Skip to main content

The Executive's AI Glossary

Introduction​

The AI industry is plagued by engineering jargon that obscures business value. This glossary defines key concepts not by how they work technically, but by what they do for your business. Understanding this vocabulary is the first step to Adaptability.

Core Concepts​

LLM (Large Language Model)​

  • The Definition: A prediction engine that generates text based on patterns.
  • Business Metaphor: Think of an LLM as a very well-read improvisational actor, not a calculator or a database. It is incredibly creative and articulate, but it doesn't "know" facts—it predicts what sounds like a fact.
  • Why it matters: You use LLMs for drafting, summarizing, and reasoning, but never as a rigorous source of truth without guardrails.

Hallucination​

  • The Definition: When an AI confidently presents false information because it looks plausible.
  • Business Metaphor: A smooth-talking intern who makes up an answer during a meeting to avoid saying "I don't know."
  • Why it matters: This is the primary risk in AI deployment. It requires strict validation layers (see our Hallucination Prevention Protocol).

Agent​

  • The Definition: An AI that can do things, not just say things.
  • Business Metaphor: A digital employee with hands.
  • Why it matters: A standard ChatGPT session is a chatbot. An Agent has access to tools—it can send emails, query your database, update your CRM, and book meetings.

Technology Stack​

RAG (Retrieval-Augmented Generation)​

  • The Definition: "Open-book testing" for AI. Instead of relying on its internal training memory, we force the AI to look up answers in your specific company documents before answering.
  • Business Value: This is how we make generic AI models (like GPT-4) experts in your specific business data without the massive cost of retraining a model.

Token​

  • The Definition: The fundamental unit of text for AI processing.
  • Rule of Thumb: 1,000 tokens is roughly 750 words.
  • Business Value: This is the currency of AI. Your costs are calculated per token. Inputting massive documents costs more; concise prompts save money.

Vector Database​

  • The Definition: The "Long-Term Memory" for your AI.
  • Business Value: Traditional databases search for exact keyword matches. Vector databases search for concepts (e.g., searching for "Revenue" will also find documents mentioning "Sales" and "Income"). This powers the search capability of RAG systems.

Prompting Strategy​

Zero-Shot vs. Few-Shot​

  • Zero-Shot: Asking the AI to do a task without giving it any examples. ("Write a marketing email.")
  • Few-Shot: Asking the AI to do a task while providing 3-5 successful examples of past work. ("Here are three of our best performing emails. Write a new one in this style.")
  • Business Value: Shifting from Zero-Shot to Few-Shot is the single highest-ROI action a user can take to improve AI reliability. Always provide examples.

Context Window​

  • The Definition: The amount of information an AI can hold in its "short-term memory" at one time.
  • Business Value: If your conversation or document exceeds the context window (e.g., 128k tokens), the AI will "forget" the beginning of the text. Managing this window is critical for processing large legal contracts or technical manuals.

Strategic Note​

Do not be intimidated by the acronyms. The tech stack is commoditizing rapidly. The lasting competitive advantage lies not in the models you use, but in the clarity of your data and the quality of your workflows.