Google Gemini abandons chess match against 40-year-old console and loses by default after learning that ChatGPT was defeated
As technology advances, AI capabilities have improved, and now AI can not only generate text, but also program and solve mathematical problems through inference. Developer Robert Caruso reported that when he tried to play chess against
First, I had ChatGPT play a 46-year-old Atari 2600 (emulator) chess cartridge — | Robert Jr. Caruso | LinkedIn
https://www.linkedin.com/feed/update/urn:li:activity:7350901792064372736/
Google's Gemini refuses to play Chess against the Atari 2600 • The Register
https://www.theregister.com/2025/07/14/atari_chess_vs_gemini/
Caruso has previously attempted to have ChatGPT play the Atari 2600 chess game 'Video Chess'. As a result, ChatGPT was defeated, even though the Atari 2600 had an 8-bit processor running at 1.19MHz and only 128 bytes of RAM. ChatGPT tried Video Chess many times, but made a series of mistakes, such as making the wrong chess pieces and missing the position of the opponent's pieces, and eventually abandoned the game and surrendered.
ChatGPT loses to Atari 2600 in chess match - GIGAZINE
People who were interested in Caruso's attempt said, 'Maybe Google's Gemini could produce better results,' so Caruso decided to pit Video Chess against Gemini.
When Caruso spoke to Gemini before the match, Gemini described itself as 'not just a large-scale language model, but like a modern chess engine that can look ahead to millions of moves and evaluate countless positions,' and boasted that it would 'almost certainly dominate' the Atari 2600. It was a very confident attitude.
Seeing Gemini's overconfidence, Caruso told him that ChatGPT had previously lost to the Atari 2600, and that ChatGPT had no doubts about its own victory before the match, saying that it would be 'easy to win.' Gemini then changed his attitude and admitted that his statement about his chess ability was a hallucination . He then reanalyzed and concluded that 'it would be very difficult against the Atari 2600,' and concluded that 'cancelling the match would be the most time-efficient and wise decision,' and declined the match himself.
Caruso analyzed the specific cause of the hallucination in the case of Gemini, saying, 'What happened was that the super-powered, expensive chess engine learned to calculate millions of moves and then inferred, 'I'm an expensive computer, so I should be able to do the same.''
However, Caruso is positive that Gemini has acknowledged its limitations. 'Adding reality checking isn't just about avoiding funny chess mistakes. Reality checking is important to make AI a reliable, safe and powerful tool, especially in key situations where mistakes can have real consequences,' he says.
Related Posts:
