pacific-design.com · v2 · ongoing
Artificial intelligence,
section by section.
Pacific Design is a working map of AI — how the models actually work, how to use them well, and where the field is going. It grows one section at a time until the whole territory is covered. All fifteen planned sections are live; the map now moves with the territory.
New here? Pick a path Start at 01 · Foundations Enter the Lab Glossary
the pathFifteen live sections, in reading order
The order is a path, not a pile — each section leans on the ones before it. Jump anywhere; the numbers tell you what's assumed.
- What a neural network is
- How models learn
- The kinds of learning
- Where data comes from
- Overfitting & generalization
- Models grading models
- Learning by building
- Keeping up
- Calibration & uncertainty
- Evaluating models
- Tables still rule
- Features & leakage
- Recommender systems
- Forecasting
- Imbalance & anomalies
- ML for science
- Transformers & attention
- Tokens & tokenization
- How LLMs are trained
- Fine-tuning in practice
- Models in other languages
- Context windows & memory
- Hallucination
- Reasoning models
- One model, many senses
- Beyond the dense transformer
- Scaling laws & small models
- Anatomy of a prompt
- Few-shot & chain of thought
- Structured output
- Context engineering
- Designing for uncertainty
- Optimizing prompts
- Why prompts fail
- What makes an agent
- Tool use & function calling
- MCP: a standard for tools
- Computer use
- Agent memory
- Multi-agent systems
- Coding agents
- Evaluating & containing agents
- Diffusion models
- Image generation in practice
- Video, voice & music
- 3D & world models
- Judging generated media
- Provenance & detection
- How machines see
- Detection & segmentation
- Vision-language models
- Video understanding
- OCR & document AI
- Vision in the wild
- What alignment means
- How models are aligned
- Bias & fairness
- Jailbreaks & prompt injection
- Privacy & memorization
- Evaluating safety
- Interpretability
- When it goes wrong
- The long game
- Learning from consequences
- Value functions & policy gradients
- Reward design & specification gaming
- RL for language models
- Learning from logs
- When RL wins
- Why robots are hard
- Sense, plan, act
- Learning to move
- Manipulation & the humanoid bet
- Self-driving
- Robots at work
- The hardware layer
- Training at scale
- Serving & inference
- Local & edge inference
- What it costs
- What it costs the grid
- The ML lifecycle
- Securing the stack
- LLMOps
the labDon't just read it — hold the knobs
Working instruments, live in your browser: train a real neural network, teach a tokenizer your own text, steer an attention head, find out what the temperature knob actually does. No installs, no accounts — everything runs on this page and nothing leaves it.
roadmapThe whole territory
Every major section of artificial intelligence this project committed to is now live. The map isn't finished — the field moves, and entries are revised and added inside sections as it does.
- Machine Learning Foundations
- Applied & Classical ML
- Large Language Models
- Prompt Engineering
- AI Agents
- RAG & Embeddings
- Generative Media
- Computer Vision
- Speech & Audio
- AI Safety & Alignment
- Reinforcement Learning
- Robotics & Embodied AI
- AI Infrastructure & MLOps
- Policy & Regulation
- History of AI
aboutWhat this is
A free place to learn how artificial intelligence actually works — and to keep learning as it changes. A reference, not a feed: entries are short enough to read in one sitting and precise enough to argue with, every one ends with the way it fails, and the Lab lets you run the mechanisms yourself rather than take our word for them.
No account, no paywall, no newsletter, nothing tracked — plain HTML that loads on anything. If you don't know where to begin, the reading paths will pick a route for you, and the glossary defines every term the site uses. Illustrations are AI-generated: this is a site about the technology, made partly with it.