The smartest light bulb: warm NotebookLM

Disclaimers: The thoughts expressed in this article are my personal opinion, which may differ from Google's point of view. What the robots say in the video is the result of AI calculations, which tried very hard to adhere to official press releases, but could lie. It was important for me not to edit or select the generated podcast.

In general, gathered as Vanya and Zhenya somehow, threw the prompt into NotebookLM (three times, like Pushkin) and came the wav with such a podcast:

For those who whom Google banned, I send the same video by telegram.

First, let's deal with the intellectual part. If we come home and it's dark, we take and turn on the light. This is such a simple and routine thing that we usually do not think about how many complex technologies and brilliant solutions are behind making everything so simple and safe. In this art project, I wanted to achieve the same thing: take an incredibly complex technology that was science fiction a year ago and use it like my grandmother. So I took NotebookLM, gave it links to the Gemini 2.0 press releases and wrote a prompt:

Prompts can and should be polished to get a good result, and the output of the neural network needs to be checked and edited. But it was important for me to keep the authorship for AI, so I made only three iterations.

I also needed to insert my remarks into the video. Fortunately, the AI watched a lot of podcasts and decided that without a break in the middle - no way (to show ads, drink water, or vice versa...). In general, it fit well.

NotebookLM speaks in two voices: this is much more engaging. I spread them across two channels using another neural network: clicked with the mouse in audacity. I was taught this at Yandex: sometimes prototypes don't even need to be coded. But then I needed to make two avatars. The first one I almost had...

Zhenya Spiral'kina

The previous article was about how the talking lamp works. So here it's only about the visual image. This is what the second prototype looked like:

The first prototype was held together with spit and crocodiles, and for humanitarian reasons I didn't photograph it. Special thanks to @vvzvlad for introducing me to terminal blocks, I recommend everyone who wasn't at tekkix in 2012 to read it. The first pancake turned out to be quiet. The bulbs were too tightly fixed, the horn was not in the right plane. The magnet was not serious... In the office it turned out to be not heard at all.

So I put the racks on hinges, bought a bigger magnet, took a photo of my beloved wife and went to bend the wire in her image and likeness...

In general, I was lucky with my wife: she is understanding and forgives a lot. But I decided not to roll this friendly smile into production. In the end, simple forms and minimalism won. And instead of hinges, I took even thinner wire.

Just watching him tremble slightly is pretty boring. So we turn up the bass to the max and add a sub-tone of a few hertz. This immediately hits the lumberjack with emotions. There's still the lightbulb head - I controlled it manually. Of course, you can attach automation - but then it will be an installation, and so - a performance: only today, only for you.