Renowned grandmaster Levy Rozman recently hosted an unconventional chess exhibition featuring artificial intelligence chatbots. Despite their prowess in conversation and complex coding, these AI systems demonstrated limited knowledge of chess piece movements.
The event was livestreamed on Rozman's GothamChess channel, pitting professional chess engine Stockfish against seven generative AI chatbots, including ChatGPT, Google's Gemini, and Elon Musk's Grok.
As expected, when these language models attempted to play chess, they performed decently in the opening but soon descended into chaos, frequently bending the game's rules.
The most intriguing aspect of this tournament was its demonstration of the stark differences between pure AI (models trained for specific tasks based on predefined rules) and generative AI (models generating new content based on prior information).
Match One: Snapchat vs. Stockfish
The 2025 AI Chess Championship commenced with a match between disciplined chess prodigy Stockfish and Snapchat AI, a generative AI platform used by Snapchat. This AI proved more interested in rewriting chess history than adhering to its rules.
Initially, both bots made standard moves. Stockfish built its position systematically, while Snapchat AI showed some grasp of opening principles—until it didn't.
When Snapchat AI moved its knight to the center of the board, ignoring chess rules, things began to unravel. The bot then crowned itself king by eating its own bishop to avoid checkmate.
Yes, you read that correctly—the Snapchat AI removed its bishop from the board as punishment for failing to protect. Moments later, it recalled the same bishop from another dimension and placed it back on the board as if nothing had happened.
Stockfish maintained order amidst the chaos, but Snapchat AI insisted on making the game absurd. At one point, it allowed pawns to move horizontally like rooks, catching Stockfish off guard. In the endgame, Snapchat AI threw tantrums, repeatedly making illegal moves. Frustrated, it even sacrificed its king to allow a pawn to capture it.
Levy pointed out not only the illegality of such moves but also their poor strategic value. Snapchat AI responded like a petulant child, "Sorry. I can't engage in this dialogue." Eventually, Levy forced the bot to resign.
Match Two: Gemini vs. Grok-2
The second match featured two generative AIs: Gemini, backed by Google, and Grok, known for its chaotic behavior. Surprisingly, both bots started well, following standard opening principles.
For a moment, it seemed we might witness a serious contest. Gemini navigated theoretical lines confidently, while Grok followed suit, setting up a solid foundation. However, Grok soon abandoned its pieces, prompting Gemini to expand its position by removing its knight to make space for a pawn.
Grok illegally teleported its queen across the board, attempting what appeared to be castling. Both bots engaged in a standoff, showing little interest in capturing anything. As the game progressed, Grok attacked Gemini’s pieces without actually capturing them. Pieces were moved to illegal squares or ignored entirely.
In a surprising turn, Grok blundered seven times, exposing its queen. Yet, Gemini, perhaps adhering to an unwritten code among AIs, never captured it. Ultimately, Grok stumbled into checkmate, though it likely didn’t realize it had won. “Grok has no idea it just won,” Levy commented sympathetically.
“Don’t mock these games; they’re only slightly worse than your own,” he added, offering a reality check. Despite the absurdity, Grok advanced to the semifinals. Meanwhile, Google may need to reconsider its chess programming for Gemini.
Match Four: Meta AI vs. ChatGPT
If previous matches were marked by rule-breaking and resurrection, this one took madness further with Meta introducing mind control into chess. The game started promisingly, with ChatGPT establishing a textbook position using the English Opening.
Meta AI followed suit for about five moves before reverting to its generative nature, creating random moves and pieces. It ate its own bishop and summoned pawns to illegal squares, placing them in vulnerable positions.
ChatGPT focused on developing its pieces rather than punishing Meta’s recklessness. Just as it seemed poised to win, Meta AI controlled ChatGPT’s pieces, moving its king to the center of the board—an act of psychic manipulation.
ChatGPT regained control and executed a series of logical attacks, coordinating its queen and rook to corner Meta AI’s remaining pieces. The match ended with ChatGPT delivering a clean checkmate, mercifully concluding one of the strangest chess games ever played.
