ChatGPT lost a game of chess to the Atari 2600 Video Chess game
By ForGeeks A Citrix engineer conducted an experiment in which OpenAI’s ChatGPT chatbot lost a chess game to a program from a 1970s video game console.
In a curious test, engineer Robert Caruso ran an emulator of the classic Video Chess game released for the Atari 2600 in 1979 and set up a game against ChatGPT. Despite all the technological advances, the result was not in the neural network’s favor.
An engineer reported serious errors by ChatGPT during the game
Caruso said in a post on LinkedIn that ChatGPT was “confusing rooks with bishops, overlooking simple pawn combinations, and getting lost in piece placement.” At first the chatbot referred to abstract Atari icons, but after switching to standard chess notation, the situation didn’t improve.
At first, the chatbot referred to abstract Atari icons, but after switching to standard chess notation, the situation didn’t improve.
“ChatGPT made so many gross errors that it would be laughed at even in a third-grade chess class,” Caruso wrote. He said the AI has “completely failed even at the entry level.”
The experiment lasted 90 minutes. During this time, ChatGPT asked to start the game again several times, but the whole game passed with systematic errors on the part of the chatbot.
Results confirm ChatGPT’s limitations in chess
Since the late 20th century, chess has become the benchmark way to evaluate the intelligence of computers. The greatest resonance was caused in 1997, when IBM’s Deep Blue system defeated grandmaster Garry Kasparov. But Caruso’s experiment underscores that a language model such as ChatGPT is not designed for strategic tasks at the level of specialized chess engines.
Developers have previously created a ChessGPT plugin designed to play chess using ChatGPT. However, Caruso himself is sure that it’s better to use the AI to discuss chess strategies rather than to participate in games.
The ChatGPT loses to Atari 2600 Video Chess was first published on ITZine.ru.
