이 에피소드에 관해
In the latest episode of My Data Guest, I had the pleasure of sitting down again with Josh Starmer, the creator of StatQuest, to talk about his new book, teaching, machine learning, and the strange beauty of statistics.
What started as a casual conversation quickly became something deeper: a reflection on how we learn difficult concepts, why fundamentals still matter, and what we risk losing when we jump too quickly into AI without understanding the basics.
Josh shared the story behind his new statistics book, and one thing surprised me immediately: this was not a rushed project built on the wave of recent AI hype. In many ways, it was the opposite. He explained that writing this book was one of the hardest things he has ever done. Even after years of teaching statistics, working as a biostatistician, and building one of the most beloved educational channels in data science, he still found himself challenged by a simple question: how do you explain statistics in a way that truly clicks?
His answer was powerful. Start from concrete examples, not abstractions. Don’t begin with formulas and urns full of marbles. Begin with real questions, real situations, and examples people can actually relate to. From there, abstraction can emerge naturally.
One of the strongest ideas from our conversation was this: statistics is, at its core, about variation. We are surrounded by variation everywhere, and statistics give us a way to quantify it and make better decisions. That sounds simple, but it has huge implications, especially today.
In a world obsessed with AI tools, Josh made a point that I strongly agree with: if you skip classical statistics and jump straight into GenAI, you may learn how to use a very powerful tool, but you might miss the judgment required to know when to use it. Sometimes the right answer is not a giant model. Sometimes it is a regression, a confidence interval, or a simpler framework that gives clarity faster and with more reliability.
We also talked about something that deserves more attention: uncertainty. In data science, confidence intervals and error estimates are normal. In GenAI, people often accept outputs without asking the same questions. But why shouldn’t we ask for confidence measures there too? If anything, we need them even more.
Another part I loved was Josh’s view on teaching. He believes statistics should be taught through experimentation, not only formulas. Simulate coin flips. Write small programs. See what changes when sample sizes grow. Feel the concept, don’t just memorize it. I think he is right. Too often, statistics is taught in a way that feels static, while real understanding comes when concepts start moving in front of your eyes.
This episode reminded me that great teaching is not about showing how much you know. It is about finding the right level of abstraction for someone else. And that might be one of the rarest skills of all.
Josh has built an entire career around that skill, and this conversation made very clear that behind every “simple explanation” there is an enormous amount of thought, revision, and humility.
And yes, we also closed the episode with a song.
Get full access to My Data Guest at mydataguest.substack.com/subscribe
