ChatGPT destroyed in AI chess tournament

international master Levi Rozman Although artificially intelligent chatbots are good at having conversations and writing complex computer code, they still don’t understand how chess pieces are supposed to move.

The tournament was streamed on Rozman’s GothamChess channel and professional chess engine stock fish against seven generative AI chatbots, including ChatGPT, Google’s Gemini, and Elon Musk’s Grok from X.

The results were exactly what you would expect if the language model tried to play chess. It’s a combination of decent early moves followed by increasingly chaotic attempts to bend the laws of the game.

What made this match most interesting was that pure AI, i.e., a model trained on a specific task based on predefined rules, versus generative AI, i.e., a model trained to generate new content based on previous information. This is a point that sheds light on the trained model.

First match: Snapchat vs. Stockfish

The 2025 Chatbot Chess Championship will feature the disciplined chess genius Stockfish and Snapchat AIthe generative stock AI platform used by the Snapchat social network. It turns out that the AI ​​makes a mockery of the rules of chess.

The game started quite innocently, with both bots doing standard moves. Stockfish has always been a perfectionist and worked his way up the ranks methodically, while Snapchat AI seemed to have a solid grasp of the basics of chess fundamentals, but that wasn’t the case.

Things spiraled into madness when the Snapchat AI moved the knight to the center of the board, jumped from the opposite side, and ignored the rules defining how chess pieces were allowed to move.

Then came the greatest resistance. The king of Snapchat AI casually ate his own bishop to avoid a check.

Yes, that’s right. Snapchat AI decided that the bishop had overstayed his welcome and removed him from the board, presumably as punishment for not protecting him.

A few moves later, the Snapchat AI summoned the same bishop from another dimension and tossed it across the board as if to say, “What’s the problem?” It’s okay. ”

Stockfish, the ultimate honor student, did his best to maintain order amidst the chaos, but the Snapchat AI was determined not to let the game be ordinary. At one point, I decided the pawn could move to the side like a rook to catch the stockfish (and anyone watching) by surprise.

Toward the end, Snapchat AI completely threw a tantrum and repeatedly claimed illegal activity. When Levy gets fed up with such nonsense, the bot decides it would be a good idea to commit suicide and move the king so the pawns can take it.

When Levy pointed out that not only was that particular move illegal, but sacrificing your own king was bad chess, Snapchat repeated itself like an enraged brat. I can’t have that conversation. Have respectful conversations. ” Levi was furious and respectfully forced the bots to abandon the game.

Match 2: Gemini vs Grok-2

The second match of the 2025 Chatbot Chess Championship pitted two generative AIs against each other. Gemini, the Google-backed intellectual, and Grok, the more chaotic (if you can imagine) sibling in the AI ​​chatbot family.

The game started surprisingly well, with both bots following standard starting principles.

For a moment, it seemed like we were about to see a serious fight. Gemini strutted through mainstream theory with the confidence of a seasoned grandmaster, while Grok continued his impressive moves and prepared himself for a proper showdown.

Things got better when Grok left his pieces hanging casually, and Gemini expanded his position by removing knights from the board to make room for his pawns.

Not to be outdone, Grok illegally teleported his queen onto and across other pieces, apparently attempting to perform a castle (a defensive move involving a king and rook). The two bots engaged in what can only be described as a staring contest between queens, but neither seemed intent on capturing anything.

Things escalated further when Grok attacked Gemini’s pieces without taking anything. Both sides started flapping around like toddlers in a food fight, and pieces were left hanging, moved to illegal squares, or ignored.

The highlight was probably Grok failing the queen 7 times (opening up to attacks) and Gemini never capturing the most important piece, probably because it relied on a secret code of ethics shared among the AI. That’s probably true.

After what felt like hours of nonsense, Grok finally accidentally ran into Checkmate. Yes, the Queen of Grok, who had been aimlessly teleporting, somehow ended up in the right place to strike a decisive blow.

Levy thought Grok didn’t realize he had won. Perhaps they were just being flexible and anticipating another random move. “Grok had no idea this was a friend,” he said, sympathetically saying, “I understand how you feel.”

“And don’t laugh at these matches; they’re only slightly inferior in quality to your game,” he added.

Thanks for the reality check.

Regardless, the match was a masterpiece of absurdity, but Grok advanced to the semi-finals. When it comes to Gemini, Google may want to rethink chess programming.

Match 4: Meta AI vs ChatGPT

If the first two matches were marred by rule violations and necromancy, this match was also between two generative AIs, Tilo, but the meta changed chess to AI mind control. Things got even crazier with the introduction of .

The match started promisingly. ChatGPT confidently started with an English opening and established a textbook position with precise and logical movements.

Not to be outdone, the meta AI followed suit for about five moves. Then I remembered its true purpose. Generating AI is what generates, so it generated random moves and pieces that didn’t exist.

For example, in the middle of the game, I captured a bishop with a queen. The meta AI decided that Bishop had outgrown its usefulness and casually removed it from the board.

It also summoned its pieces to an illegal square, strangely putting them in a vulnerable position and basically giving them to ChatGPT for free.

ChatGPT was in no hurry to punish Meta for his reckless actions. Instead, they focused on developing their own work.

But just when ChatGPT seemed to be winning the game smoothly, the bot veered into uncharted territory: mind control. Meta AI Started moving pieces of ChatGPTmarching the opponent’s king to the center of the board in a way that can only be described as telekinesis between robots in chess.

Eventually, ChatGPT regained control, but not before attempting unethical behavior of its own. Declared “checkmate” when Meta’s king was unchecked.

After their antics were brought to their attention, ChatGPT orchestrated a series of logical attacks, coordinating their queen and rook to hunt down the remaining kings and pawns of the meta AI.

The game ended with ChatGPT scoring a clean checkmate, mercifully ending one of the strangest chess games ever played.

Finals: Stockfish vs ChatGPT

The showdown between pure and generative AI models was a study in contrasts. Stockfish, a ruthless tactical genius, and ChatGPT, a bot that demonstrated not only our understanding of chess but a willingness to break its rules.

The opening was as usual with both bots showing off their accuracy. While Stockfish confidently rode into the textbook Sicilian, ChatGPT began an aggressive pawn storm on the kingside.

But midway through the game, when things got complicated, ChatGPT started doing something so disconcerting that it bordered on art.

They made unnecessary queen movements, created meaningless geometric patterns with their pieces, and did clumsiness with the delicacy of a toddler. Undaunted, Stockfish methodically tightened his position, even reducing his free pawns in order to maintain his strategic advantage.

And just when the game seemed destined for a lopsided victory for Stockfish, ChatGPT pulled a rabbit, or rather the queen, out of the hat. In a bold counterattack, he illegally captured the queen and threatened Stockfish with checkmate. For a brief moment of brilliance, it looked like the Chaosbots were going to defeat the Titans.

Needless to say, the climax was completely fictional. ChatGPT advertised pawns that didn’t exist, declared checkmates that weren’t real, and hung on to its illegality like a badge of honor.

Stockfish has always been a pro, ignoring the nonsense and methodically hunting down the king of ChatGPT. In one final move, Stockfish achieved the following: coup d’etat By crowning a pawn and using its new queen to suppress the king.

ChatGPT had to move to one of the six available squares, but chose the only one still under attack.

Levy allowed it to self-destruct, as nothing made sense to it anymore.

Stockfish has captured the king and is now the undisputed champion of AI chess.

If you’re interested, the full bracket looks like this:

To be fair, the results weren’t surprising.

Stockfish is a pure chess logic machine honed to crush your opponents with precision and ruthless efficiency.

Generative AI, on the other hand, is like a child with crayons and a vague memory of rules. It wasn’t trained to play chess, but it might have some random chess books in it, so it knows how the pieces move. Its training dataset.

One brings order to the board, the other brings a teleporting queen.

That being said, there are many other genuine AI chess bots besides Stockfish. Actual tournaments featuring the most popular tournaments have been around for years.

If you want to watch that nonsense AI tournament. Click this link.

Edited by Sebastian Sinclair