Building Complex Research Diagrams with AI

Dear Reader,
​
Having read so much content on AI and followed the news closely, I sense a shift in the collective enthusiasm for AI's transformative potential.

From one angle, we start realising that the claims of AI's practical usefulness are overblown. An MIT study found that:

From another angle, study after study after study found that using AI erodes critical thinking skills and leads to something called "cognitive surrender", where, out of a feeling of overwhelm and AI's confidence in talking to us, we defer to its judgment, even if it makes deliberately faulty statements.

The problem, however, is that while the short-term effects of AI are apparent, its long-term effects are not. The (AI) philosopher Carlo Cordasco, in a recently published essay, compares this to the introduction of anaesthesia in surgery, or electric lighting on streets, which at the time were rejected (for scientific and logically sound reasons), but in the long run enabled technologies nobody could even have imagined at the time. From his personal experience, he found his work shifting from digging into research questions over days to asking more questions while letting AI do the research.

This is by no means a negative development, but an empowering one.

My personal observation is similar. I find AI moves us to delegating tasks and interpreting results (the conductor mindset), and it enables me to engage differently with the subject. The key difficulty I find is outsourcing information gathering, but not interpretation; outsourcing searching, but not synthesising ideas. The lines between these, however, are often easy to cross, and sensationalism pushes us to outsource everything, leading to what some scientists call "cognitive collapse".

In today's tutorial, I am trying to do just this. Maximise what the newest generation of models is capable of doing without outsourcing my understanding to the machine.

One of the things that enabled me to work through my PhD quickly is visual mind mapping (see finding research gaps). If you've joined my "Finding Research Gaps Course", you might have experienced firsthand how powerful this is. Here's an example:

These mindmaps summarise dozens of papers into a single mindmap, making it much easier to analyse a research domain or decipher complex papers.

However, building these mindmaps takes an enormous amount of time, most of which is spent moving boxes around and handling Draw.IO's clunky UI.

So in this tutorial, I built an entirely new skill to generate the first drafts for me. This eliminates the friction of starting from scratch and is genuinely useful. (I tried the same approach in 2023; if you're curious, you can compare how puny the outputs were back then).

Summary: The workflow uses three components to build an AI skill capable of generating complex diagrams. (1) You need a legend or visual language that assigns meaning to colours and shapes in your diagrams, (2) examples of existing diagrams you made by hand, and (3) a technical description of the diagram format, which you can download online. These three components can be used by an AI agent like Claude Code or OpenAI Codex to generate very impressive diagrams based on your notes or a single paper (Examples in Post). Treat the output like you would academic writing from AI: as a first draft that you need to critically vet and build upon.

Lit Review Course Update: Paperpile AI

If you've been following me for a while, it's not a secret that Paperpile is my favourite reference manager (minus the Obsidian integration). Today I recorded a video on its new AI integration.

The approach Paperpile takes is quite innovative.

Rather than being a chat-with-your-PDF interface like everyone else, Paperpile sends the PDF(s) to an LLM of your choice, along with well-crafted, complex prompts, allowing you to extract concepts, compare papers, build a lit review matrix, and much more. The output contains links to individual highlights inside the PDF, allowing for a back-and-forth navigation.

The update is free for everyone who has purchased the lit review course. Just follow this link to the lecture.

Since this is the third update to the lit review course and I am planning more updates, the price will increase with the next one. If you want the upcoming updates for free, get the course soon, while the price is low.

The lit review course is one of my flagship courses with over 350 students currently enrolled and 1000s across the preceding webinars.

In today's email, I'm sharing a bit more background on what's happening in the academic AI debate right now. This makes the emails longer and maybe irrelevant to some. What do you think?

Thanks for being part of EA!

Wishing you a wonderful weekend,

Ilya Shabanov, The Effortless Academic​

The Effortless Academic is a community, newsletter and learning place for the modern academic, where Ilya Shabanov writes about information management, AI, and research tools.
​
If you want to be more productive as an academic, start with the foundation of Digital Information Management and the AI Lit Review Course to learn about tools for finding, reading and synthesising academic papers. Or get the all course bundle at a discount.