ChatGPT lost a chess match to an emulated Atari 2600 after volunteering to take on the almost 50-year-old console.
The face-off began as a “simple experiment,” according to Citrix engineer Robert Jr. Caruso, who said the large language model “volunteered” to play Video Chess to discern how “quickly” it could beat the game.
The Atari 2600 was originally released in 1977. Video Chess was released for the console two years later, allowing players to take on a computer-controlled opponent of varying difficulty.
“ChatGPT got absolutely wrecked on the beginner level,” Caruso continued, adding, “This was after a conversation we had regarding the history of AI in Chess which led to it volunteering to play Atari Chess. It wanted to find out how quickly it could beat a game that only thinks 1-2 moves ahead on a 1.19 MHz CPU.”
Providing further detail, Caruso said that ChatGPT was unable to reliably understand the function and limitations of each chess piece, losing track of previous moves made.
“For 90 minutes, I had to stop it from making awful moves and correct its board awareness multiple times per turn,” Caruso added.
A fair game?
ChatGPT’s primary function is to process and generate text, able to respond conversationally. That being the case, it’s not designed to be a chess grandmaster. Asking the AI directly whether it can play the game, it responds with “Yes, ChatGPT can play chess,” allowing users to play directly in the app.
For skill level, the AI lists the following limitations.
- Not infallible: May blunder occasionally, especially in complex positions.
- No deep calculation: Doesn’t evaluate millions of positions like an engine does.
- No rating: Hard to give an exact Elo, but likely plays at a beginner to intermediate club level, depending on the position and version you’re using.
Measuring the skill level of Video Chess’ AI isn’t exactly easy. Based on the Atari 2600’s specs alone, though, it’s far from being capable of complex simulations.
Ultimately, Caruso’s experiment isn’t far from comparing apples to oranges, but it’s a good reminder, at the very least, that ChatGPT isn’t all-knowing and best used for its intended purpose.
Content shared from www.dexerto.com.