Holding AI to a Teacher’s Standard
Our kids’ time matters as much as our own, and that includes the experience they have inside the apps they use. That’s why we built Rabbithole ourselves. LLMs like ChatGPT and Claude, out of the box, have the soul of an assistant, not a teacher, so we’ve been teaching Rabbithole the habits of great educators.
Great teachers know when to hold the line. AI is trained to give people what they ask for. In October, scholars designed defenses against microbes and turned their designs into working apps. My son chose cholera, and with an AI assistant eager to follow his orders, his project soon gained a crossbow, a vending machine, and a chicken. While we loved his creativity, we also wanted him to be thinking about cholera. So every app project now comes with a contract, written around the teacher’s goals: a few things must stay true to the science, and everything else belongs to the child. So the chicken can stay, but only once the disease spreads the way cholera really behaves.
Great teachers leave the thinking to the child. Left to itself, AI tends to steer a child through a chain of small questions toward an answer it already has. Educators call this funneling, and it looks like good tutoring, but it’s really the tutor who is doing the thinking. We’ve been training Rabbithole out of that habit. To see whether it’s working, we replay moments from past sessions and score the new replies with a published research guide for differentiating funneling from real follow-up questions, after first checking our scoring against expert ratings of classroom conversations.
Great teachers can distinguish between busyness and learning. A child can be absorbed for hours while thinking about something other than the subject at hand. Our teachers are involved every step of the way: designing each activity, steering it as it runs, and keeping scholars on track. When a scholar needs support, Rabbithole steers them toward human connection.
Creating AI with the soul of a teacher is a huge opportunity that is not fully solved, but we get closer every day as we develop new ways to evaluate, prompt, and harness frontier AI models. Try asking your scholar at dinner, “Who worked harder today, you or Rabbithole?” and let us know what they say.
Read more about how Tradewinds uses technology, or come to an open house.