Mate in 3: 7 minutes for SF 17

[RubiChess]

Well, I wrote that “strong”, used correctly, is about what’s needed to break something. SF breaks on fortresses, null move and others.

Well, feel free to write some "strong" engine that "breaks SF" by creating fortresses, exploits null move weakness "and others".
I'm very curious.
(Google people probably could do it by waving their magic wand without writing a single patch.)

Regards, Andreas

[Viz]

AZ doesn’t have “patches”, it appears to have understanding

Guess they did quite literally nothing between AlphaGo and AlphaZero and then decided to create it one day, doing nothing for 3 years is my motto indeed.
Also
"Stockish doesn't have "patches", it appears to have understanding tactics and endgames (or any other areas where it's stronger than leela or other MCTS-PUCT based engines)", or how does it work?
Also yeah, if you put sf 15 on a custom cluster hardware and sf dev on some old xeon server sf 15 will "destroy" it, clearly it's a superior engine.

[chessica]

Code: Select all

stockfish_070622			  89 
Swordfish 15.1				  87
Stockfish-16.1				  83
Stockfish-17				  82
Blue Marlin 15.1                          82
ShashChess36				  81

[JVMerlino]

Back to the topic at hand, The King from CM9000 (released in 2002) finds Mate in 3 instantly:

Code: Select all

Time	Depth	Score	Positions	Moves
0:00	1/3	-7.30	1868		1...Ne2+ 2.Kh3 Rxa4 3.Kg2 Rg4+	4.Kf2
0:00	1/4	-7.35	4802		1...Ne2+ 2.Kh3 Rxa4 3.Kg2 Rg4+	4.Kf3
0:00	1/4	-7.36	5514		1...Rxa4 2.Kg3 Kf5 3.Rd1 Ra3+ 4.Kf2
0:00	1/5	-7.45	12235		1...Rxa4 2.Rf1 Kf5 3.Rd1 Re4 4.Kg3
0:00	2/6	-7.50	41479		1...Rxa4 2.Kg3 Kf5 3.Kf2 Ra3 4.Rd1 h4
0:00	3/7	-7.52	89029		1...Rxa4 2.Rf1 Kf5 3.Rf3 Rd4 4.Kg3 a4
0:00	3/7	-Mate03	210840 		1...Rc3 2.Rh2 Kg6 3.Rf2 Rh3#

[chrisw]

Back to the topic at hand, The King from CM9000 (released in 2002) finds Mate in 3 instantly:

Code: Select all

Time	Depth	Score	Positions	Moves
0:00	1/3	-7.30	1868		1...Ne2+ 2.Kh3 Rxa4 3.Kg2 Rg4+	4.Kf2
0:00	1/4	-7.35	4802		1...Ne2+ 2.Kh3 Rxa4 3.Kg2 Rg4+	4.Kf3
0:00	1/4	-7.36	5514		1...Rxa4 2.Kg3 Kf5 3.Rd1 Ra3+ 4.Kf2
0:00	1/5	-7.45	12235		1...Rxa4 2.Rf1 Kf5 3.Rd1 Re4 4.Kg3
0:00	2/6	-7.50	41479		1...Rxa4 2.Kg3 Kf5 3.Kf2 Ra3 4.Rd1 h4
0:00	3/7	-7.52	89029		1...Rxa4 2.Rf1 Kf5 3.Rf3 Rd4 4.Kg3 a4
0:00	3/7	-Mate03	210840 		1...Rc3 2.Rh2 Kg6 3.Rf2 Rh3#

Yes, it would do. Most engines do. The interest in the position is that it shows just how much pruning there is in the SF search, pruning that seems almost stupid, but which overall works, SF being the world’s strongest engine in terms of Elo.

[chessica]

Back to the topic at hand, The King from CM9000 (released in 2002) finds Mate in 3 instantly:

Code: Select all

Time	Depth	Score	Positions	Moves
0:00	1/3	-7.30	1868		1...Ne2+ 2.Kh3 Rxa4 3.Kg2 Rg4+	4.Kf2
0:00	1/4	-7.35	4802		1...Ne2+ 2.Kh3 Rxa4 3.Kg2 Rg4+	4.Kf3
0:00	1/4	-7.36	5514		1...Rxa4 2.Kg3 Kf5 3.Rd1 Ra3+ 4.Kf2
0:00	1/5	-7.45	12235		1...Rxa4 2.Rf1 Kf5 3.Rd1 Re4 4.Kg3
0:00	2/6	-7.50	41479		1...Rxa4 2.Kg3 Kf5 3.Kf2 Ra3 4.Rd1 h4
0:00	3/7	-7.52	89029		1...Rxa4 2.Rf1 Kf5 3.Rf3 Rd4 4.Kg3 a4
0:00	3/7	-Mate03	210840 		1...Rc3 2.Rh2 Kg6 3.Rf2 Rh3#

Yes, it would do. Most engines do. The interest in the position is that it shows just how much pruning there is in the SF search, pruning that seems almost stupid, but which overall works, SF being the world’s strongest engine in terms of Elo.

Hi Chrisw, do you know your answer is stupid? If that were the case, then
You would only need a single rating list. But there are many of them and
is not in a leadership position for everyone.

Engines are built for users, not just a few scoreboard creators or programmers.

There are around 10^8 users worldwide who analyze their games or GM games.
or watch online tournaments. And the Elo values ​​have no relevance at all.

[Jouni]

But this is a rare position where even 5 piece tablebases make huge difference!

1...Rc3 2.Rh2 Kg6 3.Rf2 Rh3#
-+ (-#3) Depth: 34/21 00:00:02 4214kN, tb=42157

[Eelco de Groot]

Am I right then that Lc0 does not need 5 men tablebases anymore, because it will always recognize Zugzwang because no nullmove, even in more complicated positions. IMO big simplification and the SF people should get on this immediately (tongue in cheek mode) Away with the nullmove

[Eelco de Groot]

A Stockfish without nullmove would also not need a symmetrical evaluation anymore (was necessary for the verification search I believe, or something like that?) so for fans of that the old Glaurung style "Aggressiveness" and "Cowardice" if I remember the names correctly, could come back.Not that I am sure if at the moment the symmetry is not broken already because I have read that somewhere at several places. So that 'rule' was already broken then, without any Elo loss, apparently?

[Hai]

[d]8/8/5k2/p6p/P1r2n1K/8/8/7R b - - 0 1
This was posted in discord.
1...Rc3 2.Rh2 Kg6 3.Rh1 Ng2#
-+ (-#3) Depth: 48/62 00:07:44 1426mN
There are other programs to struggle like Obsidian.

...

If you all have such a terrible hardware, then you should start to buy Apple devices

Even the very old, 5 years old, MacBook Pro 16-inch M1 MAX shows:


0:00/245-6: 1...Rc3, 459,2K (14,3M nps), 4,8K tbhits, 32000MB Hash, 8 Threads
#-3/245 1...Rc3 2.Rh2 Kg6 3.Rh1 Ng2#


0:00/245-6: 1...Rc3, 12,1M (20,5M nps), 207,1K tbhits, 32000MB Hash, 8 Threads
#-3/245 1...Rc3 2.Rh2 Kg6 3.Rh1 Ng2#


0:00/245-6: 1...Rc3, 4,2M (21,1M nps), 63,9K tbhits, 32000MB Hash, 8 Threads
#-3/245 1...Rc3 2.Rh2 Kg6 3.Rh1 Ng2#

The only problem I see are weak chess players, sitting in front of the screen with wrong hardware/software.