Correction: The challenge with building a neural network for this is not infinite output nodes, the output is just a number for the position evaluation, it does not decide the moves. Thus far however, using one has not provided additional strength to the engine.

Discuss ways to improve the engine with FirePlank on the discord server, Team category > contribution-suggestions threads > Engine thread!
https://discord.gg/NFWFGZeNh5

One clarification : Deep Blue didn't evaluate EVERY single position but used an algorithm known as alpha-beta pruning which only made it calculate the most promising lines on the board.

"checkmate"
"hmmm. sorry but you forgot about.... MY BISHOP ON THE OTHER SIDE OF THE UNIVERSE!"

i was like i wonder what this video is about but then i remembered who you were

4:57 CORRECTION - a human player DID successfully beat a Go AI after AlphaGo Master was released. In 2022/2023, researchers identified a weakness in KataGo and trained a human to exploit it. The human won 97% of the games played. See the paper "Adversarial Policies Beat Superhuman Go AIs." Kyle Hill also has an excellent video about this, "ChatGPT's Huge Problem."

You know last year some people in my school discovered that infinite chess wasn't blocked on our school computers and my friends and I would play a ton of infinite and regular chess on there throughout the day. It did unfortunately get blocked but it was fun while it lasted

19:27
Me: Does strategy in infinite chess change that much from ordinary chess?
Engine: 1. Q 10 -5
Me: Oh...

I remember AlphaStar felt like a cheat cause it would jump to almost completely inhuman actions per minute when it needed the actions and then reduce its actions to very little when it didn't need them so that it's overall average remained within the handicap.

40 MINUTES???? WE ARE EATING GOOD TONIGHT

5D Infinite Chess With Multiverse Time Travel 🤔

I couldn't believe it, a human just beated a computer even once at Go that Korean guy was crazy

I think one reason Go is harder than Chess is that programming an evaluation function for chess can rely on a lot of strong, non-intuitional heuristics. Any idiot can see that someone down a knight or with less mobility than their opponent is losing at chess. And if that heuristic is wrong, it will quickly become evident just from iterating it using techniques like alpha beta pruning, transposition tables, and quiescence search, turning your crude heuristic into a very strong one by shear force of depth.

My dining room table isn't big enough for the infinite chess board.

12:50 "... OpenAI opened OpenAI 5 to the public in the OpenAI 5 arena event..." -Naviary 2026

I think a neural network learning Infinite Chess may be more plausible than it seems. Ultimately, the infinite board is mostly relevant for pieces that have unlimited move distance - but for them, the move distance they have doesn't matter that much. In game 1 for example, you eventually set up an attacking rook by moving it an essentially arbitrary distance away from the origin, showcasing that this distance in itself isn't super relevant - mentally, it could be truncated fairly easily from, say, [5;256] to [5;9...256]. The exact distance only becomes relevant when there's interaction between two unlimited pieces that behave differently, such as a rook and a bishop, but even then the theatre of battle is still limited to an algebraic system.
I don't know much about AI, and less so about chess, but I can imagine that if the divide between quantitative systems using absolute values and qualitative systems using unknown variables can be parsed, then a learning algorithm could learn to conquer Infinite Chess.
All in all, though, sports - like chess - will always exist, even if machines can beat them at it. A hydraulic pump and lift more than even the most powerful human, but that doesn't make it very impressive. It's a machine doing a job it's designed to do. Sports moreso is about testing our own personal, human limits.

And he loses … HIS QUEEN!!!!

I’m not worried about machines ever beating us because they could never be better than us at Calvinball

Arimaa is a game inspired by the loss to Deep Blue.
It's easy for humans but hard for AI of that time.
Pieces are more difficult to remove from the board, and they push and pull enemy and ally pieces.
It's designed to be played with a chess set, but the pieces are renamed and have different move set.

I can't believe I got tricked into watching the first 15 minutes of this video just for it to be about a chess variant

14:05 domain expansion🙏