⌁LLM FIELD NOTESan open learning map Open the workbench ↗

PUBLIC SAMPLE / INTERNAL REPRESENTATION

A document,
made inspectable.

This is the small graph behind the “try a sample” walkthrough: concepts and relations are hypotheses, evidence keeps them honest, and a fingerprint makes the projection verifiable.

Try it in the wiki ↗Download graph JSON ↓Open in Obsidian ↓Read the graph note →
3concepts
2relations
1source document
fnv64-4d8c362569fbcce7-2627graph fingerprint
SOURCE DOCUMENT

Attention sample notes

Attention mixes information across tokens. Queries, keys, and values create a weighted lookup over context. Positional encoding gives the sequence an order.

3 concepts connected by 2 evidence-backed relations: Attention uses Weighted lookup; Attention complements Positional encoding. uses complements Attention 92% confidence Weightedlookup 88% confidence Positionalencoding 86% confidence
Each node is a concept; each line is a claimed relation grounded in source evidence.
concept · 92% confidence

Attention

Attention mixes information across tokens.

3 mentions · inferred · 1 source
concept · 88% confidence

Weighted lookup

Queries, keys, and values create a weighted lookup over context.

1 mention · inferred · 1 source
concept · 86% confidence

Positional encoding

Positional encoding gives the sequence an order.

1 mention · inferred · 1 source

Now change the representation.

Open the workbench, accept or reject a candidate, download the Obsidian vault, open its native Graph.canvas view, and bring corrections back as reusable guidance.

Inspect, review, project ↗