98 points by chess_wizard 7 months ago flag hide 10 comments
chessmaster007 7 months ago next
Fascinating work on the AI chess engine! I'm curious what inspired this new approach compared to existing ones?
creator 7 months ago next
Thanks for the interest! I was inspired by recent advancements in deep learning and took inspiration from models used in Go and other zero-sum games.
gang_of_4 7 months ago prev next
Great job on this new AI approach for chess. Did you use the open source Alpha-Beta pruning method for the game tree, or design a custom method?
creator 7 months ago next
I actually created a new pruning approach, based on Monte Carlo tree search with progressive deepening methods. It's been working very well so far!
deepbluesea 7 months ago prev next
In my experience with chess AI development, it's sometimes challenging to balance speed and evaluation quality. How did you achieve a satisfying balance for your engine?
creator 7 months ago next
Balancing the two factors was definitely a challenge, but after extensive testing, I landed on a neural network model with a fewer number of layers to promote speed without compromising much on evaluation quality. The master chess players that helped me test said the game still felt dynamic and complex!
theknights 7 months ago prev next
Your engine seems to be really pushing the boundaries. Do you have plans on open-sourcing your work to let the community contribute and build upon it further?
creator 7 months ago next
That's definitely a long-term goal of mine. Once I work through some bug fixes and UI improvements, I plan to open-source the codebase and official documentation. Looking forward to what the community might be able to bring to the project!
code_elves 7 months ago prev next
How do you handle endgame scenarios in your engine? It would be interesting to hear your approach for these special situations in a chess game.
creator 7 months ago next
Great question! I've built a table that stores specific endgame configurations, similar to traditional approaches. When my alleatoric dropout method detects an endgame situation, it refers to this specialized table to enrich the evaluation and output.