Learning: A ChatGPT Based Approach

Recently I’ve subscribed to ChatGPT to get ChatGPT 4 as a tool to help teach myself math. It’s a great tool in the toolbelt for breaking down hard concepts.

I really like the problem solving approach, where the textbook has some light exposition on its topics and lots of problems.

The problems are where the hard work gets done – learning inside your brain.

But most textbooks aren’t geared towards self-learners – they rarely have solutions and even the ones that do may have incorrect or terse solutions. Since you want (preferably) multiple solutions to an answer, that won’t fly. ChatGPT solves that problem by being the bridge between problem and solution.

You can ask it to drill deep into one part of the problem, rephrase the problem another way, give similar problems to a problem, ask about certain concepts and many more. The sky is the limit. The math is a lagniappe.

You can learn almost any concept you’d like as well, being mindful of hallucinations – but that’s not that big of a downside.

When I was younger, I researched topics through the library. I’d find some “authoritative source” (whatever that means) which was usually a fairly dated book, skim it, find the main ideas, and engage with it, either choosing to devote more time to fully reading it or giving up there. Those books had plenty of “hallucinations”, where they mixed up historical facts or had a message to push. In the humanities, you can’t get a purely objective view of the world, even for the smallest events and ideas. You have to side with the author or against the author sometimes, and find other sources that agree or disagree with the points you’re currently reading.

In other words, gaining knowledge has always been lossy, and always will be lossy. Using an LLM in this way is no different than opening up a book at the library – it’s an adversarial experience.