PKM apps need to get better at resurfacing information

1. I have a Daily Note which is my starting point of the day. Aside from some TODOs and some daily reflections, it also contains an image map of things that are currently important. This map gets updated (in a template) once every couple of weeks, depending on where my focus is. Currently it links to a big project for a customer, my running progress, a webdav tool I'm working on, and a book I'm reading.

2. Many of my note templates automatically include a dataview query which spits out a list of related notes. E.g. for People that will be a list of recent meetings and mentions. For Programming related notes there's a list of notes that link to it, like for Java I have ~200 other notes that point at Java, from Method References to Apache Commons libraries.

3. Lots of links. This makes the local Graph View somewhat useful for rediscovery. This includes links to things that don't exist yet (and may never exist). Like I'm probably not going to create a note about Mitochondria, but several notes link to it.

Some more serendipitous discovery might be nice. I've experimented with various Related Notes plugins, most are garbage which find similarities where none exist because they can't tell the note structure apart from its salient content.

Personally, I’m on the other end of the spectrum. I’m still up for a good organization, and I do my own simplified version of PARA[1] + Johnny Decimal.[2] However, I’m beginning to tend more and more towards how Steph Ango uses Obsidian[3] while still maintaining simplicity and ways to walk out when needed. I’m either listing a list of files based on keywords, either via the Open command or the Search feature. I, sometimes, use a Universal Search Extension for Obsidian, but I can live without it.

My final thoughts, especially on the UX, are that, eventually, when it becomes commonplace to run a local LLM, I should most likely be just using Spotlight/Alfred or something to open, search, link, and all of that while I just keep adding contents into their own ”boxes” that I have categorized at the gate while filing the documents/content.

I’m not sure if I can even explain what I’m thinking, but the above is a precursor to what I believe will happen soon enough.

1. https://fortelabs.com/blog/para/

2. https://johnnydecimal.com

3. https://stephango.com/vault

I try to solve this by running a small embedding model on device, and the Ask Octarine feature, suggests what you may need to find/look. Isn't as sophisticated as Spotify's or the recommendation algo, however it provides a nice starting point, and then allows you to chat/RAG your way around notes!

Here's a small documentation writeup & video about it, if you want to look instead of try - https://docs.octarine.app/working-with-ai/ask-octarine

I did, however, with Logseq.

On the surface they're deeply similar -- but Logseq's native handling of outlines -- including collapsing and and expanding on the fly (org-mode does this as well, but you know) is absolutely killer. It's the first one to kind of reliably "free" me from the whole "every note is a page" thing that causes lots of friction.

I think it could go well to solving op's problem as well. Quick expanding and collapsing and easy search helps me to poke around older knowledge in a fun and useful way.

(actually, I didn't "switch," I still keep day-to-day important stuff in zim, but for that storing and exploring old knowledge, I'm now all about logseq).

I know that some people use them to learn, however I've never found this method to be useful for me, so I just stick to spaced-repetition (Anki) to help me learn the relevant facts.

Done this way, I have two "stores" of information. In Obsidian I put information I don't want to memorize or at least not long term, like work related topics (terminology, the domain, etc), project planning, project details, roadmaps, and in Anki I put information I care about learning long term, like facts about X technology or language, vocab, etc.

Point is, i think obsidian is not the right tool for learning. SRS sounds like a better fit. I'd suggest taking the effort to learn how to make good clozes (card type).

Yet as far as I know, the only thing an Obsidian account really unlocks, is a graph search. Which would seem like a missed opportunity

The LLMs are quite good at writing one-off Obsidian plugins, in my experience.

> If I’m not careful, my PKM apps become black holes where information goes to die.

Well, that's because just writing notes != learning. Obviously the information get forgotten.

I can't believe the author were so obsessed with PKM, but there zero mention of spaced repetition to help them learn and remember information. (Anki, Supermemo, or at least some Obsidian plugin with same function)

For tags, I use it much like a file navigator. In fact, it sits exactly where the files tree is in the default config. But, instead of files, its hierarchal tags.

1. https://github.com/scambier/obsidian-omnisearch

The process of organization with the app on my phone is itself mindless enough that I can focus on the meaning of what I'm looking at.

That's the friction point in a lot of these PKM solutions - the process of using them is set up for Big Important Tasks. The process of using the interface interferes with the flow of thought.

I've emulated some of the features the author wants in my obsidian using Dataview, but it's more the human ritual of review that accomplishes this for me.

I do agree, there is a good product there.

Among other things, it breaks your notes apart into facts and entities, then stores those along with embeddings. This helps surface things you've seen before since it makes and surfaces links based on individual things, instead of the text as a whole.

In case anyone is interested, I wrote a bit about the process here: https://nsavage.substack.com/p/facts-arguments-theses-buildi...

If you want to resurface them, you can show them in hierarchy(show children, or descendants in any level), search or filter by keywords, sort them by created time, URL, order, or other fields.

Resurfacing from current context can be very complicated, you should create some fields to track your activities, create many routines to handle these information automatically, and create relative UI. These are advanced features based on the current basic features.

[1] https://metaessen.com/

The author’s own gap fillers that he mentions sound cool, but his fixation on randomness makes me curious about what the purpose of his system is.

As for rediscovery of old stuff, the best solution is to get comfortable with the fact that certain notes won't resurface and to let them die. This is actually a good thing. It allows you to keep focus on your interests as they evolve. Even the legendary Niklas Luhmann, whose paper zettelkasten was arguably the inspiration for the current lmk craze had what we called "black holes" in his system—there are several notes he filed that never saw the light of day again. This was by design. These systems are supposed to be dynamic mechanisms for helpings you think, not immemorial retrieval mechanisms. If that's what you want you might as well make every note chronological.

Trying to use a PKM strictly as a "remember anything at any time" is largely missing the point. The goal is to create knowledge by stringing relevant ideas together and, as Luhmann did, to use the system to produce surprises. The best way to do this for obsidian in particular is to use terminology and link liberally whenever some term occurs in a note even if it at first seems completely unrelated—that's what you want, you want the system to surprise you as you navigate h the graph of notes.

[1]: https://thinktype.app

For now I'm keeping (and even adding!) folders because (1) AI is still not good enough at finding what it needs (2) context window is too limited with current gen LLMs and reading everything doesn't work at scale (3) few folders imply too many files in a folder, name clashes, etc.

I'll revisit that choice later, but for now my approach works, scales well and helps AI find what it needs (in general).

I've documented/shared my approach in great length in my course (https://knowledge-management-for-beginners.com), but it boils down to:

- A clearly defined set of note types (eg daily note, meeting note, literature note, etc etc)

- A specific tag for each note type

- A template for every note type (consistency for type tags + metadata)

- A root structure that combines PARA + Johnny Decimal (mostly shared across tools & platforms; not only Obsidian to reduce cognitive overhead & consistency)

- A structure that maps note types (e.g., folders for periodic notes, folder for meeting notes, etc). This works great because a note only has one type and the number of types is very limited

- Daily notes as the entry point of EVERYTHING. Every idea gets in there first

- Periodic reviews as the way to extract raw notes in daily notes into separate notes with tags, links, metadata, etc ...

I have 10+K notes and I can find almost anything in seconds. And with AI, I combine the REST API plugin with an MCP server & docs to clearly describe the note types/structure, conventions, etc. That way AI can find everything easily, create notes where they belong with the right templates, ...

With v3 of the starter kit, I'm adding new types, but most importantly, adding Obsidian Bases for all sorts of needs. This is the killer feature of Obsidian to leverage knowledge at scale. I'm busy publishing different posts about the design over here in my community: https://www.knowii.net/c/show-off-your-garden/

My goal with knowledge management is not to hoard stuff and waste time. It supports everything I do, and I intend to keep/leverage the knowledge for the rest of my life.

Also, I'm not fiddling with the system most of the time. V3 of my starter kit is the first time I do a major overhaul in the last 3 years. It's been super solid/stable. Just now, I believe there are major opportunities to better support/leverage AI, Bases and visual thinking.

I'll later revisit the structure again once the API for Bases is available, as it's going to be a huge enabler for simplifying things; cfr: https://x.com/dSebastien/status/1966456879606861883