Bengaluru: Vidit Gujrathi’s thoughts seem to race ahead of his words. The 31-year-old Indian grandmaster is chatting about hackathons, building apps and AI tools, and a recent shout-out from NVIDIA. The chess professional has built two apps - a web app for blindfold chess and a puzzle-solving app based on real tournament positions. His latest creation, Kibitz, has him fielding calls from tech insiders and potential investors.
Technology, Vidit points out, has always fascinated him. He mentions experimenting with different kinds of software back when he worked as a second to Anish Giri. “I was limited by my technical knowledge back then. I didn’t know how to code. In 2023-24, when LLMs became much better, I started getting more hands-on with them and creating things. It felt magical,” he told HT.
Chess players were among the first to confront the rise of AI. Long before Generative AI entered the mainstream, they had learned to train with machines that could calculate faster and see deeper than any human. The watershed moment came in 1997, when IBM’s Deep Blue defeated Garry Kasparov in their New York rematch, marking a turning point not only for chess but also for artificial intelligence.
Two weeks ago, Vidit was playing the World Teams Rapid and Blitz Championship in Hong Kong when he came across a call for a submission from NVIDIA on X for a hackathon. “It was at the back of my mind. After the World Teams, I was preparing for the E-Sports World Cup, dividing 75 per cent of my time for chess and 25 per cent on building an AI tool. I wanted to solve a real problem rather than just create something to submit to the hackathon. I have always believed that chess engines don’t really convey to the viewer what’s going on in a grandmaster’s mind. The human side of it. We often look at the move prediction and go ‘oh it’s such an easy move, why isn’t he playing it?’ I thought I should train a model from scratch to predict how a grandmaster would think. It turned out to be a really good idea. I’m surprised no one else thought of it, especially the serious broadcast folks. It was a no-brainer. Kibitz, the model I built, can predict what grandmasters are likely to play. Broadcasters can use it, along with AI commentary.”
{{/usCountry}}Two weeks ago, Vidit was playing the World Teams Rapid and Blitz Championship in Hong Kong when he came across a call for a submission from NVIDIA on X for a hackathon. “It was at the back of my mind. After the World Teams, I was preparing for the E-Sports World Cup, dividing 75 per cent of my time for chess and 25 per cent on building an AI tool. I wanted to solve a real problem rather than just create something to submit to the hackathon. I have always believed that chess engines don’t really convey to the viewer what’s going on in a grandmaster’s mind. The human side of it. We often look at the move prediction and go ‘oh it’s such an easy move, why isn’t he playing it?’ I thought I should train a model from scratch to predict how a grandmaster would think. It turned out to be a really good idea. I’m surprised no one else thought of it, especially the serious broadcast folks. It was a no-brainer. Kibitz, the model I built, can predict what grandmasters are likely to play. Broadcasters can use it, along with AI commentary.”
{{/usCountry}}The tool’s first-move accuracy is currently 56 per cent, while its accuracy for the first three moves stands at 86 per cent.
The accuracy, Vidit believes, can be improved upon. “I tested it on past games and the data set was fairly small. I tried it on the recent Grand Chess Tour Rapid and Blitz and got a 60 per cent accuracy on the first move. But those were rapid games. I want to test it on a larger game set and see if it holds up.”
His skills are self-taught, ones that he picked up by reading blogs, articles, papers and watching YouTube videos of industry experts. “I follow the top AI, ML experts and founders on X. I found OpenAI co-founder Andrej Karpathy’s YouTube videos excellent. He explains how these things work at the fundamental level, the neural network. I read papers which are released and try to understand things at the foundational, first-principles level. Sometimes my X followers offer great suggestions. Paul Graham, who founded Y Combinator and is one of the finest writers on the start-up world, was one such suggestion. I try to acquaint myself with all the tools that are available. I try them out, brainstorm using existing models and learn a lot by doing things.”
Even when he was playing tournaments actively as a professional, Vidit found himself curious about experimenting with tech. During his 2024 Candidates tournament preparation, he hired software engineers to help him develop tools. “I would visit forums, speak to domain experts and see how I can leverage technology in my preparation. For example, I wanted to work on tactics. Most websites which offer chess tactics are aimed at combinations where you can say, win a piece. But elite chess requires you to train in a different way. The difference may not always be between winning and equal; more often, it is between equal and slightly better.
A primary reason for Vidit finding the time to delve into this passion project is because he decided to take a step back from competitive chess after the 2024 Candidates tournament. In the year leading up to the tournament, the grandmaster from Nashik played some of the best chess of his life. In early 2024, he touched his peak rating – 2747, climbing to No 14 in the world. “Around 2021-22, I told myself that I will give my all to chess for the next couple of years. I shut my YouTube and social media accounts and single-mindedly worked on my chess. I wanted to have no regrets and see how far I could go in realising my potential. I qualified for the Candidates tournament but when it didn’t go as well as I would have liked it to, I realised I didn’t have it in me to try again with the same sincerity and effort. I lacked motivation, was burnt out and wondering what I was going to do next if it’s not fighting for the biggest titles in chess. Also, it was tough for me to sustain that lifestyle. It wasn’t chess or technical skills, but not handling pressure and nerves well that cost me big opportunities.”
Vidit, who will play in the Global Chess League in September, has scaled back his competitive appearances and still has a couple of tournament invitations he is undecided about. He is currently ranked No. 33 in the world, with an Elo rating of 2697.
“I have a problem with chess. I feel that the level I’m playing at right now, is not my level. I feel a lot of dissatisfaction and believe I’m better than this. At least, if I’m going to be a semi-professional, I don’t want to be at this level. It’s the one thing that bothers me. All of these tech and AI experiments, I started doing with the problems that I face in chess. I can create solutions which are better than others because they don’t know that these problems exist. In general, every person who is an expert in their field should try to solve problems with the technology access that we have today. India should have its own frontier models.”
In time, he believes the “universe will extract” a mentoring role out of him, for a younger chess player. He’s eager about the possibilities that tech and AI hold for him and longs to visit San Francisco, the home of generative AI.
“In chess, you’re always playing a zero-sum game. Somebody has to lose for me to win. In tech and AI, I believe there are greater opportunities for collaboration. Everyone can win.”