notebook LM | 小灰的笔记

type

status

date

slug

summary

引言

Here’s a [rule of thumb:经验之谈] for using notebook LM:

If your use case matches the following three [criteria:条件], notebook LM will perform better than other AI tools.

First, you have a very low tolerance for hallucination.

Second, you're working with information [scattered across:散落] different locations, different formats, like documents and [slides:幻灯片], and/or across different mediums (text, video, and audio).

And third, you want a quick and [reliable:可靠的] way to transform all that [fragmented:碎片] information into a [cohesive:内聚的] and meaningful output.

In the next 15 minutes, I’ll go over key features, real-world use cases, and [pro tips:专业建议] very few people know about. Let's get started.

初次使用指南

[Right off the bat:一开始], it’s a bit [overwhelming: 让人不知所措的] for most users to see this when creating their first notebook. There's a lot going on. So I recommend pressing escape, clicking the logo in the top left corner to go to the notebook LM homepage. Now that we’re oriented, we want to bookmark this for easy reference in the future.

Change to list view – that's just a personal preference of mine – and sort by title. Now let’s create a new notebook.

Here, I’m first going to press escape again, because I want to first name this notebook "Health reports 2." You’ll see why in a bit.

Click the plus icon here to start uploading sources.

Here, I’m just going to upload my three most recent annual health checkup reports, along with a PDF version of Tools of Titans by Tim Ferris.

And while all that’s loading, I’m going to add another source: YouTube video.

I am first going to add a video on the topic of uric acid. I promise this will make sense in a little bit.

One more source: YouTube video.

This is going to be a video by Andrew Huberman on the topic of fasting.

使用示例：健康报告分析

Once all sources have been uploaded and processed, we can now click into individual sources.

I’ll choose my most recent health report.

I’m going to blur this to not overshare, but ladies, just know that I’m at peak physical health.

Under Source guide, we see that notebook LM has produced a concise summary of a dense medical report.

I would definitely not have read word for word.

On the right, clicking into one of the key topics, like "abnormal results," actually prompts notebook LM to expand on that topic based on all the selected sources on the left-hand side.

Even though the key topic was just from one source, right?

So that's the first important thing to keep in mind: Everything in this chat interface takes into account all selected sources.

Meaning, if we want notebook LM to ignore a source, we need to first deselect it from the source list.

By the way, don’t worry about me. The only abnormal result from my health reports is my absolutely huge capacity to make dad jokes.

功能介绍：快速总结与扩展

The Notebook guide feature down here is like a quick start guide for beginners.

There’s a summary of all the sources added to the notebook, pre-created templates like FAQs and briefing docs that are very situational. I’ll give examples later.

And suggested questions here to help users get started.

And don’t worry, we’ll also go over audio overviews today.

Alright, we’re going to start interacting with the notebook, but since the answers take a while to generate, I’ll switch over to the actual Health reports notebook I’ve been using.

互动与数据分析：获取健康趋势

First, I asked for the top 10 health trends based on my last three reports.

A pretty time-consuming task, even for doctors, right?

But just after a few seconds, notebook LM gives me a list of 10 observations, one of which is a fact that I’ve had elevated uric acid levels over the last 3 years.

Which is unfortunately true because I eat a lot of red meat.

And because I added a video on uric acid as a source, I can now ask, "What are the top three things I can do to lower my uric acid levels?"

I’m told I can do these three things.

Okay, but just to be safe, what exactly did the video say?

Clicking the inline citation brings up the video transcript, and Dr. Burke says, "Potassium citrate can knock out uric acid pretty fast."

Okay, nice.

防止数据丢失与保存输出

Now, pay attention to this next part, because I got burned by this.

Whenever you see a good output you want to refer back to, click "Save to note."

If you don’t do this, the output disappears the next time you reload the notebook.

According to Google, this happens because notebook LM is not trained on any of the data that we upload, including our conversations.

Meaning, if we close a chat without saving the note, all that data disappears.

健康报告与个人化建议

Next, I fast for 36 hours every week, and since both the Huberman video and Tools of Titans talk about fasting, I can ask:

"Does anything from my health reports suggest I shouldn’t fast for 36 hours every week?"

And again, notebook LM does a great job sharing context, telling me my blood work is normal and I can continue fasting with no issues.

Without notebook LM, yes, I can obviously go through the exact same information and draw my own conclusions, but that’s very manual, very time-consuming, and I might miss key information.

In contrast, notebook LM is able to quickly and efficiently reference multiple sources, connect the relevant dots, and produce a good enough output in significantly less time.

增强功能：深度分析与生成答案

Finally, I’m going to throw a curveball and prompt this notebook with:

"What do the sources say about incorporating deload weeks into my workout routine?"

And as expected, notebook LM says none of the sources say anything about deload weeks, which is true.

But the output does mention related topics.

This actually illustrates a very important point:

Even though notebook LM and Google Gemini might use the same underlying model, notebook LM is fine-tuned to hallucinate less, but as a result is also less creative.

Whereas Google Gemini, while prone to hallucination, is optimized for speed and creativity.

保存和转换笔记

Wrapping up this example:

Not only can we manually add a note, but we can also select multiple notes or just simply select all and choose to convert all these notes to a standalone source.

Pro tip:

Once that new source is created, we can click into it, select all the text here, copy, and paste it to use somewhere else.

By the way, if you want to cut through all the hype and master essential AI skills, you might want to check out my free AI toolkit.

I’ll leave a link down below.

使用案例：焦点知识检索

Use case number one is something I call "Focus knowledge retrieval."

Starting off with a simple example, I have a notebook titled "Equipment Manuals," where I’ve added all the user manuals for all my filming equipment.

This allows me to ask questions like:

"Hey, how do I update the firmware for this monitor?"

Or, "How do I enable this one specific setting in my camera?"

And notebook LM is able to retrieve that relevant information from my sources.

To replicate this, you can just Google the product you have, followed by "user manual" type PDF.

Side note: This is also how I found the PDF version of Tools of Titans.

And even if you can’t find a PDF version of the user manual, remember, you can add the website directly as a source.

Pro tip:

There are some websites that actively block notebook LM from adding them as sources, but we can easily get around this by adding another source,

and just copy-pasting the text from that website right here.

使用案例：税务与财务管理

Moving on to another example:

I have a "Tax and Accounting" notebook for my business, where I’ve added documentations like tax codes from the government and audit reports from my accounts in preparation for tax season.

I can ask this notebook questions like:

"What are my tax obligations last year?"

"What are some notable trends in my financial statements?"

And very personalized questions like:

"Hey, do I qualify for offshore tax exemption since I travel a lot?"

使用案例：招聘与面试

Last example for this use case:

Since I interview candidates as part of my full-time job, I have a recruiting notebook.

This is just an example where I add sources like guidelines from HR, performance rubrics, question banks, candidate resumes, and interview notes.

I can now prepare for these interviews more effectively by asking questions like:

"What are the key achievements and relevant skills of this candidate based on their submitted documents?"

Or, "Based on what we look for in product marketing managers at this level, give me 10 questions to ask this candidate."

And, "What are the key strengths and areas for improvement for this candidate based on the interview notes?"

Pro tip:

If you try this yourself, remember to only select the documents from the candidate you're currently interviewing, or else notebook LM might incorporate information from other candidates as well.

使用案例：项目管理与协作

Next up, we have the "Project context engine"

use case.

Put simply, I have a notebook for each project I’m responsible for at work, and I add meeting notes, project plans, and documents from similar projects as sources.

Side note:

Project and program managers benefit massively from notebook LM because, by definition, their job requires them to:

Work with information scattered across different locations.

Synthesize that information in an easy-to-digest format.

And notebook LM is designed to do exactly that.

结束语

Remember, suggested templates found under "Notebook Guide" – well, here we can take that a step further to create perhaps a high-level briefing document for senior leaders, a campaign timeline, and even an FAQ document.

I have a lot more use cases to share, so let me know if you want more content like this.

In the meantime, check out my AI playlist and…