China’s AI circle has a deep-rooted disease, which we can call: AI commercialization fear and AI application lack syndrome.

The specific manifestation is that after an AI technology became popular, everyone first exclaimed, "Everyone has been transformed by AI" and so on. Then, after this track really heats up, it will turn into questioning: Can the AI ​​you make be commercialized? Why does AI still have no super applications? Why am I still not using AI applications?

It seems that AI always cycles between great expectations for technology and constant disappointment in applications. Is it true that big models make big noise but little rain, and are difficult to bring about applications with high growth and commercialization capabilities?

The recent popularity of an application on the Internet has given us some new ideas. Not long ago, the resignation of NotebookLM's core team sparked heated discussions. Team leader Raiza Martin announced that she would leave the company along with core designers and engineers to prepare for the creation of a new large-model application. The reason why this news triggered discussion is that NotebookLM is one of the most popular large-scale model applications in Europe and the United States this year. Even the reason why it is out of the circle is not at all as a basic function of AI notebooks, but as an early adopter additional function - AI podcast. generate.

Through the popularity of NotebookLM, we can talk about the commercialization possibilities of AI + voice, and how to create AI applications with high growth potential.

It may be said that more and more popular applications are proving that AI application deficiency syndrome can be cured.

NotebookLM was originally an AIGC writing tool created by a team at Google. There are many competing products around the world, and it has been launched for several years without any splash. But what recently made this seemingly mediocre app explode into popularity is the addition of a feature: podcast generation.

This updated feature allows users to upload documents, text, web pages, and YouTube videos to generate audio podcasts produced by AI. Different from other audio functions, NotebookLM does not generate a simple podcast that reads out the text, but a conversational podcast with two hosts.

The two anchors generated by AI will analyze and discuss based on the content provided by the material, and even tease each other and laugh. For example, if you input a history book, NotebookLM can generate multiple episodes of historical decryption conversation podcasts. If you enter your own information, it can also generate chat content in which two AI anchors blow rainbow farts on you. They analyze and discuss how great you are in a well-reasoned and serious manner.

This novel content generation model gives netizens great motivation to create and listen, and they can play more and more tricks. Some even believe that this may be the first AI to gain recognition not due to model capabilities, but through application experience.

And in NotebBehind ookLM is Google's Gemini 1.5 Pro large language model. It can handle up to 1,500 pages of content at a time and supports multi-turn conversations with extremely long context. NotebookLM's audio function is based on Google's voice model.

At this point we will find that NotebookLM is a standard AI native application that cannot be more standard. It is implemented based on multiple large models; it utilizes very representative AI mechanisms such as AIGC and multi-modality; and creates a service experience that is completely impossible to achieve with traditional mobile applications.

It can be seen that as long as the capabilities are right and the market is identified, it is entirely possible for AI applications to support traffic and commercialization.

Of course, the trick of NotebookLM is that its market entry point is very precise. The market demand for podcasts in the United States is huge, even surpassing short videos. Most American users, especially young users, have the habit of listening to podcasts, and celebrities, athletes, and talk show actors also generally record podcasts as a way of expression.

In this market with high attention and high demand, AI has brought an unprecedented application experience of two-person conversation podcasts, which made NotebookLM explode in popularity within a few months. Out of the circle.

If you only focus on AI podcast generation, then the meaning of NotebookLM may be cut out a lot. First of all, the podcast market in China is far less popular than in Europe and the United States. Secondly, the commercialization potential of AI podcast generation is huge. It's limited.

If you want to deeply utilize the significance of NotebookLM, you should first see that the AI ​​audio track it represents has a wider commercialization space. For the widespread business anxiety in the AI ​​industry, AI+audio is at least a relief. A large amount of content that is too long to watch in life can be converted into audio to improve reach efficiency. For example, within Google, the Google advertising team is studying the generation capabilities of NotebookLM to create sales Q&A to train new employees. In this way, the original heavy information can be simulated into a conversation between sales staff and customers. New salesmen can directly learn how to communicate with customers and how to respond and explain each question.

There are many similar examples, such as the audio version of product manuals and the question and answer mechanism of online education. Many times we are faced with the dilemma of having too much information and not looking at it for too long. At this time, we often want to ask a few questions from knowledgeable people, but now it may be enough to ask AI.

By extension, NotebookLM can also be combined with machine vision. For example, when you go to the gym, the job of a personal trainer is actually to watch you train, then point out the correct and incorrect movements, and answer questions from the bodybuilders. If NotebookLM is equipped with visual capabilities, perhaps AI podcasts can become AI personal trainers.

These explorations prove that an AI application that goes out of the circle oftenHas a chain reaction. Not only can it be popular among users in this circle, but it also has the feasibility of continuing to spread business influence.

Even if you don’t listen to AI podcasts, you have no interest in the AI ​​audio track. NotebookLM also still shows the lowest level of excellent AI applications: simple input, production of surprises, and closeness to users.

Nowadays, large models can do many things that have never been done before, but there are always no explosive applications that can trigger discussion and spread. The popularity of NotebookLM tells us that a big reason comes from the lack of product design.

Compared with other AI applications, NotebookLM has a distinctive feature at the product level, that is, the overall UI design tends to be as simple as possible. AI podcast generation itself is only part of the NotebookLM function, and the threshold for users to open it is naturally high. In this case, the project team tried to simplify its overall UI as much as possible. For example, uploading documents is a step that many AIGC applications have. However, the steps to upload files are trivial and have low fault tolerance, which often gives users an unpleasant product experience. NotebookLM supports one-click upload of source documents, and can cover multiple file sources and file formats such as documents and videos. This very friendly user interface design makes this innovative AI application have an extremely low threshold. Users will not be put off by trivial operations and complex file requirements at the UI level.

What is proportional to keeping the input end as simple as possible is that NotebookLM provides very complete and complex content on the output end. Using AI to generate podcasts is not new today, but NotebookLM can generate two-person conversation podcasts, and the conversation content has a tone, rhythm, and even a sense of humor. This sense of surprise and unexpectedness contrasts with the minimalist content input, and then becomes the driving force of the product that can make it out of the circle.

In addition, there is another key factor in NotebookLM’s success, which is that it has found podcasting, a field with great market appeal in Europe and the United States. Precisely because podcasts themselves have market demand and receive high attention, the AI ​​surprise brought by NotebookLM can naturally activate user recognition. And NotebookLM has once again proven that finding young people who are more receptive to AI and being close to the life, entertainment, and consumption patterns of the young market are the keys to the success of AI applications.

Summarizing the story of NotebookLM, we can find that a successful AI application needs three elements: 1. Minimalist user interface and generation rules. 2. AIGC effect with surprise and shock. 3. Close to the market focus of users’ attention, especially the attention of young people. Starting from the capabilities of the large model, if we find these three conditions, we should be able to outline a similar AI application. The so-called AI application deficiency syndrome is a huge problem from the positive side, but it presents many concrete opportunities from the negative side.