OpenAI’s ChatGPT lost a game of chess against the iconic retro console Atari 2600, which has only 128 bytes of RAM. The experiment showed that even the most powerful language models are unable to play chess consistently without a specialized chess engine.
OpenAI’s ChatGPT lost a game of chess against the iconic retro console Atari 2600, which has only 128 bytes of RAM. The experiment showed that even the most powerful language models are unable to play chess consistently without a specialized chess engine.
American infrastructure architect Robert Caruso conducted an experiment in which he compared the ChatGPT artificial intelligence with the ancient Atari 2600 gaming console, released in 1977. As a rival for ChatGPT, he chose the game Video Chess, which ran on the Stella console emulator. Caruso detailed the results in his LinkedIn post.
«ChatGPT was completely defeated at the initial level of difficulty,» Caruso wrote. Despite being given a starting chess position and an explanation, the bot confused pieces, including the rook with the bishop, missed obvious moves by the opponent, and lost its orientation on the board. The engineer even tried switching to standard chess notation, but it didn’t help: OpenAI’s language model «forgot» previous moves, made blunders, and couldn’t play the game consistently.
The Atari 2600, with its 8-bit 1.19 MHz processor and only 128 bytes of RAM, was a cult console from the late 1970s, one of the first mass-market gaming systems in the world. Video Chess is considered its technological breakthrough, as the developers created a functional chess algorithm despite limited resources. The game had a reputation for being relatively difficult for beginners and providing a decent challenge for average players.
Caruso, who initially expected a «nostalgic retro-style stroll,» instead watched as the AI failed to even follow basic chess rules. According to him, ChatGPT regularly asked to «start over,» promised to «learn from mistakes,» but for 90 minutes it was unable to stabilize the game. The game ended early with Atari winning.
ChatGPT is impressive in its ability to conduct dialogues, write texts, and solve problems, but it does not have a built-in chess engine on the level of traditional programs like Stockfish. Even with knowledge of chess rules, the AI is not able to reliably control the course of the game in long-term games without a graphical interface or integration with a real engine.
And while this comparison is more of a curiosity than a performance test, Caruso’s experiment clearly demonstrates the limits of modern AI in tasks that go beyond text generation. As the author himself wrote: «Have you played Atari today? ChatGPT would prefer you not to.»
We also recently wrote about a graduate of the business department at Northeastern University in the US who filed a formal complaint and demanded a refund of over $8,000 for a course when she discovered fragments generated by ChatGPT in her instructor’s handouts, even though students were officially prohibited from using AI.
