Tony Yum

Games, fighting and programming

As a kid, Tony loved three things: computer games, martial arts and programming. His childhood diary lists what he wanted to do when he grew up. Two of them were to become a programmer (a game programmer, to be exact) and to become a good martial artist.

At Imperial College London he studied advanced computer graphics, 3D animation and advanced artificial intelligence.

Startups and banks

Tony has always been drawn to startups. At Bloomberg, his first job, he met a friend, and together they started their first company. It was built on the Google Maps API: a social map of places with a Reddit-like ranking system. You could vote places up, and your friends could see what you had rated and which restaurants you recommended.

But competition in startups is tough, and in the UK it is harder than in the US. The two founders went their separate ways. His partner moved to Silicon Valley and kept going, startup after startup, and is now very successful. Tony was getting married at the time and chose more practical work. He joined Goldman Sachs, and has since worked at several investment banks, including JP Morgan and Bank of America. Along the way he studied part-time for a master's degree in Mathematical Trading and Finance at Cass Business School.

Time out to train

It was never just a career. Tony has quit his job more than once to train. The first time, he went to Thailand to train Muay Thai, and fought professional fights there. Another time he focused on Brazilian Jiu-Jitsu, and earned his purple belt.

During that time he was also running algorithmic trading on a highly liquid crypto market. The machine learning models did not try to predict the market. Instead they chose what position to hold, from fully long to fully short or anywhere in between, based on market indicators such as price and volatility. The error function was designed to minimise risk and maximise return.

In the same period, alongside the Jiu-Jitsu, the algorithmic trading and programming games, he went back to Imperial College London to start a PhD in the Department of Computing. The research was on crypto trading with neural networks, machine learning and reinforcement learning, close to what he had already been doing as a trader. Unfortunately, the funding fell through, and with so little time to spare, he could not continue.

Two gyms, run by software

With a friend, Tony started Artillery Kai, a gym in Woolwich, London, that teaches Muay Thai, Brazilian Jiu-Jitsu and MMA. When he moved to Japan, he left the gym to be run by software he wrote. Artillery Kai later merged with Diesel Gym to becomeDiesel AK. The merger brought staff and coaches who could teach, and he has other people on the ground to help run it. But he never stopped running the gym. The marketing, the finance, the accounting, the software and, in time, the artificial intelligence that helps manage the gym are all looked after by Tony.

He then started a second gym, in Tokyo, with a focus on teaching kids. By then he was head of artificial intelligence at his company, and his passion had grown from games, fighting and programming to include AI.

Teaching martial arts to an agent

Tony has spent a lot of time reading about machine-learned motion and producing it with various frameworks. But time is always short. With a full-time job, a family and two gyms to run, there is very little left over.

In the age of AI agents, there is now a way to carry on that work: to teach martial arts to an agent. And doing it feels not so different from teaching a human. You can use natural language to explain how to tuck the elbow in or rotate the shoulder, the same things you would tell a student. You can give reference images that show how you would execute the technique. And over time, the agent learns.

That journey has now begun, and this blog records it. As well as learning techniques, the hope is that agents will one day fight each other, the way Go programs learned by playing themselves, and so learn to fight better, and perhaps create techniques of their own. There are unknown unknowns. Maybe we will discover something in martial arts that nobody has thought of yet.