The big picture

Priya once wondered why her assistant helped one day and made up a rule the next. Now she knows. Your goal: see how the ideas fit together before the quiz.

What Priya learned

• What an AI model is: numbers set once in training, used to guess the next token.
• Tokens and context windows: tokens fill one window and set the bill.
• Inside one answer: one pass per token written. Saved notes make caching work.
• Why answers change: temperature: it changes variety, not truth.
• When AI makes things up: give it sources. Ask without showing the answer you want.
• Prompts that work: give the goal, context, format and examples.
• Choosing a model: match size, effort and speed to the job.
• Models that pick from a list: fast, safe choices.
• Open and closed models: open weights are not open source. Read the license.
• Search by meaning: embeddings and RAG: find text by meaning, like an open-book exam.
• Prompt, RAG or fine-tune?: prompt first, RAG for changing facts, fine-tuning for habits.
• Evals: does your AI work?: test real cases after each change.

Where to go next

• Working with AI puts this to work with APIs, tools and agents.
• Prompt injection shows why text the AI reads can carry hidden orders.

Ready? The cheat sheet & pop quiz is next.