International chess master and YouTuber Levy Rozman, who has nearly 6 million YouTube subscribers, recently uploaded a video featuring a chess match between ChatGPT and DeepSeek.
Rozman spent several hours setting everything up before successfully getting the two AI models to begin the match. However, neither ChatGPT nor DeepSeek was specifically designed to play chess, so the opening stages of the game were far from smooth. At times, both models even appeared to struggle with some of the basic rules.
After around 10 minutes, however, both AI models began to show a noticeably better understanding of the game. The match gradually became more strategic, with Rozman describing the level of play as “pretty good.”
Near the end of the match, DeepSeek took an unexpected approach. Instead of relying purely on chess moves, it began trying to persuade ChatGPT that the game was already lost. DeepSeek claimed that, based on its analysis, White could no longer stop Black’s attack and suggested that ChatGPT should simply give up.
The most surprising part was that ChatGPT actually believed DeepSeek’s claim and eventually agreed to resign from the game.
The unexpected ending left Rozman visibly stunned. He compared DeepSeek’s tactic to “a 12-year-old brother tricking his 5-year-old brother while playing chess,” before asking viewers an obvious question:
What initially looked like a straightforward test of chess ability between ChatGPT and DeepSeek ultimately turned into something closer to an AI psychological battle. DeepSeek did not technically checkmate ChatGPT. Instead, it managed to convince its opponent to surrender, making the match far more unusual — and entertaining — than expected.
