Welcome — How to Use This Primer
This primer is a non-mathematical, practical guide to large language models for technical teams: product managers, engineers, and technical leaders. It covers the full stack from tokens to agents — what these systems are, how they're built, where they're brittle, and how to make good decisions about them.
Who it's for. You don't need a machine learning background. You do need enough technical fluency to care about the difference between a base model and an instruction-tuned one, or why context window size affects your product's cost structure. If that sentence made sense, you're in the right place.
What it is not. This is not a prompt engineering tutorial, a comparison of vendor APIs, or a guide to building ML models from scratch. It's the conceptual foundation that makes all of those things easier.
Three Reading Paths
Every page has an estimated_minutes value in its frontmatter. Pages also carry a pm_read label: full (read the whole page), skim (read intro and summary), or reference (consult as needed). Use these to navigate your path.
PM / Product Path (3–4 hours)
All five Basics pages, plus a curated selection of Intermediate and Advanced chapters. Covers everything you need to make product decisions, read benchmarks, estimate costs, and talk fluently with your engineering team. See the Reading Order page for the full table.
Engineer Path (6–8 hours)
All Basics, all Intermediate, and your choice of Advanced topics. After this path you'll understand training pipelines, scaling trade-offs, RAG architectures, and agentic systems well enough to design and evaluate them. See the Reading Order page for the full table.
Researcher Path (12–15 hours)
Full primer plus the paper reading guides linked throughout. Every page cites primary sources with explicit PM reading instructions (what to read, what to skip). The Papers reference collects all citations in one place.
How to Navigate
Sidebar. Categories appear in reading order: Basics → Intermediate → Advanced → Use Cases → Resources. Within each category, pages are sequenced to build on each other.
Reading time estimates. Every page shows an estimated read time. These are calibrated for careful reading, not skimming. If you're comfortable with a concept, move faster.
PM badges. Pages marked pm_read: full are on the PM path. Pages marked pm_read: skim are optional depth. Pages marked pm_read: reference are reference material — consult them when you need them, not front to back.
Choose your own adventure. After completing your base path, Advanced topics are independent. Pick based on what's relevant to your work.
Reading Path Diagram
A Note on Currency
This field moves fast. Model capabilities, pricing, and best practices shift on a monthly cadence. The conceptual foundations in Basics and Intermediate are stable — they describe how these systems work, not which vendor is currently ahead. Use Cases pages are marked with last-updated dates and reviewed monthly.
When specific numbers or rankings matter (benchmark scores, context window sizes, pricing), verify against current vendor documentation before using them in decisions.