✨Tutorial: 4 Ways to Use LearnLM as a Professor

Welcome to AutomatED: the newsletter on how to teach better with tech.

Each week, I share what I have learned — and am learning — about AI and tech in the university classroom. What works, what doesn't, and why.

In this ✨Premium edition, I present a Tutorial explaining 4 ways to use Google’s LearnLM (a version of Gemini fine-tuned for learning). I cover how to use LearnLM

• to create sophisticated assessments that promote learning

• to develop clearer and more effective assignment instructions

• to provide more constructive feedback on student work, and

• to support student learning through guided tutoring

🧪👀 Why You Should Try LearnLM

Today's generative AI models often struggle to truly teach. They excel at presenting information but fall short of guiding students through the learning process. Over the past few years, many studies have found that this is the primary problem with students having access to these tools (see, e.g., here, here, or here), as anyone in the classroom already noticed. Effective teaching requires more than just providing answers; it requires understanding how people learn and adapting to individual student needs.

Google's LearnLM, a version of Gemini fine-tuned for learning, offers a promising solution. Unlike general-purpose models, LearnLM is designed to understand and apply pedagogical principles. However, it doesn’t assume that its creator’s theory of learning is correct; it can adapt its teaching approach based on specific instructions, a method Google calls "pedagogical instruction following."

For example, you can instruct it to use Socratic questioning or provide feedback that encourages metacognition. This flexibility allows it to support a wide range of teaching styles and learning contexts.

Early testing from Google and their collaborators suggests LearnLM outperforms other AI models in areas like inspiring active learning, managing cognitive load, adapting to learner needs, and stimulating curiosity. (You can read the lengthy white paper from the LearnLM team here.)

So, LearnLM is worth exploring as a tool to help (higher) educators, especially since it’s free to use (more on access below).

There is one catch: while LearnLM is a version of Gemini — most notable because of its massive context window — LearnLM is limited to 32,000 input tokens, not 2 million like most other versions of Gemini. (See the glossary at the end of this piece for explanations of terms like these.) I expect this to change over time, but for now it is an annoying aspect of LearnLM.

Still, LearnLM is very useful for educators for the above reasons, especially when combined with other LLMs that can condense large swathes of information to give it as inputs, as I explain below.

This Tutorial explores four key ways professors can use LearnLM to enhance their teaching.

I'll show you how to…

• create sophisticated assessments that promote learning

• develop clearer and more effective assignment instructions

• provide more constructive feedback on student work, and

• support student learning through guided tutoring.

These are the four “cases” I cover below. With each case, I provide detailed, practical guidance for working with LearnLM, including how to structure your course materials, craft effective prompts, and refine the model's outputs.

As long-time readers might expect, the net result is a very long block of text with extensive explanations and examples.

Since I know you probably don’t have time to read it all, check out my tip below for a way to quickly apply the Tutorial to your needs…

❓❓ Case 1: Create Quizzes and Tests from Course Content

Creating assessments that both measure and support student learning takes significant time and expertise. While we want our assessments to help students integrate concepts and think critically, it’s often the case that we lack the bandwidth to create ideal assessments. (I’m assuming I’m not alone in this…)

Let's explore how to use LearnLM to create effective assessments, from organizing your materials through refining the final product.

Step 1: Organize and Prepare Your Source Materials

To prepare your materials effectively, follow these six sub-steps, which I'll explain in detail below:

1. Gather all relevant course materials, from teaching content to supporting documents

2. Convert everything to digital, machine-readable formats

3. Structure your documents with clear headers and section markers for LearnLM to reference

4. Break down large materials into topic-focused chunks

5. Use other LLMs to condense lengthy content when needed

6. Remove sensitive information and organize content logically

First, gather your course content comprehensively. This means collecting whatever is relevant to the assessments you are trying to build. For instance: lecture recordings, presentation slides, class handouts, readings, homework assignments, and supporting documents. Include materials that define your educational goals, such as learning objectives, curriculum maps, and assessment criteria.

The goal is to give LearnLM access to both what you teach and how you want students to learn it. However, keep in mind LearnLM's 32,000 token context window — you'll likely need to break this content into smaller, focused chunks organized by topic or learning objective.

Next, make your materials accessible to LearnLM by converting them to easily parsed digital formats and managing their size. LearnLM can work with various file formats, but focused, text-based content chunks are most effective for assessment creation. Here’s a table for your reference:

The third step — structuring your documents — is crucial for both LearnLM's understanding and managing content within its 32,000 token context window. Use clear, consistent headers and section markers throughout your materials. These act as navigation points that help you both refer to specific content when creating questions and divide content into manageable chunks for LearnLM to process.

For example, start main sections with descriptive headers like "Unit 3: Macroeconomic Policies - Fiscal and Monetary Tools" or "Week 5: Quantum Mechanics Foundations - Wave-Particle Duality." Use subsection headers to mark important concepts within each main topic. This hierarchical structure makes it easier to split content into focused segments that fit within LearnLM's context window while maintaining coherence.

Order topics logically, whether chronologically or by complexity. Cross-reference related concepts explicitly in your section headers or summaries, but keep in mind that you may need to feed related sections to LearnLM separately due to its context window limitations. The goal is to create well-structured material chunks that help LearnLM understand both individual topics and their connections to other segments you'll introduce separately.

Fourth, and relatedly, consider enhancing your document structure with brief introductory summaries at key points (read: use brief prompts to get an LLM like ChatGPT to do this). These summaries are particularly valuable when working with LearnLM's context window limitations, as they help maintain coherence across separate chunks of content. They serve multiple purposes: they state learning objectives, highlight essential concepts, and most importantly, make explicit connections between different parts of your course that may need to be processed separately.

If you think the linkages are self-evident within each content chunk, this last sub-step isn't necessary. You could also explain the linkages in your prompts (more on this below).

Step 2: Access and Configure LearnLM

Begin by accessing Google AI Studio and logging in with your Google account. Create a new prompt and select LearnLM from the model dropdown menu (it will be under “Preview” and may be labeled “LearnLM 1.5 Pro Experimental”). The interface shows four main components: a system instruction box (at the top), an upload area (marked by a plus symbol), a prompt input window (“Type something”), and a “Run” button. There are other settings on the right side panel but you can ignore them.

At this stage, upload your prepared course materials using the plus symbol, but don't click "Run" yet — we'll craft an effective system instructions and prompt first in Step 3. The interface will list your uploaded documents in the upload area.

Remember that LearnLM will work with the internal structure you created in Step 1, using your headers and markers to understand relationships between concepts. (Or if you didn’t do enough of that, prepare to flag it explicitly in your prompts.)

Step 3: Craft Structured Instructions

Like Gemini and other LLMs, LearnLM works best with structured instructions that guide it through complex tasks. Your instructions have two components: “system instructions” and “prompts.”

The system instructions provide big-picture context about the role LearnLM is playing, who you are, and what you need. Here’s a schematic template for such a system instruction with filler bracketed parts.

After you insert this system instruction, input a prompt that has several layers, including any relevant background information, assessment question examples, desired output formats (what you want from the LearnLM beyond the questions themselves), and more detailed instructions for how LearnLM should proceed.

Here again is a schematic version of such a prompt, which I explain at greater depth afterwards:

The first layer of the prompt explains the structure and content of your uploaded materials. This helps LearnLM understand how to navigate and use your course content effectively. Describe the organization and relationships between your files, referencing the headers and markers you created in Step 1.

If you can or want, it is an option to paste the documents’ contents into the prompt itself, providing headers along the way (you can use markdown to demarcate structure, as I am in the above examples; #x for H1, ##y for H2, etc). This is more work but may get better results because it guarantees the textual content is in a format that LearnLM can parse.

After establishing the overall context and providing your files as background information, give several concrete examples that demonstrate the types of questions you want. These examples serve as models for LearnLM to follow, showing both the structure and the level of sophistication you expect.

Next, provide a clear format for the output. Ideally this would match the format of the examples above, but it isn’t necessary (this is extra work; I don’t do this here). This structure ensures consistency and makes the assessment easier to review and implement. The format should capture both the question itself and its pedagogical purpose:

Conclude with step-by-step instructions that explain how a generalist pedagogy expert (LearnLM) should proceed through what you provide to achieve your goals.

Step 4: Review and Refine Generated Assessments

Review the assessment both holistically and question by question. Does it create a coherent learning experience? Check how LearnLM has referenced your course materials and explained its pedagogical choices.

When you find areas for improvement, structure your revision requests clearly. For example:

You can request multiple improvements simultaneously:

Given its context window size, LearnLM remembers previous versions and feedback throughout refinement, allowing you to reference specific elements when requesting changes. Through this process, you can develop an assessment that both measures and supports student learning.

(If you go on for a while, or if you have a lot of documents or files, you will reach the end of the context window and may need to start a new conversation to ensure your initial prompt is retained.)

📄✏️ Case 2: Improve Pedagogical Quality of
Assignment Instructions and Rubrics

Clear assignment instructions and well-designed rubrics are essential for student success, yet crafting them requires balancing detail with clarity. Students need to understand both what to do and how their work will be evaluated, while instructors need rules or guidance for consistent grading across many submissions (and many levels of fatigue).

What makes LearnLM particularly valuable for this task is its understanding of pedagogical scaffolding and assessment design. The model can analyze your current materials and suggest improvements that clarify expectations, highlight learning goals, and create better alignment between instructions and evaluation criteria.

Subscribe to Premium to read the rest.