Augmented Learning Basics

I've wanted to write about AI-assisted learning, and working with AI on projects, for a while now.

I'm not quite ready to go into the details on how AI has helped me accomplish my goals faster than I could have imagined, the foundation of how I use it can be explained quite succinctly so here we are.

The focus of this article will be on chatbots. It's important to understand what they are and what they're good at first.

What's a LLM ?

Noob answer: A large language model is a REALLY REALLY REALLY sophisticated version of the predictive text on your phone. It takes your prompt and predicts the next word to your answer one chunk at a time.

Intermediate answer: A LLM is a meaning calculator. It takes your prompt, translates it into arrows in one-zillion-dimensional space, then based on how it interprets what you wrote it comes up with a response, which it translates back into text.

How to work with LLMs

A classic mistake I see all the time is to see working with AI as a single transaction.

"Imagine you're the best marketing consultant in the world. Come up with a product I can sell and write a marketing campaign for that product."

This example is absurd but hyper sophisticated versions of this mistake substitute this short sentence with hundreds of pages worth of text and take the singular answer as the end result.

The key to working with AI is to see it as an iterative process. You and the bot are exchanging tokens with each other, mainly in the form of text, until the bot writes the text you needed to read to get the insight you were looking for, or the idea or the document or the code, etc…

In the iterative process, the chatbot mainly plays one of three roles.

The Tutor, The Intern, and The Rockstar

The Tutor

AI can serve as a personal teacher to help you explore whatever topic you want to learn more about. Asking it questions until you understand the topic enough to get a basic grasp of it to get inspired or get the skill you were looking for is an amazing process.

A good way to learn is to apply the theory in a project. Projects have all sorts of tasks. Some fun, some boring. Some of the boring ones can be handled by…

The Intern

LLMs write whatever you ask them to (within TOS 😊). You can offload the boring stuff to the bot. Something to always keep in mind is that LLMs have a stochastic element to them. If you ask a bot to do something you couldn't do yourself you can't evaluate whether or not it did a good job.

If you could do the job yourself but would rather have the bot do it you can just ask it to. Meanwhile you can handle the fun tasks yourself.

But like all projects you run into problems. The tutor can help you learn what you need to learn to solve them. If the solution is beneath you you can have the intern do it.

If you understand the solution well enough to do it, but the effort required would be substantial enough that you don't have enough time to do it manually, that's when you turn to…

The Rockstar

You have the blueprints, you understand them, you'd enjoy turning them into reality but you gotta get it done NOW. Now's the time to call in the rockstar.

The rockstar and the intern serve the same purpose. You give fun-but-hard tasks to the rockstar.

Putting It All Together

Whether you're trying to learn something by building something or building something for other reasons, AI can help get you up and running at lightning speed when used right!

Brainstorming with AI is a different thing than working with it per se. If needed you brainstorm with the bot until you get the insight you need and then you can either start working with the bot right there or start a new chat.

When working with AI it's vital to give it tasks that you can evaluate the result of yourself.

If you couldn't tell if the problem was solved properly or not you ask the Tutor questions about it until you decide whether or not you'd rather solve it yourself or have the Intern/Rockstar do it.

When using AI from that perspective, what you write in any given prompt matters less. Skillful prompting helps enormously but as you build working relationships with the bots you get better at doing so over time.

That's my approach in a nutshell. I hope to add a visual to this soon.

Syd