Something I've noticed is even older chess puzzles never seem to age, somehow they even seem more masterful and demanding then the puzzles created today, but what if the solution for a puzzle is refuted by a program such as AlphaZero, or when quantum super computers are made, stronger engines, different engines?Will there be any shift with established puzzles, or do we just get to enjoy new theory and different puzzles?
Our modern chess engines are already perfect for solving tactical puzzles. @Bad_Bishop's example is an easy problem for Stockfish.
If you're talking about artificial deep chess problems (not puzzles), even AlphaZero might have problems. Some chess engines (e.g. Houdini) has a "tactic" mode for those problems.
No. Nothing much will change with stronger engines.
How about composing new problems with the help of engines? This should be much easier now than in the old days, Feb 15, 2018 at 9:07
"perfect" is a bit exaggerated here. "extremely strong" is more accurate, for instance I don't think any software to be capable yet of solving any mate in 50 or 100 moves like the ones created from the Blatny matrix. Most probably most cooked puzzles from history are already within the grasp of today's computers (and even of computers from 2005) , and that's why new softwares won't probably cause any major changes, but it doesn't mean that they wouldn't bring some marginal corrections.– EvargaloFeb 15, 2018 at 9:16
Most probably a wide majority of cooked puzzles from history are already within the grasp of today's computers (and even of computers from, say, 2005), and that's why new softwares shouldn't bring any major changes, but it doesn't mean that they wouldn't bring some marginal corrections.
A funny example is given by Tim Krabbé in entry 324, "A cooked, correct study", of his chess diary.
[FEN "8/8/7r/3N4/8/8/7P/R3K2k w - - 0 1"]
White to play and win - E. Pogosyants, EG 1978
From the human point of view, the single solution is the composer's intention: 1.Ne3 Rxh2 2.0-0-0#.
From the tablebase point of view, since castling is not implemented (at least, such was the case in 2006, it might have changed...), this solution is impossible. However, there is also a single winning move: 1.h4! after which the KRNKR endgame is a mate in 33. Less poetic, sure.
So this is a study with two solutions. A human solution that is beyond the grasp of the tablebase, and a tablebase solution that is beyond the grasp of humans.
The main but pointless point is that the puzzle has changed because of the computer assessment.
What happens if after Ne3 black moves the rook somewhere? Still be winning for sure, but not pretty either, is it? Feb 15, 2018 at 9:04
Depends on the "somewhere". In most case 2.Nf1 just keeps the Ph2 with a winning material advantage. If 1...Rh3 2.Kf2 Kxh2 3.Ng4#. It remains only 1...Re6, pinning the knight, but 2.h4 should do the trick. 2...Rxe3+ 3.Kf2+ and other moves are met by 3.Ra4 or 3.Kf2+ (and if need be 4.Ng2). Sure, not that pretty, but since it is a secondary line of the study it doesn't spoil it. In any case, not in the way the comp's 1.h4! spoils it...– EvargaloFeb 15, 2018 at 9:12
This puzzle is easy for current engines. Stronger engines wouldn't do anything better, which is the question.– ABCDFeb 15, 2018 at 10:52
1FYI, although egtb doesn't include casteling, most engines by now have support for probing the tables in a way to make it work. Jan 16, 2019 at 18:29