Tuesday Noon: AI Vocabulary Deep Dive
By midday your brain has warmed up — this is a great time for vocabulary that requires a bit more nuance. This session focuses on AI-specific jargon: not just what these words mean, but how to use them accurately in conversation, and how to explain them to colleagues who aren’t deeply technical.
The ability to translate AI concepts into plain English is a superpower for tech leads. It earns trust from product managers, executives, and new team members who need to understand your decisions.
Word of the Day
hallucination /həˌluː.sɪˈneɪ.ʃən/ — ảo giác / bịa đặt (trong AI: mô hình tạo ra thông tin sai nhưng trông có vẻ đúng)
Pronunciation breakdown:
- hə — unstressed schwa
- luː — long “oo” sound
- sɪ — short “i”
- NEI — stressed syllable, rhymes with “day”
- ʃən — “shun”
Full word: hə-LOO-si-NAY-shun (stress on the 4th syllable: NAY)
What it means in AI: When a language model generates text that is factually wrong but stated with confidence. The model isn’t “lying” — it doesn’t know it’s wrong. It’s pattern-matching in a way that produces plausible-sounding but false information.
Examples:
“The model hallucinated a citation that doesn’t exist — we need to add a fact-checking layer.”
“Hallucination rates are lower in RAG-powered systems because the model is grounded in retrieved facts.”
“The biggest risk with customer-facing AI is hallucination — users might act on incorrect information.”
Explaining it simply: “It’s when the AI makes things up and presents them as facts. Like a very confident person who hasn’t checked their sources.”
Vietnamese tip: Cụm từ chính xác là “model hallucination” hoặc “AI hallucination”. Không nói “model bị ảo giác” — hãy nói “model hallucinate” (dùng như động từ) hoặc “có hiện tượng hallucination.”
Vocabulary Table: 5 AI Terms You Need to Explain Clearly
| Term | IPA | Plain English Explanation |
|---|---|---|
| fine-tuning | /ˈfaɪn.tjuːnɪŋ/ | Taking a general model and training it further on your specific data to make it better at your task |
| grounding | /ˈɡraʊndɪŋ/ | Connecting the AI to real, factual sources so it doesn’t make things up |
| context window | /ˈkɒntekst ˈwɪndəʊ/ | The maximum amount of text the model can “see” and remember at once during a conversation |
| temperature | /ˈtemprɪtʃər/ | A setting that controls how creative vs. predictable the model’s output is (higher = more creative/random) |
| system prompt | /ˈsɪstəm prɒmpt/ | Hidden instructions given to the model before the user’s message — sets its role, behavior, and constraints |
Pronunciation Guide
“hallucination” — 5 syllables. The hardest part is the stress: ha-LOO-ci-NA-tion. Practice: tap the table twice on “LOO” and “NA.”
“fine-tuning” — two words blended. FINE-TOO-ning. The hyphen means it reads as one concept. Say it smoothly, not as two separate words.
“grounding” — GROUND-ing. The “ou” sounds like “ow” in “crown.” Not “GROO-nding.”
“temperature” — 4 syllables but native speakers often say 3: TEM-pra-cher. Both are acceptable in technical speech.
“system prompt” — stress on both words: SYS-tem PROMPT. When saying it fast in conversation: “sys-tem-prompt” as a compound noun.
Exercise 1: Explain in Plain English
Your product manager asks: “What’s the difference between fine-tuning and a system prompt?”
Write a 2-3 sentence explanation using plain English. Then compare with the model answer below.
Model answer
“A system prompt is like a set of instructions you give the model before each conversation — it tells the AI how to behave, what role to play, or what not to say. Fine-tuning is different: you actually train the model on your own data, so it learns your specific patterns and knowledge permanently. Think of system prompt as ‘briefing an employee before a meeting’ and fine-tuning as ‘months of on-the-job training.’”
Exercise 2: Spot the Hallucination
Read this AI-generated paragraph. It contains one hallucination. Find it.
“Claude Sonnet 5 was released in early July 2026 by Anthropic. It supports a context window of 200,000 tokens. The model was trained using a technique called Constitutional AI, first described in Anthropic’s 2022 paper. Claude Sonnet 5 achieved a score of 92.4% on the MMLU benchmark, making it the highest-scoring model on that test as of its release date.”
Identify the hallucination
The claim “making it the highest-scoring model on that test as of its release date” is likely a hallucination — benchmark rankings change rapidly and this specific claim cannot be verified from the model’s training data alone. AI models frequently generate superlatives (“highest,” “first,” “best”) without factual grounding.
The other facts (Sonnet 5 release, 200K context window, Constitutional AI paper) are grounded claims that can be verified. When reviewing AI output, watch especially for comparative claims and rankings.
Idiom of the Day
”under the hood”
Meaning: The internal workings of a system — what’s happening technically, beneath the surface that users see.
In AI context:
“Under the hood, the chatbot is using RAG to pull from our knowledge base before generating each response.”
“The product looks simple, but under the hood there’s a lot of prompt engineering and output validation.”
Origin: From cars — “under the hood” = inside the engine. In English-speaking tech culture, this phrase is used constantly.
Vietnamese: Tương đương với “bên trong kỹ thuật” hay “cơ chế bên dưới.” Nhưng dùng nguyên “under the hood” trong tiếng Anh kỹ thuật nghe rất tự nhiên.
Mini Dialogue: Explaining AI to Your PM
Read this dialogue aloud, playing both roles. Focus on natural pacing — not too fast, not too robotic.
PM (Alex): I got a complaint from a user — the AI assistant told them we offer a “24/7 phone support line.” We don’t have that. What happened?
You (Tech Lead): That’s a hallucination. The model generated a plausible-sounding response, but it wasn’t grounded in our actual documentation.
Alex: How do we stop it?
You: Two options. Quick fix: tighten the system prompt — add a constraint like “only answer based on provided documentation.” Better fix: implement RAG so the model retrieves from our actual support articles before generating a response.
Alex: Which do you recommend?
You: Both, actually. System prompt is a day’s work. RAG pipeline is a sprint. I’d do the system prompt now as a guardrail, then plan the RAG work for next sprint.
Practice tip: Record yourself saying the Tech Lead lines. Listen back. Focus on the pronunciation of “hallucination,” “grounded,” “system prompt,” and “RAG.”
Challenge
Find one real example of AI hallucination in the news, your own usage, or a colleague’s experience. Describe it in 2-3 sentences of English. Use the word “hallucination” and at least one other word from today’s vocabulary table.