Gen AI Lingo
Definition: The basic unit of data processed by a language model. A token is typically about 4 characters or roughly 0.75 words in English.
Everyday analogy: Think of tokens like syllables or puzzle pieces. Models don’t read whole words or ideas—they calculate numbers assigned to these text chunks.
Workplace tip: Understanding tokens explains why models occasionally miscount letters in a word or stumble on simple math.
Explore interactive practice: Chapter C25: From Text to Small Pieces →
Definition: The maximum number of tokens an AI model can consider at one time, including your prompt, uploaded files, previous messages, and its own response.
Everyday analogy: Imagine a desk with limited surface area. Once the desk is full of papers, putting a new sheet on top pushes the oldest sheet off.
Workplace tip: When a long conversation seems to forget early instructions, it has exceeded its context window.
Explore interactive practice: Chapter C06: What the App Can Actually See →
Definition: A configuration setting that controls how much randomness is introduced when the model chooses each subsequent token.
Everyday analogy: Think of temperature like a risk dial. Low temperature (e.g. 0.2) picks the most probable, conservative words; high temperature (e.g. 0.8) allows less likely choices.
Workplace tip: Use low temperature for data extraction, summarization, and coding; use higher temperature for creative brainstorming.
Explore interactive practice: Chapter C05: How an Answer is Made →
Definition: The practice of designing, structuring, and refining inputs to guide an AI model toward the most accurate, useful, and formatted output.
Everyday analogy: Writing a prompt is like briefing a capable junior colleague: clear context, explicit constraints, and examples yield far better results than vague instructions.
Workplace tip: Always include the desired role, task, background context, format constraints, and a sample of good output.
Explore interactive practice: Chapter C07: Give a Useful Brief →
Definition: When a generative AI model generates false, fabricated, or unsupported information while presenting it with complete confidence and fluency.
Everyday analogy: A dream state where the model predicts words that sound grammatically and stylistically plausible, but have no basis in real-world facts.
Workplace tip: Never rely on an AI model for citations, historical dates, or complex arithmetic without independently verifying against primary sources.
Explore interactive practice: Chapter C10: Check an Answer →
Definition: Providing one or more worked examples of inputs and desired outputs directly inside your prompt before asking the model to perform the task.
Everyday analogy: Showing someone three completed expense reports before asking them to fill out their own.
Workplace tip: Adding just 2–3 clear examples inside your prompt is the single most effective way to lock in output formatting.
Explore interactive practice: Chapter C04: Learning from Examples →
Definition: Background instructions set at the beginning of an AI session that define the assistant’s persona, behavioral boundaries, and default output rules.
Everyday analogy: The employee handbook or job description given to a worker before they start taking customer requests.
Workplace tip: Use system prompts to enforce constraints like “Always respond in JSON” or “Never guess if the answer is not in the document.”
Explore interactive practice: Chapter C34: From a Base Model to a Useful Assistant →
Definition: A technique that retrieves relevant passages from external documents or databases and inserts them into the prompt before generating an answer.
Everyday analogy: An open-book exam: instead of relying on memory, the student searches their textbook for the exact chapter and answers based on that text.
Workplace tip: RAG dramatically reduces hallucinations by grounding the AI’s response in your specific proprietary documents.
Explore interactive practice: Chapter C37: Give a Model Useful Outside Information →
Definition: An AI system equipped with tools (search, code execution, APIs) and the autonomy to plan, execute multi-step workflows, and evaluate its own progress.
Everyday analogy: A project assistant who doesn’t just write an email draft, but also checks your calendar, looks up recipient addresses, and schedules the send.
Workplace tip: Always maintain human-in-the-loop approval gates for consequential actions like payments or sending external emails.
Explore interactive practice: Chapter C16: Tools and Agents →
Definition: An open standard developed by Anthropic that allows AI models and assistants to securely connect to external tools, databases, and local file systems.
Everyday analogy: USB-C for AI: a single universal connector that allows any AI model to interact with any compatible external data source or tool.
Workplace tip: MCP enables tools like Claude Code and desktop assistants to safely read repositories and execute terminal commands.
Explore interactive practice: Chapter C55: MCP Connects an Application to Capabilities →