short time to make an assessment, so they might have changed their “minds” had they worked for a longer period of time — but this experiment nonetheless gives an indication of the huge difference between them.
New ...
New Game
Edit Game
Setup Position
Open...
PGN
FEN
Share...
Share Board (.png)
Share Board (configure)
Share playable board
Share game as GIF
Notation (PGN)
QR Code
Layout...
Use splitters
Swipe notation/lists
Reading mode
Flip Board
Settings
Move
N
Result
Elo
Players
1.e4
1,185,960
54%
2421
---
1.d4
960,101
55%
2434
---
1.Nf3
286,728
56%
2440
---
1.c4
184,987
56%
2443
---
1.g3
19,897
56%
2427
---
1.b3
14,604
54%
2428
---
1.f4
5,958
48%
2376
---
1.Nc3
3,917
50%
2383
---
1.b4
1,791
48%
2379
---
1.a3
1,250
54%
2406
---
1.e3
1,081
49%
2409
---
1.d3
969
50%
2378
---
1.g4
670
46%
2361
---
1.h4
466
54%
2382
---
1.c3
439
51%
2425
---
1.h3
289
56%
2420
---
1.a4
118
60%
2461
---
1.f3
100
47%
2427
---
1.Nh3
93
66%
2506
---
1.Na3
47
62%
2476
---
Please, wait...
1.Nf3Nf62.c4c63.Nc3d54.e3a65.Qc2b56.b3e67.Bb2Be7
7...Nbd78.d4SF 0.7ish7...Bd68.g4!?Very much the move of the
moment. You could hardly imagine Shakhriyar Mamedyarov passing up the chance
to play it.h68...Nxg4is definitely a bad idea after Rg1, but as ususal
the two oracles vary enormously:9.Rg1SF +1.8! H +.79.Rg1g5!?
Trying to block, but a broken blockade is much worse than no blackade at all.10.h410.c5is probably a move I'd hardly think of. SF likes it, intending
to control the centre and then attack afterwards:a510...Bxc511.Nxd511.d4SF+1 H +0.210...gxh411.Ne5Rg812.f4Bd613.Be2Bb714.0-0-0
SF +1.7 H +0.7Bxe5?!This does very bad things to the dark squares, and SF
goes up to 2.5.after the recapture and Houdini to +1.215.fxe5Nfd7
16.g5!This nice move takes g5 away form the black queen.16.Qh7Qg516...hxg517.Qh7Rf818.cxd5Using our human noddles, Matthew and I disliked
this wanting not to give the black knight c6 and concentrating on Ba3
immediately.18.Ba3Qa518...c519.cxd5b420.dxe6and unsurprisingly
it will be "splat", though of course in a game you'd have to prove it. SF +3.4
H +219.Bd6!Rather than take the excahnge, White cements the black
square bind and will start punching next move. SF is now at > +7 (!) and H at
> +4.5.18...cxd519.Ba3Nc620.Bxf8Nxf821.Qg7Qc722.Kb1Qxe5
Watching at the time, I thought that Black might be able to fight, but we
then decided that in fact after Rxg5! there were too many immediate threats
for Black to be able to get organised. However SF - and in this position it is
a very valuable resource - disagrees.23.Rxg5Qxg724.Rxg7
24...d4?
Now the knight gets in and Black is unable to create serious counterplay.24...Ne725.Bh5d426.exd4b427.Bxf7+Kd828.Ne2Nf5and Black is
fighting hard. SF +1.2 H +125.Ne40-0-026.Nc5Ne527.exd4Rxd428.Rg8Rd829.Rc1Kb830.d4Neg631.Bh5Ne732.Rh8Bc833.Bxf7Nf5
SF +5.1
H +2.634.d5A nice human move liquidating (though computer assessments
actually go down).exd535.Bxd5Ne636.Rxd8Nxd837.Kb2h338.Ka3a539.Kb2Ne740.Bh1Ndc641.Na6+Kb742.Nb8Clever but apparently a bad idea.42.Nc5+Kc743.Ne4SF +3 H +1.342...Kxb843.Bxc6b4?43...Nxc644.Rxc6b445.Kc2Bf5+46.Kd2Kb747.Rh6Kc7You'd imagine at first glance
that White could win this somehow, but the assessment of both computers, which
will involve some very long lines, is that White is less than a pawn up so
presumably Black should hold.44.Be4Bd745.Rc5a446.bxa4Bxa447.Rh5Kc748.Rxh3+-SF + 3.6Kd649.Rh5Nc650.Rd5+Kc751.Bc2Ne752.Re5Nc653.Rh5Bxc254.Kxc2
If Black could now play Kb6 and Na5, then this
would be difficult, though tablebases tell us that it is winning. As it is,
the white king is able to advance at once so it's fairly easy.54...Kd655.Kb3Ke656.Kc4Kd657.Rh6+Kd758.Kc5Ne759.Kxb4Nd5+60.Kc4Nc761.Kc5Ne6+62.Rxe6!
Throughout my playing career I have found the Hedgehog one of the most difficult type of positions to master. The basic aim of this video is to improve understanding of these complex positions and to help tournament players score better.
Jonathan SpeelmanJonathan Speelman, born in 1956, studied mathematics but became a professional chess player in 1977. He was a member of the English Olympic team from 1980–2006 and three times British Champion. He played twice in Candidates Tournaments, reaching the semi-final in 1989. He twice seconded a World Championship challenger: Nigel Short and then Viswanathan Anand against Garry Kasparov in London 1993 and New York 1995.
How do you play the Queen's Gambit Accepted? Does White have promising variations or can Black construct a water-tight repertoire? The Powerbook provides the answers based on 300 000 games, most of them played by engines.
The Queen's Gambit Accepted Powerbase 2025 is a database and contains a total of 11827 games from Mega 2025 and the Correspondence Database 2024, of which 240 are annotated.
Rossolimo-Moscow Powerbase 2025 is a database and contains a total of 10950 games from Mega 2025 and the Correspondence Database 2024, of which 612 are annotated.
The greater part of the material on which the Rossolimo/Moscow Powerbook 2025 is based comes from the engine room of playchess.com: 263.000 games. This imposing amount is supplemented by some 50 000 games from Mega and from Correspondence Chess.
€9.90
We use cookies and comparable technologies to provide certain functions, to improve the user experience and to offer interest-oriented content. Depending on their intended use, analysis cookies and marketing cookies may be used in addition to technically required cookies. Here you can make detailed settings or revoke your consent (if necessary partially) with effect for the future. Further information can be found in our data protection declaration.
Pop-up for detailed settings
We use cookies and comparable technologies to provide certain functions, to improve the user experience and to offer interest-oriented content. Depending on their intended use, cookies may be used in addition to technically required cookies, analysis cookies and marketing cookies. You can decide which cookies to use by selecting the appropriate options below. Please note that your selection may affect the functionality of the service. Further information can be found in our privacy policy.
Technically required cookies
Technically required cookies: so that you can navigate and use the basic functions and store preferences.
Analysis Cookies
To help us determine how visitors interact with our website to improve the user experience.
Marketing-Cookies
To help us offer and evaluate relevant content and interesting and appropriate advertisement.