God this is the most idiotic thread ever and gives a bad name to the chess forum IMO.. Carlsen has as much of a chance against Rybka as my left nut sack. You gotta be a fu**en moron to think otherwise.

It's incidental that the numerical score does not matter in this position. The fact that white thinks it's crushing here is a huge issue. That means given the opportunity to, white would jump at obtaining this position - even if black was lost, but had more material, just a few moves prior.

Can you try and construct a position prior to this where a win existed for white, but where Rybka (not Chessmaster 1.0 or some other garbage) decided to go for this line instead?

I will put up some money against if you'd like to make it a bet...

If you'll put money up, I'll likely be more than happy to do it. Otherwise it's not really worth my time.

How much would it take? I will escrow up to $10k to anyone reliable on FTP, as long as you do the same, at 1:1 odds.
If you are prepared to offer odds we can do the same bet where Chessmaster or some other crappy program has white.

Here are the terms as I see them:

The final position that Rybka / Chessmaster gets to has to be one with this exact pawn structure, the black king on f8 or e7, and the white king and bishop anywhere inconsequential.

The original position has to be winnable for white. Rybka/Chessmaster is allowed to think about their move for 3 minutes. We agree on someone reliable to run Rybka latest version / chessmaster latest version on their machine, which has to be reasonably competent (doesn't need to be 8 cores, but some single core 2ghz crap is obviously not a reliable test of what the chess engine would actually do).

Not much. I'm not doing this for profit. I just want some incentive since it's probably not going to be most thrilling exercise!

Ok, well I have faith in computers so I'll back them for 'not much' then

You shouldn't have so much faith in them. Go to the Rybka site itself for a discussion with tons of positions where Rybka just totally sucks, particularly in endgames unsurprisingly.

I understand that, but we were discussing this specific position I thought.

No, I was discussing your "faith in computers."

As for the specific position, after a whopping 5 minutes I do not see an obvious way to force a blockade tactic that forces the identical pawn structure.

Yeah I get that, but my faith in computers was that Rybka would never turn a winning position into this position.

I also have faith that Rybka would never lose a game of chess played without handicap, and at regular time controls against any human opponent playing without computer assistance of any kind, but since such a match is never going to happen it's about as meaningful as having faith in God.

I understand that it is possible that there do exist positions where Rybka would currently draw after a given amount of time to think that some human players could win with the same thinking time. The last similar example I could find was a game from 2006 where an old version of Rybka swapped off rooks into a drawn B+pawns vs N+pawns ending where after lengthy analysis it turned out the rook ending was won.

Pyromantha, I don't have Rybka, only Fritz8, but I put the white king on a1, the bishop on d3 and added a white knight on d5, a black rook on e7 and a black pawn on a2. After 5 minutes Fritz still wants to play Nxe7? though Nf4 wins.

I don't believe that any program ever will be unbeatable. There is always a possibility to outprepare the program in a sharp variation like the Botvinnik. By outpreparing I mean preparing a novelty in a sharp variation and having a look beyond the horizon beforehand.

Funnily enough, my old Rybka 2.1 does want to play Nf4 here, and as a human player I don't see the win at all after it (not sure if Rybka does either actually but for whatever reason it likes that move). I only have the demo version anyway at the minute so it stops at ply 20

Something just occurred to me.

The top GMs are still much stronger than the top engines.

They just aren't playing long enough time controls because of tradition, which is silly. The next human player to play against a top engine should demand a match of 12+ hour games with several days' rest in between games.

Sure, I was making a weaker version of Pyro's claim: that the programmers may have specifically programmed the engine to avoid these types of "bad bishop" blockades. It's plausible that they might do so without trying to alter the evaluation function too much, much like they don't care how Rybka evaluates 5-man endgames because the tablebases take care of that.

I don't have a modern copy of Rybka, but I wouldn't be shocked if Rybka could stumble into this. It does seem like something the developers could fix though.

Do you have any idea how hard that would be for the humans? Humans get tired, engines do not. Fatigue would be a huge factor.

I agree with you that at correspondence-length time controls humans are probably still better, but otherwise your claims are hard to swallow. For what you propose: Carlsen probably doesn't even think he would win such a match.

The fatigue factor would be less than the current "blunder factor", especially if there were a 3-day break between games.

10k are you on the pipe, you'd likely lose this bet for sure. I'm sure you could just make it white to move, put a pawn on a4, move the bishop back to like a2. put a black bishop on b5 and a black pawn on a6....I'd be surprised if Rybka doesn't just take the bishop off, thus leading to insta-draw.

Seriously warning for everyone here if you aren't like at least a master of chess, don't make random chess bets with people who are much better than you, they will almost always win.

Is this bet open to anyone? I think 1-1 is actually pretty fair odds, maybe you have slightly the best of it assuming you aren't going up against an experienced endgame composer.

I might take the bet for the challenge, wouldn't consider putting up the whole 10k though.

Also this is an absurd bet because people can work on it before starting the bet, know the answer, and then accept the bet

Yeah, if this bet is serious there should have been a strict time constraint for accepting it, like 20 minutes from when it was originally posted.

I take your point, but Rybka plays a5 on my machine, and realises it's best after about 20 secs. Maybe you can add extra pieces to confuse it such that it would take more than the stipulated 3 mins to find, but obviously given long enough to think it is not going to fall for those kind of tricks.

Yes I realised that shortly after I posted it - would still be interested if anyone can find such a position but no cash at stake

Well given that it thinks the diagram position is totally winning, pretty sure there are some tricks it'll fall into.

What I was trying to get at and failing was that it shouldn't matter if it thinks the diagram position is winning if it sees an actual win in another line. For example in the position that you suggested it sees that it can queen it's a-pawn and gives a5 +20 or so after a while. If there is a genuine win in some other similar position, why wouldn't it be able to find that as well?

Is that wrong?

Because computers are terrible. If it can't see a "mate in x" then it will go for whichever position has the higher evaluation, or in more practical terms - which position its ahead the most material in. All one needs to do is to setup the position such that black in an unclear position, sacrifices material to get to this position. Since the computer is too bad to realize this position is drawn, it will jump on the 'free' material.