Tagged: knowledge-graph

Video: I implemented Karpathy's LLM wiki method on my digital garden

I rebuilt Karpathy's LLM wiki on top of my digital garden content, with interesting results.

Symbolic AI for the win

Thematic-TAO: a three-pass method for post analysis

Extracting meaningful entities and relations, combining thematic analysis with TAO topic mapping in three passes.

Experiment 2: thematic analysis with 1M context

Testing the 1M context window by loading the entire garden and running thematic analysis, comparing fresh analysis against graph-augmented analysis.

Thematic analysis of the garden: graph-augmented

What happens when you start from embeddings instead of reading? Comparing cluster-based discovery against the fresh thematic analysis.

External content discovery

Using the garden's own knowledge graph to find relevant external content, papers, and articles.

Stable map positions: giving the garden roots

How the explore map got stable positions and pinned territories, so the landscape doesn't shift on every rebuild.

Typed relations as garden infrastructure

How hub/develops relations replace tag-based project structure with an explicit ontology, and what that means for the three kinds of connections in the garden.

How content chunk size influences embedding representations

Why embedding models cluster documents by format instead of topic when content lengths vary, and what to do about it.

Explore page: a visual browsing UI for the garden

Design document for an /explore page that lets visitors walk through the garden visually, using semantic maps, serendipity, and trails.

Full-scale key phrase extraction: results

Running LLM-based key phrase extraction on all 92 garden items. What the results reveal about phrase overlap, short-text limits, and the role of key phrases in the knowledge graph.

Key phrase extraction for the garden

What key phrase extraction is, how it evolved from word counting to LLM prompting, and which approach fits a small multilingual knowledge garden.

Tuning the similarity threshold: from 2,771 candidates to useful suggestions

How to go from "everything is related to everything" to a useful set of link candidates. Data analysis, filtering strategies, and the research behind them.

Key phrase extraction prototype

Extract key phrases from 10 garden items using Ollama, compare LLM-based extraction with YAKE.

Embedding models: a survey for the garden

A non-coder's guide to text embedding models, from word counting to semantic understanding, surveyed for use in a personal knowledge garden.

Knowledge graph for the garden

Research into making the garden smarter about its own connections, from automatic backlinks to a full world model.

Reading notes: Apple's Saga knowledge graph

Reading notes from Apple's Saga papers on building knowledge graphs at scale, and what transfers to a personal digital garden.

Embeddings for knowledge gardens: a research gap

Academic research on embeddings hasn't caught up with how PKM tools actually use them. There might be something worth writing here.

Saga: a platform for continuous construction and serving of knowledge at scale

Apple's hybrid batch-incremental knowledge graph platform. Key inspiration for building smarter connections in this garden.

Setting up local embedding models

How to install and run embedding models locally using Ollama, compare three models on garden content, and run a full-scale embedding pass.

European chatbot conference: 10 talks that I'm going to rewatch

My personal highlights from the European Chatbot and Conversational AI summit.

7 new skills to learn for conversation designers in 2022

Looking to fill up your convo design toolbox? Here are 7 skills to take your conversational AI career to the next level.