[Organizational AI Utilization #166] What should you put into the 'brains' of NotebookLM or custom AI (Gem)? If you're lost, just go to a bookstore.

Hello! This is Terada.

Currently, I serve as the Representative Director of the AI Digital Community (ADC), a community for AI practitioners such as AI promoters at digital-related companies, as well as the Representative Director of FURIKAKE Partners Inc., which supports the 'xAI' transformation of client businesses, and AI Portalize Inc., which provides products that support organizational AI utilization. I support organizational AI utilization from various perspectives! Also, until recently, I served as the Representative Director of D-Marketing Academy Inc., a corporate generative AI & digital marketing talent development training service, and worked on promoting organizational AI utilization at the CARTA HOLDINGS AI Promotion Office. I have been deeply immersed in supporting practical AI utilization for the past two or three years.

What to put into AI knowledge

Recently, I have the impression that opportunities to utilize tools like "custom AI" and "NotebookLM" in the workplace have increased dramatically. Until now, it was just chatting with AI. But now, the mainstream way of using it is to have it read proprietary knowledge and then have it think and answer based on that.

In other words, mechanisms like RAG can now be easily built at hand, even by non-engineers. I think this is a very significant change. Of course, it is a convenient feature, but once you can create your own AI like this, many companies hit a wall.

※Please also see this article regarding how to divide "notebooks" for NotebookLM operation.

The problem is, "So, what should I have it read?" Therefore, this time I would like to write about how to think when you are lost in choosing the source that will become the "AI's brain," and some unexpected solutions.

Organize knowledge by separating it into "internal" and "external"

When thinking about the knowledge to have AI read, I do not recommend mixing it all together. I think it is best to first classify it into two broad categories. One is the knowledge within the company.

• Past meeting minutes

• Internal regulations and manuals

• In-house business workflows and know-how

By having it read these, the AI will act like a new staff member who is knowledgeable about your company. The image is that of an excellent assistant who understands all the internal circumstances.

The other is knowledge outside the company.

• General business frameworks

• Theories in specialized fields

• Trends and case studies in the world

By incorporating external knowledge in addition to internal common sense, the AI can have the perspective of an external consultant or expert. Furthermore, after you finish inputting internal documents, you may find yourself stuck wondering what to create next. I think this is almost certainly a case of being lost on how to choose this external knowledge.

Bookstore shelves are a "knowledge list" for AI bots

One way to come up with ideas is to list existing tasks one by one. However, that is a laborious task, and ideas tend to remain an extension of current work. Therefore, what I strongly recommend is going to a bookstore.

If you're lost and don't know what to build, instead of groaning in front of your PC, it's a good idea to visit a large bookstore nearby. This is because a bookstore is a place where human wisdom is systematized and packaged by genre. In fact, I believe that the arrangement of those bookshelves is a treasure trove of project proposals for what kind of AI you can create by inputting specific knowledge.

As you walk from one end of the bookstore to the other, try imagining what would happen if there were an AI containing the knowledge from this book. For example, suppose you stand in front of the SEO strategy shelf. If you look at the table of contents or the back cover of the specialized books there, you will see words like "keyword selection," "competitor analysis," and "secrets of rewriting."

If you could make an AI understand the SEO evaluation criteria written in such a book, you could come up with an idea for an "SEO Editor-in-Chief Bot" that would critique your blog posts just by having you paste them in. Also, if you go to the HR and evaluation shelf, there are books on "360-degree feedback" and "management by objectives." Looking at these, you might think that if you teach an AI the framework of evaluation theory, you could create an "evaluation interview assistant" that would edit your phrasing just by having you input the feedback content for your subordinates.

In this way, I get the impression that going to a bookstore makes it easier to visualize knowledge and functions as a set. Isn't it overwhelmingly more efficient to look at the spines of books that have already been systematized than to think from scratch in front of a PC?

Using "unique perspectives" that AI doesn't know as a source

However, you need to carefully consider copyright. I told you to go to a bookstore, but acts like cutting up and scanning a book you bought and feeding the whole thing into an AI to deploy externally carry the risk of copyright infringement. The premise is to use it only as a seed for ideas.

Finally, I will share a small tip for creating sources. When getting hints at a bookstore, there is no need to go out of your way to prepare sources for general term definitions or famous concepts. The reason is that current AI has often already learned basic knowledge.

What is worth loading as a source is information that the AI does not yet know. In other words, isn't it the sharp, unique perspectives of that book, its application in specific contexts, and the latest niche information? If you can successfully extract parts that the AI likely doesn't know and create a unique source, I believe your work efficiency will definitely accelerate.

First, load the information you have on hand, and if you get lost, head to a bookstore

First, try to get a feel for loading the information you have on hand using NotebookLM or custom AI. In short, actually touching and operating it is the shortest path. As a result, by knowing how the AI reacts, you will see what information you want to put in next.

Conversely, if you change the quality of the input, the quality of the output will also change dramatically. And if you are lost about what to put in next, please head to a bookstore. There, countless seeds for future AI bots that will help your work should be lined up.

※ Please also see this article regarding NotebookLM and Gem integration, which turns website information into AI knowledge.

First, tomorrow, on your way home or in your spare time, please take a quick look at a large bookstore near you. I hope you will imagine your own AI that will update your company's work while looking at the bookshelves.

Thank you for reading!
Click here for past AI-related articles!

By the way, I have also published a book summarizing the promotion of AI in organizations. Please take a look if you like.