neural / embeddings
Every word is a coordinate.
Meaning is geometry.
We're so early on understanding what the latent space actually encodes. The emergent structure of high-dimensional token embeddings is the most underexplored surface in AI-native development. This is a live map of the neighborhood.
latent space — 2d projection
hover to identify · click to pin · clusters = semantic neighborhoods · edges = cosine proximity
dimensions
1,536
text-embedding-3-small. Each token is a point in 1,536-dimensional space. You can't visualize it. That's fine.
cosine similarity
0–1
The only distance metric that matters in high-dimensional space. Euclidean distance lies. Cosine tells the truth about angle.
nearest neighbors
k=5
RAG is just kNN with extra steps. The retrieval quality ceiling is set by your embedding model, not your prompt.
we're so early
100%
Embedding-based memory, semantic routing, latent-space reasoning — this is the agentic stack nobody has fully shipped yet.
embedding laws
king − man + woman ≈ queen. This arithmetic works. It's been working since 2013. We're still figuring out what it means.
Similar concepts cluster. Analogies form parallelograms. The geometry of meaning is real and measurable.
The model doesn't 'know' anything. It knows distances. Everything else is projection — yours, mine, the attention head's.
Fine-tuning moves points. RLHF reshapes the manifold. Every training step is a coordinate transformation on meaning itself.
the agentic implication
If you're routing agent actions through keyword matching, you're leaving 90% of the signal on the floor. Embedding-based routing is AI-native. The agent reads the semantic neighborhood, picks the nearest action cluster, executes. No regex. No switch statements. Pure emergent dispatch from a coordinate lookup. This is the architecture. We're so early.
→ context window laws