Overview

Wiki Overview

High-level synthesis of captured knowledge.

LLM Wiki Pattern

The Kaparthy LLM Wiki

Andrej Karpathy’s idea: LLM maintains Persistent Wiki from raw sources, vs RAG.

Aspect	RAG ${[}^1]$	LLM Wiki
Retrieval	Query-time${[}^1]$	Pre-compiled
Accumulation	None	Compounding
Maintenance	Manual/human	LLM-automated
Scale	Rediscovery cost grows	Touches 10-15 pages/ingest

In Obsidian

• Copilot as LLM agent.
• Graph view, clipper, plugins.

Karpathy’s Paradigms

Andrej Karpathy - AI Ascent 2026

Paradigm	Description
1.0	Explicit rules/code
2.0	Learned weights (data/train)
3.0	Prompting (LLM interpreter)

(Grows with ingests/queries.)

10 Core AI Concepts (2026)

10-AI-Concepts

1. LLMs
2. Tokens & Context Window
3. AI Agents
4. MCP
5. RAG
6. Fine-tuning
7. Context Engineering
8. Reasoning Models
9. Multimodal AI
10. Mixture of Experts

Practical Applications

Ways-to-Use-AI-in-Daily-Life

• 101 Use Cases: Grouped by productivity, work, daily life, finances, learning, career, relationships.

• Key Insight: Foundational tools (ChatGPT/Claude/Perplexity) + prompt engineering > specialized subs.

• Progression: Simple (image ID) → advanced (Vibe Coding, agents: n8n, workflows).

• AI Industry 2026 AI Industry 2026: Global spending >$2.02T (+36%); infra software +83%.¹

Segment	2026 ($B)	Growth
AI Infra SW	230	83%
App SW	270	57%
Platforms	29.3	14.2% CAGR to 2035

Key Trends:

• Inference Economics (deployment > training; +83% infra)
• Great Divergence (physical industries lead)
• Agentic Commerce (40% B2B agents EOY)
• Multi-Agent Systems, Physical AI, Multimodal Reasoning
• EU AI Act (Aug 2026 full enforcement)
• Anthropic Economic Index (uneven geo adoption, AUI)
• Self-Healing Workflows (anti-fragile enterprise agents)
• Multimodal Reasoning (sensory synthesis for real-world)
• Three-Tier Hybrid Models (cloud/edge/on-prem infra)
• EU AI Act Enterprise Compliance: 7% fines → rebuild stacks.

Knowledge Graphs & AI Foundations

Stanford CS520 2020

Theme	KG Role
Current AI	Symbiotic: Powers apps (search/recommenders); built by NLP/ML (extraction/linking/inference).
Graph Data Science	Graphs + feature eng/ML for relational predictions (finance etc.).
Long-term Vision	Explicit reps (semantic nets → description logics) for reasoning, explainability, commonsense (vs black-box LMs).

Key: KGs essential; WordNet example explainability.

Footnotes

1. The Artificial Intelligence Industry 2026 ↩