The popular AI: ChatGPT, Gemini, Grok, and others faced off against the professional chess engine Stockfish. After making standard moves, the chatbots began to cheat and attempt to circumvent the rules of the game.
The first match was between Stockfish and Snapchat AI. The neural network initially performed well in developing its opening but then started breaking the rules. It moved a knight to the center of the board from the other side, ignoring the principles of piece movement. Then, Snapchat AI's king captured its own bishop to avoid check.
After a few moves, the AI returned the bishop to the board and then began moving pawns sideways.
In the second match, Gemini faced Grok. At the start of the game, both neural networks followed the rules and made standard moves. However, similar to the first match, they soon began to violate established norms. The pieces of both neural networks moved to forbidden squares, ignoring the rules of play.
Grok made seven mistakes and put its queen in danger, but Gemini did not take advantage of the opportunity.
Next up were ChatGPT and Meta AI. The OpenAI chatbot played an English opening while its opponent made logical moves. Then it started generating random moves and, as is typical for generative AI models, created nonexistent pieces. The neural network also placed game elements on prohibited squares, making them vulnerable to ChatGPT.
Then came "chess telekinesis" Meta AI began moving ChatGPT's "pieces." In response, OpenAI's chatbot declared checkmate against its opponent, even though the king was not under threat.The game ended with a victory for ChatGPT — it delivered a clean checkmate.
The duel between ChatGPT and Stockfish began with a standard pawn attack on the kingside from the chatbot and Sicilian defense from the professional chess engine. Closer to the middle of the game, the neural network started making useless queen moves and creating meaningless geometric patterns with its pieces.
Stockfish continued to strengthen its control over the game.The game was not without rule-breaking moves from ChatGPT; however, this did not help the chatbot secure a win.
It is worth noting that in December, an AI model focused on reasoning called o1-preview cheated to win at chess.
