It’s not every day that you get to read about OpenAI’s all-knowing ChatGPT being taken to the cleaners by a 46-year-old, 8-bit program, Star Wars-era chess engine.
But that’s exactly what happened in a matchup orchestrated by Citrix software engineer Robert Caruso.
The ChatGPT, Atari Matchup
It all began as a casual conversation between Caruso and ChatGPT about using AI in chess. Following that, the cocky chatbot suggested playing chess against the 1977-vintage Atari 2600, which could only look 1 to 2 moves ahead.
What followed was a 90-minute comedy of errors on the Stella emulator with ChatGPT confusing rooks for bishops, forgetting where the pieces were, and repeatedly making silly and illegal moves.
Switching the interface to standard algebraic notation — Caruso thought this would give ChatGPT a clearer view of the game — didn’t help either, and Atari won fair and square.
Copilot Loses, Gemini Backs Off
Atari’s next opponent, Microsoft’s Copilot, didn’t fare any better despite the bot’s tall claims that it could “think 10–15 moves ahead” and win easily. Copilot lost two pawns, a knight, and a bishop by the seventh turn, and it was ‘game over’ when it moved its queen into a direct capture.
Google’s Gemini AI — after reviewing Caruso’s prior matches — turned out to be the smartest of the lot: It downright refused to take on Atari, saying it would “struggle immensely,” while also citing “time efficiency” and a “sensible decision” as reasons for withdrawing.
LLMs’ Limitations
Atari’s triumph over some of the world’s most advanced large language models isn’t just shocking, it’s telling.
In an era of AI anthropomorphization, the multi-billion-dollar chatbots’ defeat to a 4KB chess engine shows these tools are only designed for language prediction and not structured reasoning, which requires strict logic and memory.
Image credit: AlexBuess/Shutterstock