Tiny Practice and AI Tip -- Create Your Topical 'A Happy PhD AI Companion'

Have you ever wished for advice from this blog, tailored to your specific questions or situation? For getting a tl;dr of our blog posts on a certain topic? For an “A Happy PhD” podcast? In this post, I detail step-by-step an easy way of using AI to get information and advice tailored to your questions, about the blog’s very topics and contents.

If you have been on this planet during the last year, you will have heard of “generative artificial intelligence” (GenAI) tools like ChatGPT. Like everyone, I have been dipping my toe in the ocean of GenAI tools to see how they could help my research and this blog (see an example). Here is an easy way of using AI that is both useful, (relatively) ethical, and relevant to this blog.

The goal of this simple process is to have a little “AI companion” that digests, synthesizes and is able to respond to questions and give advice about “A Happy PhD” topics, in a way that is (hopefully) aligned with what we say in the blog, while citing the original sources in case you want to dig deeper into the blog’s posts. Google’s NotebookLM uses retrieval-augmented generation (RAG), allowing Google’s Gemini model to ground its responses in your provided documents (in this case, our blog posts on a specific topic).

Here is the step-by-step process to build such “A Happy PhD AI Companion”, on a topic of your interest, say, doctoral dropout:

Go to https://notebooklm.google.com/ and log in with your Google/Gmail account.
Create a new NotebookLM.
Go to the blog’s tag page for your topic of interest, e.g., doctoral dropout (see the blog’s topics page).
In the “Sources” tab, add as sources the URLs of the posts in the topic page in step #3 (yes, this is manual and a bit boring).
Enjoy asking questions to our blog posts (or getting an AI-generated podcast about them). Be mindful that it may make mistakes!

The tool can generate a quick summary of the whole topic with links to the different posts, and it can even generate a podcast-like audio with some of the posts ideas (which is a bit lame sometimes but not horrible). Sadly, you will not be able to insert there the full blog archive in this way. While NotebookLM’s 50-source limit might seem restrictive, specialized topic-specific uses of these AI models indeed are often more effective.

Of course, you can also use this with other sources for your thesis, to help you wade through literature and a semi-infinite to-read pile. However, use it with care! Chatting with a paper/AI in this way is not the same as reading the actual paper (in terms of learning and understanding)! I would rather use it to prioritize a reading list, so that sources that appear central to your questions are read first.

If you give this one a try let me know how that went and whether it helped you. I’m really curious!

Header image by MidJourney¹.

The prompt used was: “An chubby doctoral student in a labcoat, with a set of tiny robot companions sitting in her shoulders, shot from behind, lab equipment in the background, as drawn by Alphons Mucha in art deco style –chaos 15 –ar 3:2 –stylize 250 –v 6.1”. I am still trying to get these artistic AIs to give me more “normal looking” characters (as in, not uniformly caucasian and physically attractive). But it is hard to do that without getting caricaturesquely ugly results :-/ ↩︎

Luis P. Prieto