During a somewhat tedious journey, Carlsen came up with the idea of playing chess against OpenAI's chatbot. They played by inputting chess notation for each move into the chat window. The Norwegian grandmaster didn't need a physical board, and the AI didn't display one either. Carlsen also kept his identity as the world's top player hidden.

Playing black, Carlsen opted for the Philidor Defense (1. e4 e5 2. Nf3 d6). Philidor, a French composer and the strongest chess player of his time (1726-1795), was the namesake of this opening. ChatGPT praised the 34-year-old's handling of the Philidor Defense as easy and effective.

Carlsen playing chess against ChatGPT. Image: ChatGPT

After 53 moves, ending with Carlsen's bishop capturing a piece on h5 (Bxh5), White resigned. The AI explained its resignation by stating it had lost all its pawns. Carlsen, in contrast, hadn't lost a single piece. "Such a systematic and sharp style. Well played," the AI commented.

Carlsen replied that the AI played the opening well, even executing an impressive sacrifice, but failed to handle the position correctly afterward. The five-time world chess champion then inquired, "What do you think of my chess skills?"

The AI assessed Carlsen as having a good understanding of openings, patience with a positional style, tactical awareness, and strong endgame technique. ChatGPT also noted that Carlsen had identified some of White's illegal moves and allowed it to take them back.

Surprisingly, ChatGPT estimated Carlsen's Elo rating to be between 1,800 and 2,000 if he played standard chess regularly. This rating range corresponds to players outside the world's top 50,000, while Carlsen currently holds the number one spot with a 2,839 Elo rating.

Carlsen shared the conversation on X (formerly Twitter), attracting over one million views. Many people expressed hope that he would play against other chatbots, like xAI's Grok. Chess.com jokingly commented, "1,800 Elo is really strong. Good for you, buddy."

While AI chatbots often play openings well, they tend to make weak, even illegal moves, or confuse piece positions in the middlegame. ChatGPT's chess skills have improved significantly in the past few years but are still no match for grandmasters, let alone Carlsen. ChatGPT's assessment of Carlsen at 1,800 to 2,000 Elo suggests its own playing strength is significantly lower than that range.

Despite dominating ChatGPT, Carlsen is still no match for specialized chess engines like Leela Chess Zero (Lc0) or Stockfish. Stockfish's Elo has surpassed 3,700, and Lc0 is also over 3,600. In a friendly match on 1/7, Lc0, giving a rook handicap, decisively beat world number two Hikaru Nakamura (Elo 2,807) with a score of 9.5-4.5.

Xuan Binh