Google explosive learning artifact update, a key to interrupt AI dialogue, where will not ask where

#News ·2025-01-08

The note taking tool that amazed the entire AI circle on its debut - Google NotebookLM has evolved again!

Recently, they have updated a blockbuster feature: Join discussions between AI at any time.

图片

Now, not only can you generate podcasts from a variety of sources in real time, but you can also click the Join button to join an online panel discussion at any time.

You can interrupt the podcast at any time to ask questions, skip parts you're not interested in, and ask for more details, similar to raising your hand in a web meeting.

NotebookLM team member Steven Johnson also gave a personal demonstration, asking NotebookLM to generate suggestions for his books.

图片

After a bit of flattery, NotebookLM really gave some constructive comments, and Steven was very satisfied.

The NotebookLM team previously demonstrated this feature at the Google Developer Conference, and recently it finally went live in Beta.

Netizens have also praised: the best AI tool of the year, much better than ChatGPT advanced voice function (doge)

图片

图片

It is a "ChatGPT" grade product that has been praised by Kapasi and Ultraman and other big guys.

图片

3 steps to generate live interactive podcasts

In just three steps, you can build your own interactive podcast with NotebookLM:

1. Upload the source

Create or open a notebook in NotebookLM.
Then add any source you want (documents, slides, PDFS, online videos, website links, etc.).

图片

2. Generate podcasts

Load the Audio in the Audio Overview section.

图片

You can also Customize podcast content around topics before it is generated, making it more focused on certain topics.

图片

3. Real-time interaction

Select "Interactive mode" and start the audio.

图片

Then, you can always click the Join button to ask questions or discuss!

图片

The measured results are very reliable

What is the actual effect? Qubits have also helped you measure a wave.

Let's first look at the ability to interpret a paper:

Take, for example, a recent Stanford paper examining LLM's mathematical abilities
Putnam-AXIOM: A Functional and Static Benchmark
for Measuring Higher Level Mathematical Reasoning, we generated a 16-minute podcast using NotebookLM.

图片

Then in interactive mode, we try to get it to enumerate and explain specific mathematical topics that appear in a paper.

It responded within seconds and explained the Fibonacci sequence problem in the paper with an example:

Short essays are fine, but what about longer ones?

We uploaded another 76-page summary of the prompt words and asked NotebookLM to generate a custom podcast with the following requirements:

Focus on tips for multimodal language models.

Pay attention to tips and tricks related to role-playing.

图片

图片

Next, in interactive mode, we asked two related questions:

When generating video, how can I more accurately describe the subject of the picture and the camera movement?

I am working for a financial employee in a securities company to generate a year-end report. How can I describe this position with more precise prompts?

In response to these two specific problems, NotebookLM also gave very professional advice immediately.

For example, when generating video description prompts, it suggests that we can give more specific descriptions of people, such as replacing "a lady" with a woman in a red hat, and more precise terms for camera movements, such as pan left, zoom in, and bird's eye view.

For the second problem, NotebookLM also returned a very reliable suggestion.

Not only did it ask us to try a role description for the big model, "Let's say you're an employee at a securities firm," but it also suggested that we could add more additional information, such as the projects done in the past year, the type of client, the relevant rank, and the number of years worked.

图片

One More Thing

In terms of function, there are still two small defects in this real-time interactive voice function:

  • Do not drag the audio progress bar while interacting
  • Currently, only English is supported

In addition to the interactive audio overview, Google has also introduced a new interface for NotebookLM.

The new interface organizes content into three areas: a "Source" panel on the left for uploading information, a "Chat" panel in the middle for discussing sources with AI chatbots, and a "Studio" panel on the right for creating audio overviews and study guides.

图片

NotebookLM also released a new Plus subscription plan.

NotebookLM Plus subscribers will receive several benefits, including the ability to generate five times as many audio overviews, notebooks and sources for each notebook; Customize the style and length of your notebook replies; Sharing team notebooks and usage analysis; As well as additional privacy and security guarantees.

图片

Reference link:
[1] https://x.com/stevenbjohnson/status/1869881743232381397.

[2] https://x.com/itsPaulAi/status/1875605628800135536.

[3] https://blog.google/technology/google-labs/notebooklm-new-features-december-2024/.

TAGS:

  • 13004184443

  • Room 607, 6th Floor, Building 9, Hongjing Xinhuiyuan, Qingpu District, Shanghai

  • gcfai@dongfangyuzhe.com

  • wechat

  • WeChat official account

Quantum (Shanghai) Artificial Intelligence Technology Co., Ltd. ICP:沪ICP备2025113240号-1

friend link