Magnus Carlsen beats ChatGPT in chess game without looking

Magnus Carlsen beats ChatGPT in chess game without looking

Avatar of thepuppylover2016
| 0

Hi today we are going to see a  shocking game Magnus played it is Magnus vs chatGPT and more shockingly Magnus was playing blindfolded and still won

World number one Magnus Carlsen defeated ChatGPT at a Freestyle Chess Grand Slam side event in Las Vegas while seated with his back to the board, even allowing the AI to make several illegal moves.

OpenAI's chatbot played black, with a human assistant entering Carlsen's moves and executing ChatGPT's on the physical board. Early on, ChatGPT called Carlsen's opening "strange," to which he smiled and agreed, but he soon described the AI's approach as "passive."

Unlike top chess engines such as Stockfish, ChatGPT occasionally offered very human-like ideas, and sometimes non-existent piece locations or proposed illegal moves. In the spirit of the game, Carlsen let some stand.

Midgame, Carlsen made a move which ChatGPT warned was "risky." When Carlsen asked if it saw the danger, the bot appeared confused and he took advantage of this. A human assistant filtered out a few of ChatGPT's illegal move suggestions and replaced them with legal ones, with Carlsen's consent.

In the endgame, Carlsen jokingly "moved" his king like a knight, an illegal move that ChatGPT nevertheless praised as strong. From there, the board turned into chaos as ChatGPT revived pawns, moved rooks into odd postitions and captured pieces without a logical reason. Carlsen eventually delivered the checkmate by having his queen leap over his own king, which is also illegal, but the result was accepted given the exhibition setting and the AI's prior absurd moves.

The match took place after what organizers billed as the first AI chess tournament, held Aug. 5–7, where OpenAI entered the competition with two representatives: the o3 large language model and the faster o4-mini.

Powered by o3's reasoning, ChatGPT won the event 4–0 in the final against Elon Musk's Grok 4, with o3 making significantly fewer errors than the model used in Carlsen's exhibition. The o4-mini model is a speed-optimized variant available free to ChatGPT users, while access to the more capable o3 is limited for free accounts to a small number of uses per month.

Thanks for listening

Hi

Thanks for visiting my blogĀ 

hope you like them