Citrix engineer Robert Caruso Jr. told how he brought together the ChatGPT 4o neural network and the chess engine of the Atari 2600 game console, introduced in the late 70s, in one chess game. The opponents were unequal, but the loser, oddly enough, was the “omnipotent” AI.
According to Caruso, the idea to play the game came from ChatGPT itself during a conversation about the history of AI in chess. The chatbot wanted to know how quickly it could win, since the Atari 2600 processor with a frequency of 1.19 MHz can only calculate the game 1-2 moves ahead. Caruso followed the bot’s hint, launched Atari Video Chess through the Stella emulator and organized the match, but something went wrong.
“…ChatGPT confused rooks with bishops and constantly lost track of where pieces were – at first it blamed the Atari icons, which were too abstract to recognize, but it didn’t improve even after switching to standard chess notation. It made enough mistakes to get kicked out of the third-grade chess club…”, the engineer said.
During the game, the bot sometimes made “impressive” moves, but often the decisions were absurd. For example, the AI would sacrifice a knight for a pawn. The engineer concluded that ChatGPT’s ability to preserve the basic state of the board from move to move was “very disappointing” compared to the Atari 8-bit engine, which “just did its job.”