ChatGPT loses at chess to 1977 Atari game console

OpenAI's ChatGPT lost a chess game against an old Atari 2600 game console released in 1977. This console has only 128 bytes of RAM.

An experiment conducted by American infrastructure architect Robert Caruso showed that even the most modern language models cannot consistently play chess without a specialized chess engine. Caruso said this in his LinkedIn post.

ChatGPT "suffered a complete defeat at the initial level of difficulty." Despite being given a starting chess position and an explanation of the rules, the bot confused pieces (for example, the rook with the bishop), missed obvious moves of the opponent, and lost orientation on the board. An attempt to switch to standard chess notation also did not help: OpenAI's language model "forgot" previous moves, made gross errors and could not play the game stably.

The Atari 2600 is a cult console of the late 70s, one of the first mass-produced gaming systems in the world. It is equipped with an 8-bit processor with a frequency of 1.19 MHz and only 128 bytes of RAM. The game Video Chess for this console was considered a significant technological achievement of its time, as the developers created a functional chess algorithm, despite extremely limited resources. The game had a reputation for being quite difficult for beginners and was a worthy challenge for ordinary players.

Although ChatGPT can conduct dialogues and solve problems, it does not have a built-in chess engine at the level of traditional programs such as Stockfish. Even knowing the rules of chess, AI is unable to reliably control the course of the game in long-term games without a graphical interface or integration with a real chess engine.

As a reminder, on Tuesday, June 10, users around the world encountered disruptions in the work of OpenAI services, in particular ChatGPT, the Sora tool, and the API.