I have a problematic relationship with organization/knowledge management tools. I’ve wanted to use them to make keeping track of my life easier. But I have a recurring cycle of trying organizers and then abandoning them for plain textfiles scattered across my home folder. The organizers I try all suffer from what I informally call the “input problem.” I’m not sure how much of it is a problem with the organizers themselves or a unique issue with how I experience them. Nevertheless, the difference isn’t practically meaningful.

The input problem

So what is the input problem? There’s low-hanging fruit that is relatively easy to put into a structured representation like “remember to schedule dentist tooth filling” or “grocery shopping checklist.” But so much of my life — or how I intuitively experience it — is dispersed piecemeal across many overlapping but distinct formats and contexts. Not all of them intuitively fit into the organization tool and need to be modified in some fashion to be compatible with it. The tedium of translating, encoding, converting, storing, and organizing quickly adds up.

Keeping the organizer current requires endless work that detracts from the time and effort needed to perform the tasks being organized. The work feels pointless because the end product is a fragile and static representation of fluid circumstances. Sometimes it even feels like an authority figure threatening punishments for non-completion of daily data inputs. I tried using the organizers as planning tools, but that didn’t really work out either. I ran into variants of my original input problem and quickly gave up.

I feel like most personal organizer/notebook/wiki tools are stripped down versions of software originally intended for enterprise clients. They don’t seem like they’re really intended for single-user small-scale problems. I’ve also tried graphical tools — everything from GANTT charts to mind maps. There’s a chicken-and-egg problem: if the problem I’m trying to create plans for was well-organized enough to be charted, diagrammed, or mapped out I wouldn’t need the graphical software in the first place.

A part of me still thinks I just haven’t given the graphical tools enough time. It would be really cool if I could visually represent something I’m struggling with as dynamic, interactive, and entrained objects and their dependencies. I’m the kind of guy that would get really excited making a UML diagram or engine block schematic out of a relationship problem I’m having. But I still haven’t found a way to do that without it looking like one of those military spaghetti slides people keep mocking. Maybe I’ll eventually figure it out, but I can’t right now.

i finally broke the cycle after unexpectedly having an epiphany about my organizational issues. I probably would never overcome the input problem because forcing my life into a tool always felt too much like homework. I’d just have to do my best to keep track of things in the way I’m used to. Finding a total life organizing schema wasn’t what I actually wanted out of organizing tools. I wanted something very different: a substitute for chatting with a friend. Talking to people about things is how I end up imposing conceptual order on them.

I’ve never been able to keep a diary because “getting it out” on paper (or bits) doesn’t have therapeutic benefits for me. When something is too amorphous to make sense of and feels too intractable to fix, conversation makes it feel less oppressive and more concrete. But people aren’t always around when you need them. How could I replicate that calming experience in “single-player” mode? Was that really possible? Not if you interpret that question literally. If interpreted more liberally, many possibilities open up.

Writing as alien voice

Nothing can ever really replace talking to friends, family, and loved ones. Other people have their own minds and personalities. They also have preexisting relationships with me that allow them to understand me even if I don’t necessarily understand myself and respond productively. I cannot really replace that and I am not fully attempting to do so. However, breaking a “feels bad issue” into a structured hierarchy of notes worked when I started treating it like a conversation. 1 Albeit a conversation with a strange disembodied version of myself.

The separation of thoughts from the body — what Plato abhorred about writing — is precisely what made a difference.

[F]or Plato, it was writing itself that threatened to undermine the human subject of knowledge by outsourcing its most fundamental attribute – what he describes as “the living, breathing discourse of the man who knows” – to an uncanny simulacrum that “you’d think was speaking” yet remains “solemnly silent.”…Plato noted [an]…eeriness in the way in which written texts mimic the flow of live human speech,

That very uncanny feeling of externalized speech made “talking” to the organizer neither self-expression nor rational organization. Displacing my thought into structured writing amounts to constructing a stripped-down counterpart that extends myself while still being in some way meaningfully separate from me.

The problem in my head was “my voice”, and what I produced in the organizer was some alien imitation of it responding to me. It was something listening to me and helping me make sense of a problem despite being unable to really “talk back” to me in the same way another person would. It’s me but not me, but it also feels like someone else. Despite being nothing more than an inert tree of markdown files. 2 There’s a useful tension in that process and the active labor involved is a nontrivial component.

It is possible that I am falling prey to the cognitive illusion that self-made objects are more valuable. However, there is some use for the input problem as long as it is not overly tedious. Educators have always understood that you cannot separate mental labor from learning, whether it is rote memorization or high-level problem solving. Everything has tradeoffs, and one should not raise or lower the amount of effort and granular focus without logical justification. For now, I feel the tradeoff is pragmatic.

What about LLMs?

I suppose you might get something similar from chatting with an LLM. So why not just do that? This is not the place for a discussion about LLMs, but I’ll just make one broad observation about them from the standpoint of late 2024. 3

Some people have crudely analogized LLM outputs to outputs from interpreters or compilers. A high-level program — written in a human-comprehensible structured operational language — is fed into a mechanism that eventually results either in direct execution or the production of an executable. LLMs are far less precise but also require some kind of strict operational subset of human-comprehensible language to yield optimal results.

It seems pointless to “talk” to an instrument unless you already have a well-formed idea of what you’re going to say and what it ought to tell you in return. That might be useful for some problems but it serves a different purpose than what I have in mind. I’m not ruling out the possibility that LLMs could fulfill that purpose, but I’m wary of asserting too much without putting in enough time tinkering to find out.

.

Footnotes

  1. For full transparency, the organizer I’m using is Obsidian, which is also the visual editor I use to author content on this blog. I don’t think its features are particularly relevant to my argument, as you might get the same effect from any number of other possible tools.

  2. The computer in general is “on the border between inanimate and animate…both an extension of the self and part of the external world.”

  3. As an extended aside, LLMs are advertised as instruments theoretically capable of assuming any generic identity you arbitrarily assign them given sufficient guidance and resources. There are many reasons it might fall short of the ideal: the competence of the user, the “safeguards” built in by external parties, the architecture and mechanism of the LLM itself, the data it has access to, infrastructure bottlenecks (everything from GPUs to API limits), and so forth. Human beings are far less flexible. Aside from gifted actors and intelligence operatives, we cannot help but behave as we are even if we aren’t always predictable. The LLM at least superficially seems to be a true “voice from nowhere.” When revising this post I came across an intriguing essay Snav wrote about the distinction between “tokens and coins” with respect to the truth-value of LLM inputs and outputs. There’s some overlap with the “view from nowhere” angle so I may return to it later.