OpenAI Assistants, Google NotebookLM are going to change things a bit.

A Rag Doll cat working on his AI start-up, DALL-E

First some AI Career Advice: If you want to look like this week’s leading AI trailblazing innovator, quick as you can, go grab some documentation and upload it to GPT or NotebookLM before everyone does.

RAG, of course, stands for Retrieval Augmented Generation. When you ask a question of your LLM, the software looks for relevant documents to feed to the LLM as well so the AI can answer the question. For example, if you ask a question about Mr. Darcy from Pride & Prejudice, the chatbot would first feed the book to the LLM before passing along your question. This makes the AI look like it knew this, when literally it just read it.

A year ago, RAG was the subject of academic videos discussing the future. But now, just last week, Google introduced a RAG-centered tool, Google NotebookLM. The idea is that you upload documents to a notebook, and then you can ask it questions.

This is somewhat akin go the “GPTs” that OpenAI introduced last month, where you can (among other things) upload a variety of documents and ask questions against them.

Way back in August, when I wrote an earlier post on RAG, you had to do a fair amount of heavy lifting on your own to implement RAG; there was no direct support for it in GPT or other LLMs. How do you manage the documents? How do you figure out which ones to retrieve? How do you feed it to the AI? None of those are trivial question. That limited its usage to those who could code (or at least LangChain, I suppose). But now?

If you’re in school, and been assigned to read a classic book, go to Project Gutenberg, grab the text, upload it to one of these services, and ask questions … or ask it to write your paper! Whatever technological advantage we thought we had growing up is going to be insignificant in comparison.

I predict (always a perilous thing to do) that 2024 will be the year where everybody figures out how to upload documentation to make their own “copilots” for anything and everything. Like with everything AI, it will be far harder to make things work consistently and reliably than to piece together an amazing demo (I’m looking at you, Googs).

As I mentioned above, my advice to you is to go grab some documentation, upload it to GPT or NotebookLM, and proclaim yourself a pioneer NOW, before everyone does it!


  1.” was taken, probably dooming our poor rag doll’s startup.
  2. Old AI: “Once it works, it works all the time.”
    New AI: “Once it works, it randomly doesn’t”

Comments are closed.