The tool generates a podcast called DeepWho presents a man and a female voice discussing everything you have downloaded. The voices are breathtaking realistic – the episodes are laid down with small sentences with human consonance as “man” and “wow” and “oh right” and “hold on, let me do things well”. The “hosts” even stop.
To test it, I copied each story of MIT review of MIT technology‘s 125th anniversary number In Notebooklm and made the system generate a 10 -minute podcast with the results. The system has chosen a few stories to focus on, and AI hosts have done an excellent job to transmit the high -level general essential of what was the problem. Listen.
MIT Technology Review 125th Anniversary Number
The AI system is designed to create “magic in exchange for a little content”, Raiza Martin, the product leader for notebooklm, said On X. The vocal model is intended to create an emotional and engaging audio, which is transmitted in a “hyper interested tone”, said Martin.
Notebooklm, which was initially marketed as a study tooltook his own life among users. The company is currently working to add personalization options, such as modifying the length, format, votes and languages, said Martin. Currently, it is supposed to generate podcasts only in English, but some Reddit users have managed to obtain the tool to create an audio French and Hungarian.
Yes, it’s cool – increasingly delicious, even – but it is not safe from problems that afflict generative AI, such as hallucinations and biases.
Here are some of the main ways that people use Notebooklm so far.
Podcasts on demand
Andrej Karpathy, member of the founding team of Openai and previously director of AI in Tesla, said on x that Deep is now his favorite podcast. Karpathy created His own Podcast AI series entitled Stories of Mysterieswhich aims to “discover the most intriguing mysteries in history”. He says he has been researching on subjects using Chatgpt, Claude and Google, and used a Wikipedia link from each subject as a source material in Notebooklm to generate audio. He then used Notebooklm to generate episode descriptions. The entire Podcast series took him two hours to create, he said.