Learning in the age of AI
Yesterday, I was out for a walk with my wife and my dog. Her hair salon recently started offering IV vitamin drips for $300. I asked if there was any science behind it, or any proof that a vitamin drip actually did anything for you, and my wife wasn’t confident that there was.
So I pulled up ChatGPT, put it in voice mode, and asked it. I had it find the exact salon my wife goes to, read its claims about the drips, and compare them to what the FDA and research papers say. The short version is that there isn’t much evidence it does anything for healthy people, since your body just gets rid of the extra vitamins it doesn’t need. By the end of the walk, we were both more educated than when we started.
Learning with AI is one of my favorite use cases. The internet made it way easier for anyone to go online and learn almost anything, and I think AI is doing that again. Being able to pull up a live conversation while I’m doing dishes or walking the dog and learn something interactively is something I find amazing. You get a hyper-personalized conversation about whatever you want, whenever you want, which wasn’t possible a few years ago.
A friend of mine recently started a new job and wants to learn Ruby because that’s what his company uses. I told him to have AI teach him Ruby by comparing it to TypeScript, which he already knows really well. This type of personalization is what I think makes learning with AI so powerful.
With any positive comes a negative, and I think this has no exception. Like a calculator on steroids, AI can let you skip learning entirely. Universities are already struggling with students generating essays or writing entire programs without actually learning anything. I agree that’s not good long term.
In my head, I split my time into two modes. There are times when I want to learn and times when I want to produce. University is about learning, not producing the most output. Work is about producing value, not learning (though it’s great when you can do both at the same time). That’s why at work, I believe you should use AI for all your coding. The goal is to ship, and AI lets you ship a lot faster. But in places where the whole point is to learn, like university or picking up something new outside of work, AI should help you learn and not cheat you out of it.
But can you trust the answers AI gives you? This one is hard to figure out.
On one side, AI can hallucinate, and it sounds just as confident when it’s wrong as when it’s right. That’s what makes it risky for learning. When you’re learning something new, you don’t know enough yet to notice when something is off, so you can end up learning the wrong thing and not realize it. It’s also easy to walk away thinking you understand a topic when you’ve only heard a simplified version of it.
On the other side, what’s the alternative? On that walk, my options were the salon’s own marketing, a quick Google search full of sponsored results, or just not knowing. The AI can read more sources than I ever would have and gave me a clear answer. It isn’t perfect, but it’s often better than what I would have found on my own.
My current philosophy is to trust AI by default for low-stakes learning. I still want a doctor to diagnose me, but I’d be okay with a doctor using AI, since they’re qualified to question it. For the vitamin drip question, I never fact-checked it, and I’m fine with that. I believe frontier models have reached a point where they can accurately summarize the right sources for questions like this with near perfect results. And I wasn’t going to make any decisions off of it. I was just curious.