What kind of game is chess?

[Peter Berger]

It is a tactical game, basically ?

Or is it a strategic game? Or if it is a tactical one, is it mostly about Zugzwang?

I was considering a post with a lot of interesting positions, but you guys do spend +way+ more time with these things than me, so why spend time? And I haven't seen this question discussed here.

I have watched a lot of slow games by now on my old Laptop, mainly facing the BLAS version of LC0 aganst Crafty. The latest 6+ nets get about 22NPS here at very slow time controls. And LC0 is incredibly weak tactically, it really is. Yesterday I watched a game and I suddenly realized: if LC0 really plays this move, this is mate in 4. It did, and right I was; as Crafty pointed out immediately..

But it does win quite a lot of games, too. When it is about games against Crafty, nullmove situations hurt Crafty IMHO. LC0 creates a lot of situations where there is some kind of zugzwang. Is there sth like a Stockfish without Nullmove?

[hgm]

LC0 is weak at tactics. But it knows that, and has learned it should better avoid tactically complex positions.

Conventional engines are good at tactics. But they are not aware of that. Their evaluation is just as happy in tactically quiet positions as in tactically complex ones. They never learned that tactics is good for them, because they were tuned by playing opponents that also are good at tactics, offsetting the advantage.

So they naively allow LC0 to take them to tactically quiet positions, where they cannot use their major (only?) weapon.

[Uri Blass]

LC0 is weak at tactics. But it knows that, and has learned it should better avoid tactically complex positions.

Conventional engines are good at tactics. But they are not aware of that. Their evaluation is just as happy in tactically quiet positions as in tactically complex ones. They never learned that tactics is good for them, because they were tuned by playing opponents that also are good at tactics, offsetting the advantage.

So they naively allow LC0 to take them to tactically quiet positions, where they cannot use their major (only?) weapon.

I wonder how lc0 learned that is should avoid tactically complex positions.

I understood that it learned only by playing against itself and not by playing against opponents that are stronger than it tactically so based on result of the games it had no reason to believe tactical positions are bad for it.

[hgm]

If it gets in tactically complex positions during self-play the result would be like a blind gamble. This pushes the result expectancy towards 50%. Which would be bad is it is ahead. This at least should learn it to avoid tactics when ahead.

After it knows this, it shouldn't often get into tactically complex positions anymore even when it is behind, because the opponent would actively try to prevent that. What it does when it is behind is of less important anyway, because it is already likely to lose, and not much is at stake.

[alchemist888]

I am absolutely sure that author of the topic doesn't mean the tactical strength of LCZero, but he appreciated her feature when she doesn't realize the advantage quickly. For example an AB engine sees checkmate in 4 moves, but LCZero makes another 50 moves and gives checkmate in endgame like KRK, KQK, KBBK, KBNK etc. The same thing with forced wins. Many people still don't understand that this is a feature of the MCTS. It doesn’t matter to MCTS how quickly you wins, the main thing is hight winning probability

In fact, the tactical capabilities of LCZero are superior to most AB engines. I am sure that LCZero will significantly outperform Crafty in tactical tests, if in every puzzle a tactic will be the only way to win.

[Uri Blass]

when it is behind it should learn to try to force tactically complex positions based on the same logic.
behind is not only when you already expect to lose and behind can be also when the expected result is 30% or 40%.

There are positions when the stronger side cannot keep the advantage and also avoid tactically complex positions so lc0 may learn not to suffer in a quiet line that it make 40% against itself but go to some dubious tactical line that make the expected result against itself higher but does not make the expected result against ab engines higher.

[jp]

For example an AB engine sees checkmate in 4 moves, but LCZero makes another 50 moves and gives checkmate in endgame like KRK, KQK, KBBK, KBNK etc. The same thing with forced wins. Many people still don't understand that this is a feature of the MCTS. It doesn’t matter to MCTS how quickly you wins, the main thing is hight winning probability

This excuse for Lc0's endgame play has been made many times on this board, but Lc0 can also draw winning endgames and lose drawn endgames.

[hgm]

And this has nothing to do with MCTS; it is just a consequence of a faulty evaluation (failing to reward progress). Which again is a consequence of a flawed training procedure.

A-B engines suffer from exactly the same problem, once you similarly corrupt their static evaluation (e.g. by not having a centralizing PST for the Kings in KBNK).

[alchemist888]

For example an AB engine sees checkmate in 4 moves, but LCZero makes another 50 moves and gives checkmate in endgame like KRK, KQK, KBBK, KBNK etc. The same thing with forced wins. Many people still don't understand that this is a feature of the MCTS. It doesn’t matter to MCTS how quickly you wins, the main thing is hight winning probability

This excuse for Lc0's endgame play has been made many times on this board, but Lc0 can also draw winning endgames and lose drawn endgames.

Any engine can draw winning endgames and lose drawn endgames. My post is not about that at all. If LCZero finds that her chances are very high, she almost always wins. I remember the only one case when Terminator net missed technically winning endgame in KRNPvKRP. Do you have more examples?

[Look]

It is a tactical game, basically ?

Or is it a strategic game? Or if it is a tactical one, is it mostly about Zugzwang?

I was considering a post with a lot of interesting positions, but you guys do spend +way+ more time with these things than me, so why spend time? And I haven't seen this question discussed here.

I have watched a lot of slow games by now on my old Laptop, mainly facing the BLAS version of LC0 aganst Crafty. The latest 6+ nets get about 22NPS here at very slow time controls. And LC0 is incredibly weak tactically, it really is. Yesterday I watched a game and I suddenly realized: if LC0 really plays this move, this is mate in 4. It did, and right I was; as Crafty pointed out immediately..

But it does win quite a lot of games, too. When it is about games against Crafty, nullmove situations hurt Crafty IMHO. LC0 creates a lot of situations where there is some kind of zugzwang. Is there sth like a Stockfish without Nullmove?

In Chess one needs basics of tactics and strategy. Both of them . For new learners book I have read and suggest "My System" from Nimzowitsch.