LLM Wikis are becoming popular, and my biggest concern about using them is that they reduce the germane load in PKM.
In the early days of the ChatGPT craze, something felt off to me, and I wrote the newsletter An Ability You Must Never Outsource to AI: Summarizing. Generalizing a bit further, the problem seems to be that it reduces germane load.
The LLM wiki I'm targeting in this piece is not the LLM Wiki used as long-term memory for AI agents. I'm talking about people combining Obsidian with an LLM to use their PKM as an LLM Wiki. In other words, it's the problem with Vibe Noting.
Germane Load in PKM
Using a PKM involves a certain germane load. There is little benefit to the user in eliminating it, yet certain AI usage patterns remove it.
It's like doing a squat where I hold the barbell at the bottom, a coach lifts it all the way up from below, and then I shoulder it again. The process itself is what matters, but this tries to get the result while skipping the process. These days people skip exercise and take Mounjaro or Wegovy, and this gives me the same feeling, which makes me uncomfortable.
The germane load that arises in PKM comes in the following kinds.
The first is links. You can't leave links out of a PKM. Connecting one note to another is an important form of germane load, because it matches exactly what cognitive psychology calls elaboration. A question like "What is this note related to?" pulls my existing notes out of long-term memory, triggers the encoding process as well, and helps me remember the note I'm writing now more firmly. Connecting notes from different contexts to form schemas is also an important part of gaining insight when using a PKM.
But if you link notes only as the AI recommends, the germane load shrinks. The process itself is easy and fast, so it's easy to build a network of interconnected notes. It's easy to produce a graph view worth showing off online. But those concepts aren't actually connected in my head, so the notes become no different from material found on the internet. They are my notes, but not really mine.
The second is extracting key concepts. You could also call it summarizing. For this, An Ability You Must Never Outsource to AI: Summarizing is a good reference.
How to Cope
The first approach is to accept desirable difficulty. This is a mindset approach: don't avoid the parts of PKM that feel hard and annoying, but accept them. Writing and maintaining notes is clearly a process with a heavy cognitive load. But if you rely on AI to escape this pain, you end up with a PKM that is effectively useless. So first, let's acknowledge that the pain of maintenance is unavoidable. (Of course, this doesn't mean we should accept extraneous cognitive load too.)
The second approach is to navigate the notes by continually digging into linked notes. (Keeping Yourself Looking at Your Notes) For a PKM to work as a second brain, it has to be an add-on to my actual brain, my first brain. But the wider the gap between the PKM and my brain, the more useless a second brain it becomes.
To see where the gap can be narrowed, I want to borrow elements from the extended mind hypothesis. For an external tool to act as an extension of my mind, it must be (1) accessible at any time and (2) trusted. I take this "accessible at any time" to mean more than carrying a smartphone around so I can open Obsidian. It means being able to reach the note I want quickly, whenever I want. Looking often at notes I've made, whether the AI or I made them, makes the notes I want faster to reach. In the process, my skill at finding notes improves, and I may also restructure the notes to make them easier to find. And looking at notes increases fluency, which in turn increases trust itself. 1
The third is Progressive Summarization. For convenience, it's hard to give up using AI to create notes altogether. But after they're created, you can generate germane load through progressive summarization. When summarizing, you shouldn't simply delete unneeded sentences. You have to paraphrase in your own words. For example, delete a whole paragraph and write its core in a single sentence yourself.
Footnotes
-
It is commonly said that reading text over and over increases fluency but not actual understanding. Here, though, I actually want to use that increased fluency to gain the benefit of building trust. ↩