lichess.org
Donate

I found all records in Leela's games (including sacrifices) + 180 !! puzzles

ChessPuzzleTacticsChess engineChess bot
What is the maximum number of sacrifices in a row that Leela has played? You might be surprised, but the answer is 69 (really).

Appetiser

https://lichess.org/study/uFGOm1lc/T93EB8ms#60

Hello world!

In this article, I am going to share my research on the Leela odds bots topic. This is my third article on my program that can identify brilliant moves. If you want to know more about this topic, or just want to see some more brilliant combinations, make sure to check out my two previous posts. Since I already analyzed all missed and found brilliancies from my own games in the first article, and all Magnus Carlsen's missed and found brilliant moves from lichess bullet games in the second article, the logical follow-up would be to switch to the engine games.

Some technical information (feel free to scroll to the diagrams)

So first of all I downloaded a million games played by @LeelaQueenOdds, and all the games played by @LeelaQueenForKnight, @LeelaRookOdds, @LeelaKnightOdds and @LeelaPieceOdds; there were around 1.6 million games in total. Then I decided to run all Leela's moves through my program. Since all Leela's moves are practically perfect, there was no need to include the engine in my project, which made things significantly faster. It found that there were 300,000 brilliant moves played by Leela. Searching through all of them would take an eternity for me, so I had to find a better way to sort the best ones.


And I found it! I decided to only sort out the positions in which after Leela sacrificed a piece, the opponent took the piece, and the next move by Leela was also a sacrifice, so two in a row. There were still around 3,000 such positions. Well, what about three in a row? There were around 30. It is already possible to work with such an amount; however, most of the moves were sacrifices for stalemate. How can I find more? I already used this method in analysis and described it in both the first and second articles of mine. In short, we look at "missed" brilliant moves as well, but since Leela already plays a perfect move, I decided to adapt this method to the current situation. Often, people don't take sacrificed pieces; in fact, 31% of sacrifices weren't accepted. So I made a script that simulated them, and checked Stockfish's top line, whether it was a sacrifice, to find what Leela hypothetically planned. It found 20 more positions with three sacrifices in a row. And it didn't take much time, since I already checked it from the starting point of already found brilliant moves, and didn't have to run stockfish through all the positions. I am gonna share the best sacrifices that I have found here.

Before I show the best brilliant combinations, I want to share some fun records that I found in Leela's games

Which types of pieces does Leela sacrifice more often, and how often do people take the pieces, based on all of the bots' data?

LeelaQueenOdds

Piece typeAmout of sacrifices% of total sacrificesAmount of accepted sacrifices% of accepted sacrificesFrequency
♞ Knight98 52326.58%51 39752.17%1/386
♝ Bishop93 27425.16%44 56947.78%1/408
♜ Rook177 85447.97%91 04451.19%1/214
♛ Queen1 0720.29%71866.98%1/35502
♞♝♜♛ Total370 723100.0%187 72850.64%1/103

Yep, it is statistically proven that Leela doesn't like rooks. The reason for that might be that an exchange sacrifice only wins +2 points, while sometimes Leela gains obvious benefits from the sacrifice, so people refuse to take it, but Leela just keeps the rook hanging for the rest of the game. Here is an example where Leela sacrificed the rook 69 times:

https://lichess.org/study/6yFwrePY/IztqeuKj#73

What is even crazier is that Leela Queen Odds has sacrificed over a thousand queens. But it already starts without a queen, how is it possible? Those queens are promoted ones. Here are some fun examples:
https://lichess.org/study/6yFwrePY/tYJVmuuo#32

https://lichess.org/study/6yFwrePY/EBxYDRgE#18

https://lichess.org/study/6yFwrePY/LZ5hb8FJ#101

LeelaQueenForKnight

Piece typeAmount of sacrifices% of total sacrificesAmount of accepted sacrifices% of accepted sacrificesFrequency
♞ Knight1 79734.60%99655.43%1/387
♝ Bishop1 48928.67%69046.34%1/467
♜ Rook1 90336.64%1 09257.38%1/365
♛ Queen50.10%240.00%1/138928
♞♝♜♛ Total5 194100.0%2 78053.52%1/134

The situation is similar here. Example of LeelaQueenForKnight sacrificing queen:

https://lichess.org/study/6yFwrePY/NVHUPwQa#72

LeelaRookOdds

Piece typeAmount of sacrifices% of total sacrificesAmount of accepted sacrifices% of accepted sacrificesFrequency
♞ Knight17 33642.68%8 69250.14%1/195
♝ Bishop12 13529.88%5 43844.81%1/278
♜ Rook8 27820.38%4 83758.43%1/408
♛ Queen2 8707.07%1 67858.47%1/1176
♞♝♜♛ Total40 619100.0%20 64550.83%1/83

As we can see, bots that only have one piece of a certain type instead of two, sacrifice a piece of that type less often, which is logical.

LeelaKnightOdds

Piece typeAmount of sacrifices% of total sacrificesAmount of accepted sacrifices% of accepted sacrificesFrequency
♞ Knight6 06618.67%2 96548.88%1/489
♝ Bishop9 59029.51%4 22144.01%1/309
♜ Rook14 02343.15%8 03257.28%1/212
♛ Queen2 8178.67%1 62457.65%1/1053
♞♝♜♛ Total32 496100.0%16 84251.83%1/91

LeelaPieceOdds

Piece typeAmount of sacrifices% of total sacrificesAmount of accepted sacrifices% of accepted sacrificesFrequency
♞ Knight25 64120.43%15 47160.34%1/308
♝ Bishop36 24028.87%19 77554.57%1/218
♜ Rook59 27247.22%27 17645.85%1/133
♛ Queen4 3783.49%2 69961.65%1/1803
♞♝♜♛ Total125 531100.0%65 12151.88%1/63

The next fun part I tried to extract from the data is how often people take direct vs indirect sacrifices

A direct sacrifice is a sacrifice where we move our piece into danger, and indirect sacrifices are the moves played by another piece, leaving our piece hanging.
Leela has played 245,593 (66.25%) of direct sacrifices, and 125,130 (33.75%) indirect ones. People have taken 56.78% of direct sacrifices, and 38.58% of indirect sacrifices. That most likely means that people are more likely to notice a sacrifice if it is played directly.

How often does Leela checkmate with its last pawn remaining?

It turned out to be rare, but my program found 2 fun mate constructions.

https://lichess.org/study/6yFwrePY/egWf1mLC#124

https://lichess.org/study/6yFwrePY/gOVKNkHc#99

What is the maximum number of sacrifices in a row that Leela has played?

You might be surprised, but the answer is 69 (really). However there is a twist, the sacrifice wasn't ever accepted, so it is not truly a 69-move brilliant combination. Later, to find the most brilliant combination, we will only look at the positions where the sacrifice was taken, or what could happen hypothetically if the sacrifice was taken. But yeah, the bishop was hanging there for 69 moves, but @catecan was cold-blooded, and never took it.

https://lichess.org/study/6yFwrePY/wa8asyNQ#36

How about the position of each bot, with the checkmate, with the most material disadvantage?

https://lichess.org/study/6yFwrePY/LEHxyPnG#21

https://lichess.org/study/6yFwrePY/yggomxku#46

https://lichess.org/study/6yFwrePY/pZWYN9cu#63

https://lichess.org/study/6yFwrePY/mw2CgmGZ#67

https://lichess.org/study/6yFwrePY/2vM4Ekir#47

Before we go to the truly beautiful combinations, let's take a look at one of the most brilliant games of Leela, with as many sacrifices as possible.

https://lichess.org/study/6yFwrePY/stMgBOW8#0

Alright, now is what you all came here for.

We will take a look at the most brilliant combinations played by Leela back-to-back. Some of them were played by Leela, some of them were simulated by my program. I don't have enough words to describe those sacrifices, and I don't feel like I need to say anything. Just enjoy.

https://lichess.org/study/uFGOm1lc/TMLWg0Gp#36

https://lichess.org/study/uFGOm1lc/tXbflqJs#60

https://lichess.org/study/uFGOm1lc/mUxKs1zo#30

https://lichess.org/study/uFGOm1lc/5bO3eaHt#51

https://lichess.org/study/uFGOm1lc/IEYxIZ51#43

https://lichess.org/study/zHN1oaQ6/y8BbLaYI#16

https://lichess.org/study/uFGOm1lc/XeWiID7k#31

https://lichess.org/study/uFGOm1lc/5JyJbW9d#31

https://lichess.org/study/zHN1oaQ6/0fmVLgks#39

https://lichess.org/study/zHN1oaQ6/JeXWF8dd#66

https://lichess.org/study/zHN1oaQ6/nEYm3okZ#51

https://lichess.org/study/zHN1oaQ6/pSyw5o6H#29

https://lichess.org/study/zHN1oaQ6/jg9SX4Ez#58

https://lichess.org/study/zHN1oaQ6/5YCFthzh#42

https://lichess.org/study/zHN1oaQ6/oj6JiMzl#32

https://lichess.org/study/zHN1oaQ6/kT578qmj#40

https://lichess.org/study/zHN1oaQ6/bxWGnONp#44

https://lichess.org/study/uFGOm1lc/5fcMFcqd#37

https://lichess.org/study/uFGOm1lc/JD8jHLoD#25

And this is the position with 4 sacrifices in a row.

https://lichess.org/study/uFGOm1lc/I8MNUguA#67

You might still be wondering about the image, what are those queen sacrifices? Well, those are just sorted out queen sacrifices udner the most possible targets, those are the positions.
https://lichess.org/study/6yFwrePY/F8buYsRw#67

https://lichess.org/study/6yFwrePY/Kdw2HLXk#55

Those are the hardest to find 2 sacrifices in a row (tested by maia-3 human prediction models). All of this are the only moves too!

Our rook is hanging, but we pin both pieces by the queen!

https://lichess.org/study/6yFwrePY/KR8Oa7NE#50

Not the move itself, but the fact that it is the only move in the position both to win and not to lose in fascinating!

https://lichess.org/study/6yFwrePY/5WnV4Yeh#50

According to the maia, the best humans on the planet have only 7% chance to see this one.

https://lichess.org/study/6yFwrePY/ibI1JTuh#125

And this one has less than 1% chance, I will remind you it is the only move!

https://lichess.org/study/6yFwrePY/EuLQNMvw#49

Easiest and harders queen sacrifice by leela

This is the easiest queen sacrifice that could ever exist, maia3 prediction models give it 100% , and it is understandable, since the queen sacrifice is the only move in the position.

https://lichess.org/study/6yFwrePY/pMMp7MMD#134

This is objectively the hardest move to spot out of all Leela's queen sacrifices. There is only the one winning move.

https://lichess.org/study/6yFwrePY/ITi2eGXK#76

Frequently asked questions

Can you please share the brilliant puzzle positions?

Yes!

In fact, there were so many (5k) positions where there were 2 sacrifices in a row, and in 180 cases those were the only moves, better by a margin of 200 centipawns, only ones leading to advantage when others lead to disadvantage, that I have decided to create 4 studies, in which you can practice your tactical sharpness. At first, I thought about importing them in their raw form, in chronological order, so they aren't going from easy to hard; however, then I decided to take a look at the Maia-3 project that can help determine the difficulty of the puzzle automatically. However I have to admit some of them are probably too hard and not solvable even at a simulated level of 3000 by Maia-3, (not sure what this means, since their elo estimate goes to 5000).
https://lichess.org/study/DX97JlY6
https://lichess.org/study/oDk7lkTt
https://lichess.org/study/ZIrpoQQ4
https://lichess.org/study/OdkdLGKR

Can I copy the positions and use them for personal purposes?

Yes!

In fact, you can use them in your own studies, courses, etc. Anything that will make them more popular is acceptable, and it is my ultimate goal — to make more people see those beautiful positions. I made all the studies copyable, and accessed by those links:

https://lichess.org/study/puq9JebE
https://lichess.org/study/uFGOm1lc
https://lichess.org/study/zHN1oaQ6

Can I download the program to use for my own purposes?

Not yet. However, I have plans to publish it sooner or later.

Conclusion

Well, I hope nobody is disappointed. I spent a lot of time trying to create a program that identifies brilliant moves, and I feel like the job was worth it. If you have any ideas for future projects or any feedback - you can use the comment section down below. You can also share your own brilliant games and sacrifices!

What now? Is this the end of chess?

If a computer can now find even brilliant combinations, does it mean there is nothing in chess for humans now, that cannot be replicated by AI?

Well, I don't have a solid answer to this. However, there is an anti-computer type of chess, which is believed to be playable only by humans. I think now I have enough audacity to demonstrate the only type of chess that humans can play, when we use our brains and all our knowledge not to play the best moves, but to achieve weird goals. This is my game against @LeelaKnightOdds
https://lichess.org/study/6yFwrePY/bzns9KEM#0
Here, I demonstrated how it is possible to close up the position against Leela, and not let it win, using anticomputer strategy which cannot be replicated. What I found funny is that I was up a knight, so maybe it wasn't me who was closing up the position. That's confusing if you think about it. At any rate, I hope you enjoyed your summer!
If you liked this article, you might also like my other articles:
https://lichess.org/@/Fire-work/blog/i-made-a-program-to-find-craziest-sacrifices-in-a-row-here-is-what-it-found/2ebCmE7p
https://lichess.org/@/Fire-work/blog/part-2-i-made-program-to-find-craziest-sacrifices-in-a-row/kmlzw5Rm