A Grandmaster Trained an AI in Three Hours

Vidit Gujrathi built a model that guesses your chess rating from a single game, knowing no chess, in an afternoon. Here’s forty minutes on vibe coding from someone whose sport lost to a machine first.

The model saw 400,000 chess positions, cannot play a single move, and still guesses ratings closely. From there it becomes a conversation about working with agents: the tools he moves between, a personal assistant he wired into his own WhatsApp that answered someone on his behalf while he was at a wedding, and his estimate that this gets you 60 or 70 per cent of a project today. The chess parallel is the sharpest part, because the players who refused engines were left behind and the strongest players gained the most from them.