LLM Ontology¶

A personal knowledge base mapping artificial intelligence end-to-end as a multi-level ontology. The files stay flat (01-…26-); this page is the reading lens: theory → build → operate → ecosystem. The four groups below are navigation, not a strict taxonomy — a few branches deliberately straddle several.

I. Concepts — what AI is¶

#	Branch	Covers
01	Artificial Intelligence	The root field; paradigms (symbolic vs statistical) and capability classes
02	Machine Learning	Learning from data, partitioned by supervision signal
03	Deep Learning	Neural-network mechanics and architecture families
04	Math & Data Representation	The numeric substrate: tensors, embeddings, similarity
05	Foundation Models	Broadly pretrained, adaptable models by modality and role
06	Language Models	Text-specialized foundation models; lineage and post-training role
07	Transformer Architecture	The forward path of a token through attention
08	Model Internals	Static anatomy + the runtime token pipeline

II. Building — how models and AI systems are made¶

#	Branch	Covers
09	Data & Datasets	The raw material: lifecycle, dataset types, data governance
10	Training & Post-Training	Acquiring capability, then shaping behavior
11	Evaluation & Testing	Measuring capability and failure modes
12	Inference	A single model call — knobs and I/O
13	Reasoning & Test-Time Compute	Buying quality with compute at run time
14	RAG	Grounding generation in retrieved documents
15	Knowledge & Memory	What the system knows and remembers, by lifespan
16	Agents	Goal-directed iterative action with tools
17	Tools, Skills, Commands & Protocols	Capabilities and the wiring that connects model to world
18	AI Engineering	Building reliable systems around a fixed model
19	Model Lifecycle	Idea-to-retirement timeline

III. Operating — keeping systems usable, safe and economical¶

#	Branch	Covers
20	Safety, Governance & Alignment	Defensive controls, governance principles, alignment
21	Security & Threat Model	The attacker's-eye view; agent-specific risks
22	Cost & Economics	Token/training economics, cost levers, unit economics
23	Infrastructure & Runtime	Compute and serving machinery
24	Interfaces	How humans and systems reach the model

IV. World — ecosystem, vendors and historical lineage¶

#	Branch	Covers
25	Vendor & Model Ecosystem	The organizations and product families (incl. OpenClaw)
26	People & Research Lineage	The humans and intellectual history

Full tree¶

AI / ML / LLM Ontology
│
├── I. CONCEPTS
│   ├── 1. Artificial Intelligence — symbolic vs statistical; generative/predictive/multimodal/agentic
│   ├── 2. Machine Learning — supervised · unsupervised · self/semi-supervised · RL · Bayesian
│   ├── 3. Deep Learning — NN mechanics · CNN · RNN · Transformer · diffusion · autoencoders · GANs
│   ├── 4. Math & Data Representation — scalar→tensor · embedding · latent space · similarity metrics
│   ├── 5. Foundation Models — language · vision · audio · video · multimodal · embedding · reranking · frontier
│   ├── 6. Language Models — statistical→neural→SLM→LLM · base/instruct/chat/code/reasoning
│   ├── 7. Transformer Architecture — embeddings · attention · FFN · residual · norm · encoder/decoder
│   └── 8. Model Internals — weights · layers · heads · context window · tokenizer · logits · sampling
│
├── II. BUILDING
│   ├── 9. Data & Datasets — lifecycle (source→clean→label→curate) · dataset types · data governance
│   ├── 10. Training & Post-Training — pretraining · SFT · instruction tuning · RLHF/RLAIF/DPO · distillation
│   ├── 11. Evaluation & Testing — benchmarks · human/automated eval · red teaming · eval-driven dev
│   ├── 12. Inference — prompts · tokens · temperature/top_p · structured output · streaming
│   ├── 13. Reasoning & Test-Time Compute — CoT · thinking budget · planning · reflection · verification
│   ├── 14. RAG — parse→chunk→embed→vector DB · semantic/hybrid search · rerank · grounded generation
│   ├── 15. Knowledge & Memory — short-term → long-term/user/project → knowledge graph → ontology
│   ├── 16. Agents — goal→plan→action→observation loop · tool use · multi-agent · (Claude Code, OpenClaw…)
│   ├── 17. Tools, Skills, Commands & Protocols — tools · skills · function calling · MCP · OpenAPI · SDK
│   ├── 18. AI Engineering — prompt/context/RAG engineering · spec/eval-driven dev · observability
│   └── 19. Model Lifecycle — research → train → eval → deploy → monitor → deprecate → retire
│
├── III. OPERATING
│   ├── 20. Safety, Governance & Alignment — guardrails · sandboxing · policy · privacy · responsible AI · alignment
│   ├── 21. Security & Threat Model — prompt/tool injection · data leakage · agent risks · personal-agent runtime risk
│   ├── 22. Cost & Economics — token/training economics · batching · routing · unit economics · build-vs-buy
│   ├── 23. Infrastructure & Runtime — local/cloud/edge · GPU/TPU · quantization · serving · vector DB · orchestration
│   └── 24. Interfaces — chat UI · API · CLI · IDE · browser · mobile · voice · background worker
│
└── IV. WORLD
    ├── 25. Vendor & Model Ecosystem — OpenAI · Anthropic · Google · Meta · Mistral · DeepSeek · xAI …
    │       └── local/open — LM Studio · Ollama · llama.cpp · vLLM · Hugging Face · OpenClaw
    └── 26. People & Research Lineage — symbolic · ML · DL · Transformer authors · LLM · alignment · founders

Cross-cutting notes¶

Knowledge & Memory (15), Agents (16), and Model Lifecycle (19) intentionally straddle several groups — they are operational lenses across concepts, building, and production, not strict taxonomic categories.

OpenClaw is tracked deliberately as its own class — a personal agent runtime: not a model, vendor API, or plain local inference tool, but a layer connecting an LLM, tools, memory, chat channels, cron/background tasks, and real user accounts. It appears under Vendor & Model Ecosystem → local/open, as an example in Agents, and as a risk class in Security & Threat Model.

How to read this KB¶

The tree is the index; cross-links are the structure. Follow the "Related" links at the bottom of each branch.
Branches age differently. Concepts (1–8) and most of Building are stable; Vendor (25) churns monthly; People (26) grows by accretion. See CLAUDE.md.