> ## Content Index > Fetch the complete content index at: https://www.thedailyconstraint.com/llms.txt > Use this file to discover other available public pages before exploring further. # AI Terms Every Operations Manager Should Know: A Working Glossary - URL: https://www.thedailyconstraint.com/ai-terms-every-operations-manager-should-know-a-working-glossary/ - Published: 2026-08-28T13:21:26.000Z - Updated: 2026-08-28T14:26:49.000Z - Author: Andrea Bullock This is a reference, not a story. No Line 4 this time. Every article in this series has used a few of these words without stopping to define them. This is that stop. It's grouped by theme, not alphabetically. New to AI? Start at the top and read straight through. Already using these tools daily? Skip to the section you need and grab what's useful. Six sections: Talking to AI. When AI Gets It Wrong. Using AI Responsibly. How AI Is Built. Getting Better Results. Where This Is Heading. **Talking to AI** ![](https://storage.ghost.io/c/6a/5a/6a5a81c1-c40d-4db0-af99-e0cedc71c0c2/content/images/2026/08/B1.png) ● **Model:** The AI system itself. Claude, ChatGPT, and Gemini are all models. “Model” is the general word for the thing you're talking to. ● **Copilot:** A common brand name for AI built into other software, like Microsoft Copilot. It's not a different category of AI. It's a model wearing a different company's name tag. ● **Prompt:** What you type or say to ask AI for something. The quality of your prompt shapes the quality of the answer, the same way a vague question to a new hire gets a vague answer back. ● **System prompt:** Instructions set before you ever type a word. It shapes how the AI behaves for the whole conversation, quietly, in the background. ● **Context window:** How much text the AI can hold in mind at once. Go past it and older parts of the conversation start falling out the other end. [Here's a plainer breakdown](https://en.wikipedia.org/wiki/Context%5Fwindow?ref=thedailyconstraint.com) if you want more detail. ● **Token:** The small chunk of text AI reads and writes in. Roughly three-quarters of a word. This is why AI tools sometimes charge or limit by “tokens” instead of words. **When AI Gets It Wrong** ● **Hallucination:** When AI states something false with total confidence. It doesn't know it's wrong, which is exactly why you check its work the way we've covered all series long. [More detail here](https://www.cloudflare.com/learning/ai/what-are-ai-hallucinations/?ref=thedailyconstraint.com) if you want it. ● **Bias:** Skewed patterns baked in from whatever data the AI learned from. Garbage in, skewed out, same as any analysis built on bad source data. ● **Overfitting:** When a model memorizes its training examples instead of learning the underlying pattern. Great on the data it already saw. Unreliable on anything new. ● **Guardrails:** Built-in limits meant to stop AI from doing certain things. Not foolproof. Not nothing, either. ● **Grounding:** Anchoring AI's answer to a real, verifiable source instead of whatever it recalls from training. A grounded answer can point to where it came from. A hallucination can't. **Using AI Responsibly** ![](https://storage.ghost.io/c/6a/5a/6a5a81c1-c40d-4db0-af99-e0cedc71c0c2/content/images/2026/08/B2.png) ● **Human-in-the-loop:** A person reviewing and approving AI output before anyone acts on it. This is the practice this entire series has been built around, whether or not it used those exact words until now. ● **Knowledge cutoff:** The date after which a model doesn't know what happened. Ask about something more recent than that, and it's either guessing or telling you it can't help, depending on the tool. ● **Data privacy:** Whether what you type into a tool gets stored, reviewed, or used to train future models. Worth checking before pasting real production numbers or personnel information into any AI tool. ● **Prompt injection:** Hidden instructions buried in a document, email, or webpage, aimed at tricking AI into doing something the actual user never asked for. More relevant as AI tools start taking actions on your behalf instead of just answering questions. **How AI Is Built** ● **Machine learning:** The broad field of teaching computers to find patterns in data instead of following rules a person wrote out by hand. Every other term in this section lives somewhere underneath this umbrella. ● **Generative AI:** The branch of machine learning that creates new content, text, images, audio, instead of just sorting or predicting from fixed categories. This is the actual category this whole series has been about. ● **LLM (Large Language Model):** The type of generative AI behind tools like Claude and ChatGPT. Trained on massive amounts of text to predict what comes next, one piece at a time. ● **Foundation model:** A large, general-purpose model trained on broad data, meant to serve as a starting point. Fine-tuning and specialization happen on top of one of these, not from scratch. ● **Training data:** The material AI learned from before you ever touched it. If something wasn't in there, the model is guessing, not recalling. ● **Pretraining:** The first, massive round of training on general text, before any specialization happens. This is what makes a model capable of holding a conversation at all. ● **Fine-tuning:** Extra training on top of a general model, aimed at making it better at one specific job. ● **Parameters:** The internal settings adjusted during training. More parameters usually means more capacity to learn patterns. It does not automatically mean better answers. ● **Inference:** The moment AI is generating your answer, as opposed to being trained. This is the part you're waiting on. **Getting Better Results** ![](https://storage.ghost.io/c/6a/5a/6a5a81c1-c40d-4db0-af99-e0cedc71c0c2/content/images/2026/08/Glossary-3.png) ● **Prompt engineering:** The skill of writing clearer, more specific instructions so AI gives you what you need the first time. [Anthropic's own guide](https://docs.claude.com/en/docs/build-with-claude/prompt-engineering/overview?ref=thedailyconstraint.com) is a solid place to start if you want to go deeper. ● **Few-shot / zero-shot:** Giving AI examples first, versus asking cold with none. Few-shot usually helps. Zero-shot is faster when the task is simple. ● **Chain of thought:** Asking AI to show its reasoning step by step instead of jumping straight to an answer. Slower. Often more reliable, and easier to catch a mistake in. ● **Temperature:** A setting that controls how predictable or creative the output is. Low for consistency. High for brainstorming. ● **Latency:** How long you wait between asking and getting an answer. Matters more than people expect when you're troubleshooting a line that's down right now, not writing a report due next week. ● **Structured output:** Getting AI to answer in a specific format, like a table or a numbered list, instead of a loose paragraph. This is exactly what built the hypothesis matrix and risk register earlier in this series. **Where This Is Heading** ● **Agent:** AI that can take multi-step actions on its own, not just answer a question. Think “go do this” instead of “tell me about this.” ● **RAG (Retrieval-Augmented Generation):** AI that looks up real documents before answering, instead of relying only on what it learned during training. [IBM's explainer](https://www.ibm.com/think/topics/retrieval-augmented-generation?ref=thedailyconstraint.com) covers the mechanics well. ● **Multimodal:** AI that can work with images, audio, or documents, not just typed text. Feed it a photo of a defect, not just a description of one. ● **Reasoning model:** A newer type of model built to work through harder problems in visible steps before answering, closer to chain of thought by default than earlier models. ● **Open-weight model:** A model whose internal settings are published for anyone to download and run themselves, instead of only being accessible through a company's paid service. ● **Small language model (SLM):** A smaller, more specialized model built to do one thing well and run cheaply, instead of trying to do everything the way a large general model does. That's the list for now. It will grow. If a term trips you up somewhere else in this series, or you keep hearing one that's not here, *comment below* and I'll add it.