Answer: Best ways to do just-in-time learning. Part 3: getting a teaching summary with NotebookLM

 Last time... 

A woman studying a complex anatomy topic. P/C Pexels.com by RF._.studio _

      

we talked about asking your favorite LLM to give you a summary (say 500 words on a specified topic).     

But maybe there's a way to be even MORE specific.  Remember our Challenge:  

1. What are the best AI-inspired (or AI-driven) new Micro- and Meso-learning that you've found?  Have you actually tried this method?  How well has it worked out for you? 

Asking for an essay, especially one that you can have read aloud, is pretty handy.  

The downside is that you're getting the full LLM, and that may (or may not) be what you're looking for.  

Using Google's NotebookLM lets you start with a specific set of resources--presumably ones that you've hand picked to be reliable, accurate, and on-topic--and then create a summary in text, audio, or video formats!  

Here's what I did.  

1. Start a new Notebook.  Visit NotebookLM and create new notebook.  Then add sources by clicking "+ Add" in the upper left corner.  As you can see, I've added 5 papers I found by using regular Google Scholar 

The papers are summarized in the main Chat window (in the center).  In short, they're: 

1. Selgrath et al. (2024) reconstruct kelp forest changes over centuries in central California

2. Leichter et al. (2023) examine the impact of nutrient availability

3. Garcı́a-Reyes et al. (2022) demonstrate the predictive power of winter oceanographic conditions on summer bull kelp canopy cover in northern California

4. Butler et al. (2021) document the soundscapes within and around a marine protected area off La Jolla 

 

(A minor hassle here in the design of NotebookLM.  The references in the Chat window are all correct, but it's hard to know which of the links in the Sources window map onto the refs in the Chat.  You have to click on the source to see which is which, and it's not an obvious mapping. It also didn't tell me that source 2 and source 4 are duplicates, which is why there are only 4 sources listed in the summary. Argh!) 

In any case, if you click on the "Video Overview" in the upper right corner, NotebookLM will create an 8 minute video summary of the contents of the sources.  Link to the YouTube video.  

You'll see the effect in  a minute or so... 

Let me repeat--the summary (while great to use for learning)--is NOT a replacement for reading the papers.  There's all kinds of detail that's not covered in the video overview.  That's why it's an overview, and not the papers themselves. 

As we mentioned earlier, you can also generate an audio overview.  Here's the M4a file of the summary of these papers. (14 minutes)  This is handy for putting onto your phone and listening as you walk.  (One minor annoyance--Google speech doesn't know how to pronounce the city name "La Jolla."  It's NOT "la-jala," it's "la-hoya.")  

ALSO note that you can customize both the video and audio overviews.  Click on the 3-dots in the upper right and you'll get a set of customization options: 

Obviously, you can get the overview in another language (handy, if you're studying that language) OR you can get it focused on a particular learning task, such as preparing for a quiz on the topic or a specific sub-topic within the collection of papers.  

I have to admit, even though I know a fair bit about this topic, I learned something by listening to the audio summary.  NotebookLM really does a good job of synthesizing something coherent and useful from the sources. 

 Last time we also mentioned using the Feynman technique for learning.  Using the prompt: 

[tell me about the Feynman technique for effective learning] 

.. this will cause Gemini to jump into the "Feynman teaching mode" on the topic of California kelp.  

If you give NotebookLM this prompt, it will give you a summary of the material, which you can then use to guide your own Feynman method. 

How use the Feynman Technique with NotebookLM

After you've uploaded your sources (documents, articles, etc). You can select a topic you'd like to understand in more detail.  

Prompt NotebookLM to explain: Ask the AI to explain the concept as if to a child or a beginner to force simplification and the creation of analogies. For example, you can use a prompt like, "Explain the relationship between purple urchins and the loss of bull kelp in California using the Feynman technique, as if teaching it to someone with no background in the topic."  (Don't worry about marking off the concept from the query--NotebookLM will figure out what you mean.)  

Evaluate the explanation: Review the AI's explanation for clarity and accuracy. Does it all make sense to you?  If not, you need to... 

Identify gaps in your understanding: If the explanation is still unclear or if it highlights weaknesses in your own understanding, make a note of these areas.  Once you've identified these issues... 

Return to your sources: Go back to your original documents within NotebookLM to find the information needed to fill those knowledge gaps. 

Refine your explanation: Go back to NotebookLM and re-explain the concept, incorporating the new information you've and simplifying the language further. 

Repeat the cycle: Continue this process of explaining, identifying gaps, researching, and refining your explanation until you fully grasp the concept. 

Make sense?  That's the value of having your own tutor on the topic you choose.  

Of course, your next step would be to find additional resources to extend your learning.  You could ask NotebookLM to: 

     [ recommend other papers to read on this topic] 

Or, you can use the Discover button (upper right of the Sources panel) to add additional sources.  

 

SearchResearch Lessons

This is Part 3 of our "how to learn rapidly" series.  We learned that: 

1. Using NotebookLM can give you control over the content you're trying to learn. Add the materials you need to learn to the Notebook, then ask for a summary, or ask questions of it as you wish.  

2. You can create audio or video summaries of the Notebook sources. This is remarkably handy as a way to learn-on-the-go.

The NotebookLM method promises to give you much higher quality summaries since it's basically answering questions (or generating your summaries) from the materials that you've selected. (Which you presumably selected for their accuracy!)  

There are many more AI-augmented learning methods out there.  Keep your eyes open and let us know if you find any that are particularly effective for you.  Keep those comments coming! 

Keep searching!    

Answer: Best ways to do just-in-time learning. Part 2: conversation with a chatbot

  As mentioned... 

A woman studying a complex anatomy topic. P/C Pexels.com by RF._.studio _

      

There are a huge number of methods to quickly learn something, especially today.  This is the next in a series of posts about methods I've tried.    

Remember the Challenge: 

1. What are the best AI-inspired (or AI-driven) new Micro- and Meso-learning that you've found?  Have you actually tried this method?  How well has it worked out for you? 

Last week I wrote about just searching for pre-existing tutorial content.  That often works quite well and is a good first place to start.  

But sometimes you want something that's bespoke and tailored to you interests at the moment.  Here's a way to use the chatbots to do that.  

You can put both ChatGPT and Gemini (and many others) into chat mode that lets you have a conversation with the AI. Here's my example using Gemini.  To start the conversation, click on the button with the vertical lines (lower right) to get started.  

The key idea here is to ask the chatbot to read you an essay on the topic of interest.  Here, I continue the quest to learn more about kelp with my prompt: 

[Give me a 500 word essay on the natural history of kelp along the California coast]

Note that it reads around 120 words/minute, so I'm asking for a 4-minute reading. Ask for 1000 words if you want a more detailed version. 

Be sure to be as specific as you need.  Note that I wanted info about the "natural history" and specifically to the "coast of California."  (I know that kelp lives in many oceans, but I have a parochial interest in local kelp. See picture at the end of this post.)  

NOTE: As you listen to your topical essays being read aloud, note that you can interrupt it and ask a follow-up question.  (Such as, "what is the role of pollutants in damaging the kelp coverage?")  You don't need a special hot word (like "Hey Google!"), just start speaking when you want to drill into the topic. 

After you've done that,  you can say something like "... let's go back to the previous topic."  It really does feel like a conversation with a subject-matter expert. 

Of course, each of the chatbots will create a somewhat different essay. Here's the one that Gemini created and the essay ChatGPT created.  

Here's a visual for you (typography by me):  

Click to see in full-size

 
Notice that Gemini wrote 407 words, while ChatGPT wrote 530 words.  Gemini neglected to mention bull kelp at all, while ChatGPT gives it a prominent place.  ChatGPT also included information about Seasonal Cycles and the historical Human Use of kelp in California as well.  For my money, a much better job. But some might prefer the lighter touch of Gemini.    

I've found myself asking for short essays on topics that I'd like to know more about, especially when I'm about to do something (visit a museum, attend a seminar, go to a particular part of the city).  

As remmij pointed out, you can even use this chatbot conversational style to ask questions about learning method.  When I asked the bot: 

[tell me about the Feynman technique for effective learning] 

Gemini not only told me what it was, but then jumped into "Feynman teaching mode" on the topic of California kelp.  

But we'll talk more about that next time.  

 

SearchResearch Lessons

This is Part 2 of our "how to learn rapidly" series.  We learned that: 

1. Chatbots can read on-topic essays to you. Use this as a way to delimit what you'd like to learn.  

2. Think of a chatbot as a way to have a conversation with the material. In particular, you can drill down deep into a topic in a very natural, conversational way.  

But AS ALWAYS validate what it tells you.  Thus far, the conversations I've had all seem pretty high quality and accurate.  But that's not a guarantee.  If you have questions or hear something that sounds a bit...pardon me... fishy... be sure to check!  

Diving in the giant kelp in the Channel Island, California. P/C Dan.

Keep searching!  

Answer: Best ways to do just-in-time learning. Part 1 - search for tutorial content

  What's a great way to... 

A woman studying a complex anatomy topic. P/C Pexels.com by RF._.studio _

... learn something relatively quickly?    

As I said in the Challenge last week, in my line of work, I often have to be a fast learner and become an instant expert (or as close as possible within the time I've got).  

This motivates the Challenge: What are some new (largely AI-driven) ways to rapidly learn something?  

So here was the Challenge: 

1. What are the best AI-inspired (or AI-driven) new Micro- and Meso-learning that you've found?  Have you actually tried this method?  How well has it worked out for you? 

There are SO many ways to answer this Challenge that I'm going to give it in parts.  Here's Part 1.  I'll give some more ideas next week (instead of another Challenge).

 I AM mindful that real learning takes work.  As Andrej Karpathy wrote (Feb 10, 2024): 

Learning is not supposed to be fun. It doesn't have to be actively not fun either, but the primary feeling should be that of effort. It should look a lot less like that "10 minute full body" workout from your local digital media creator and a lot more like a serious session at the gym. You want the mental equivalent of sweating. It's not that the quickie doesn't do anything, it's just that it is wildly suboptimal if you actually care to learn.

That's all true.  But there are times when we need to learn something rapidly.  How can we do that?  I'm happy to sweat and do focused work, but sometimes I need a micro-learning or meso-learning experience.  How can we do that? 

Here's my approach.  (Next week we'll talk about other approaches.)  

1. Identify what you're trying to learn. Does it have a name?  Do you have a fairly clear learning goal? Just flailing around watching short videos isn't the same as having a plan.  This is often the hardest part of the learning task--especially if you don't know what you're really trying to learn.  As a way to clarify the task... 

2. Write down your goal. What I do is to write down (usually in pen on paper!) what I'm trying to learn about.  

Example: Last week, after meeting a kelp researcher on the beach in northern California, I wanted to learn about the different kinds of kelp that grow locally. I literally wrote out, "Learn: kinds of kelp in N. California. How many varieties? Ecology. Threats. Current condition."  

(That's Russell-ese for "Learning Goal: To learn about the different kinds of kelp that grow off-shore in California. Learn about the different species of kelp, what kinds of ecosystems do they grow in, what are the current threats to their growth, what's the current condition of kelp beds in 2025?")

In essence, I'm outlining a topic and a few relevant sub-topics. In the past, I would have started classical searching, and pulled together a few articles to read, then spent a happy couple of hours reading them, digging deeper as interest led me.  

Here's my first search.  Note that I've opened the AI Overview as a way to get a first-glance at the topic:   

This quick search tells me a lot. There seem to be only 2 species of kelp along California, NOAA is doing research, there is an effort in kelp forest restoration because of multiple threats.  

3. Do a quick search for pre-existing edu materials.  It's worth doing this first--you never know when someone out there will have already made exactly the right tutorial for you.  No matter how obscure the topic, someone might have already put together a package for you.  As my friend Leigh says, "Dan, it's the internet... there's always someone who is interested in that."  Time has proven him right. 

Point is: Check for pre-existing tutorials (or lectures, or lessons, or lectures) first. 

Looking at the videos, it seems there are a couple of potentially useful ones already.  

To scan a video like this, I'll typically open the video and look at the transcript. (Sometimes the link to the transcript is a little buried. See below for a useful tip.)  

Then, once that panel is open, you can click to open the transcript. 

Then, once that panel is open, you can click to open the transcript. Click on the magnifying glass to search in the transcript.  (Or use Control-F / CMD-F.) 

I then do a few searches inside of the text to make sure that the video really is on-topic and discusses what I want.  (For instance, does the word "ecosystem" appear in the video?)   

Alternatively, you can ask Gemini to summarize the video (again, to see if it's really what you want to learn.. keep in mind that you want to watch the whole video, not just read the summary).  Looking at the summary below, it seems like a great video to watch--it's nearly an hour long, and probably worth the investment of time.  

When you've scrolled down a bit, you might well find the "People Also Ask" section... which is another way to see what relevant topics you might learn.    

Of course educational materials can be more than just YouTube videos.  Using the term "lesson" is often useful, though I've found that they're often oriented for 6-12 grades.  (Which might be what you're looking for.)  

For more advanced learning units, I'd use the terms "university" and "class."  That will often find entire semester-long classes, with a syllabus and everything.  

4. Create a quiz to help test your knowledge. As we know, self-tests are a very useful way to ensure that your studying has paid off.  In a great tip from Ben Gomes at Google, he suggested using Gemini to make a quiz to check.  You can do this with either a video OR a longer text document.  

One of the documents about kelp I read was this one from California State Fish and Game - you can see what I asked Gemini to make a quiz for me (look on the left hand side... I said "Here's a document... <URL> ... create a study quiz from this file." 

You can do the same thing with videos.  Here, I dropped in a link to a YouTube video about kelp.  

SearchResearch Lessons

This is Part 1 of our "how to learn rapidly" series.  We learned that: 

1. Identify your task. Use this as a way to delimit what you'd like to learn.  

2. Write down your learning goals. A great way to do this is to literally write down what you'd like to learn.  As you write, you'll learn what you don't know--identifying knowledge gaps is a great method to figure out how to proceed. 

3. Search for already existing learning materials on your topic. Don't skip this step!  

4. Create self-test quizzes to ensure that you've learned the material. 

Keep searching!  

SearchResearch Challenge (8/6/25): Best ways to do just-in-time learning?

 It happens constantly... 

P/C Pexels.com by RF._.studio _

... maybe it happens constantly to you too.  

I suddenly need to learn something I know little-to-nothing about, and learn it fast.  

In essence, I have to be a fast learner and become an instant expert (or as close as possible within the time I've got).  

That's what this Challenge is all about.  What are some new (largely AI-driven) ways to rapidly learn something?  

I'm curious what you've found are the best ways to pick up a topic quickly.  

In the research world, people have written about "micro-learning," tiny episodes of learning that you fit in between the other moments of your day.  (Example of an academic paper on this topic)

But I also want to know about "meso-learning" (intermediate scale topics), and even if you've found good ways to organize "macro-learning" (that is, big topics that take a long time to acquire).  

Let's divvy it up this way:  

Micro-learning - instruction usually taking 10 minutes or less. (There's even a Wikipedia article on Microlearing.)  

Meso-learning - between 10 minutes and 1 hour.  

Macro-learning - more than an hour, and up to the rest-of-your-life.  

(I also realize that people use Microlearning to do macrolearning--like learning a language or do vocabulary practice. We'll ignore that distinction now and return to it next week.)  

So here's your Challenge: 

1. What are the best AI-inspired (or AI-driven) new Micro- and Meso-learning that you've found?  Have you actually tried this method?  How well has it worked out for you? 

Example:  I'm in the process of brushing up my Spanish (because I live in California, so it's handy) and German (because I go to Switzerland fairly often).  So I've been using Duolingo as a Micro-learning method.  It's not super-AI-ish, but they keep adding AI components to it.  I've also been trying AI-driven language tools (e.g., MakesYouFluent) to exercise my conversational skills.  (I'll give you my thoughts about each next week.)

But YOU might have a few tricks up your sleeve.  For instance, I know of some people who create a Google NotebookLM, drop in a bunch of content, and then have it generate a podcast they can download and listen to as a way of learning something.  Have you tried this?  Did it work well for you? 

What tactics and strategies do you employ to understand a topic area quickly?

Most importantly, what do you do? 

Summary next week, along with some thoughts on my part.  

Keep searching!  

Answer: Wake up and smell the 19th century coffee?

Interesting challenge... 

An AI generated image that's intentionally slightly misleading. 
(So don't waste time mining this image for clues.  Read the Challenge below!)  

As I mentioned, this is one of those Interesting Challenges.  Usually, the Interesting Challenges are some fascinating aspect of the world--or, as in this case, of history.  

Here's the Challenge: I came across a claim in my reading that struck me as so strange and bizarre that I had to SRS the claim.  Is this true?  If so, can you find reasonable evidence for it?  

1. I read that during 19th century Germany, it was prohibited to roast coffee beans without official royal approval. One couldn't import, roast, grind, or sell coffee to the masses.  What's more (and even stranger, which is what makes this a fun Challenge), the government employed people whose job was to literally sniff out illicit coffee production.  Since it's hard to hide the smell of coffee roasting, it seems unlikely!  So the big question is:  Was this a real thing?  If so, who, when, where, and most of all, WHY?  

After doing my SearchResearch (described below), the bottom line is this.... 

Yes, it's all true--except it was in Prussia during the 18th century.  (Note sure where the error crept in, but it's a good point to remember--the basic question might have incorrect information built into it!) 

In Prussia, coffee became widely popular in the mid-1700s. The government became concerned with the economic implications of coffee consumption because coffee had to be imported, costing Prussia some serious money which they could ill afford to lose. 

Frederick II (aka, Frederick the Great; ruled 1740–1786) created strict regulations with royal monopolies and prohibitions on private roasting and selling of coffee beans.  Yes, average Prussian could have and drink coffee, but the government didn't want people roasting their own beans--that was something they wanted to control.  

The situation escalated dramatically in 1777 when he issued his famous anti-coffee manifesto, reaching the point of peak coffee suppression by setting up an agency to find illicit coffee roasters. They used around 400 disabled war veterans as "coffee sniffers" (German: Kaffeeriecher or Kaffeeschnüffler), so it was also a vets jobs program!  

What I did to find this... 

I have to admit that I did my research the old-fashioned way, with queries like [German coffee sniffer 19th century].  

That way worked pretty well, but as Regular Reader Arthur Weiss pointed out, one can copy/paste the entire Challenge into a LLM (he used Perplexity) and generate an answer.  You can read Perplexity's answer to the Challenge here.  To his credit, Arthur then verified the claims by examining the links given in the article.  2 go to Wikipedia, and 1 goes to a German museum, and 1 goes to a Mashable article that's well-written, with links to resources of its own.  

Motivated by Arthur's example, I did the same thing with Google Gemini and got very similar results. (Link to Gemini's answer.)  I also checked the citations, but there were only 2... and while they were fine, they left me wondering if we couldn't improve the situation.  

So, yeah--the AIs are doing a pretty good job with this Challenge. 

Is there something else one could do to improve the SRS process?  Sure! One obvious thing occurs to me.  Since this is a German language topic, perhaps it would make sense to search in German!  

Just searching for the term Kaffeeriecher on the English language Google doesn't do much that's different.  But if you change the LANGUAGE of the search results, you do get different results.  

Here's how to change the search language:  Under the TOOLS menu (on the right side), go to "Advanced Search," which will then give you the option to select your "Narrow your results by..."  Then select "German."    

This gives you only German language results.  In this case, the German-only results led me to a few additional results from reputable German-language sources.  (Including some contemporary ones, which were fairly amusing.)  

This also led me to read (in translation) the German Wikipedia article, which points to The Coffee Noise of Paderborn in 1781.  The article tells us that  “On February 25, 1777, he issued an edict which forbade the bourgeoisie, peasants, and lower officials from buying and using coffee, and declared it henceforth only a privileged beverage for the nobility, the clergy, and higher officials.”  This was the Edict of Paderborn.  

“His Royal Majesty Himself had been raised on beer soup as a youth, so people could just as well be raised on beer soup"; it was much healthier than the coffee to which "every farmer and common man" had now become accustomed. To achieve his goal, he introduced a fairly high coffee tax and established a special coffee administration, whose officials were popularly called 'coffee sniffers.' (Kaffeeriecher)"

The edict also includes the penalties for illegal coffee use:  

"5) If it is found that one or another person, who is allowed to use coffee, has given away coffee either publicly or secretly for money, or sold it in portions, or had it given away or sold by his family, he shall also [ 97 ] incur a penalty of 10 Reichsthalers each time, and shall be liable to immediate execution."  

Pretty serious stuff. 

As you might have predicted, “Coffee parties flourished…” as the locals objected to the increased price of coffee and the imposition of the Kaffeeriecher.  

Out of curiosity, I also translated the Challenge into German and asked Gemini for its response.  Happily, it gave the response in German as well, though I didn't learn anything different on this attempt.

I also did an Image search for [Kaffeeriecher] which gave me several images that I've montaged into a single image here: 

Various images of coffee sniffers breaking in on women's coffee parties, entering a home to search for illegal coffee, and a cartoon showing a Kaffeeriecher smelling out all forms of coffee.  

You can see where the Image Gen model got its inspiration!  The top two images are the most common pictures of Prussian coffee sniffers out there.  

SearchResearch Lessons

There are a couple of lessons to take away... 

1. When you've got a longish text for the Challenge, it might well be a great AI question to pose.  Both Perplexity and Gemini did decent jobs on this Challenge. 

2. For topics that are obviously in a specific language, consider switching to that language to search for results.  You can often find much more content, and will certainly find more original content than you will find in translation.  

3. Remember that errors can creep into the statement of the Challenge.  The Kaffeeriechers weren't from the 19th century, but creatures of the 18th century.  It's a big difference.  

4. Consider looking at Images for additional ideas.

Keep searching! 

SearchResearch Challenge (7/23/25): Wake up and smell the 19th century coffee?

There's a small shift in focus for SearchResearch... 

An AI generated image that's intentionally slightly misleading. 
(So don't waste time mining this image for clues.  Read the Challenge below!)  

... it's been brewing for a while, and is one that lends itself to this week's Challenge.  There are changes afoot at the old SearchResearch Rancho, and I want to keep you in the loop.  

As you probably noticed, the process of doing online research has been rapidly changing with the advent of AI methods.  Consequently, the things we talk about here are changing as well.  

Each week I'll try to focus on either:  

A. Something so interesting we have to talk about it.  You's seen these Challenges before--topics like Hummingbird Hawk-moths or why some clocks use IIII rather than IV.   

There are really two reasons to do these "interesting" Challenges:  (1) you still need to have the basic search skills to find a good, high quality answer; and (2) I'm trying to get everyone to learn to see the world around them with a bit of a curious eye.  I hope everyone will get into the habit of asking "why is this like that?" or "what is that thing anyway?"  Learning to see--rather than just look--is a fundamental skill for SearchResearch.    As Betty Edwards says, “We mostly see what we have learned to expect to see.” (In:  Color: A Course in Mastering the Art of Mixing Colors)  I want to encourage everyone to see rather than merely look.  

B. Discussions of new (mostly AI-based) ways of doing online research and evaluating what you find.  That's what last week's Challenge was all about: So what ARE LLMs good at? What are they bad at?   

And I expect that we'll alternate the Interesting with the Discussions posts as we go forward.  

This week is one of those Interesting Challenges... 

I came across a claim in my reading that struck me as so strange and bizarre that I had to SRS the claim.  Is this true?  If so, can you find reasonable evidence for it?  

1. I read that during 19th century Germany, it was prohibited to roast coffee beans without official royal approval. One couldn't import, roast, grind, or sell coffee to the masses.  What's more (and even stranger, which is what makes this a fun Challenge), the government employed people whose job was to literally sniff out illicit coffee production.  Since it's hard to hide the smell of coffee roasting, it seems unlikely!  So the big question is:  Was this a real thing?  If so, who, when, where, and most of all, WHY?  

As mentioned, the point in these Interesting Challenges is to find the answer, but JUST as importantly, to learn a method for finding the answer and validating it.  

What can you do with the Challenge this week?  Can you sniff out the answer?  

Be sure to let us know what you did to find the answer!  (Yes, all methods are good--but you should say what you did, and give some background information on why you find this plausible.)  

Keep searching!