Carlsen had an easy victory over ChatGPT without even looking at the chessboard - Photo: ChessBase India
Despite easily defeating the chatbot without losing a single piece, Carlsen was rated by the AI as having an Elo rating of only 1,800 – a number that is outside the top 50,000 players in the world .
Carlsen came up with the unique idea while playing chess with OpenAI's ChatGPT during a long commute in a car. The game was played by both players typing their moves into the chat box.
Accordingly, Carlsen does not need a chessboard, the AI does not display images and the "Chess King" starts the game while hiding his true identity.
Carlsen took black and chose the Philidor Defense opening (1. e4 e5 2. Nf3 d6) – a position named after the great French composer and chess player Philidor.
ChatGPT even praised Carlsen, noting that he handled this opening "easily and effectively".
The funny story was shared by Carlsen on social networking site - Photo: social network X
After 53 moves, with Carlsen's bishop capturing the h5 piece (Bxh5), White (ChatGPT) resigned, saying he had lost all his pawns. Meanwhile, Carlsen kept all his pieces on the board intact.
"A very methodical and sharp play. Well played," commented the AI, looking impressed. Carlsen later responded that the AI played very well in the opening, even making some impressive sacrifices. However, ChatGPT seemed confused and did not know how to handle those sacrifices properly.
The climax of the surprise was in the "scoring" of ChatGPT's level. When Carlsen asked: What do you think of my chess level?
The AI gave a pretty detailed assessment, Carlsen had good opening understanding, patience with persistence, good tactical awareness and endgame technique. ChatGPT even "bragged" that Carlsen recognized some of its illegal moves and allowed it to move again.
But then, ChatGPT came up with a number that made the entire chess world laugh: "If Carlsen played standard chess regularly, he could reach an Elo rating of 1,800 to 2,000."
For a player who is currently ranked No. 1 in the world with an Elo of 2,839, and a rating of 1,800 that only corresponds to players outside the Top 50,000, this is indeed a confusing evaluation from the AI. This interesting conversation was shared by Carlsen on the social network X, attracting more than a million views.
Many fans are eagerly waiting for the "Chess King" to continue testing his skills with other chatbots like xAI's Grok. The Chess.com platform also did not miss the opportunity to tease: "Elo 1,800 is really strong. So happy for you, my friend."
While AI chatbots have made significant strides in the past few years, they are still generally strong in the opening game. In the middle game, AI often shows weaknesses, sometimes making “silly” or even illegal moves, confusing the positions of pieces.
ChatGPT's chess skills have improved, but they are still no match for grandmasters, let alone a genius like Carlsen. The fact that ChatGPT rates Carlsen as having an Elo rating of 1,800-2,000 means that the chatbot's actual level is much lower than that.
Despite completely dominating ChatGPT, Carlsen is still no match for specialized chess software developed specifically for this purpose such as Leela Chess Zero (Lc0) or Stockfish.
With Elo ratings exceeding 3,700 for Stockfish and over 3,600 for Lc0, these AI “monsters” are still on a whole other level.
The proof is that in a friendly match in early July, Lc0 even "gained a handicap" but still easily defeated world number 2 player Hikaru Nakamura (Elo 2,807) with an overwhelming score of 9.5-4.5.
Source: https://tuoitre.vn/chatgpt-danh-gia-vua-co-magnus-carlsen-chi-la-tuyen-thu-nghiep-du-20250719075032337.htm
Comment (0)