Raubfisch

[Stephen Ham]

Hello AR,

I respectfully disagree. My selected lines were sharp, unbalanced and asymmetrical, and tactically rich while avoiding massive piece exchanges. In my openings, pieces were developed rapidly and then stayed on the board. Many games lasted over 100 moves and retained a sharp nature throughout. Did you even examine the lines I provided?

Instead, the lines you suggested involve items that that chess engines don't evaluate properly, such as Black's advancing pawn wall in the King's Indian. That's aggression and not tactics. Besides, chess engines there fail to exhibit the correct technique to execute pawn storms without an opening book guiding them into the middlegame and beyond.

Are you joking about the Piano and Queen's Indian? Piano is an Italian word for "quiet" as it leads to dull and "quiet' chess. Similarly the Queen's Indian is a "safe" defense, not known for tactics and instead has a very high draw rate. Sharp Dragons and Najdorfs often lead to drawing lines after piece exchanges. I know the Dragon - I've played it my entire life. Those lines are indeed aggressive, but not as tactically rich as the lines I selected.

Regards,
-Steve-

[supersharp77]

Hello AR,

I respectfully disagree. My selected lines were sharp, unbalanced and asymmetrical, and tactically rich while avoiding massive piece exchanges. In my openings, pieces were developed rapidly and then stayed on the board. Many games lasted over 100 moves and retained a sharp nature throughout. Did you even examine the lines I provided?

Instead, the lines you suggested involve items that that chess engines don't evaluate properly, such as Black's advancing pawn wall in the King's Indian. That's aggression and not tactics. Besides, chess engines there fail to exhibit the correct technique to execute pawn storms without an opening book guiding them into the middlegame and beyond.

Are you joking about the Piano and Queen's Indian? Piano is an Italian word for "quiet" as it leads to dull and "quiet' chess. Similarly the Queen's Indian is a "safe" defense, not known for tactics and instead has a very high draw rate. Sharp Dragons and Najdorfs often lead to drawing lines after piece exchanges. I know the Dragon - I've played it my entire life. Those lines are indeed aggressive, but not as tactically rich as the lines I selected.

Regards,
-Steve-

If those lines were as "tactically rich" as you say..you would have different results my friend....engines avoid these complex lines for a reason..they tend to "trade out" often quite early...unbalance is the key and the avoidance of early exchanges thats why the Piano and the queens indian are not to be underestimated..why do you think the Super Gm's chose such lines?
Now the dragon has been overanalyzed along with some of the sicilians but there is still quite a lot of bite in them..The lines you "produced" never heard of them my friend and I've seen them all...They did not "Produce"....

[Dann Corbit]

Are you joking about the Piano and Queen's Indian? Piano is an Italian word for "quiet" as it leads to dull and "quiet' chess. Similarly the Queen's Indian is a "safe" defense, not known for tactics and instead has a very high draw rate. Sharp Dragons and Najdorfs often lead to drawing lines after piece exchanges. I know the Dragon - I've played it my entire life. Those lines are indeed aggressive, but not as tactically rich as the lines I selected.

Regards,
-Steve-

Computers have an awful time with any Indian opening. I guess that they simply do not score fianchetto bishops correctly, but there may be some other reason. If I were going to play against a computer with a goal of highest score achieved, I would use Indian openings. But I always play Orangutan, because I'm quirky like that.

Anyway, I would not begin to trust Indian opening analysis till it is at least 40 plies, and I would not feel comfortable below 45 plies.

[Stephen Ham]

Hi Dann,

The fact that the games often ran over 100 moves confirms that the openings led to positions with material retention. There were no early "trade outs". I've presented the selected opening lines in an earlier post. Those lines are shallow, terminating quickly to force the engines to solve problems immediately. And, their terminal positions are all tactically rich and unbalanced. Again, I can send you games to examine so you can see that the draws at termination had nothing to do with the openings nearly 100 moves earlier.

You've made my case with regard to the King's Indian defense, Dann - we agree that chess engines don't understand it. Therefore, please note that I didn't use any in the match.

However, engines don't have problems with all Indian lines, nor do they have general problems with fianchetto bishops. For example, I play the Gruenfeld and Dragon Sicilian as black and find that most engines evaluate them correctly. Instead, it's the pawn storms in King's Indians and Benonis that chess engines fail to evaluate and play correctly.

Again, the Queen's Indian is a "safe" defense while the Giuoco Piano is Italian for "quiet game". These are not openings one would ever select for tactical tests. So Dann, "super GMs" never choose these openings for tactics.

I've had my chess engines play hundreds of Dragon games to study that opening. But the resulting positions were not nearly as sharp as those in the opening lines I just tested. Again, please examine those lines - there's nothing drawish about them.

So the mystery remains - why such a low decision rate? I believe chess is a draw with best play and the top engines are now approaching best play. We see that too in TCEC. But I doubt the openings selected in my match represent best play. Still, the low decision rate surely is not related to the openings which occurred many moves earlier. I'm auditing an 89 move draw now and find that the moves played match those selected now by Stockfish even after allowing it to search much deeper than in the game.

All the best,
-Steve-

[Graham Banks]

.......it's the pawn storms in King's Indians and Benonis that chess engines fail to evaluate and play correctly.

Engines nowadays are a lot better in this regard.

[Stephen Ham]

Indeed, they are better, Graham. Nonetheless, they still have problems.

In one of my latest ICCF games, I was white in a King's Indian. The position was equal, but all my engines claimed significant advantages for my side, even after searching very deeply. This persisted until Black's pawn storm dissolved. A "normal" position resulted and only then did the evaluation drop to equality.

-Steve-

[Graham Banks]

Indeed, they are better, Graham. Nonetheless, they still have problems.

In one of my latest ICCF games, I was white in a King's Indian. The position was equal, but all my engines claimed significant advantages for my side, even after searching very deeply. This persisted until Black's pawn storm dissolved. A "normal" position resulted and only then did the evaluation drop to equality.

-Steve-

Yes - I still see this happen too.

[Dann Corbit]

Instead, it's the pawn storms in King's Indians and Benonis that chess engines fail to evaluate and play correctly.

I adore pawn storms.
Even when they don't work, they are boiling with excitement.
That's also why I love opposite castles. And I am bummed by an opposite castle that does not lead to a pawn storm.

Which reminds me of another engine weakness:
long castles. They have some kind of inner phobia for long castles. Or, at least. some do.

[Graham Banks]

Dear gents,

I stopped the match after 200 games. The result was a 15 victory to 12 loss result in favor of Raubfisch X40 over Stockfish 3/31/19.

Strictly on the basis of performance, this engine should be noticed. Statistically, this result could have been obtained by playing Stockfish against itself. After all, Raubfisch's mysterious "author(s)" admits it's derived from Stockfish.

Nonetheless, Raubfisch apparently has more code. I also perceive that it has a more accurate and stable evaluation function and slightly different move selection. But search speed and depth are virtually identical for both.

However, I remain VERY surprised that only 14% of the games were decisive, especially when the opening book was ultra-sharp and largely terminated after 10 moves or less. After all, I wanted the engines to solve their tactical situations as early as possible.

I can make all games available to interested parties.

Sincerely,
-Steve-

I'd be interested in looking at them.

[supersharp77]

Hello AR,

I've learned from previous engine tests that my openings extended too deeply - they solved much of the tactics for the engines before reaching the middlegame. Instead, I wanted to reach sharp and unbalanced positions as quickly as possible to force the engines to solve matters on their own. This was meant to be a test of tactical superiority. So, I created a shallow opening book with the following opening lines:

1 d4 d5 2 c4 Bf5 3 Qb3 e5 4 cxd5 exd4 5 Nf3 Be4 6 Nxd4 Bxd5 7 Qg3
5... Bc5 6 Nxd4 Bxd4 7 Qa4+ Nc6 8 dxc6 b5 9 Qb3 Ne7 10 e3 Bf6 11 Bxb5
2 Nc3 Nf6 3 e4 dxe4 4 f3 exf3 5 Nxf3 Bf5 6 Ne5 e6 7 g4 Be4 8 Ne4 Nxe4 9 Qf3 Qh4+ 10 Kd1
1... Nf6 2 c4 e5 3 dxe5 Ng4 4 Nf3 Bc5 5 e3 Nc6 6 b3 0-0 7 Bb2 Re8
1... b6 2 c4 Bb7 3 Nc3 e6 4 e4 Bb4 5 Bd3 f5

1 e4 e5 2 f4 exf4 3 Nf3 g5 4 Bc4 Bg7
3 Bc4 Qh4+ 4 Kf1 d6 5 d4 Nf6 6 Nf3 Qh6 7 Nc3
2 Nf3 Nc6 3 d4 exd4 4 c3 dxc3 5 Bc4 cxb2 6 Bxb2 Bb4+ 7 Nc3 Nf6 8 Qc2 d6 9 0-0-0 0-0 10 e5 Ng4 11 Ncxe5 12 Ng5 g6 13 Nce4 Bf5 14 Qb3 Qe8 15 f3

I've since modified my book to add 8 Be2 to the Budapest Gambit (line 4) as I was shocked to see that both engines invariably played 8 h3?!, wasting a tempo to force Black to do what he's going to do anyway...recapture the pawn. I subsequently forced Stockfisch to calculate to see if it ever selects 8 Be2 on its own. It required 42 minutes on my old quad before selecting 8 Be2 and then staying with it for several hours. But this is a move that most humans would play automatically. Even 8 Bd3 is superior to 8 h3. Somehow the chess engines failed to see that instead of losing a tempo, they should develop and castle.

Some of the above lines variously favor White or Black. But since both engines alternate colors with each opening line, then matters were fair.

Both engines often claimed large advantages, frequently winning advantages, only to have the game draw. Some of the overstated evaluations were no doubt a result of Contempt. At times, both engines simultaneously claimed the advantage for itself. Clearly both engines need improved evaluation functions.

Based on the openings used it looks like you are using Stockfish engines (ie Raubfisch) to analyze the mysteries of 'Gambits' now when you referenced your post you mentioned "Tactics" I discussed 'Tactical positions' and 'Complications' I would venture this... there is a difference between 'Gambits' and General 'Tactics' Most chess engines are good with 'Tactics' but have extreme problems with "Gambits" [the concept of giving up something (1-2 pawns) early in a opening for strong (10-20 moves later maybe) compensation]. How do you program that in? It's got to be quite difficult...A few quite famous engines were great (or pretty good ) at such positions..for example Chessmaster..Junior 7, Deep Junior, Szint Ippolits, etc. many engines won't do it at all plus with "best defense" these gambits have spotty results with a few exceptions (Volga Gambit etc) Now let me tell you a story... when I was young in junior High school (I learned chess at a young age 7-8 ..my first book Fischers 60 Most memorable games) I used to occasionally skip school to play chess at the local coffee house where they had a few chessboards books & clocks...The owner of this chess coffee House was a famous chess master gambler, and Gambit Player.. Mr Jack W. Jack W was a former state champion and constantly smoked and drank coffee and kibbitzed while he played Gambit after gambit...Jack knew them all...E-4 was his first move He sacked pawns when he got up in the morning and pieces for lunch..he gave knight and rook odds if the "money was right". Watched him win game after game..to compete with Jack you had to study openings and Gambits...So I had to study..Hard...Occasionaly Jack would lose but most of the time he came out on top against most. Only the top players could best him. Most can never be Gambit players..Most chess engines never will uncover the "Mystery Of The Chess Gambits"... LC0 (Leela0) does seem to have that kind of flair and attitude for that type of "Old School" chess...