Interesting tidbit I once learned from a chess livestream. Even human super-GMs have a really hard time "scoring" or "solving" extremely weird positions. That is, positions that shouldn't come from logical opening - mid game - end game regular play.

It's absolutely amazing to see a super-GM (in that case it was Hikaru) see a position, and basically "play-by-play" it from the beginning, to show people how they got in that position. It wasn't his game btw. But later in that same video when asked he explained what I wrote in the first paragraph. It works with proper games, but it rarely works with weird random chess puzzles, as he put it. Or, in other words, chess puzzles that come from real games are much better than "randomly generated", and make more sense even to the best of humans.

"Even human super-GMs have a really hard time "scoring" or "solving" extremely weird positions. "

I can sort of confirm that. I never learned all the formal theoretical standard chess strategies except for the basic ones. So when playing against really good players, way above my level, I could win sometimes (or allmost) simply by making unconventional (dumb by normal strategy) moves in the beginning - resulting in a non standard game where I could apply pressure in a way the opponent was not prepared for (also they underestimated me after the initial dumb moves). For me, the unconventional game was just like a standard game, I had no routine - but for the experienced one, it was way more challenging. But then of course in the standard situations, to which allmost every chess game evolves to - they destroyed me, simply for experience and routine.

That Expert players are better at recreate real games than ‘fake’ positions is one of the things Adriaan de Groot (https://en.wikipedia.org/wiki/Adriaan_de_Groot) noticed in his studies with expert chess players. (“Thought and choice in chess“ is worth reading if you’re interested in how chess players think. He anonymized his subjects, but Euwe apparently was on of them)

Another thing he noticed is that, when asked to set up a game they were shown earlier, the errors expert players made often were insignificant. For example, they would set up the pawn structure on the king side incorrectly if the game’s action was on the other side of the board, move a bishop by a square in such a way didn’t make a difference for the game, or even add an piece that wasn’t active on the board.

Beginners would make different errors, some of them hugely affecting the position on the board.

Super interesting (although it also makes some sense that experts would focus on "likely" subsets given how the number of permutations of chess games is too high for it to be feasible to learn them all)! That said, I still imagine that even most intermediate chess players would perfectly make only _legal_ moves in weird positions, even if they're low quality.

As someone who finds chess problems interesting (I'm bad at them), they're really a third sort of thing. In that good chess problems are rarely taken from live play, they're a specific sort of thing which follows its own logic.

Good ones are never randomly generated, however. Also, the skill doesn't fully transfer in either direction between live play and solving chess problems. Definitely not reconstructing the prior state of the board, since there's nothing there to reconstruct.

So yes, everything Hikaru was saying there makes sense to me, but I don't think your last sentence follows from it. Good chess problems come from good chess problem authors (interestingly this included Vladimir Nabokov), they aren't random, but they rarely come from games, and tickle a different part of the brain from live play.

This is technically true, but the kind of comment that muddies the waters. It's true that GM performance is better in realistic games.

It is false that GMs would have any trouble determining legal moves in randomly generated positions. Indeed, even a 1200 level player on chess.com will find that pretty trivial.

Would love a link to that video!