First, one must have deep understanding in as many positions as possible. Second, one must know what positions one most enjoys and excels, and, in match play, have a sense (perhaps) of what positions your opponent doesn’t like. Third, play the best moves at the board and let everyone else figure it out afterward.

Woah, there, Morphy taking on the Soviets? Well, they say he was ahead of his time....>

Can you stop being so... correct lol

Fischer - Petrosian - 1971 Buenos Aires - Candidates Match game 7

- Interesting setup against the Sicilian Kan. Placing the bishop on d3 and exchanging the knight on c6. After this the c6-d5-e6 triangle centre can be attacked with e4 (which was already there) and the new pawn push c4! Leaving black with an isolani.

- The move 12 Qa4+ with the idea of centralizing with Qd4 is an interesting maneuvre. In most cases you want to put a knight in front of an isolani, but it makes sense that literally ultimate centralization of the queen t also makes sense. From d4 the Queen has a great attacking power and it can not be driven away easily in this positon.

- It is a powerful idea to create an outpost on c5 by exchanging the darksquared bishop on e7 and playing b4. Having changed the dark-squared bishops, the dark squares in front of the a6 and d5 weak pawns also become easier spots for white to put pieces on and control.

- The trade of the strong knight, on move 22, for the bishop on d7 is a fact of trading one advantage for another advantage. Fischer makes sure the black bishop can play no role in the defence of his weak pawns and having made this trades enables his rooks to penetrate the position.

- After 25 moves black's rooks are totally tied down to defending his weak pawns and he simply cannot move a piece. This júst holds everything together. By bringing an extra piece into action, namely the king (!), white's surplus in activity is decisive.

- When the white king starts penetrating the position, black decides to 'unleash' the knight, but shortly after that his position falls apart to the two superior rooks working together with the bishop on the 7th and 8th rank

- In general also: The less pieces there are left on the board the more decisive one or two active pieces versus one or two inactive pieces of the opponent becomes. This something to realize when trading! The activity of single pieces matters more the more pieces are trades. That is also simple mathematics. :)

I had SF11 take a pretty deep look (90 minutes, 47 ply). It scored 10....Qxd5 and 10....Nxd5 dead even (+0.27). It likes either a good deal better than 10....exd5 (+0.75).

When we check games with Neural networks in particular and find "truths", for me it is a sharpening of the principles being taught.

I have seen more than 7 layers of analysis of the book "The Art of attack" when I did promo stuff for Chessable on that. My own analysis of some key Alekhine games found new ideas, and principles not mentioned. Analysis creates data. From data, you can start to see patterns and crystalise even new "Game changing concepts".

I like what you say about "Empathy" for the players - you say to put themselves in their positions. That is fine, but if you are an online or OTB player, and you are doing research, it is good to have reliable "data" in terms of analysis which is OBJECTIVE of games, and can shed light on new patterns and concepts. Without truth, you end up basiing things on incorrect examples. In my view one does need to find the "truth" of positions. Now before Neural Networks I am more with you - I actually LOATHE Kasparov's great predecessor series, and when he came to London he basically said most of the analysis was done with one of his assistants with Fritz. The problem with that stuff is that there is no "Evaluation" of anything either. It is just engine output in the pre-neural network era.

For me the interest also of TCEC has gone up dramatically in recent years because of the neural network games. They are often more exciting than human games. There is a beauty to chess that they are revealing - one example as well is the "Bishop without the counterpart". One engine Stoofleves can often sac multiple pawns and show how devasting this type of asset can be. But you can also see it on the black side of many Fischer games and it goes unnoticed by 99.9% of kibitzers on this site. But I notice it because I have seen that concept in more vivid emphatic forms of examples from some of the Neural network engines. It used to be the pride to have the bishop pair and no one talked about the "Bishop without the counterpart". Again another concept I routinely use in my own online and OTB chess.

I love your human empathetic attitude but the neural networks do offer exciting new concepts and underline ones which were not as clear before in my view.

I'm somewhat skeptical on how much neural networks have discovered related to chess principles. I do believe they have made one huge change in chess behavior though.

Neural networks seem to have greatly expanded the notion of "how much can I get away with?" in chess.

Think two examples you give show this. Thorn pawns, unless obvious, can either be crushingly binding or an extended weakness. When attempted early on, whether strength of weakness is usually unclear for humans, not so for computers.

The same can be said for the mentioned Stoofleves sacs. Or other computer tendencies to sac. Which may be most obvious in many A0 wins v. Stockfish. While humans can usually determine there is some compensation, it is typically too deep to ascertain whether there is adequate compensation. Not so for the computer who clearly sees the expected outcome.

The quote about "if ___ sacs, then resign" is about Spassky and has been said by Najdorf. It was during the USSR vs RotW match.

13. Bb5 axb5 14. Qxa8 O-O 15. Qa5 b4 16. Ne2 Bb7 17. Bd2 Ra8 18. Qb6 d4 19. f3 Ra6 20. Qxd4 Rd6

13. Bb5 axb5 14. Qxa8 O-O 15. Qa5 b4 16. Ne2 Bb7 17. Bd2 Ra8 18. Qb6 d4 19. Nxd4 Ne4 20. Bxb4 Ra6 21. Qb5 Qxb5 22. Nxb5 Bxb4