We tested Google NotebookLM: This AI could revolutionize work and learning

In July 2023, Google announced a service described as: “a notebook artificial intelligence” (IA). Originally released under the name Project Tailwind NotebookLM.

The company’s idea is to present An AI-powered work or study environment It helps you with various tasks. The platform has the ability to summarize texts, whether it is a report or an academic article, highlight the most important points, explain concepts and even create new content.

NotebookLM now Available in Brazil and Portuguese And TecMundo tested the trial version of the service. But can it really change the way we handle our notes, research, and presentations? Or is this another AI project that promises a lot but delivers content with little consistency?

How does NotebookLM work?

NotebookLM is a work environment based on Google’s own Gemini 1.5 Pro language model. It is a type of digital assistant that does the “heavy lifting” for those who need to deal with large amounts of reading.

The user’s first task when creating a new project (called “notebook” in the service) feed it a database. The files will be the “inspiration” for the AI ​​to work, which means it is the most important phase of the project.

NotebookLM currently works with: Google Docs or Slides files, PDFs, text documents, copied text, and website URLs. Adding a limited file base reduces rendering possibilities as the AI ​​doesn’t have all the internet or other text it can use.

For the same reason he less risk of making mistakes or inventing content, although it is not perfect. After making the selection, you need to wait a few minutes for the notebook to update.

The NotebookLM interface is very intuitive. On the left are the documents used as the basis of the “ledger”. At the hub you have a selection of ready-made content that you can request from the service.

The interface is extremely user-friendly and shows some of the possible AI tasks in the area “Notebook Guide”. Only then do we have an idea of ​​how the service can help us with these materials.

It is also important to keep privacy in mind here. Personal data is not used to train Google’s artificial intelligence, but “Real human reviewers can review your commands, loads, and model responses.” In some cases.

Therefore, the company itself recommends you: Do not send information that cannot be seen by a human reviewer.

NotebookLM’s capabilities

In our tests, we filled a “notebook” full of academic articles about digital influencers on social networks. The more similar the contents in the database are, the more coherent and complete the material presented by artificial intelligence.

Pre-programmed content possibilities are as follows:

• Common questions: the service creates a FAQ (Frequently Asked Questions) with possible content-related questions based on the main topics covered in each text;
• Study Guide: content helps you make an assessment by isolating essay questions about texts and providing answers. Teachers can use this function to prepare tests;
• summary: an organized summary of the main topics covered in the database, in a structure of points that could already be part of an outline, and even with a suggested title;
• Timeline: Artificial intelligence searches for numbered facts and dates in articles and creates a chronology of important events. It also organizes a list of “characters” or frequently cited people (such as theorists or people involved in the subject);
• Summary document: A report or white paper that summarizes the elements of the database. Its structure is in the form of an article consisting of introduction, development and conclusion parts.

There is also a text box at the bottom of the screen where you can enter requests for specific content or ask questions.

This prompting system is also useful: Just like with traditional Gemini, you can ask different questions by typing or choosing topics. In the answers, footnotes indicate exactly where the underlying content is located in the sources.

In general, the results obtained They don’t go beyond the basics in terms of vocabulary, but they are impressive. The feeling is one person reads all uploaded files and generated summaries in different formats.

Useful but has limitations

We encountered a few bugs in our testing. In one attempt, the service failed to generate a digest and returned a file with a title but no content. But overall NotebookLM is quite stable for an experimental product.

However, it should not be forgotten that it is still in the testing phase, so problems are expected and may be encountered. HE It also brings some limitationssuch as the ability to upload only 50 fonts per notebook and read slides only via Google Drive.

Also, although it works with a user-limited database, still able to interpret incorrect content. As with generative AI content, don’t trust everything you read: the result should always be reviewed by a human.

please take note of this The AI’s work is only as good as the quality of the text in its database. If articles, websites or texts are not objective or leave room for different interpretations, the same applies to the content produced.

NotebookLM does not work miracles and can improve a lot. But it is already impressive across a range of roles and certainly opens up many possibilities in the academic research sectors and the job market.