what can programs see that 2700 GM cannot

[syzygy]

so the question is how quick do the programs evaluate this as a blunder?

I think the better question is how quickly top GMs recognise 24...cxd4 as a blunder.

I am not sure what this proves, because you will find blunders in almost every engine game where one engine loses to the other.
Dan Corbit

Oh really?
I check most long time control games by programs and I don't find any.

You have to use a 3700 Elo engine to find the blunders.

[syzygy]

A more philosophical view :
It depends how deep one search : a "2 points" mistake becomes a "3 points" mistake when you search deeper. And all "mistakes" are "mate mistakes" if one search deep enough.

Exactly.

The only real difference between blunders is that the worst lose a full point and the rest only half a point.

[Dann Corbit]

A more philosophical view :
It depends how deep one search : a "2 points" mistake becomes a "3 points" mistake when you search deeper. And all "mistakes" are "mate mistakes" if one search deep enough.

Exactly.

The only real difference between blunders is that the worst lose a full point and the rest only half a point.

A one pawn advantage is devastating.
Consider the following SQL query:

Code: Select all

select sum(white_wins) as ww, sum(black_wins) as bw, sum(draws) as dr from Epd where Epd like '% w %' and ce > 100 and games > 0 
union all
select sum(white_wins) as ww, sum(black_wins) as bw, sum(draws) as dr from Epd where Epd like '% b %' and ce > 100 and games > 0 
union all
select sum(white_wins) as ww, sum(black_wins) as bw, sum(draws) as dr from Epd where Epd like '% w %' and ce < -100 and games > 0 
union all
select sum(white_wins) as ww, sum(black_wins) as bw, sum(draws) as dr from Epd where Epd like '% b %' and ce < -100 and games > 0 

And here is the result set:

Code: Select all

ww	bw	dr
875666	84520	298181
18203	576252	256026
8240	683142	53898
906904	42100	91539

[Laskos]

You have to use a 3700 Elo engine to find the blunders.

(of top engines).

Yes, and the problem of top GMs is that most of the times one needs a 2400 Elo engine to find many of their blunders. One can't deny that the aura of mystics regarding top GMs (still remember the adjourned games and overnight GM kibitzers) is gone for good.

[syzygy]

A one pawn advantage is devastating.

From the point of view of perfection any winning advantage is equivalent to any other winning advantage.

[syzygy]

You have to use a 3700 Elo engine to find the blunders.

(of top engines).

Yes, and the problem of top GMs is that most of the times one needs a 2400 Elo engine to find many of their blunders.

Sure, but at the same time any GM will be able to point out positional mistakes in games between top engines (especially if you give them a 2400 Elo engine to verify their analysis). The OP likely will not notice such errors when analysing engine-engine games, because the analysing engine won't tell him.

One can't deny that the aura of mystics regarding top GMs (still remember the adjourned games and overnight GM kibitzers) is gone for good.

Yes, if you're using Stockfish to analyse a live game between two humans, that is like watching a boxing match between two blind players.

But then again the same might apply to games between top engines if you analyse using either an engine that is simply much stronger, or some other kind of chess playing entity that excels in a particular area where the engines are still lacking.

I think it is normal to find tactical mistakes in games between humans, because a human can afford to take tactical risks when playing another human if he thinks he can calculate better. Humans are certainly much better at adapting to opponents than chess engines are.

[syzygy]

A one pawn advantage is devastating.

From the point of view of perfection any winning advantage is equivalent to any other winning advantage.

And, somewhat unfortunately, from the point of view of perfection there is no difference between a drawing move to which any patzer can find a sufficient reply and a drawing move that will trick the average GM into blundering away his draw.

So, ignoring "real blunders" (that change the game-theoretic value of the game), any grading of the "quality" of moves is subjective. Top engines will tend to have similar subjective evaluations of such moves, but why should the subjectivity of top engines be the measure for evaluating the quality of human moves?

[rabbits23]

Possibly Uri possibly. I don't have enough knowledge or interest in
chess to explore any deeper. The comments I made are simply my response to the position.
Allan

[carldaman]

check out 23.Qf3, a 2600 move. another blunder from chessbase.
now show me a one move program blunder, I have shown you two but I have a thousand more

By your definition in every game that a program lose there is a blunder assuming the opening position is a draw(if you use the perfect player to analyze).
One move change the real score from draw to mate against the losing program.

If you use programs of the future they will show you the blunder and the evaluation may be changed from draw to mate in 30 against the program in one move.

I think that definition of a blunder by the number of pawns you lose based on the engines of today on the hardware of today is clearly wrong.

For me a blunder is a move when it is easy to understand why it is wrong.
A move that lose even if it lose the queen for nothing based on engines is not a blunder if humans cannot understand fast why it lose the queen.

Uri

I like your definition, but I would add to it.
A move is a blunder if one computer sees instantly that it loses but the other computer does not see it.

A blunder should also jeopardize the outcome of a game, by (potentially) turning a win into a draw or loss, or a draw into a loss.

Regards,
CL

[PaulieD]

But there are many examples of where blunders were made and it did not effect the overall score. Probably because of a newly coined term (co-blundering) ...giving a blunder back.