Lesson 1 — How Intelligence Represents Information

The Geometry of Meaning

Before an intelligence can reason, it must represent. Data is mapped into a vector space where distance encodes difference and direction encodes meaning. This is the core idea: embeddings turn symbols into locations.

1 · From Symbols to Vectors

Tokens, pixels, or audio frames become coordinates (vectors). Co-occurrence pulls points together; rarity pushes them apart. Clusters behave like concepts.

• ML: tokenization, embedding, cosine similarity
• Philosophy: understanding as location
• Quantum: measurement collapses potential meaning to a point
• Cognitive: perception organizes the field

2 · Learning as Compression

Training adjusts vectors to reduce loss — compressing regularities, separating confusions.

3 · Attention as Perspective

Context reshapes representation via attention: relevance as weight, awareness as computation.

4 · Emergent Concepts

Dense regions = ideas; directions = relations ("king − man + woman ≈ queen" as a classic analogy).

Hands-On: Give Meaning Coordinates

Goal: embed a small text and visualize semantic neighborhoods using the Foundry API proxy.

Step 1 — Prepare a sample

Create a file notes.txt (a few short paragraphs you wrote).

Step 2 — Embed it

curl -X POST /api/foundry/embed \
  -H "Content-Type: application/json" \
  -d '{ "doc_id": "lesson1-notes", "content": "PASTE a few sentences here", "metadata": {"source":"lesson-1"} }'

Step 3 — Semantic search

curl -X POST /api/foundry/search \
  -H "Content-Type: application/json" \
  -d '{ "query": "core idea of my notes", "k": 5 }'

Step 4 — Visualize (optional UI component)

Reflection — The Anchors of Meaning

If meaning has coordinates, what anchors yours? Write 5–8 sentences mapping a small "personal embedding space": three ideas you often connect, and one you rarely link but perhaps should.