The vocabulary: 15 terms you'll see everywhere
A one-line dictionary of the words the whole field assumes you already know.
In plain terms
Every AI article throws around 'inference', 'parameters', 'hallucination' as if you were born knowing them. Learn these 15 once and everything else in this roadmap reads twice as fast.
Why it matters
Jargon is the #1 reason beginners think AI is harder than it is. Most terms are simple ideas wearing intimidating names.
How it works
Model - the trained 'brain' (a giant file of numbers). Parameters/weights - those numbers; more ~= smarter but slower. Training - adjusting weights by showing examples (done by labs, not you). Inference - running the model to get an answer (what you do all day). Prompt - the text you send in. Completion/response - what comes back. Token - a word-chunk, the model's unit of text. Context window - how much text the model can 'hold in mind' at once. Hallucination - confident wrong answer. Fine-tuning - extra training on your own examples. Embedding - text turned into a list of numbers capturing meaning. RAG - letting a model look things up before answering. Agent - a model in a loop that uses tools. Latency - time until the answer. Throughput - answers per second.
When you use it
Read now, revisit when a term surprises you. Each term gets its own full node later in the map.
Common mistakes
- Confusing training with inference - as an AI engineer you almost never train models, you use trained ones.
- Thinking hallucination is a bug that can be fully 'fixed' - it's inherent; you engineer around it.
- Assuming 'agent' means something magical - it's a loop with tool calls.
Best practices
- Keep a personal glossary file; add every new term with your own one-liner.
- When reading anything, refuse to skim past a word you can't define.
Try it yourself
Close this panel and explain 5 of these terms out loud to a rubber duck (or a friend). The ones you stumble on are your reading list.
Resources
- Jay Alammar - The Illustrated Transformer (skim visuals) Bookmark now, read properly in Phase 01.