"What is ChatGPT?" "I heard AI can write my lesson plans" "How do I create a good prompt?" These are real concerns we hear from teachers every day. To address these questions, I gave a talk at a staff meeting to help teachers with their understanding of AI. Given the positive response, we decided more of you might appreciate the content. Let's dive into understanding AI.
What is ChatGPT?
Think of ChatGPT like a super-powered predictive text system. Just as your phone learns to predict what you might type next, ChatGPT predicts what words should come next based on a massive amount of training data - but at a much more sophisticated level.
Let's break down what ChatGPT stands for.
Chat
You're interacting with a chatbot! You say something and it responds.
Generative
There are three types of AI that you may interact with on a daily basis. Reactive and Predictive AIs are far more popular, but Generative AI is what you're probably interested in!
Reactive AI
Like a traffic light changing based on the presence of cars.
Responds to specific triggers with pre-programmed actions.
Predictive AI
Like a GPS predicting your destination based on your usual routes.
Makes educated guesses about what you'll want next.
Generative AI
Like a travel agent creating a custom vacation itinerary from scratch.
Creates new content based on what it knows.
A Generative AI can generate a response based on a huge amount of training data and your initial prompt. The amount of information it uses to create its response is what really separates it from other types of AI.
Pre-Trained
It uses the huge amount of data it's been trained on to predict an answer. The answer is usually in the same format as the data it's trained on, be it text, images and, more recently, video.
Pre-training also includes reinforcement from humans. Those who make the model, tell it what a good answer is and what a bad answer is by rating responses during the training period.
Just How Much Training Data Are We Talking About?
Transformer
This is a pretty complex idea that needs a bit more time to explain. In short, a transformer is a type of neural network designed to seek the most relevant parts of an input. In other words, it can figure out how words in a sentence are related, and which parts are more important than others.
Want to know know more? Watch this video from Google Cloud Tech.
ChatGPT is a Generative Pre-Trained Transformer. That means it can Generate new text from the massive amounts of data that it was trained on, and it can do it while sounding like a human!
Who else is making AI Chatbots?
Making an AI, as we just layed out, requires a huge amount of data and a huge amount of power. Only the largest tech companies are creating these complex models right now.
- ChatGPT - Made by OpenAI who has major investments from Microsoft.
- Claude - Made by Anthropic who has major investments from Amazon.
- Copilot - Made by Microsoft.
- Gemini - Made by Google.
- LLama - Made by Meta (Facebook).
What about all the other AI tools like MagicSchool?
All of the other AI tools you might find on the internet, use the services of one of the major companies. MagicSchool, actually states that they use multiple AI services and choose the "best model" for the task at hand. Unfortunately, they won't tell you which system they use.
What does prompting mean?
Prompting is how we ask an AI to do something. You may find that when you ask it to do a simple task, you receive an answer that is too general or unfocused to be helpful. We can use specific strategies to get better responses.
An effective strategy is to use the Role, Task, Format (RTF) Technique.
- Set a role. 📋 Tell the AI who you want it to be.
"Act as an educational coach and lesson planner. You specialize in tech integration, inquiry, and project-based learning, following the BC curriculum. You are an expert in identifying opportunities for cross-curricular learning..." - Define the Task. 🪜 Tell the AI what you want it to do and how to do it.
"...Guide me through developing a Grade 3 Social Studies lesson plan by asking scaffolded questions, one at a time, about skills, learning milestones, assessments, and project structure. Challenge me on any blind spots you notice and suggest improvements..." - Define the Format. 📝 Tell the AI how you want it to format it's responses.
"...Keep asking questions until you can summarize the project with a detailed breakdown of key milestones, assessment goals, lesson structure, rubrics, and technology tools for integration."
Want to learn more about prompting? Check out this slide deck: AI 101 for Teachers: Large Language Model Prompting Guide (make a copy)
More Resources
- Great Read: How Large Language Models Work - From Zero to ChatGPT
- Great Watch: How LLMs Work - IBM Quicky explainer!