Notes
A compiled wiki on AI tooling and method, built from primary sources. Articles are longer synthesis pieces; concepts are short definitions used as cross-references. About describes the process.
Articles
- 2026-05-20 The Inspectable Archive — The Wiki's Three-Layer Compilation Method This wiki has a compilation method. Article
- 2026-05-14 The Spectacle and the Digital Ouroboros — AI, Knowledge Production, and the Recoverable Tension Marshall McLuhan in 1966 and Guy Debord in 1967 each diagnosed the shape of the present before its substrate existed — McLuhan through media form, Debord through commodity-form. Article
- 2026-05-13 Marshall McLuhan in 1966 — Anticipating the Information Economy In a 1966 Canadian television interview, Marshall McLuhan described — without using any of the words for them — personalised information retrieval, products-as-services, attention-as-product, and an advertising-saturated culture. Article
- 2026-05-13 Ned Block on ChatGPT and the Watch Face A small failure case — ChatGPT cannot reliably read or generate the time on an analogue clock — opens onto Ned Block's larger argument that current LLMs operate without the perceptual layer that grounds human spatial reasoning. Article
- 2026-05-08 Neurosymbolic AI — Sound Reasoning, Knowledge Reuse, and the Third Wave Neurosymbolic AI integrates neural networks with symbolic logic to address the reliability problems that pure LLMs cannot fix by scale alone. Article
- 2026-05-06 The 2025 Cognitive Bellwether — Convergence, Cost Collapse, and the Meta-Skill Shift Between summer 2025 and spring 2026, three layers of evidence — hard capability events, insider acknowledgement, and lay-observer dissonance — converged on a single conclusion: the relationship between human and cognitive work had structurally changed. Article
- 2026-05-04 Obsidian as a Second Brain — Capture, Compile, Publish Most personal-knowledge systems collapse under their own weight because they treat capture and compilation as the same activity. Article
- 2026-05-02 Karpathy in 2026 — The Capability Gap, Cognitive Cores, and the Harness Mindset Across April 2026, Andrej Karpathy made three closely-related arguments that together describe how to think about LLMs this year: the discourse is split between people on free or year-old models and… Article
- 2026-04-29 Prompt Anatomy for Claude — Structure and Negative Prompting Anthropic's recommended prompt structure for Claude is closer to a short technical brief than a sentence. Article
- 2026-04-29 Retrieval-Augmented Generation — Architecture, Limits, and Practice Retrieval-Augmented Generation is not a feature you bolt onto an LLM. Article
- 2026-04-29 Skill Composition — Nesting, Progressive Disclosure, and Specialization Standalone skills are fancy prompts. Article
- 2026-04-29 The LLM Knowledge Base — Karpathy's Wiki Compilation Pattern Andrej Karpathy's "LLM Knowledge Bases" workflow turns an LLM from a chat partner into a wiki compiler. Article
- 2026-04-06 Context Window Management - Designing Around the Constraint Every design decision in Claude Code — concise CLAUDE.md files, on-demand skills, sub-agent isolation, LSP integration — traces back to one physical constraint: the context window is finite.… Article
- 2026-04-04 Claude Code Architecture - The 4-Layer System Claude Code's real power emerges when treated as a 4-layer architecture — memory, skills, hooks, and agents — rather than a single prompt interface. Article
- 2026-04-04 Claude Code for Research and Knowledge Work Claude Code isn't just a programming tool — its architecture of persistent memory, skills, and sub-agents maps directly onto research workflows like literature review, data analysis, experiment design, and knowledge synthesis. Article
- 2026-04-04 CLAUDE.md — The Configuration File Pattern CLAUDE.md is a project-level configuration file that gives Claude Code persistent context about a repository's purpose, structure, rules, and workflows — turning it from a generic assistant into a project-aware collaborator. Article
Concepts
- 2026-05-01 Agent Harness The deterministic code that wraps a language model and turns it into an autonomous agent — handling retries, tool dispatch, memory, and context shaping so that the model itself can focus narrowly on next-step reasoning. Concept
- 2026-05-08 Auto-regressive Network A network that generates output one element at a time, feeding each prediction back as input to predict the next. Concept
- 2026-05-08 AutoFormalization The unsolved subproblem at the heart of LLM-to-symbolic neurosymbolic AI. Concept
- 2026-04-04 Autoresearch An open-source framework created by Andrej Karpathy for automating the scientific experiment pipeline, from project planning through execution to reporting. Concept
- 2026-04-04 Claude Code Anthropic's terminal-based AI coding agent that runs directly in the command line, reads and writes files, and can spawn sub-agents for parallel task execution. Concept
- 2026-04-17 Claude Routines Anthropic's natural-language workflow primitive: a triggered job that runs a Claude prompt against connected tools (Gmail, Slack, CRMs) on a schedule, API call, or webhook — replacing visual no-code chains like n8n with a numbered SOP written in plain English. Concept
- 2026-04-04 Context Window The fixed-size token buffer that an LLM uses for each interaction, constraining how much information can be included in any single conversation or task. Concept
- 2026-05-23 Dataism The view that the universe is best understood as data flows and that the value of any entity — human or otherwise — is determined by its contribution to data processing. Concept
- 2026-05-08 Fidelity (Neurosymbolic AI) How well an extracted symbolic description approximates the trained network it came from. Concept
- 2026-04-04 Hooks (Claude Code) Deterministic automated guardrails that fire at specific trigger points to enforce safety patterns without requiring explicit instruction each time. Concept
- 2026-04-29 Hybrid Retrieval A retrieval strategy that combines dense semantic vector search with sparse keyword search (and often a reranking pass), then merges the rankings — designed to recover queries that either method alone would miss. Concept
- 2026-05-08 Knowledge Extraction Pulling structured symbolic descriptions out of trained neural networks. Concept
- 2026-04-29 LLM Knowledge Base A markdown wiki written and maintained by an LLM rather than a human, compiled incrementally from raw source material so that an agent can later query its own structured notes instead of stuffing everything into context. Concept
- 2026-04-19 Model Indirection Layer An architectural layer between application code and the model API that decouples business logic from any single provider, prompt format, or model version — turning model swaps and prompt changes into configuration rather than code rewrites. Concept
- 2026-04-29 Negative Prompting The technique of telling an LLM what not to do, in addition to what it should do — used to override the model's default tendency to produce safe, generic, hedged output. Concept
- 2026-04-29 Nested Skills A skill that acts as a router rather than a single instruction set, dispatching to one of several sub-skills based on the task at hand — the architectural pattern that allows skill libraries to scale past a few dozen entries without becoming unmanageable. Concept
- 2026-05-08 Neurosymbolic Cycle The defining process of neurosymbolic AI. Concept
- 2026-04-29 Progressive Disclosure A skill design pattern in which the skill itself loads as a short instruction file, and supporting reference material is fetched on demand only when the current task requires it — minimising token cost while preserving full expert depth when needed. Concept
- 2026-04-29 Retrieval-Augmented Generation A pattern in which an LLM is given relevant external documents at query time, retrieved from a corpus, so that the model can ground its answer in knowledge it was not trained on. Concept
- 2026-04-29 Semantic Collapse An informal name, popularised on social media, for the failure of single-vector embedding-based retrieval at scale — formally established by Weller et al. Concept
- 2026-04-04 Skills (Claude Code) Reusable packaged instruction sets stored in .claude/skills/ that encode expert workflows and are invoked on demand rather than loaded permanently. Concept
- 2026-04-29 Sleep-time Compute Compute that an LLM agent spends between user turns — pre-computing likely answers, reorganising memory, and updating indexes — so that test-time queries are faster, cheaper, and more accurate. Concept
- 2026-04-04 Sub-agents Child processes spawned by a main Claude Code agent to execute tasks in parallel while preserving the parent's context window. Concept