Author: Peter Pappas

Posted on

AI Magic: Create Podcast-Style Summaries with STORM and NotebookLM

Image by Marco Migorança Migorança from Pixabay

Imagine this: you’ve got a mountain of research, articles, PDFs, and notes staring back at you. Instead of sifting through it all, what if AI could pull the best bits and hand you a podcast-style summary? Magic, right? Well, thanks to tools like STORM and NotebookLM, it’s more than possible—it’s a breeze!

In this post, I’m going to show you how to harness the power of these two AI wizards to take your research from the page (or PDF or website) to your earbuds—turning hours of reading into minutes of listening. And here’s the kicker—it took me less than 10 minutes to pull off a podcast-style summary using both tools.

Ready to dive in? Let’s explore how these tools can transform your workflow, whether you’re a teacher looking for new ways to engage students, a student cramming for exams, or just AI-curious and want to supercharge your content game.

What Exactly Are STORM and NotebookLM?

STORM is an AI genius built to help you gather and organize massive amounts of info. (It’s a project of Stanford Open Virtual Assistant Lab). Think of it as your personal research assistant, whipping up “Wikipedia-style” summaries, complete with footnotes. It’s perfect for digging into deep topics, whether you’re writing a post, preparing a lesson plan, or pulling together materials for your next class.

On the other hand, NotebookLM—Google’s brainchild—takes all that research (or any other text you throw at it, like webpages, PDFs, or slides) and creates engaging, podcast-style audio summaries. It’s like turning your research into a conversation—perfect for auditory learners, multitaskers, or anyone who just prefers listening over reading.

But the real magic happens when you combine these two tools. STORM sets up the content, and NotebookLM turns it into something you can listen to on your next walk.

Here’s How to Use STORM and NotebookLM (a.k.a. The Dynamic AI Duo)

Step 1: Research and Create with STORM

Let’s say you’re working on a project, lesson plan, or maybe an essay. Fire up STORM and let it gather and synthesize all your research into a tidy, well-organized summary. It’s like having an assistant that never takes a break.

  • Pro Tip: Keep your prompts simple. You don’t want to confuse your AI buddy!

STORM does all the heavy lifting by pulling in key information, then creating a Wikipedia-style summary (footnotes and all!). Once you’re happy with your summary, download it as a PDF, and you’re ready for the next step.

Step 2: Summarize and Transform with NotebookLM

Now that you’ve got your research sorted, it’s time to add a little more AI magic. Upload your PDF to NotebookLM (it also works with public YouTube URLs and audio files directly into your notebook, and PDFs, Google Docs, Slides, websites – add as many as you want). From here, NotebookLM can create a variety of summaries. Including:

  • FAQs
  • Study guides
  • Table of contents
  • Timeline
  • Briefing document (main theme, most important ideas, conclusion)

But my favorite output is: NotebookLM turns your content into an audio summary that sounds like a conversation between two chirpy AI hosts.

On Dec 13, 2024 Google announced some updates to the look and function.

From the start, we wanted NotebookLM to be a tool that would let you move effortlessly from asking questions to reading your sources to capturing your own ideas. Today, we’re rolling out a new design that makes it easier than ever to switch between those different activities in a single, unified interface.

The redesign organizes NotebookLM into three areas. The “Sources” panel manages all the information that’s central to your project. The “Chat” panel lets you discuss your sources through a conversational AI interface with citations. The “Studio” panel lets you create new things from your sources with a single click, like Study Guides, Briefing Docs and Audio Overviews.

This design also fluidly adapts to your needs. For example, you can expand the source viewer and notes editor side by side if you want to take notes on an important document, or you can ask questions in the chat while listening to an Audio Overview.

    Join and interact with the AI hosts in Audio Overview

    Today, we’re starting to roll out the ability to “join” the conversation. Using your voice, you can ask the hosts for more details or to explain a concept differently. It’s like having a personal tutor or guide who listens attentively, and then responds directly, drawing from the knowledge in your sources. Here’s how you can use it:

    • Create a new Audio Overview.
    • Tap the new Interactive mode (BETA) button and hit play.
    • While listening, tap “Join.” A host will call on you.
    • Ask your question!

    Fun and Creative Use Cases

    Let’s get creative! Here are some ways to use STORM and NotebookLM beyond just summarizing research:

    1. Study Aids: Turn your notes or textbook summaries into quick audio sessions to review while you’re on the go. Perfect for prepping for exams without sitting at your desk.
    2. Lesson Recaps: Teachers, imagine turning your daily lesson summaries into a podcast for your students. They can catch up on the go, ensuring they stay on track.
    3. Interactive Class Discussions: Students can upload class notes and turn them into audio summaries to share with classmates—perfect for group study sessions or reinforcing what’s been covered in class.
    4. Meeting Recaps: Don’t have time to re-read meeting notes? Turn your minutes into audio form and catch up while you’re commuting or making dinner.
    5. Book Summaries: Summarize key insights from books you’re reading (whether for class or pleasure), so you can “read” while you’re doing something else.

    Why Audio Summaries are a Game-Changer

    Turning written content into audio isn’t just about convenience—it’s about making your work more accessible and engaging. Podcasts are perfect for multitasking (commuting, anyone?), and the conversational format keeps listeners hooked. With NotebookLM, you can even involve your audience by integrating voice messages, making the experience more interactive.

    With STORM organizing your research and NotebookLM turning that content into an engaging audio experience, the days of drowning in text are behind you. Whether you’re a teacher, student, or just looking for a way to streamline your learning workflow, these tools will give you a fresh way to interact with your content.

    Posted on

    Hands on With AI

    I’m offering a workshop for educational leaders at AI Empowered EDU conference at the University of Portland on May 14. My session is called “Hands on with AI.”

    I created this website to support the workshop.

    The site gives tips on how to create prompts, best AI assistants, and sample prompts. The prompt library offers sample prompts useful for students, teacher and educational leaders.

    It also includes an excellent how-to videos.

    Here’s the workshop agenda.

    Are you curious about the potential of artificial intelligence in education? Join our interactive workshop to explore how AI can enhance teaching and learning.
    Visit the LLMs Page to see suggested AI assistants.

    – Hands-On Exploration: Bring your own laptop and dive into practical activities. Collaborate with fellow participants in small groups as we explore AI tools and applications.
    – Use Cases: Discover real-world examples of AI from the perspective of student, teacher and administrator. You’ll explore prompts and scenarios that improve efficiency, spark creativity and enhance problem-solving.
    – Tool Showcase: The session will feature custom website that showcases AI tools and methods relevant to education.

    AI-created featured image above

    Posted on

    When Google Couldn’t Help: How AI Helped Me Track Down an Elusive Image

    George Tooker's "Government Bureau" (1956)

    Recently I found myself in a situation where “Googling” was useless. I was trying to recall a work of art without knowing the artist, title, or country of origin.

    Note: At some point in my life, I had seen George Tooker’s “Government Bureau” (1956) – a striking image (above) that had burrowed into my memory. But I lacked the descriptors that would work in a traditional text-based search.

    Armed with nothing but a vague sense of the era and the artwork’s elusive “feel,” I turned to Meta AI for assistance. What followed was a bit magical—a journey through memory cubicles that led me to the masterpiece I sought.

    It began with this prompt

    You’ll notice that my prompt tells Meta AI “who it is.” That’s a critical step in getting any AI to approach the task from a more specified perspective.

    Next, I outlined the Q and A process we would use. This was essential, since I wanted the AI to ask follow up questions to narrow in on a solution. For fun, I gave it a bit personality – “genial and collaborative.” Also note that I incorrectly say it’s a black and white image.

    Here’s a screenshots of Meta AI’s first response- the questions were logical – style, time period and where I might have seen it.

    My response gave a date range but really focused on how the “feel” of the image in my memory – I said “Eastern European Kafka-esque bureaucratic workplace.”

    In its second response, Meta AI noted my “Kafka-esque vibe” (a clever rephrasing). It pointed me to artists like Grosz and Dix. (I checked those artists and didn’t see any possibilities) And it posed a few more questions on how I remembered the faces and tone.

    The best I could recall about the faces was they were blank and fearful. I added that the faces were “realistic with an overlay of artistic enhancement.” (Not the best description I’ve ever crafted.)

    By Meta AI’s third response, it had correctly identified Tooker as a possibility.

    I suddenly remembered the artist and asked to see some samples. Meta AI then pointed me to Google search of Tooker works (below) and there it was in upper left.

    I was so impressed, I congratulated Meta AI for solving my query in such short order. It seemed like the thing to do. After some pleasantries, we signed off.

    Comparing AIs

    After this success with Meta AI, I thought I’d try the same prompt with other leading AIs. Note: I pretended not to know the answer and kept my responses limited to just the information I gave to Meta AI on the first run.

    I was using the free versions. Here’s how they compared:

    • Meta AI (above)- Identified artist in 3rd response and linked to “Government Bureau” in 4th.
    • Microsoft CoPilot – Identified artist and suggested it might be “Government Bureau” in 3rd response.
    • Anthropic Claude – On it’s 9th response, it mentions Tooker but hallucinated a title he did not create.
    • Google Gemini – Keeps suggesting artists and on 6th response, it gave up and wished me luck.
    • Perplexity – gives up after 5th response. (But it did begin with a fun opening response: “clears throat and adjusts monocle. Ah, I believe I may know the work you are describing, my dear fellow! Let me ask you a few questions to jog your memory”
    • ChatGPT 3.5 – patiently guessed artists and styles. But I quit after 11th response when its suggestions began to drift off track.
    Here’s why AI beats a traditional search engine in this use case:

    A traditional search engine like Google relies heavily on keywords and metadata. When you search for an image, it looks for matches based on text associated with images across the web, such as titles, descriptions, tags, and other textual content. This means you need to have some specific information or keywords related to the image to start your search.

    AI-driven image retrieval systems use machine learning algorithms to understand the content of the images themselves. They can analyze visual elements such as color, shape, texture, and even abstract concepts like the emotion an image might evoke. This allows them to find matches based on visual similarity rather than relying solely on text descriptors.

    This process is often referred to as content-based image retrieval (CBIR), which is a feature of AI that can search and retrieve images from large data sets based on user-generated cues, even if they are as abstract as feelings or vague visual recollections.

    So, if you’re using AI for this purpose, you don’t necessarily need to know specific details about the artwork. You can provide a description of what you remember, and the AI can interpret that description to find images that are visually similar to what you’re describing. It’s a more intuitive process that mimics the way humans might try to find something based on vague recollections, rather than the precise keyword matching used by traditional search engines

    Posted on

    Getting Started with the AI Toolkit

    AI-generated with Lexica. I used prompt: “A steampunk tool box in a mysterious setting“

    I’ve been exploring AI use for teachers and students for the past view months. I recently was interviewed by Aaron Maurer for his educator’s podcast “Peter Pappas And How AI Is Challenging Educators To Rethink Practices Of Teaching And Learning“

    Listen on Apple Podcast

    Listen on Spotify

    Here’s an AI Toolkit that I designed to assist students and teachers with integrating AI into teaching and learning. This is reposted from my course website: EdTechMethods

    Since the recent launch of ChatGPT – Artificial intelligence (AI) tools are in the news. Here’s a brief list of AI tools and resources. All are free (or have free intro promotions). We will be integrating them into our course over the semester.

    OUr Course AI Policy

    You are free to use AI tools in this course for your weekly assignments. Some assignments – like this week’s – will require it. You are free to use AI in creative ways. But be sure to always site AI use as noted in #1 below. See my example at the end of this page.

    Be aware of the limits of ChatGPT and other text generating apps:

    1. AI is a tool, but one that you need to acknowledge using. Please include a paragraph at the end of any assignment that uses AI explaining what you used AI for and what prompts you used to get the results. Failure to do so is in violation of academic honesty policies.
    2. If you provide minimum effort prompts, you will get low quality results.
    3. You will need to refine your prompts in order to get good outcomes. This will take work.
    4. Don’t trust anything it says. If it gives you a number or fact, assume it is wrong unless you either know the answer or can check in with another source. You will be responsible for any errors or omissions provided by the tool. It works best for topics you understand.
    5. Be thoughtful about when this tool is useful. Don’t use it if it isn’t appropriate for the case or circumstance.
    6. AI is a tool. It is not always the right tool. Consider carefully whether, given its weaknesses, it is right for the purpose to which you are planning to apply it.
    7. There are many ethical concerns you need to be aware of. AI can be used to infringe on copyright, or to cheat, or to steal the work of others, or to manipulate. And how a particular AI model is built and who benefits from its use are often complex issues, and not particularly clear at this stage. Ultimately, you are responsible for using these tools in an ethical manner. Be transparent about how you use AI, and take responsibility for the output you create ~ adapted from Ethan Mollick
    Start Here: How to use AI to do Practical Stuff: A New Guide

And  Seven Ways of Using AI in Class
    writing Tools – type in prompt / get a text response
    • Access GPT4 without a key or invitation on Hugging Face. | Link
    • ChatGPT | Link. First create an OpenAI account to log in (registration may require a phone number)
    • Best option if ChatGPT is down, use  OpenAI Playground
    • Perplexity | Link Better at looking information up than generating new ideas. But shows sources and lets you edit them out or add new sources. [Click 3 dots in upper right of the response.]
    How to Create writing Prompts
    AI Plagiarism Detectors
    • GPTZero. Built for educators. Paste in text and see if its created by AI | Link
    • GPT-2 Output Detector – Paste in text and see if its created by AI | Link
    image Tools – Type in descriptors (or upload image) / get an image response
    • Lexica | Link Requires sign up (my favorite of this group)
    • DALL-E | Link Requires sign up
    • Midjourney | A powerful image generator on Discord. Link to QuickStart guide
    • Image Maker | Link Free to use Stable Diffusion-powered text to image generator
    • Karlo Generator| Link Can be a bit wonky and only produces low resolution images, but no sign up needed.
    • Image Mixer | Link Combine multiple images into one. No sign up needed
    How to Create Image Prompts
    • How to use AI image prompts to generate art (this post references Microsoft Designer, but tips work with all image generators) | Link to article
    • DALLE-E Prompt Book | Link
    Audio / Video Tools
    • AssemblyAI – extract transcription, summary or topic detection from a video | Link
    • Google Speech-to-Text | Link | Free sign up
    • Google Text-to-Speech | Link | Free sign up

    AI note for this page: I elected to use AI to generate a featured image for this page. Thought it would be proof of concept. I used Lexica and this prompt: “A steampunk tool box in a mysterious setting.” It took me a number of tries to add descriptor prompts to get something I wanted. The final image included some elements that make no sense – what’s the “screen” at right and cables in foreground? Mysterious?

    Our first AI Lesson
    Class 3: Exploring AI in the Classroom
    Posted on

    Mapping Inequality: Exploring Personal History in Redline Maps and the 1940 Census

    Designing Inequality

    The Detroit Eight Mile Wall is a one-foot-thick , six-foot-high  separation wall that stretches about 1⁄2 mile. It was constructed in 1941 to physically separate Black and white homeowners on the sole basis of race. The wall no longer serves to racially segregate homeowners and, as of 1971, both sides of the barrier have been predominately Black. Photo by John Vachon / Library of Congress

    This is repost of a lesson I did in my social studies methods class at University of Portland. It explored the largely forgotten government policies that segregated nearly every major American city and metro area. We also looked at how those unconstitutional actions fostered inequality in America since their enactment in the 1930s.

    Scroll down for student projects. They are very creative and worth a look!

    Overview

    In preparation for this class, students watched the video Segregated by Design.

    Class opened with the brief film Housing Segregation and Redlining in America: A Short History. I then put students into breakout groups and (using a Jamboard) they created mind maps of the legacy of redlined neighborhoods based on what they had seen in the two introductory videos. They brainstormed and developed visual comparisons that explored multiple factors such as – income, wealth, health, schools, policing, services, employment, environment.

    Next I introduced the tools they would use for the lesson – digitized version of  1940 US Census and the website Mapping Inequality: Redlining in New Deal America .

    Mapping Inequality

    Mapping Inequality allows the user to explore the “redline” maps created by agents of the federal government’s Home Owners’ Loan Corporation (HOLC). The maps are displayed over contemporary maps that make it easy to view where redline districts are today.

    HOLC Map of Portland Or See interactive map here.

    The HOLC created redlining maps in the 1930s as part of a federal effort to assess the risk of insuring mortgages in urban areas. Neighborhoods were graded from “A” (Best – most desirable) to “D” (Hazardous – least desirable), with “D” areas often shaded red, hence the term “redlining.” These assessments were based on factors such as housing quality, economic stability, and racial composition, often marking predominantly Black or immigrant neighborhoods as high-risk regardless of actual financial conditions. This practice systematically denied residents in redlined areas access to affordable loans and investment, entrenching racial and economic inequalities that persist today.

    Sample “Best” and “Hazardous” neighborhoods in Portland

    “Best” A7 Arlington Heights District- Census page from 229 SW Wright Ave. in A7 | 229 SW Wright Today

    Clarifying Remarks from HOLC map: “Deed restrictions have expired but single-family residential zoning and topography give ample protection. Infiltration of Subversive Races is remote. Foreign-born families: 0% Negro: none. Residents: Executives, business and professional men, retired capitalists, etc.“

    “Hazardous” D8 Southwest River District – Census page from 931 SW 1st Ave in Portland in D8. | Rooming house torn down. 931 SW 1st is now the World Trade Center

    Clarifying Remarks from HOLC map: “This is distinctly a workingmen’s rental district and approximately half of the population being employed in the industrial plants along the river.
    Infiltration of Subversive Races has occurred. Foreign-born families 35%; Orientals 750 Chinese, 250 Japanese, 35 Filipinos, Negro 30%    “

    How to search in Census of 1940

    Next I took them to the the 1940 US Census and showed them how to look up street addresses and navigate to specific neighborhoods. I shared census pages from some Portland neighborhoods.

    Note: it helps to refer to Blank 1940 Census form showing categories

    You will “search by location where the person lived.” Begin by narrow search by state, county, city, street. 

    After you click search, you have the option to add a cross street. 

    It will take you to multipage document that may not exactly zero in on the intersection you wanted. You may need to scroll through some pages to get to that location. You can find the streets and house numbers on far left. (Street written vertically) 

    The assignment

    I wanted the student to draw a personal connection between the HOLC categories (best, still desirable, definitely declining and hazardous) and the data from the 1940 census. To demonstrate, I shared my mother’s family entry from 1940 census then showed that my grandparents and family were living in a HOLC designated “Definitely Declining” neighborhood. I shared some family photos of the house and neighborhood. Then we looked at the family home value in 1940 and shared how the neighborhood had steadily declined to the point that the home today is currently valued at less than one-half of the inflation-adjusted value listed in the 1940 census.

    I knew that not all students would be able to find a family listing in the census nor might they ever have lived in a city mapped by HOLC. So I left the final product very open ended and invited them to make connections that worked for them.

    Sample student work

    A number of students were able to find family members on the census and family home in the HOLC maps

    Rooted in History: Finding Family History in the 1940 census
    A Bohemian Rhapsody: The Life of a Good Man
    In The Slums of Italian Town

    A few students focused on the history of homes they lived in.

    The Changing Face of NE Portland
    In This House 100 Years Ago…
    My Beautiful Home? In a Red Zone?

    Some students looked how the HOLC influenced residential options for immigrant groups

    Life in Detroit: My Grandfather’s Hometown
    Where in Portland was I allowed to live in the 1940s?

    Others looked at redlining’s impact on property values and neighborhood development

    The Difference 1900 Feet Makes: A Look Into Portland’s Redlining Past
    A Two-Pronged Effort to Stop Redlining in Inkster, Michigan