2025-03-01 –, Main Hall (LH 111)
Most programmers think notebooks are about combining code with text and graphics. But that's like saying programming is about combining symbols with semicolons. The real power of notebooks - like their ancestor, the REPL - lies in enabling a particular kind of conversation with our data. When we understand this, we can use notebooks not just as a place to write code, but as a tool for developing the kind of clear thinking that leads to better research. This talk explores what that means, and why it matters.
The hardest problems in data science aren't about code or algorithms. They're about understanding: understanding your data, your models, and why they behave the way they do. Notebooks promise to help with this understanding. But the way we use them usually makes things worse.
The core problem is that we've confused two different things: exploring data and building understanding. When you're exploring, you want to try things quickly. But understanding requires something else - it requires thinking clearly about what you're doing and why. The REPL pattern, which notebooks are built on, was designed for this kind of thinking. But we've buried it under layers of state, hidden dependencies, and complex interfaces.
This talk examines what actually helps us understand complex datasets. We'll look at why certain notebook practices that seem convenient actually make understanding harder. Then we'll explore patterns that work: ways to separate exploration from explanation, manage complexity, and build reliable insights.
The key insight isn't about notebooks themselves - it's about how we think when we're trying to understand something complex. Once you see this, you'll approach interactive computing differently. Not as a way to make exploration easier, but as a tool for thinking better about hard problems.
Beginner
Category:Data Science/Analysis/Engineering
Clark Urzo is the Strategic Director of WhiteBox Research, the first AI interpretability lab in the Philippines. Previously, he participated in the virtual workshops of the Stanford ML Alignment Theory Scholars (MATS) program under John Wentworth and was also a facilitator in BlueDot Impact’s AGI Safety Fundamentals course. In a past life he was the co-founder of Veer, one of the first virtual reality startups in Manila.
Clark Urzo also won a grant from Pioneer.app in 2018, a selective program in Silicon Valley run by Daniel Gross and funded by Stripe and Marc Andreessen. In his spare time, he collects weird films and used clothes and spends his evenings playing 90s-era survival horror games.