Oracle is the first chess engine that plays like a human, from amateur to super GM. She can play like a 2800-rated player in classical or an 800-rated player in blitz, or at any other level, in any time control.
- Human-like Play: Oracle adapts her predictions to any rating level from amateur to super GM.
- Time Control Flexibility: Oracle adapts her predictions to 4 different time controls: bullet, blitz, rapid, or classical.
- Expected Score Evaluations: Instead of using centipawns, Oracle evaluates positions by giving White's expected score out of 100 games. This expected score takes into account the position, the likeliest next moves, the rating of both players and the time control.
- Two Modes:
- Oracle_one_move: Takes a PGN game as input and predicts the likeliest next moves.
- Oracle_pgn_file: Takes a PGN file with multiple games and predicts every move of every game, useful for creating data to test Oracle.
- Oracle_one_move: Set your openAI API key and the path to your Stockfish at the top of the file then run the file. Past the PGN up to the move you want to predict into the console, type END and press Enter
- Oracle_pgn_file: Set your openAI API key, the path to your Stockfish, your input pgn file, and your output csv file, and then run the file. Oracle will write her predictions for every move of every game of the PGN into the csv file.
Position after 58...a3 in the last tie-break of the 2023 World Championship. Stockfish shows 0.00, but considering it's rapid, Oracle only gives white a 18.50% expected score. Nepo ended up blundering with 59. Qc7??, which was the likeliest move according to Oracle.
Position after 11...Nc6 in the infamous "Danse of the Knights" pre-arranged draw between Daniil Dubov and Ian Nepomniachtchi at the 2023 World Blitz Championship. Despite it being an obviously bad move, Oracle predicts 12. Nb1 that was actually played by Nepo and gives it a high 72% likelihood.
Oracle uses GPT-3.5 turbo instruct and Stockfish to deliver human-like chess moves. Users need an OpenAI API key and a version of Stockfish, which can be downloaded here.
By default, Stockfish is set to a time limit of 1.3 seconds, a depth limit of 20 plies, uses 8 threads and 512 MB of hash. You can change these settings in the analyze_moves function.
On my computer, Oracle takes on average about 2 seconds per move. This duration could be reduced with an optimized code and better hardware.
Because Oracle uses GPT for her prediction, she is costly! The average cost is ~400 predictions per $1 but can vary greatly with the length of the prompt (up to 10.000 predictions per $1 for opening moves with a short header)
I am a FIDE Master and Woman International Master with no previous coding experience, so the code might contains mistakes or improper formulations.
I've decided to name my chess engine Oracle because just like the Oracle from The Matrix, her predictions feel magical even though they are just pure calculations performed by a program. For that reason, Oracle should be referred to as she/her.
Because I'm new to coding, Oracle's code should be improvable. The next significant step for Oracle would be the creation of an open-source LLM trained on full PGNs with headers to replace GPT-3.5 Turbo Instruct, making Oracle completely free. Following this, Oracle could be turned into a user-friendly executable file and used on a large scale for broadcasts, training, opening preparation, anti-cheating, bots creation, and so on.
I have dedicated several hundred hours to this project and invested a significant amount of money. As a professional chess player and coach, my resources are limited. While I am happy to offer Oracle to the chess and scientific communities for free, any donation would be greatly appreciated.
If you value my work and wish to support Oracle and me, please consider making a donation.
I would be very thankful. 🙏
I've written an extensive article on my personnal website
This project is licensed under the MIT License. See the MIT License file for details.
© 2024 Yosha Iglesias. All rights reserved.