lichess.org
Donate

What if chess would be solved?

I absolutely fell in love with your posts! I'm happy for you for the ad revenue that will be pouring in, I know we're much better at paying our bloggers here on lichess compared to the website that shall not be named. I thought of hearting some of them but thought that was too many bytes and my internet bill might go up, I fear I can only afford this reply. I hope you appreciate it!

I absolutely fell in love with your posts! I'm happy for you for the ad revenue that will be pouring in, I know we're much better at paying our bloggers here on lichess compared to the website that shall not be named. I thought of hearting some of them but thought that was too many bytes and my internet bill might go up, I fear I can only afford this reply. I hope you appreciate it!

If chess was solved we'd all have to grapple with the fact the only objective way to play for a win would be double grog

If chess was solved we'd all have to grapple with the fact the only objective way to play for a win would be double grog

You are a great comedian @DaBassie :)

And your readers gotta learn the difference between "sarcasm" and "irony" ...

You are a great comedian @DaBassie `:)` And your readers gotta learn the difference between "sarcasm" and "irony" ...

@chesspawnrookking said ^

Interesting article I was also looking into analyzing wit weaker engines with low dept just to represent the human playability. I would stick to at least the NNUE versions I believe sf 13+ since most of the ones before that are quite bad positionally at low depts tat modern ones can actually still be quite good. Part of why the Leela odds bot works so well is it is purposefully programmed to not think much so it doesn't calculate and et desperate as mentioned in the article.

I very remotely followed the rise of the first Neural Network engine(s) in TCEC, and a common theme back then was that they constantly sent winning positions into drawn opposite-color-bishop endgames (that they falsely assumed were winning).
So positionally a lot of progress has been made there as well, I guess.

@chesspawnrookking said [^](/forum/redirect/post/SiD9T0dW) > Interesting article I was also looking into analyzing wit weaker engines with low dept just to represent the human playability. I would stick to at least the NNUE versions I believe sf 13+ since most of the ones before that are quite bad positionally at low depts tat modern ones can actually still be quite good. Part of why the Leela odds bot works so well is it is purposefully programmed to not think much so it doesn't calculate and et desperate as mentioned in the article. I very remotely followed the rise of the first Neural Network engine(s) in TCEC, and a common theme back then was that they constantly sent winning positions into drawn opposite-color-bishop endgames (that they falsely assumed were winning). So positionally a lot of progress has been made there as well, I guess.

@forksforbreakfast said ^

I absolutely fell in love with your posts! I'm happy for you for the ad revenue that will be pouring in, I know we're much better at paying our bloggers here on lichess compared to the website that shall not be named. I thought of hearting some of them but thought that was too many bytes and my internet bill might go up, I fear I can only afford this reply. I hope you appreciate it!

Thx. Yeah these bytes are getting really expensive nowadays.
My lichess paycheck seems to be delayed, but i'm sure they're fixing that behind the scenes.

@forksforbreakfast said [^](/forum/redirect/post/E6q6jApY) > I absolutely fell in love with your posts! I'm happy for you for the ad revenue that will be pouring in, I know we're much better at paying our bloggers here on lichess compared to the website that shall not be named. I thought of hearting some of them but thought that was too many bytes and my internet bill might go up, I fear I can only afford this reply. I hope you appreciate it! Thx. Yeah these bytes are getting really expensive nowadays. My lichess paycheck seems to be delayed, but i'm sure they're fixing that behind the scenes.

@MillenniumBug said ^

You are a great comedian @DaBassie :)

And your readers gotta learn the difference between "sarcasm" and "irony" ...

Haha comedian... I'm not spontaneous enough for that kind of stuff, but writing gives time to think, that helps me a lot. (Talking is like hyperbullet chess for me, and writing is like classical)...

Sarcasm is traditionally reason for trouble on the Internet, so it's underused in my humble opinion. Some people just won't get it, never. But I feel like the silent majority (surely >90%) gets it.
Also I believe people on lichess are somewhat smarter than the average Internet user... Which might be a very dangerous statement.

@MillenniumBug said [^](/forum/redirect/post/1egwu8wy) > You are a great comedian @DaBassie `:)` > > And your readers gotta learn the difference between "sarcasm" and "irony" ... Haha comedian... I'm not spontaneous enough for that kind of stuff, but writing gives time to think, that helps me a lot. (Talking is like hyperbullet chess for me, and writing is like classical)... Sarcasm is traditionally reason for trouble on the Internet, so it's underused in my humble opinion. Some people just won't get it, never. But I feel like the silent majority (surely >90%) gets it. Also I believe people on lichess are somewhat smarter than the average Internet user... Which might be a very dangerous statement.

@DaBassie said ^

Haha comedian... I'm not spontaneous enough for that kind of stuff, but writing gives time to think, that helps me a lot. (Talking is like hyperbullet chess for me, and writing is like classical)...

Same here. I've seen some clips of your stream and you seem to be doing well. Oh man it sucks when people don't understand satire or sarcasm. And the last thing you want to do is explain a joke. It simply never works.

@DaBassie said [^](/forum/redirect/post/Xx4vWSYK) > > Haha comedian... I'm not spontaneous enough for that kind of stuff, but writing gives time to think, that helps me a lot. (Talking is like hyperbullet chess for me, and writing is like classical)... > Same here. I've seen some clips of your stream and you seem to be doing well. Oh man it sucks when people don't understand satire or sarcasm. And the last thing you want to do is explain a joke. It simply never works.

I still remember when checkers got solved. Even though an optimal strategy exists, people still play it.

I still remember when checkers got solved. Even though an optimal strategy exists, people still play it.

I propose a stage-aware NNUE design that guides a chess engine like a GPS. Use the number of captures as a simple stage signal (0–10 early, 11–20 mid, 21–30 late), train three NNUEs on positions from each range, and blend their outputs smoothly based on the capture ratio. Share feature extraction for efficiency, and let tablebases override the late-stage net when available.

The idea treats chess as a monotonic funnel of decreasing exchange entropy: many possible exchanges early, fewer later. Each net learns the objective distribution of its exchange tier without relying on traditional phase labels.

The goal isn’t to solve chess, but to improve practical strength and efficiency—better static guidance, less unnecessary deep search. Tactical correctness still requires quiescence, extensions, selective deepening, and tablebases in reduced positions.

Using Leela Chess Zero self-play to generate stage-specific data—and even to run three specialized critics—could accelerate training, provided switching artifacts and distributional collapse are controlled via soft blending, shared backbones, and data diversity.

I propose a stage-aware NNUE design that guides a chess engine like a GPS. Use the number of captures as a simple stage signal (0–10 early, 11–20 mid, 21–30 late), train three NNUEs on positions from each range, and blend their outputs smoothly based on the capture ratio. Share feature extraction for efficiency, and let tablebases override the late-stage net when available. The idea treats chess as a monotonic funnel of decreasing exchange entropy: many possible exchanges early, fewer later. Each net learns the objective distribution of its exchange tier without relying on traditional phase labels. The goal isn’t to solve chess, but to improve practical strength and efficiency—better static guidance, less unnecessary deep search. Tactical correctness still requires quiescence, extensions, selective deepening, and tablebases in reduced positions. Using Leela Chess Zero self-play to generate stage-specific data—and even to run three specialized critics—could accelerate training, provided switching artifacts and distributional collapse are controlled via soft blending, shared backbones, and data diversity.

As a mathematician (who also plays chess), I have been saying this privately for years. It's good that someone is taking the time to make the topic more accessible.

GM Jacob Aagaard has a funny way of putting it: Stockfish 17 (or 18 now) is a terrible engine to analyse with; instead, you want Fritz 6! His point is that the older engines were strong enough to know when there was trouble (like a human), but weren't strong enough to immediately solve all of the problems in the position (like a human!). It's a very good point, but I still use Stockfish to analyse, because I never used those older engines and I've now trained myself to be calibrated to modern evaluations. Except, of course, when a new model comes out, and I have to re-adjust...

As a mathematician (who also plays chess), I have been saying this privately for years. It's good that someone is taking the time to make the topic more accessible. GM Jacob Aagaard has a funny way of putting it: Stockfish 17 (or 18 now) is a terrible engine to analyse with; instead, you want Fritz 6! His point is that the older engines were strong enough to know when there was trouble (like a human), but weren't strong enough to immediately solve all of the problems in the position (like a human!). It's a very good point, but I still use Stockfish to analyse, because I never used those older engines and I've now trained myself to be calibrated to modern evaluations. Except, of course, when a new model comes out, and I have to re-adjust...