When looking at alternative moves or continations, how do you guestimate how the opponent plays?

He might have been a little more condensed in how he said that, but that's roughly what I took him to mean.  Maybe he'll come along and say otherwise.

I assume he was being precise in saying "mislead," which is different from saying the engine is wrong.  It's not about whether the engine is objectively right about a position -- it's about when it is and isn't a useful teacher (and in some positions it is -- I'm guessing he'd agree on that).  I take that to be his point in the second instance too.  I don't think he meant the engine would be wrong about *that* particular position, but rather that it can be really hard to move from what the engine says about one position (which is based on really intricate wheels-within-wheels calculations whose nuances it can be almost impossible for us to fathom), and generalize it as a pattern we're supposed to recognize and learn from in future positions.

 Well said.

On the specific question of the usefulness as Engine as Teacher: I am wonder if others use the 3200 Max engine as I do when practicing against it; and that is, decide on my best move, and before I commit to it, to try to guess what move the computer would select before I make my response move - and hit the button to reveal the computer's idea of the best move.

When my decided on move exactly matches the engine upon reveal, I figure I am doing well.

When the computer's answer differs from my own, I stop, study, and try to understand why the computer's move is superior - assuming that is it superior - and see what I can learn.

QUESTIONS:

1. I would like to know if this is how others practice.

One thing that is very odd:

When the engine will reveal a "best move" different from my own; and yet, occasionally when I play the computer selected move, the review immediately returns a "Blunder" evaluation for it's own selected move... ;

AND,

2. Does anyone know what the story is with that oddity?

In my opinion, one of the main defects of engines (and one of the main arguments AGAINST using engines for strategic advice rather than for blunder-checking) is that they seem to have no understanding of the concept of counter-play.

When a Human player manages to get a winning material advantage (let's say, a piece ahead for near-zero compensation), his first priority will be to defuse the situation on the board and eliminate the opponent's active counter-play. Doing this will make sure that the opponent cannot muddy the waters, and will ensure that your winning advantage is preserved instead of being thrown away in confused tactics.

Computers fail to recognize this basic principle, and continue to play the same way after gaining a winning material advantage as they do in equal positions.

Yes, that's well put.  It must be nice being able to play insane positions without any risk.

A corollary is that they're not very helpful in suggesting practical ways to play for chances if you're in a losing position.

Or to put it another way, I think a 2500 human would be a lot more likely to beat me with rook odds than a 3500 computer.

I think you are wrong about this. First of all, if the position contains tactics that the engine can't calculate through, the engine won't go in there. Engines are not known for intuitive attacks, that is a human thing. So the engine isn't risking the advantage "being thrown away in confused tactics".

If the engine (or a human) fully calculates a tactical line, there is no risk in it. There are games of Tal or Kasparov when they are already ahead of material and still continue to play tactically because they calculated it all and they are confident that they are not wrong. That is what the engine does, too.

When you write "one of the main arguments AGAINST using engines for strategic advice rather than for blunder-checking" I have a very hard time interpreting that thought too. How could you even use the engine for strategic advice? To me, it seems almost completely impossible.

The engine was not designed to teach players. An attempt to use simple methods to interpret the engine's lines and transform it to advice (like chess.com's "The Coach") is bound to be a ridiculous failure. Teaching chess is hard.

On the other hand, useful lessons can be learned from engine lines all the time. There are many examples of defending seemingly hopeless positions, and there are many possibilities of brilliant attacks too. The most useful lesson is that chess is very-very rich and deep, much more than most people realize. 

I will show you an example:

White is up a piece. The safe, human move is simply 25. Qxc8, trading queens. Stockfish likes this, it is his 3rde choice. The engine's top choice is Rxh6+ leading to a mating attack. 25 Rxh6+ Kxh6 (the only legal move) 26. Rd3! (not really a queen sacrifice, as it threatens mate-in-1. Black can't escape.

It is easy to dismiss this as an oddity, a weird engine line that has no human significance. In the real game played between 2 humans (a blindfold simul in Paris, 1858) Paul Morphy as white played 25. Rxh6+ Kxh6 26. Rd3. He was playing on 8 boards blindfolded, and still chose the ridiculous engine line because he calculated it.

And how is this useful for a 1000 player who lacks any trace of Morphy's ability?

None of this addresses the point I made above, that there are many players on this site... including some rated over 2100, as hard as that might be to believe... who honestly seem to believe that engines offer useful advice in the opening, with remarks like:

"You shouldn't play the King's Indian Defense as Black! AlphaZero gives White a +0.9, so that whole line is unsound."

Opening theory is based partly on calculation and partly on concepts. Engines are great at one, but utterly worthless at the other.

It is useful. Showing brilliant games to students has a many many advantages. It teaches players to play with an open mind, looking for unexpected tactics. Studying those tactics is also very useful at getting better at chess. Finally, it also shows that chess is not boring, despite what a superficial spectator might think.

As for your comments on opening prep, I honestly don't know how those apply here. I agree with you on what you say, but I fail to see how it is even tangentially significant in this topic.

Yes, but why?  Nobody in this topic was talking about opening prep until llama suddenly started talking about it out of the blue, completely offtopic.

Well, how do you calculate what _you_ will play when in that situation? If you can do it for your side, then turn the board around and do the same thing from your opponent's perspective. Seeing moves for your opponent that are as good as the moves you see for yourself is called objectivity. One way to develop objectivity is by playing superior players. Because they will teach you to be afraid of the move you didn't see and to look for it before it happens.

I mostly use engines to blunder check. But as for move preference, I don't concern myself with the engine's top choices. The engine will choose lines and moves that I often find uncomfortable, or too difficult to navigate, so I don't torture myself with those things.

Trying to mimic the engine is a losing game, in my opinion - it would be like seeing a forklift carry thousands of pounds of bricks, then expecting myself to do the same ...

I know my limits as a human player, so I don't try to train myself to be something I'm not. I use the engine as a "find where I blundered" tool, but not much more than that. (Otherwise, my sanity suffers. )

That example might not show what you think it shows. In a blindfold simul, every game in progress represents a cognitive load, and the possibility of error goes up as some power (greater than 1) of the number of games. So rather than being a "ridiculous engine line", the forced mate is an efficient way of terminating one of the games quickly, which makes the probability of error in the remaining games go down by rather more than the probability of having miscalculated in the game under consideration.

In other words, a master might choose a prosaic endgame in a single game against a master, and a brutal sacrifice in a blindfold simul against an amateur, in each case for the same reason -- it's the most pragmatic way to win.

By the way, Rxh6+ is the first move I looked at in the diagram. Firstly, because doing many studies, puzzles, etc. has trained me to look for flashy moves in any diagram. Secondly, I'm not sure but I think I've solved this position before.  But not being Morphy, it took me a bit to prove the mate to myself after 1.Rxh6+ Kxh6 2.Rd3 Kh5, because I started with 3.Rh3+?!. I didn't even notice the engine try 2...Bc5+ 3.Kh1 Be3!?, although no doubt Morphy did.

An instance from this evening:

Here, I'm a rook up, but neither side's king is particularly safe, and loose pieces abound.  I decided that, given my extra material, it's well worth giving back an exchange to simplify to a winning endgame, so I played Qxb7+ and won easily.  Now the engine does think this is good -- +2.5.  And it's even willing to give back the exchange.  But what it really wants to do is go Rxb7+ and then, after Qxb7, it wants to keep queens on with Qc5+ (+4 it says). 

The engine isn't wrong, of course -- with correct play, that's even more convincing and wins even more quickly,  But if I'd played it, it would have made me about eight times more likely to hang a rook, allow a perpetual, or blunder mate.

I usually disagree.

If sacing the exchange to eliminate counterplay causes your human opponent to resign within the next few moves, and alternatively, if the engine line causes your human opponent to have hope of a win and play for a long time, then the engine line is incorrect.

Sometimes it's useful to explore the engine lines, and push ourselves to accept more risk during games, but I call a move incorrect when it unnecessarily increases the opponent's winning chances.

Yeah, that type is a clear example.

Obviously there is a strong correlation between moves the engine likes and all the ways to play for a win (or draw) in a position, but what some people seem to not understand is there isn't a one to one relationship. Sometimes a 0.00 move is a win in practical play. Sometimes a mate in 10 for white is actually a loss for white. It just depends.

This isn't to say beginners (or I) shouldn't use an engine... just that sometimes it will lead us astray for reasons that are too difficult for us to understand.

It's also worth pointing out that objectively speaking a +5 position is not different from a +7 (or sometimes a +2) in the sense that they're all wins with perfect play. Centipawns are an artifact of an imperfect evaluation function. So even a person who can't understand practical play should be able to understand why 0.70 may not be better than 0.00 or why 7.00 may not be better than 5.00.

... I mean, this is so obvious to me it's hard for me to even argue about it. I think most players understand these things.

Yeah, sounds like we basically agree.  It doesn't matter so much whether you say the engine is "right" or "wrong" as it does to understand the context in which you're making that judgment -- that is, to udnerstand that "right" for an engine" might be "wrong" for a human. 

At this point, it's hard for us not to remember that in any position there exists an optimal "engine" way of playing, and maybe there's some utility to describing that as "right" (even if that "right" remains always in scare quotes) -- as long as we can distinguish it from pragmatically good ways for us falible humans to play.

Yeah, I agree with you and emu... mostly I'm posting these arguments with one person in mind