RAG without knowing RAG?

Notebooklm is a new product from Google, currently in beta. It allows the user to upload data, in the form of text, pdf, URL, or Google Drive.

It then uses Gemini 1.5 Pro to answer user's questions, by retrieving relevant information from the uploaded data. In technical terms, this is a kind of RAG (Retrieval augmented generation).

This product lets anyone, without any AI and coding knowledge, utilize the power generative AI model to provide answers based on their data (such as classroom notes, research materials, travel planning, and meeting minutes). Like an AI personal assistant, as Google suggested.

Here's a demo. I copied a text file that contains what I ate for dinner every day, for a continuous of 2 months.

Once the document is uploaded, it immediately provides a summary of this document and suggests some questions that I might have.

I tried to select one of the suggested questions - What's the common cuisine? It summarises with bullet points, with references quoted.

I then ask a random question - When did I last have chicken rice?

It says 31 July, and I cross-checked with the input source, it seems correct.

...
2023-07-29 Onion fried egg, noodle with sausage and baby carrots.
2023-07-30 Salad chicken with rice, fried green beans.
2023-07-31 Chicken rice, kaya butter toast, vegetables salad

Finally, I got it to do a summary for me, and translate it to Chinese.

Not too bad. :)