Rebel 16.2: Impressive!

[chrisw]

Oh guys depth 16 from Stockfish (and maybe Ponder OFF) - that's a joke. How can you get excited about games like this? You live on another planet!

Well, some of us engine programmers are interested in toppling the long time champion from her number one spot and to do that we have to know where her weaknesses lie. Since it is very difficult trying to work out what a chess engine is thinking/doing from incomprehensibly deep lines, it helps to play fast games and look for her problems with shallower lines. What these games show is the very high pruning and type of pruning that SF does, namely very materialistic. What does SF miss? Clearly elements related to king safety. So now we have a line of attack. Theee fast games show the line of attack working, which gives us confidence (or excitement or whatever you want to call it) to continue. Plus it’s fun.

[mclane]

...............try this game for Tal Attack. I've not see a Stockfish so comprehensively smashed in so few moves without any idea of what was happening. Look at the evaluation discrepancies .....

Fast Game UHO openings, TC=20+0.05


[pgn][Event "?"]
[Site "?"]
[Date "2023.02.22"]
[Round "126"]
[White "Stockfish_13"]
[Black "Chess-System-Tal-1.59"]
[Result "0-1"]
[ECO "A40"]
[Opening "Queen's pawn"]
[TimeControl "20+0.05"]
[Termination "adjudication"]
[PlyCount "55"]
[GameDuration "00:00:18"]
[GameEndTime "2023-02-22T20:17:41.373 W. Europe Standard Time"]
[GameStartTime "2023-02-22T20:17:22.467 W. Europe Standard Time"]

1. d4 { book } 1... e6 { book } 2. Bf4 { book } 2... b6 { book } 3. e3 {
book } 3... Bb7 { book } 4. Nf3 { book } 4... d6 { book } 5. h3 { book }
5... Nd7 { book } 6. Bh2 { book } 6... f5 { book } 7. c4 { book } 7... g6 {
book } 8. Nc3 { book } 8... Bg7 { book } 9. a4 { +0.91/22 2.3s } 9... Ngf6
{ -1.41/13 0.75s } 10. Ng5 { +1.01/16 0.23s } 10... Qe7 { -2.25/15 0.52s }
11. Nb5 { +1.36/17 0.35s } 11... O-O { -1.92/16 0.81s } 12. Nxc7 { +1.31/17
0.46s } 12... f4 { -1.69/16 0.70s } 13. Nxa8 { +1.86/17 0.23s } 13... fxe3
{ -1.30/15 0.64s } 14. fxe3 { +2.11/17 0.30s } 14... e5 { -1.89/16 0.70s }
15. Nc7 { +3.93/17 0.25s } 15... exd4 { -1.68/15 0.69s } 16. Nge6 {
+4.15/16 0.23s } 16... Ne4 { +1.34/14 0.41s } 17. Nxf8 { +4.40/17 0.27s }
17... Qh4+ { +0.25/14 0.52s } 18. g3 { +6.62/17 0.30s } 18... Qf6 {
+2.08/15 0.34s } 19. Qe2 { +6.90/18 0.26s } 19... Ndc5 { +3.52/14 0.45s }
20. O-O-O { +3.25/24 1.7s } 20... d3 { +12.61/16 0.75s } 21. Nd7 { +3.25/21
0.31s } 21... Nb3+ { +M19/33 0.30s } 22. Kb1 { -M18/20 0.69s } 22... Nc3+ {
+M17/38 0.43s } 23. bxc3 { -M16/32 0.24s } 23... Qxc3 { +M15/42 0.33s } 24.
Nf6+ { -M14/34 0.26s } 24... Bxf6 { +M13/40 0.28s } 25. Qa2 { -M12/38 0.26s
} 25... d2 { +M11/41 0.36s } 26. Bd3 { -M10/43 0.23s } 26... Bxh1 { +M9/44
0.30s } 27. Bc2 { -M8/65 0.23s } 27... Be4 { +M7/54 0.28s } 28. Rxd2 {
-M6/245 0.17s, Black wins by adjudication } 0-1[/pgn]

Excellent game!

Very nice game!! And I recognise the playingstyle off CSTall 2 from zillion years ago. Good to see that it is still / again in development.

Its the mission of grandfather Ed Schroeder and grandfather Chris Whittington to deliver king attack chess games against the premium top ten of todays chess engines.

The mission began (Chris) 1981 on the zx81 or (Ed) 1979 on the TRS80.
In these days hardware had slow CPUs and very limited RAM and ROM capacities.
The software was written first in BASIC and used assembler routines to speed them up.

Today the resources are nearly infinite.
CPUs have 64 cores and RAM and ROM is huge.
Instead of assembler you have C and the use of neural nets gives plenty of possibilities.

So lets enjoy our grandfathers work !

[mclane]

More results:

Code: Select all

 Caro_Kann_top2, 120min/40+120min/40+(15  2023

                    1  2  3  4  
1   Stockfish 15.1  ** ½½ 1½ ½½   3.5/6  9.75
2   Rebel-16.2      ½½ ** ½½ 1½   3.5/6  9.75
3   Koivisto 9.2    0½ ½½ ** ½½   2.5/6  7.75
4   Berserk 11      ½½ 0½ ½½ **   2.5/6  7.75

given Caro Cann opening,
8 cores Ponder On, Hash 8192 MB

[chrisw]

Here an example of an attack being generated out of nowhere (well, actually the inferior black side of an UHO opening). The opponent didn't seem to play any "blunders", haha! TC=20+0.02, move times are about 0.5 seconds, long enough.

9 ... Rb8, then b5, just ridiculous, but it preempted any white attack and left black with a perfect launch pad after 19. O-O, all pieces perfectly poised on the back rank ....
I am quite impressed.

[pgn][Event "?"]
[Site "?"]
[Date "2023.02.24"]
[Round "102"]
[White "Seer-2.6"]
[Black "Chess-System-Tal-1.59"]
[Result "0-1"]
[ECO "C48"]
[Opening "Four knights"]
[Variation "Spanish Variation"]
[TimeControl "20+0.05"]
[PlyCount "60"]
[GameDuration "00:00:23"]
[GameEndTime "2023-02-24T11:09:48.684 W. Europe Standard Time"]
[GameStartTime "2023-02-24T11:09:24.908 W. Europe Standard Time"]

1. e4 { book } 1... e5 { book } 2. Nf3 { book } 2... Nc6 { book } 3. Nc3 {
book } 3... Nf6 { book } 4. Bb5 { book } 4... Bd6 { book } 5. d3 { book }
5... O-O { book } 6. a3 { book } 6... Re8 { book } 7. h3 { book } 7... Bf8
{ book } 8. Bc4 { book } 8... h6 { book } 9. Be3 { +1.09/16 0.86s } 9...
Rb8 { -1.79/16 1.1s } 10. g4 { +1.54/17 0.86s } 10... b5 { -1.98/16 0.70s }
11. Bxb5 { +1.69/17 0.52s } 11... Nd4 { -1.97/17 0.61s } 12. Nxd4 {
+1.68/16 0.42s } 12... exd4 { -2.08/18 1.1s } 13. Bxd4 { +1.41/17 0.67s }
13... c6 { -1.81/17 0.67s } 14. Ba4 { +1.42/17 0.57s } 14... Rxb2 {
-1.92/17 0.61s } 15. Ne2 { +1.68/17 0.75s } 15... Rb6 { -1.48/14 0.59s }
16. Bxb6 { +1.74/17 1.0s } 16... axb6 { -1.42/16 0.40s } 17. c4 { +1.50/17
1.6s } 17... b5 { -0.54/16 0.76s } 18. cxb5 { +2.05/16 0.51s } 18... d5 {
-0.48/15 0.48s } 19. O-O { +2.14/16 0.33s } 19... Bxg4 { -0.24/14 0.54s }
20. hxg4 { +2.44/17 0.49s } 20... Nxg4 { +2.12/14 0.39s } 21. Ng3 {
+2.67/18 0.41s } 21... Qh4 { +2.40/15 0.50s } 22. Kg2 { +2.78/17 0.33s }
22... Qh2+ { +2.40/14 0.41s } 23. Kf3 { +2.86/17 0.39s } 23... Bd6 {
+2.05/15 0.33s } 24. Kxg4 { 0.00/16 0.43s } 24... Qg2 { +17.37/14 0.30s }
25. Kh4 { 0.00/17 1.2s } 25... Re5 { +M11/29 0.40s } 26. Rh1 { -46.43/26
0.55s } 26... Be7+ { +M9/39 0.27s } 27. Kg4 { -46.44/30 0.28s } 27... Rg5+
{ +M7/52 0.26s } 28. Kh4 { -46.44/31 0.23s } 28... Rxg3+ { +M5/82 0.17s }
29. Kh5 { -46.45/33 0.25s } 29... Rg5+ { +M3/116 0.027s } 30. Kh4 {
-46.45/33 0.26s } 30... Re5# { +M1/116 0.020s, Black mates } 0-1

[/pgn]

[dkappe]

Oh guys depth 16 from Stockfish (and maybe Ponder OFF) - that's a joke. How can you get excited about games like this? You live on another planet!

Well, some of us engine programmers are interested in toppling the long time champion from her number one spot and to do that we have to know where her weaknesses lie. Since it is very difficult trying to work out what a chess engine is thinking/doing from incomprehensibly deep lines, it helps to play fast games and look for her problems with shallower lines. What these games show is the very high pruning and type of pruning that SF does, namely very materialistic. What does SF miss? Clearly elements related to king safety. So now we have a line of attack. Theee fast games show the line of attack working, which gives us confidence (or excitement or whatever you want to call it) to continue. Plus it’s fun.

An interesting proposition. When one realizes that a deep search has a shallow search at the tail end, it gives one hope. My experience, however, is that SF gets dramatically better as you add on a few ply. Nevertheless, best of luck.

[Whiskers]

Oh guys depth 16 from Stockfish (and maybe Ponder OFF) - that's a joke. How can you get excited about games like this? You live on another planet!

Well, some of us engine programmers are interested in toppling the long time champion from her number one spot and to do that we have to know where her weaknesses lie. Since it is very difficult trying to work out what a chess engine is thinking/doing from incomprehensibly deep lines, it helps to play fast games and look for her problems with shallower lines. What these games show is the very high pruning and type of pruning that SF does, namely very materialistic. What does SF miss? Clearly elements related to king safety. So now we have a line of attack. Theee fast games show the line of attack working, which gives us confidence (or excitement or whatever you want to call it) to continue. Plus it’s fun.

An interesting proposition. When one realizes that a deep search has a shallow search at the tail end, it gives one hope. My experience, however, is that SF gets dramatically better as you add on a few ply. Nevertheless, best of luck.

I honestly think that breaking SF down with a good old fashioned attack is nigh impossible if it has enough time to think - the old "use search to take care of the king" doesn't work at all at depth 15, but at depth 30+ it's much more likely to be viable, especially as SF's king safety is every bit as good as other top engines'. Being able to create an engine that beats a high depth SF with an attack regularly would be one of the most incredible feats I've ever seen.

But those Rebel games are still very nice looking. A 3300 strength version of CS Tal is an amazing proposition - like Velvet but even more entertaining. Looking forward to seeing what's coming!

[mclane]

[pgn][Event "?"]
[Site "?"]
[Date "2023.02.24"]
[Round "102"]
[White "Seer-2.6"]
[Black "Chess-System-Tal-1.59"]
[Result "0-1"]
[ECO "C48"]
[Opening "Four knights"]
[Variation "Spanish Variation"]
[TimeControl "20+0.05"]
[PlyCount "60"]
[GameDuration "00:00:23"]
[GameEndTime "2023-02-24T11:09:48.684 W. Europe Standard Time"]
[GameStartTime "2023-02-24T11:09:24.908 W. Europe Standard Time"]

1. e4 { book } 1... e5 { book } 2. Nf3 { book } 2... Nc6 { book } 3. Nc3 {
book } 3... Nf6 { book } 4. Bb5 { book } 4... Bd6 { book } 5. d3 { book }
5... O-O { book } 6. a3 { book } 6... Re8 { book } 7. h3 { book } 7... Bf8
{ book } 8. Bc4 { book } 8... h6 { book } 9. Be3 { +1.09/16 0.86s } 9...
Rb8 { -1.79/16 1.1s } 10. g4 { +1.54/17 0.86s } 10... b5 { -1.98/16 0.70s }
11. Bxb5 { +1.69/17 0.52s } 11... Nd4 { -1.97/17 0.61s } 12. Nxd4 {
+1.68/16 0.42s } 12... exd4 { -2.08/18 1.1s } 13. Bxd4 { +1.41/17 0.67s }
13... c6 { -1.81/17 0.67s } 14. Ba4 { +1.42/17 0.57s } 14... Rxb2 {
-1.92/17 0.61s } 15. Ne2 { +1.68/17 0.75s } 15... Rb6 { -1.48/14 0.59s }
16. Bxb6 { +1.74/17 1.0s } 16... axb6 { -1.42/16 0.40s } 17. c4 { +1.50/17
1.6s } 17... b5 { -0.54/16 0.76s } 18. cxb5 { +2.05/16 0.51s } 18... d5 {
-0.48/15 0.48s } 19. O-O { +2.14/16 0.33s } 19... Bxg4 { -0.24/14 0.54s }
20. hxg4 { +2.44/17 0.49s } 20... Nxg4 { +2.12/14 0.39s } 21. Ng3 {
+2.67/18 0.41s } 21... Qh4 { +2.40/15 0.50s } 22. Kg2 { +2.78/17 0.33s }
22... Qh2+ { +2.40/14 0.41s } 23. Kf3 { +2.86/17 0.39s } 23... Bd6 {
+2.05/15 0.33s } 24. Kxg4 { 0.00/16 0.43s } 24... Qg2 { +17.37/14 0.30s }
25. Kh4 { 0.00/17 1.2s } 25... Re5 { +M11/29 0.40s } 26. Rh1 { -46.43/26
0.55s } 26... Be7+ { +M9/39 0.27s } 27. Kg4 { -46.44/30 0.28s } 27... Rg5+
{ +M7/52 0.26s } 28. Kh4 { -46.44/31 0.23s } 28... Rxg3+ { +M5/82 0.17s }
29. Kh5 { -46.45/33 0.25s } 29... Rg5+ { +M3/116 0.027s } 30. Kh4 {
-46.45/33 0.26s } 30... Re5# { +M1/116 0.020s, Black mates } 0-1

[/pgn]

Unbelievable game IMO.
The whole time white must ask himself WHY on earth is black doing these strange moves.
When the mate announcement came it saw the mess.

Black was playing with white like a cat plays with a prey, a mouse, before it eats the mouse.

It seems nothing has changed much from earlier days. Chess engines still count material. They don’t understand that chess is NOT about having the most material but MATING the opponent king.

[Eduard]

Oh guys depth 16 from Stockfish (and maybe Ponder OFF) - that's a joke. How can you get excited about games like this? You live on another planet!

Well, some of us engine programmers are interested in toppling the long time champion from her number one spot and to do that we have to know where her weaknesses lie. Since it is very difficult trying to work out what a chess engine is thinking/doing from incomprehensibly deep lines, it helps to play fast games and look for her problems with shallower lines. What these games show is the very high pruning and type of pruning that SF does, namely very materialistic. What does SF miss? Clearly elements related to king safety. So now we have a line of attack. Theee fast games show the line of attack working, which gives us confidence (or excitement or whatever you want to call it) to continue. Plus it’s fun.

Oh well,
you're welcome to do it, have fun! Stockfish achieves incredible search depths on modern computers and currently outperforms all other engines. On the server with 8 cores in the middle game with Stockfish I achieve a search depth of about 40 plies and more.

I ran Rebel 16.2 through my test suite (at 60s per move) but in all tactical positions Stockfish is better and faster.

You pick out a few individual positions here and are fascinated by what happens on lame hardware and bullet (16 plies). Have fun.

[chrisw]

Oh guys depth 16 from Stockfish (and maybe Ponder OFF) - that's a joke. How can you get excited about games like this? You live on another planet!

Well, some of us engine programmers are interested in toppling the long time champion from her number one spot and to do that we have to know where her weaknesses lie. Since it is very difficult trying to work out what a chess engine is thinking/doing from incomprehensibly deep lines, it helps to play fast games and look for her problems with shallower lines. What these games show is the very high pruning and type of pruning that SF does, namely very materialistic. What does SF miss? Clearly elements related to king safety. So now we have a line of attack. Theee fast games show the line of attack working, which gives us confidence (or excitement or whatever you want to call it) to continue. Plus it’s fun.

Oh well,
you're welcome to do it, have fun! Stockfish achieves incredible search depths on modern computers and currently outperforms all other engines. On the server with 8 cores in the middle game with Stockfish I achieve a search depth of about 40 plies and more.

You don't achieve it, Stockfish does. Stockfish is a very powerful and fine engine, no doubt about it.

I ran Rebel 16.2 through my test suite (at 60s per move) but in all tactical positions Stockfish is better and faster.

The games shown are not by Rebel, but by Chess System Tal

You pick out a few individual positions here and are fascinated by what happens on lame hardware and bullet (16 plies). Have fun.

Games, not positions. Games are a sequence of moves, with plan and theme.

The lame hardware is 15K Euros worth of high end AMD threadripper, 256 Gb RAM, 3090 GPU etc. etc.

Moves in these short games give about 400 or 500 ms per move, SF goes maybe 15-18 iterations deep in that time, chekc the m,ove times, they're in the PGN. Probably comparable to Deep Blue 1997, 3 minutes per move, Grandmaster strength, give or take a few hundred Elo. Give SF 20ms per move and it would probably destroy both you and me every game, and here it gets way more than 20ms, I don't buy this 'bullet' excuse.

[Eduard]

What excuse? I play on the server. Come by with your engine and your 15.000 euro computer. We play Bullet. I allow you to use all books, I run Stockfish with 8 cores without book. We play 100 bullet games. OK?

I might have tested it myself with Rebel 16.2 on the server. Unfortunately I don't understand the options for Move Overhead. The default value is 7 and can be adjusted between 0 and 25.

Is that 25ms? To be able to play on the server because of LAG I need an overhead of 300ms.

In the past more than 20 years ago there were no fast machines. However, the technology industry has made extreme progress. 20 years ago nobody could have imagined such devices as smartphones. Today everyone has a smartphone. Today everyone has a fast PC. Why shouldn't programmers use this technique to create fast engines? The neural networks are also very new. Before 2017 nobody could have imagined an engine like Lc0 or Stockfish NNUE.

Why should Stockfish be programmed differently? Stockfish uses state-of-the-art technology and is an amazing modern engine. But you wanted to bring back the past. Some here are talking about great knowledge-based engines that understand chess better than stockfish. That may be the case, but on modern computers Stockfish is worlds better, also tactically! I therefore see no point in dealing with engines that dominated the computer chess scene 25 years ago.

I think it's great what's happening with Lc0. Unfortunately, graphics cards are expensive and eat up too much power. The programmers of Komodo Dragon do it better. Great progress has been made in the last 3 years. Ethereal is also interesting. These are engines with a future.