Human-LLM-PKM Collaboration

The idea, stated plainly: the LLM’s arrival makes completeness obsolete as a goal for personal note-taking; the vault’s only remaining job is to hold what one mind is actively making, not what it has read.

That idea is sound and the piece earns it—six paragraphs, one function each, no dilution. But three things stand between this and something a skeptical reader finishes.

First, the lede overpromises. “Human thought has always leaned on tools” and the memory-palace/commonplace-book pairing set up a large claim about cognition generally; the actual argument is narrower and more useful—a specific verdict on what to keep in a vault. The historical frame earns its keep later (the “universal commonplace book” callback), so it’s not wasted, but the reader waits two paragraphs of taxonomy before the piece tells them anything they didn’t already suspect. The real start is paragraph three: “The consequence for note-taking is direct.” Everything before it can compress to two or three sentences.

Second, there is no instance anywhere in six paragraphs. Every sentence operates at the level of “the human,” “the vault,” “the LLM.” No note, no query, no moment where this division of labor actually happened to someone. A cold reader has nothing to check the claim against and no reason to believe it over any other plausible-sounding taxonomy. This is the piece’s one real hole—not a fact problem, since nothing here is checkable, but an evidence problem: assertion standing in for demonstration.

Third, the ending hedges where it should land. “And that, modestly, is the point” undercuts the sentence just before it, which was doing real work (“a workshop full of borrowed insight produces nothing without a craftsman”). “Modestly” is doing the same job “quietly” usually does in a draft—buying false humility instead of earning the close. Cut the last four words and stop on the craftsman line, or find one sharper than the hedge. Also flag “genuinely one’s own”—the adverb isn’t adding anything the sentence doesn’t already have.

The one change that would matter most: replace the historical wind-up with one concrete case of the triad actually working—a real note, a real question put to the model, a real judgment the human made that the machine couldn’t. That single instance would do more than any amount of taxonomy to make the reader believe the argument rather than just follow it.