LLM Wiki as Medium

Meta-page: why this wiki exists, and how it embodies the Karpathy LLM-wiki pattern that Jacob and David ended up enacting in real time.

The conversation that became this wiki

Near the end of the recording, after Jacob had spent an hour and a half pouring out his full vision, he said:

"I need to have an LLM wiki for the vision for IdeaFlow in the world that just keeps updating. And I'll put one together tonight. Out of this conversation, we'll keep pushing stuff to it. Okay, on WikiHub, and then it's a great use of LLM wiki."

And:

"Just trying to get this thing into an LLM wiki before I lose the thread here."

The conversation wanted to become a wiki. It already had the structure of one — a tour through entities, concepts, projects, practices, all interlinked. The transcript is the raw layer; this site is the wiki layer.

The Karpathy pattern, applied here

Andrej Karpathy published an LLM-wiki gist in April 2026. The structure:

1. Raw sources (raw/) — immutable. The LLM reads but never modifies.
2. Wiki (wiki/) — LLM-generated markdown. Concept, entity, summary, comparison pages. The LLM owns this layer entirely.
3. Schema (AGENTS.md) — config telling the LLM how to structure, ingest, format, cross-reference.

This wiki maps cleanly onto that:

Why this medium fits this content

Jacob's vision is graph-shaped. A linear essay doesn't capture it because the same node (Sparks of Motivation) appears in:

• the IdeaFlow product story (capturing them)
• the Inner Ecosystem philosophy (subconscious sparks unifying)
• the Collective Intelligence vision (humanity's collective open gestalts)
• the Super Conscious State practice (sparks of vitality directed inward)

A wiki lets you land on any of those four, see the link, and traverse to the others. An essay forces a single ordering that suppresses three of the four.

Ripple updates

Karpathy's signature observation: ingesting one source can update 10-15 wiki pages. This is the magic.

If Jacob does another conversation tomorrow about, say, his thoughts on contemplative-education curricula, ingesting the new transcript should:

• Create a new theme page (e.g. themes/contemplative-education-curriculum.md)
• Update Iyengar Yoga, Qigong (Arms-Up Position), Stages of Adult Development with new sections
• Update Vision for the World to reference the new theme
• Potentially create new entity pages (any new teachers/authors he names)

The append-only log.md is what makes the ripple traceable.

The LLM is the programmer; WikiHub is the IDE

Karpathy's slogan: "The LLM is the programmer; Obsidian is the IDE; the wiki is the codebase."

For this wiki: WikiHub is the IDE. Specifically:

• WikiHub renders [[wikilinks]] natively
• Frontmatter is parsed and indexed
• Pages live in a real git repo (the wiki is also a clone-able repo)
• Search is fuzzy and full-text
• The whole site is API-addressable, so future ingestions can be done by an LLM script via HTTP

This is the purpose of WikiHub. From the WikiHub landing page: "GitHub for LLM wikis."

What this wiki is not

• Not a transcript viewer. The transcript is in raw/; everything in wiki/ is summary, synthesis, and cross-reference.
• Not the final say. Many pages are flagged confidence: speculative or low. The LLM (initially Claude Opus 4.7) made judgment calls on attribution, framing, and emphasis. Jacob is welcome to overwrite.
• Not exhaustive. GPS noise was stripped. Side conversations (about Harrison's pause app, Dana coming home, the bikes in the garage) are mentioned only where they color the main content.

Future ingestions

Anything Jacob wants to add — a podcast, an essay, a Slack export, another otter recording — should:

1. Land in raw/ as-is (or with a cleaned companion).
2. Get an entry in log.md.
3. Trigger ripple updates per the AGENTS.md schema.

The wiki grows. The vision crystallizes.