GM says Rybka & Fritz weaker than best GMs in classical

[Terry McCracken]

How well do computers understand both the position and the game?

I saw the win faster than a computer. I saw it on the spot!

[d]r1bq1r1k/1pp3bp/8/p3pp2/2P3pn/P1NPP3/1BQ2PPP/R3RBK1 w - - 0 20



Zoltan Ribli vs Eric Lobron

If instead of Ne2 as in the game, what if Nd5 were played with the idea of Nf6 blocking the path of black's QR. Just a quick look, but looks winning for white.

There's no Ne2, there's a Nd2 and it's forced...and Nd5?? is losing fast.

1. c4 e5 2. Nc3 Nf6 3. Nf3 Nc6 4. d3 Be7 5. e3 O-O 6. Be2 d5
7. cxd5 Nxd5 8. O-O Kh8 9. a3 f5 10. Bd2 Bf6 11. Qc2 a5
12. Rfe1 Nb6 13. b3 g5 14. Bc1 g4 15. Nd2 Bg7 16. Bb2 Ne7
17. Bf1 Ng6 18. Nc4 Nxc4 19. bxc4 Nh4 20. Rad1 Ra6 21. d4 Nf3+
22. gxf3 gxf3 23. dxe5 Qh4 24. Ne2 Rh6 25. h3 fxe2 26. Qxe2 f4
27. e6 Rg6+ 28. Kh2 f3 29. Qc2 Qg5 0-1

Well there was a 'Ne2'!

But, in my haste I gave white an extra move.

Sorry I misunderstood you. Yes there is a Ne2 but I thought you were looking at something else

Anyway, if you play 24. Nd5?? then Black mates. Is that what you meant by extra move?

Please be clear.

24. Nd5 & Nf6 in a single leap.

I was very pressed for time.

Thanks! I am sorry for my reaction due to a misunderstanding. My post should have been deleted...*Sigh*

[M ANSARI]

I think a lot of you still don't understand how much chess engines combined with hardware have changed ... they have changed A LOT from one or two years ago. I recentely built a very powerful machine running 8 cores at 4.8 Ghz ... it is amazing how much more powerful Rybka and Zappa play on this hardware. The days where a GM could play better positional chess are gone IMHO ... let us face it ... Rybka or Zappa will play better positional chess than even the best GM ... I have put Rybka 2.3.2a and Zappa in a few 30 0 games and the level of play is phenomenally strong .... and to think that Rybka 3.0 is even stronger ... I think the facts are obvious. A good GM can still give engines a good run for their money if the hardware is changed ... say a single core Athlon ... but on a Quadcore or an Octa ... sorry no chance.

Hala Majd,
Is your octa machine stable enough at 4.8 Ghz
What cooling system are you using

Hala Wael ... yes 100% stable. I am rebuilding it now to try to get 5 Ghz 24/7 as I had one core 10C cooler than the other due to better insulation. For cooling I am using a custom 2 x 1hp rotary cooling system specially designed for dual socket. It gets a good Fritzmark score of 42.9, which is the highest I have ever seen so far.

[M ANSARI]

That is of course 5...50 times intentionally exaggerated !
Of what use is positional understanding if a sequence of sub-optimal tactics throws the game out of the window ?

Matthias.

Again experience over guesswork here. Under the right circumstances, such as the old correspondence games Mike Valvo used to play against Deep Thought, you can see just how wide the gap actually is. Humans make _very_ few tactical mistakes in correspondence. And there their positional judgement is the critical component...

As I said, computer tactical accuracy is what is winning games today, not positional judgement where they are still sorely lacking. However, at today's speeds, tactical accuracy is obviously quite enough to more than compensate for the ugly positional mistakes (most of the time, although humans still win games).

Again experience over guesswork here really ?
Are you saying your ridiculous 10-fold to 100-fold better posisitional play of humans (without any statistics) is not guesswork ?

Humans make _very_ few tactical mistakes in correspondence chess?
Ridiculous ! It is the strong engines used for correspondence chess that "make _very_ few tactical mistakes" and you certainly know that.

Matthias.

Exactly !!! This comparison with correspondence chess is silly ... in corr chess humans will use computers HEAVILY to look for any move that could lead to a tactical mistake. Corr Chess should now be called Centaur Chess ... Corr Chess as per pre computer Corr Chess is hard to verify and would not work. If you look at the level of corr games before computers and now ... there are mysteriously less tactical mistakes. In pre computer corr chess, you would on many occasions see a deep tactical shot that was overlooked by someone ... not anymore. Give me a single core Rybka and I guarantee you that I can win and draw quite a few games against even an Octa ... just allow me to use my time to probe the lines that Rybka likes and that I think are weak ... but really that is simply using Rybka to beat Rybka.

[bob]

You must be kidding us Robert....
Very high quality chess at 1 second per move,the humans
Even if they move the piece with an eye click,they won't be able to play a high quality chess,sorry,in this life time....

No. I guess he means very high quality Chess in comparison with what computers do at 1 ply.
And he is right of course. Human intuition and ability for planning at these low depths is crucial.

something is wrong here,you can't compare time,1 second,with 1ply of a computer,he's comparing too totaly different things,or does he mean 1 second for the humans is comparable to 1-2 plys by the machines

Of course dumping the chess engines to 1-2 plys will affect horribly the playing strength,nothing new here....

What is wrong here is you are not reading. I said that at less than a second per move, a GM is _not_ searching. Yet he comes up with incredibly strong moves. Would you like to take me on using Rybka with a real search depth of one ply? Any player that has played a large number of games will overwhelm such a weak search, since it won't be very accurate tactically, and positionally it will be terribly weak...

When a human sees that envading a b2 hole with the queen is devastating he is searching by sight. He is aware of the defending pieces movements and just sees what is possible. He has conducted a search--not a deep calculation, but a type of search none the less, Qd8c8 Qc8a6 Qa6a3 Qa3b2. A computer has no sight. It must search in order to 'see' the board and the possibilities. Take away the computers search and then ofcourse the computer is stupid because it can not see that the queen can reach b2 and what it can do there. Search can not be excluded from the definition of what constitutes a computers positional understanding.

Sorry, but he is _not_ searching. He is doing something totally different. I have played chess for almost 50 years now. And while I don't use alpha/beta, I definitely "push pieces" in my head. But I push far _less_ when playing 5 minute chess. And even less when playing 1 second games. That is the difference between how the human brain works and how the computer works when both are playing chess. I can "feel" that my king is a bit unsafe without _any_ searching and play a move to shore up the defense without taking any time at all. GMs are so much better than that it is not funny... Just compare some GM blitz games to patzer blitz games to see how much better they are. Compare GM 1 minute games to patzer 10 minute games and see which is of higher quality...

In long chess, quite often a GM picks the move he is going to play after a second or so, although he spends a lot of time making sure that it does not fall to any tactical or positional flaw. Their branching factor is impossibly low compared to a computer chess program. Because the "hardware" is so different in what it can do and how it does it...

[Matthias Gemuh]

In long chess, quite often a GM picks the move he is going to play after a second or so, although he spends a lot of time making sure that it does not fall to any tactical or positional flaw. Their branching factor is impossibly low compared to a computer chess program.

A program picks the move it is going to play even faster than a GM, although it too spends a lot of time making sure that it does not fall to any tactical or positional flaw.

A GM's branching factor is _NOT_ lower compared to a computer chess program. A GM has to invest at least the same extra relative time as a program to see one ply deeper.
As the search gets deeper, the GM has far greater problems than the program memorizing his variations as they increase in number and length. So the GM's branching factor RISES (and that is bad) more than the program's.

Bob, you know all this and shouldn't be writing the opposite.

Matthias.

[George Tsavdaris]

A GM's branching factor is _NOT_ lower compared to a computer chess program.

Sorry but this is completely ridiculous and wrong!

A GM's branching factor is thousands of times lower(more effective) than that of a top Chess engine.

A GM looks a position and "mysteriously" knows what move he should investigate and analyze. If he finds something bad for him with this move, he has another move in his mind and analyzes this.
Both 2 these moves, would almost certainly be the right path to the position.
Computer on the other hand, needs to search every single crap move of the board, even for just a fraction of a second, only to see it's crap and reject it.
The GM does NOT search any such crap moves. He rejects them automatically by intuition.

As the search gets deeper, the GM has far greater problems than the program memorizing his variations as they increase in number and length.
So the GM's branching factor RISES (and that is bad) more than the program's.

I don't understand how the first implies the second?!
I mean the difficulty a GM has, compared to a computer, in memorizing variations, what has to do with the branching factor?

[Matthias Gemuh]

A GM's branching factor is _NOT_ lower compared to a computer chess program.

Sorry but this is completely ridiculous and wrong!

A GM's branching factor is thousands of times lower(more effective) than that of a top Chess engine.

A GM looks a position and "mysteriously" knows what move he should investigate and analyze. If he finds something bad for him with this move, he has another move in his mind and analyzes this.
Both 2 these moves, would almost certainly be the right path to the position.
Computer on the other hand, needs to search every single crap move of the board, even for just a fraction of a second, only to see it's crap and reject it.
The GM does NOT search any such crap moves. He rejects them automatically by intuition.

As the search gets deeper, the GM has far greater problems than the program memorizing his variations as they increase in number and length.
So the GM's branching factor RISES (and that is bad) more than the program's.

I don't understand how the first implies the second?!
I mean the difficulty a GM has, compared to a computer, in memorizing variations, what has to do with the branching factor?

Hi George,
you seem not to understand what a branching factor is.

Measure the time t1 that a GM needs to search 10 plys deep.
Measure the time t2 that he needs to search 11 plys deep.
His branching factor is the _ratio_ t2/t1.

This ratio is not worse for programs.

Do you now see why memorizing variations has a lot to do with the branching factor?

It gets worse for a human as search gets deeper.

Matthias.

[Dr.Wael Deeb]

I think a lot of you still don't understand how much chess engines combined with hardware have changed ... they have changed A LOT from one or two years ago. I recentely built a very powerful machine running 8 cores at 4.8 Ghz ... it is amazing how much more powerful Rybka and Zappa play on this hardware. The days where a GM could play better positional chess are gone IMHO ... let us face it ... Rybka or Zappa will play better positional chess than even the best GM ... I have put Rybka 2.3.2a and Zappa in a few 30 0 games and the level of play is phenomenally strong .... and to think that Rybka 3.0 is even stronger ... I think the facts are obvious. A good GM can still give engines a good run for their money if the hardware is changed ... say a single core Athlon ... but on a Quadcore or an Octa ... sorry no chance.

Hala Majd,
Is your octa machine stable enough at 4.8 Ghz
What cooling system are you using

Hala Wael ... yes 100% stable. I am rebuilding it now to try to get 5 Ghz 24/7 as I had one core 10C cooler than the other due to better insulation. For cooling I am using a custom 2 x 1hp rotary cooling system specially designed for dual socket. It gets a good Fritzmark score of 42.9, which is the highest I have ever seen so far.

Wow,a 5 Ghz stable machine will distruct the opponents at the ChessBase server for sure
IIRC,you have a private optimized opening book for Rybka which should increase your chances to get a higher spot in the rating list....
Good luck for rebuilding your system

[Dr.Wael Deeb]

That is of course 5...50 times intentionally exaggerated !
Of what use is positional understanding if a sequence of sub-optimal tactics throws the game out of the window ?

Matthias.

Again experience over guesswork here. Under the right circumstances, such as the old correspondence games Mike Valvo used to play against Deep Thought, you can see just how wide the gap actually is. Humans make _very_ few tactical mistakes in correspondence. And there their positional judgement is the critical component...

As I said, computer tactical accuracy is what is winning games today, not positional judgement where they are still sorely lacking. However, at today's speeds, tactical accuracy is obviously quite enough to more than compensate for the ugly positional mistakes (most of the time, although humans still win games).

Again experience over guesswork here really ?
Are you saying your ridiculous 10-fold to 100-fold better posisitional play of humans (without any statistics) is not guesswork ?

Humans make _very_ few tactical mistakes in correspondence chess?
Ridiculous ! It is the strong engines used for correspondence chess that "make _very_ few tactical mistakes" and you certainly know that.

Matthias.

Exactly !!! This comparison with correspondence chess is silly ... in corr chess humans will use computers HEAVILY to look for any move that could lead to a tactical mistake. Corr Chess should now be called Centaur Chess ... Corr Chess as per pre computer Corr Chess is hard to verify and would not work. If you look at the level of corr games before computers and now ... there are mysteriously less tactical mistakes. In pre computer corr chess, you would on many occasions see a deep tactical shot that was overlooked by someone ... not anymore. Give me a single core Rybka and I guarantee you that I can win and draw quite a few games against even an Octa ... just allow me to use my time to probe the lines that Rybka likes and that I think are weak ... but really that is simply using Rybka to beat Rybka.

The most accurate sentence posted in this thread....

[George Tsavdaris]

A GM's branching factor is _NOT_ lower compared to a computer chess program.

Sorry but this is completely ridiculous and wrong!

A GM's branching factor is thousands of times lower(more effective) than that of a top Chess engine.

A GM looks a position and "mysteriously" knows what move he should investigate and analyze. If he finds something bad for him with this move, he has another move in his mind and analyzes this.
Both 2 these moves, would almost certainly be the right path to the position.
Computer on the other hand, needs to search every single crap move of the board, even for just a fraction of a second, only to see it's crap and reject it.
The GM does NOT search any such crap moves. He rejects them automatically by intuition.

As the search gets deeper, the GM has far greater problems than the program memorizing his variations as they increase in number and length.
So the GM's branching factor RISES (and that is bad) more than the program's.

I don't understand how the first implies the second?!
I mean the difficulty a GM has, compared to a computer, in memorizing variations, what has to do with the branching factor?

Hi George,
you seem not to understand what a branching factor is.

Measure the time t1 that a GM needs to search 10 plys deep.
Measure the time t2 that he needs to search 11 plys deep.
His branching factor is the _ratio_ t2/t1.

I knew that branching factor was how many child nodes someone considers in each node.
Your definition is equivalent if we assume that total child nodes per second someone thinks/calculates, is steady.

But humans don't think exactly in this way. Humans get's confused many times and slow or stop the process of searching.

Moreover, what matters is not that humans spend time to remember variations that they have analyzed-- and due to MUCH worse memory abilities from computers they have forgotten or have been confused and have to analyze the same thing again-- but that they really choose to analyze only 1-2 moves in each case discarding all others by intuition(and this intuition is correct in 99.99% of times).

This ratio is not worse for programs.

Do you now see why memorizing variations has a lot to do with the branching factor?

It gets worse for a human as search gets deeper.

Yes i understand that what you referred to, was the effective branching factor of a human, and the limited memory abilities of humans makes it much worse than the theoretical branching factor value.

But i was talking mostly about the ability of humans to eliminate almost all possible moves of a position and just search 1-2 moves that in most cases it's the right path.
This is because a human GM by just looking at a position and without searching at all, he feels what is the path that should be followed in the game.
While computer's first plies are compared to this, complete crap.