The excitement around AI has picked up again on the Web. In recent days, Internet users have become passionate about NoteBookLM, and more particularly, its “Audio Overview” functionality. This tool – currently in prototype state – left Google’s laboratories at the end of last year, but it was recently updated.
More concretely, it allows you to synthesize different documents in a few clicks, and above all, to transform them into a podcast of around ten minutes. Currently this feature is only available in English.
The Tribune tested it and the result is, it is true, often astonishing. The two virtual hosts, always with a warm and enthusiastic tone, talk about anything: a 250-page report on the future of artificial intelligence, a very in-depth scientific article on cardiology, a 2-hour YouTube video , but also any other more trivial document such as a LinkedIn profile, a bank statement, an exchange of SMS messages… Google specifies in passing that this data – sometimes very personal – shared by users, is not used to train its AI.
Protection of personal data: Google must face a new complaint in the United States
Synthesis of dozens of different sources
The Mountain View firm presents NoteBookLM, which is based on its multimodal Gemini 1.5 model, as a virtual “research assistant”. It is possible to submit several documents in Google Docs, Google Slides, PDF format (courses, notes, meeting transcripts, commercial documents, etc.).
It will thus produce different summaries, find common points, recurring people, significant dates, etc. The tool also allows you to query these documents, by asking questions via a chatbot. A “citation” functionality makes it easy to find the source of ideas noted by NoteBookLM.
“It’s a tool for understanding things”simply sums up Steven Johnson, one of the designers, on the HardFork podcast from New York Times. Note that Steven Johnson is a writer, who recently joined Google Labs with the aim of developing a tool specifically for writers, which therefore became NoteBookLM.
Other AI applications already allow this document synthesis work – like Anthropic’s Claude Project – but Google stands out for the quantity of information that can be downloaded. A NoteBook project can contain 50 sources of 500,000 words (or approximately 5 novels) each or weighing 200MB. The latest update of the tool allows you to use YouTube videos, but also audio files.
OpenIA becomes the third largest start-up in the world after fundraising
A tool that has gone viral on X and TikTok
Above all, NoteBook stands out for the generation of very realistic “podcasts”, which it has recently become possible to share with others via a URL. Enough to make the tool viral. First within the tech community: on Hacker Newsa reference site in the sector, NoteBookLM is one of the most commented topics in recent days.
On X, investors, entrepreneurs and researchers have fun publishing their “Deep Dive” experiments (the name given to these synthetic podcasts), like at the very beginning of ChatGPT. Some of these experiments are shared several million times. Ultimate recognition: Sam Altman, creator of ChatGPT, recognized during the very first OpenAI conference for developers that NoteBookLM was “very cool”.
Sam Altman, all-powerful at the head of OpenAI
Courses transformed into quick podcasts
The tool is also starting to gain a wider audience. On TikTok, students praise its merits, claiming that it “changed their lives” by allowing them to summarize hundreds of pages of lessons into podcasts lasting tens of minutes. Some of these videos already have more than 10 million views.
Raiza Martin, senior product manager for AI at Google Labs, told TechCrunch that the users, who were originally mainly from the world of education, today include 50% educators and learners, and 50% professionals.
@leahmiami this is about to change the game for studying. shoutout to the guy that sent this in our gc we all owe you #study #studyhacks #studytok #lawschooltips #lawschool #studytips ♬ original sound – leahmiami
According to SimilarWeb, NoteBookLM has already accumulated 1 million monthly users. We remain far from the records of ChatGPT, which had gathered more than 100 million users after two months of existence.
However, it is a great feat for Google, which is struggling to find its place in this frantic race for generative AI. Its Gemini model never enjoyed the glory of its competitors GPT-4 and Llama. We know him more for his slip-ups than his prowess. And it would be, according to The Informationlittle appreciated by developers, who find it less easy to use than those of its rivals.
The Nazis and Black Vikings of Gemini or how Google failed in managing AI bias
Restore your image
It was ultimately with a product launched experimentally, without much fanfare or publicity, that Google ended up distinguishing itself. For what ? Because the company offers a new way of interacting with generative AI, which does not involve chat, or even audio conversation, but by listening to content.
NoteBookLM also shows that the challenge now is not so much to develop the most powerful model, but also to find a way to package its capabilities into a product useful to the public.
It remains to be seen whether the company intends to monetize this tool, and if so, how. In HardFork, Steven Johnson assures that several new features are in preparation: the generation of podcasts in languages other than English, and the possibility of interrupting virtual hosts to ask them questions.
Related News :