Education
- 2019 – 2023
Komaba Junior & Senior High School, University of Tsukuba
- 2023 – 2025
Kingswood School, Bath
- 2025 – Present
University of Cambridge, Computer Science (BA)
Highest Achieving Part IA Student (Ranked 1 / 129)
Affiliation
- 2023 – Present
Tazaki Foundation, Cohort 7
Read the reports
Seeing the world through the lens of intelligence.
I've always been drawn to intelligence. No inquiry can outrun the limits of the subject conducting it. That's why I don't think I can truly approach any other question without first understanding intelligence itself — it's the constant beneath however far my interests wander.
Current Research
Extending Policy Gradient Methods — University of Tokyo
Reinforcement learning feels close to the essence of intelligence to me: an agent coming to understand its environment through the interaction between itself and that environment, mediated by reward. But my work on hierarchical world models made me feel how short I fell on knowledge and experience, and that I needed to build real mathematical strength and intuition by working through the underlying theory head-on, in both math and code. With the good fortune of a great mentor, I'm now working to build an essential understanding of the theory behind policy gradient methods — specifically ordergrad, a method for optimizing metrics like max@k and pass@k — and to extend it using extreme value theory.
Mastering an interdisciplinary view
The world has recurring patterns, regardless of scale or subject — insight from one field turns out to be useful in another. This graph traces my own path of hunting for those patterns. Click a circle to see some of the activity behind it.
Intelligence, and beyond.
Every activity can be framed as an interaction between an agent and its environment. An agent with a goal refines the means to reach it as it understands that environment more deeply. That "deep understanding," I think, is the agent abstracting the environment — compressing it and internalizing it. That's part of what draws me to hierarchical world models: they let me work directly with how that abstraction happens.
But that understanding has its limits. The human brain runs into material constraints — neuron transmission speed among them — along with structural constraints left behind by evolution, and perception itself is imperfect.
As a kid, I visited the Galápagos Islands and was captivated by the sheer diversity of life there. That experience led me to a way of thinking: if life takes this many forms, why wouldn't intelligence too? Neither the human brain nor artificial intelligence gets to claim absolute superiority over the other.
Artificial intelligence carries its own constraints too, shaped differently from ours. But viewed together as one whole, human and artificial intelligence can compensate for each other's limits and push our understanding of the world further than either could reach alone. That's the same interest behind the work I'm doing at the University of Tokyo, extending policy-gradient theory.
The same is true of purpose, not just intelligence. Biology's origin is self-replication, but countless sub-goals have branched off from it — avoiding pain, predicting the future — and they drive us, trading dominance back and forth. In me too: the urge to know the unknown as soon as possible, and the doubt that expanding knowledge is even the point of it all, coexist, trading places depending on the day. Purpose isn't singular, even within one person.
In recent years, AI has gotten remarkably good, remarkably fast, at producing answers. What still doesn't exist is an agent that can think alongside us about what to aim for, or what even counts as a good question. Understanding intelligence itself feels like the long way around to get there — but the only way I trust.
Contact
About the research, or about the foundation.