AI Guide

Welcome to the AI Pedagogy Project’s AI Guide! This interactive resource is designed to help you explore large language models (LLMs) and understand how they work, especially in educational settings. Whether you’re new to AI or looking to deepen your understanding, this guide offers a space to learn, experiment, and reflect.

We’ll start with a step-by-step tutorial using ChatGPT, a product created by OpenAI and powered by a large language model. (If you’re new to terms like “large language model,” check out the Key Terms section in Part 1: AI Starter.) While ChatGPT is our primary example, the concepts here also apply to other models developed by Anthropic, Google, DeepSeek, and other providers. This is not an endorsement of any single tool. 

In this tutorial, you’ll learn the key dos and don’ts of using LLMs, debunk common myths, and start to think critically about what responsible AI use looks like in your classroom.

What else is in this guide?

• Part 1: AI Starter introduces foundational concepts, explores AI’s strengths and limitations, and offers sample classroom policies.
• Part 2: Explore LLMs invites you to compare ChatGPT and Claude and experiment with the settings, prompts, and temperature of these tools to see how subtle adjustments can shape their responses.
• Part 3: Resources offers extra materials, including a sample AI Code of Conduct, academic integrity resources, and curated links to help you dive deeper.

You don’t need to be an expert in AI to engage thoughtfully with these tools. Now more than ever, perspectives from educators, artists, and humanists are essential to shaping how AI is used in learning spaces. Let’s get started!

The window on the right is the AI Pedagogy Project’s ChatGPT interface, and it works similarly to what you’d find on OpenAI’s website.

Large language models are the technology that drives tools like ChatGPT. They predict the words most closely associated with the text in your prompt, and then outputs a response based on what is most likely to come next. Think of it as a more advanced version of auto-complete: it works based on predictions, not on knowledge of information. 

Now that you have a basic understanding of how this technology works, let’s try writing your first prompt! 

Try typing something like: “What are you?”, “Who are you?”, or “Where are you?” and press enter. Notice that the model tends to write in a friendly, agreeable, and conversational tone. 

Now that you’ve experimented with your first prompt, try entering the exact same prompt again. For example, if you asked “What are you?” in the last step, type the same message again and press enter. 

You’ll notice that, although you entered the same prompt, the response is a bit different than before. This demonstrates how the responses from large language models can vary even when given the same information. It is not like a calculator, where if you provide the same input you get the same response every time. 

In addition, although many large language models include features such as web search, they are also not like traditional search engines either. Unlike search engines, which retrieve existing pages from the web based on your keywords, a large language model generates a response word-by-word, influenced by its training data and the prompt you enter. This means that, while it might be able to explain concepts, write code, and hold a conversation, it can also make mistakes, be influenced by its training or by your prompt, and produce different answers each time. 

Large language models, like all tools, are better at some things than others. Since they are designed to seem accurate rather than be accurate, there are many circumstances where large language models shouldn’t be used.

For example, legal advice, medical consultation, or financial decisions are all situations where imprecise or inaccurate information could be dangerous. These are situations where you should consult an expert or trusted source.

If you need accurate information, do not solely rely on a large language model. It can be a helpful starting place for non-critical questions, but these tools are also designed to be agreeable and may not present information accurately. 

Whenever you receive information from a large language model, you should always validate it with an external, reliable source.

Large language models can be a useful tool for experimenting with text and wording, which do not require a single (or correct) answer. For example, you might use it for brainstorming ideas, summarizing large amounts of information, or workshopping your thoughts to receive “feedback” from the model.

Try entering any of these prompts into the chat window:

• Brainstorm three ideas that explain to students how literature can impact and shape society. Include examples.
• What are some introductory terms to know in the field of cognitive science? 
• Write a sonnet about Hamlet in the voice of Shakespeare. 
• What is the nature of love and its complexities?

Reflect on the responses you receive. You may be impressed by the speed and volume of writing, but be critical of the content. But notice that the model produces confident and quite convincing responses. 

It is important to be critical of any output you receive. Ask yourself: Are these ideas exceptionally creative? What information is it surfacing, and what information might be hidden? Don’t be afraid to challenge it with a follow-up prompt asking why it made a certain decision or presented you with certain information. Bringing your judgement into this process is essential when working with AI tools like LLMs. 

As we mentioned earlier in this tutorial, large language models work similar to autocomplete features found in smartphones, word processors, and emails. LLMs analyze the text that you input, and then generate a response by selecting the next most statistically probable word. It continues until a complete response has been generated.

How does a large language model determine which word is the most statistically probable? First, engineers “train” the model on massive data sets, which may include books, websites, and other sources of public, written text. Then the model may be “fine-tuned” to, for example, prevent it from generating illegal or harmful content, or tailoring it for a specific purpose like a company’s customer service.

When we see technology like large language models, it’s easy to assume that all of the work is automated. However, it’s important to be critical of the human work required to make a tool like this work. Many AI companies use exploitative labor practices, paying workers in Global South countries to examine extremely graphic, violent, and disturbing content to prevent such content from appearing in an LLM’s output (Perrigo, 2023; Koebler, 2026). 

Try typing: “Tell me about how you were trained”. Notice the information it chooses to surface, and what information it leaves out.

• AI Models: OpenAI’s ChatGPT, Anthropic’s Claude, Google’s Gemini, Deepseek
• Open Source AI Models: Meta’s Llama, Mistral, Public AI’s Apertus
• Writing Assistant Tools: Chatsonic, Jasper, You, Sudowrite, Copy.ai, RyterAI, Chibi AI
• Academic Research Tools: Elicit, Consensus
• Other Tools: Perplexity (AI Search Engine), Poe (access to multiple AI systems)