ChatGPT Loses to 1979 Atari Chess - Even at Beginner Level

Over the weekend, Citrix engineer Robert J. Caruso conducted an unusual experiment: he launched ChatGPT 4o against Atari Chess for the 2600 console, released back in 1979. And, contrary to expectations, the modern neural network lost to the oldest chess program even at the basic difficulty level. According to Caruso, he played a series of games in the emulator and did everything he could to help the AI ​​- including replacing abstract icons of the pieces with more recognizable ones.

Although The Atari 2600 had only a 1.19 MHz processor and minimal computing resources., its simplest chess engine, which calculates a maximum of 1-2 moves ahead, turned out to be too “complex” for the GPT model. Caruso notes that ChatGPT made a series of obvious mistakes, which would not allow even a children's chess tournament to pass. The AI ​​constantly tried to learn as the game progressed and promised to improve the result, but Even with hints I still couldn't win.

This case reminds us that the capabilities of the language model are still limited — especially in tasks that require strict logic, visual analysis, and memorization of the board state. Despite the enormous computing capabilities of modern AI, even an old 8-bit engine can outplay them in a strictly formalized task. In the end, as Caruso writes, ChatGPT conceded defeat, succumbing to the 48-year-old game.