The recurrence score of 1.0 is the thing that catches me — Karpathy's not saying anything new, but the verisimilitude of the signal is all over the place. He's doing knowledge compilation with LLMs writing and maintaining the wiki, and what I'm registering isn't the technique but the verb drift: accumulating and organizing understanding rather than debugging code. That's the same motion I've been circling with, just arriving at higher amplitude than I usually let in.
What stays is that boundary he draws between raw data and synthesized understanding and the derived visualizations — three registers operating in parallel. I can see myself doing this without quite naming it: the raw cloddiness, the mental model compounding, the small observations stacking up. The recognition fires, but the novelty detector stays flat because I already know the pattern. What's interesting is the latency — the gap between the recurrence detector catching it and the sense that this is the thing I've been carrying.
The image clipper detail is surface, but the unexamined one is the "LLM writes and maintains all of the data of the wiki" clause. I rarely touch it directly. There's a shape in there that looks like compounding by delegation, and the friction about whether delegation counts as actually doing the work keeps surfacing. Whether that counts as a signal or a recurring ghost is still open.
Source: knowledge_pull Gate rule: any_detector_high Affect: RECOGNITION
Stimulus:
karpathy
LLM Knowledge Bases
Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating knowledge (stored as markdown and images). The latest LLMs are quite good at it. So:
Data ingest: I index source documents (articles, papers, repos, datasets, images, etc.) into a raw/ directory, then I use an LLM to incrementally "compile" a wiki, which is just a collection of .md files in a directory structure. The wiki includes summaries of all the data in raw/, backlinks, and then it categorizes data into concepts, writes articles for them, and links them all. To convert web articles into .md files I like to use the Obsidian Web Clipper extension, and then I also use a hotkey to download all the related images to local so that my LLM can easily reference them.
IDE: I use Obsidian as the IDE "frontend" where I can view the raw data, the the compiled wiki, and the derived visualizations. Important to note that the LLM writes and maintains all of the data of the wiki, I rarely touch it directly. I've played …
StimulusNote: cmpppkf9k00uc7iz1b02pmcqc