Well, I thought it was slow. Also one of the best examples of coordinating a 5 bishop attack I've ever seen. Makes me wonder, what was the longest game ever in terms of move count?

I'm going to have to query Rybka's 174th move there. It's an interesting game, since N. has to not only close the position, but then achieve enough of a material deficit that the engine will self-destruct rather than take the draw.

If you count informal games, I'm sure two people on ICC have played a "game" just to get the move count to break 1000. If you collude, you don't even have to worry about the 50 move rule or repetition. Otherwise, wikipedia.

59. ... Re6! and 110. ... Qe6!! are clearly a BRILLIANT pair of exchange sacs

Seriously though, I'm confused about the bug he's exploiting. I get that Rybka would make a bad move to avoid a draw when it thinks it's winning... but why would it make a losing move when it thinks the position is drawn? (180. Rh3, 181. Kb2, 182. Re1) I mean I'm looking at it with Fritz, not Rybka, but each of those three moves come in a position that Fritz evaluates at (0.00), whereas after the move Fritz evaluates black as being ahead. Does Rybka see it differently? Because those moves have me confused...

I'm assuming this is a product of it being a blitz game. Fritz sees evaluates the position at (0.00) after either 174. c4 or after any of the other moves that end things instantly in a draw, if you give it time. For the first 20 seconds or so, though, it has 174. c4 as being slightly better than drawing (0.09, then 0.06). I assume Rybka evaluates it similarly.

So us humans are not allowed to exploit the computers' weaknesses? No wonder we can't beat them. Anyhow...

I won't be able to find a slow, building attack where Naka beat Rybka but I observed it as it happened, so you can choose to believe me or not. It wasn't a BS closed position either, Naka just went for a slow kingside attack a la the Vienna or a Closed Sicilian or something similar and sacced two pieces and mated the beast. Like I said, believe it or not, but I saw it happen.

On my machine Rybka thinks White is better after 174. c4 for at least 30 seconds (+.8, then +.3). I think this particular bug is a Rybka-specific thing, so I would expect Fritz to evaluate the positions more accurately.

Of course they're "allowed" to.

But this one is a bit like if Nakamura told you he can outrun a racehorse and set it up as a prop bet. The race starts and they both start out of the gates together, then the horse panics and throws its rider.

As you are paying up your lost money, you ask him how he did it. He lets you in on the secret -- he knew this racehorse was terribly afraid of dogs, so he started making barking noises as soon as they got out of the gate and the horse panicked.

It's still impressive -- he had to figure out the horse's weakness, learn how to make a convincing barking sound, and run fast enough to keep within shouting distance of the horse for the first 20 yards or so. All that and he also had to find a sucker who would take his bet. But it's not as impressive as a guy actually running a 2-minute mile.

No, it's absolutely nothing like that. The computer has a weakness, Naka figured it out, Naka won. End.

That's what I said!

Ruff Ruff Ruff Ruff

if I had to bet money on it, I think that its >50% chance that DR3 on my octal with 48h/move could draw 99/100 games against anything the future will turn up.

Computer vs computer is only exciting with shorter time controls today. magnus has no chance of winning against rybka3 on longer time controls... ever....

It sounded like you were trying to convince me that the win was somehow hollow or illegitimate. It was nothing of the sort. Sure, it's a very strange weakness to exploit, but I don't think it's something that the programmers could just insta-fix (or they would have after seeing all of Naka's games). So it's a legitimate weakness that the GM detected and used to win. End of story.

Meow meow meow.

How the hell do you do that in three minutes??????????

Premove.

Also: no need to quote that giant wall of text, and the last 9 question marks are superfluous!!!!!!!!!!

I'm skeptical of this claim. I would guess that the top correspondence masters would still be able to beat Rybka in a game at those time controls. Does anyone know if this has been done?

I wonder, actually, if it would really be that hard. Probably not -- I'd guess that the main reason is that the performance hit isn't worth it, and that the engine is designed for longer time controls anyway. That is, would rybka make the same mistake if it had a few more seconds per move? My guess is that the developers aren't principally worried about losing a few blitz games now and then...

The top correspondence players are nowhere near as good as the top players, if they played correspondence chess.

Anyway, players would still make tactical mistakes, even given all the time in the world.

I think a computer is much more likely to make a tactical mistake than a strong human, given something like 3 days per move. It's kind of strange. The primary thing computers have going for them is their tactical calculation, but they still are very far off of their evaluation in lots of the sharpest lines such as the Botvinnik variation of the semi-slav. Although humans, in turn, rely extensively on the computers to get them through those variations that the computer's in turn completely misevaluate after the position settles.

So many paradoxes in computer chess.

This was an interesting debate 7-10 years ago, but with the top engine playing ~500 ELO above the top human at standard time controls, there really is no longer any comparison beyond digging up obscure positions or games where an engine did something "stupid".

In response to the post above me, I don't think it is the case that the top engine(s) are positionally weak anymore. Rybka plays accurately without a book in virtually every opening in my experience. And I have not seen it make any "tactical" errors given long time controls.

I also don't think computers are slowing down any time soon. Hardware improvements are certainly not slowing down, and there is still plenty of room for algorithmic improvement in chess engines in my humble opinion.

Excellent post willyc, I agree with virtually everything you wrote.

It's pointless to argue (and I don't really have an opinion on) your first claim, although the character of the games is somewhat different. Nonetheless it's irrelevant -- I only mean to say that humanity's best at these sorts of time controls, whoever that may be, is likely still better than the best engines.

Keep in mind that (generally speaking) in international correspondence chess (as opposed to, say, the US Chess Fed. rules) there are no restrictions on outside aid, and so people pay assisted by computers. So I'm sure that there's data on people who entered and just played whatever their computer told them to, and I doubt that they would win any high-caliber events.

How can you claim that the best OTB players would be better than the best correspondence players if they played correspondence chess? That's so absurd. It's like saying that the best 400m runners would be better than the best 5000m runners if they chose to run that event.

I base my statement by being part of the rybka vs hiarcs and rybka vs the gladiators 48h/move matches which both drawed. it was really strong play by both sides and the evalutation always got closer to 0.0 the deeper the evaluation went. it's really really hard to beat rybka3 at longer time controls on modern hardware. no human alive today have any chance of doing this.

Im sorry but its definitely true. If Anand wanted to be world correspondence champion he could do so easily.

[ ] good analogy

No, it's not true. Think about all the weak GMs and strong IMs who are great analysts and find holes in OTB games and stuff. The best analysts are the best correspondence players. Maybe Anand in particular could be CC world champion if he wanted to, but that's not the point.

Then what is the point? (seriously, I'm just wondering)