🤖 Tuesday Noon — AI Vocabulary Deep Dive
Good afternoon! Today we explore the vocabulary that defines the AI era — words you’ll hear in tech meetings, read in papers, and need to explain clearly to clients, managers, and teammates.
🌟 Word of the Day: Hallucinate
| IPA | /həˈluː.sɪ.neɪt/ |
| Vietnamese | Ảo giác (khi AI tạo ra thông tin sai nhưng nghe có vẻ đúng) |
| Part of speech | Verb |
When an AI model hallucinates, it generates information that sounds confident and plausible — but is factually incorrect or completely made up.
📖 Example Sentences
- “The chatbot hallucinated a fake research paper with a real-sounding author and title.”
- “We need to validate outputs carefully — large language models can hallucinate code that doesn’t compile.”
- “The AI hallucinated an entire API endpoint that doesn’t exist in our documentation.”
🔗 References
- 📘 Merriam-Webster — hallucinate
- 🎥 What is AI Hallucination? — IBM Technology (YouTube)
- 📖 Cambridge Dictionary — hallucinate
📚 AI Vocabulary Table
| Phrase | IPA | Vietnamese Meaning | Example |
|---|---|---|---|
| inference | /ˈɪn.fər.əns/ | Suy luận / Chạy mô hình | ”Inference latency dropped to 50ms after optimization.” |
| fine-tuning | /faɪn ˈtjuː.nɪŋ/ | Tinh chỉnh mô hình | ”We fine-tuned the model on our company’s data.” |
| prompt engineering | /prɒmpt ˌen.dʒɪˈnɪər.ɪŋ/ | Kỹ thuật viết câu lệnh cho AI | ”Good prompt engineering can dramatically improve output quality.” |
| grounding | /ˈɡraʊn.dɪŋ/ | Gắn AI vào dữ liệu thực tế | ”We use RAG for grounding the model in our knowledge base.” |
| context window | /ˈkɒn.tekst ˈwɪn.dəʊ/ | Cửa sổ ngữ cảnh (giới hạn bộ nhớ của AI) | “GPT-4 has a 128k context window, enough for long documents.” |
🗣️ Pronunciation Guide
Practice Sentence
“The model hallucinated a response due to insufficient grounding, despite careful prompt engineering.”
Breakdown
| Word | Syllables | Stress | Audio Tip |
|---|---|---|---|
| hal·lu·ci·nat·ed | 5 syllables | hal-LU-ci-na-ted | Stress on 2nd syllable |
| ground·ing | 2 syllables | GROUND-ing | Short, sharp first syllable |
| en·gi·neer·ing | 4 syllables | en-gi-NEER-ing | Stress on 3rd syllable |
🎧 Audio References
- 🔊 Hallucinate — Forvo pronunciation
- 🔊 Inference — Cambridge Audio
- 🎥 AI Terms Explained — Google DeepMind (YouTube)
✏️ Exercise 1 — Vocabulary in Context
Fill in the blanks with the correct AI vocabulary term:
Word bank: hallucinate / fine-tuning / context window / prompt engineering / grounding
- “The AI gave a completely wrong answer — it seemed to __________ the entire statistic.”
- “Our solution uses retrieval-augmented generation for __________ the responses in real data.”
- “After __________ on 10,000 support tickets, the model handles customer queries much better.”
- “We spent a week on __________ to get consistent, structured outputs from the LLM.”
- “The document is 200 pages — does it fit in the model’s __________?”
✅ Check your answers
- hallucinate — AI tạo ra thông tin sai
- grounding — Kết nối AI với dữ liệu thực
- fine-tuning — Tinh chỉnh mô hình trên dữ liệu cụ thể
- prompt engineering — Kỹ thuật thiết kế câu lệnh
- context window — Giới hạn bộ nhớ ngữ cảnh của AI
✏️ Exercise 2 — Translation Challenge
Translate these Vietnamese sentences into natural English using today’s vocabulary:
- “Mô hình đã tạo ra một đoạn code không tồn tại — đây là hiện tượng ảo giác của AI.”
- “Chúng tôi đang tinh chỉnh mô hình trên dữ liệu nội bộ của công ty để cải thiện độ chính xác.”
- “Kỹ thuật viết câu lệnh tốt có thể giúp giảm đáng kể hiện tượng ảo giác.”
✅ Sample Translations
- “The model generated non-existent code — this is a classic case of AI hallucination.”
- “We’re fine-tuning the model on internal company data to improve accuracy.”
- “Good prompt engineering can significantly reduce hallucinations.”
💡 Language tips:
- Use “a classic case of” to describe typical examples professionally
- “significantly reduce” sounds more formal than “a lot less”
- “internal company data” is the natural business English phrasing
💡 Idiom of the Day: “garbage in, garbage out”
| Meaning | Dữ liệu đầu vào kém → kết quả đầu ra kém |
| Vietnamese | ”Rác vào, rác ra” — chất lượng đầu ra phụ thuộc vào chất lượng đầu vào |
| Register | Professional / Tech |
Usage Examples
- “We spent months cleaning training data because, well — garbage in, garbage out. The model quality depends entirely on what you feed it.”
- “The client was unhappy with the AI responses, but honestly, their prompts were terrible. Garbage in, garbage out.”
🎭 Mini Dialogue
Context: A developer (Thuan) is explaining AI issues to a product manager (Sarah) in a sprint review.
Sarah: The AI assistant gave a customer completely wrong information about our refund policy. What happened?
Thuan: It hallucinated — the model confidently generated an answer that wasn’t grounded in our actual documentation.
Sarah: So how do we fix it? More fine-tuning?
Thuan: Not necessarily. First, we improve grounding using RAG — we connect the model to our real policy documents. Then we work on prompt engineering to instruct it to say “I don’t know” when uncertain.
Sarah: Got it. And what about the context window — is our policy document too long?
Thuan: Good point. It’s 50 pages, which is fine for GPT-4’s context window, but we should chunk it properly for retrieval. Garbage in, garbage out — if the retrieved chunks are messy, answers will be too.
🎯 2-Minute Challenge
Right now, open any AI chatbot (ChatGPT, Gemini, Claude) and deliberately try to make it hallucinate:
- Ask it a very specific, obscure question: “What was the exact sales revenue of [small local company] in Q3 2019?”
- Notice how it responds — does it admit uncertainty, or does it sound confident with a made-up answer?
- Try this prompt engineering fix: Add “If you don’t know, say ‘I don’t have reliable information about this.’”
- Compare the two responses.
Share what you find — did the AI hallucinate? Did your prompt engineering reduce it?
📊 Today’s Vocabulary Summary
| Term | Vietnamese | Use It When… |
|---|---|---|
| hallucinate | ảo giác AI | AI tạo thông tin sai |
| inference | chạy/suy luận mô hình | Nói về tốc độ/chi phí AI |
| fine-tuning | tinh chỉnh mô hình | Customize AI cho domain cụ thể |
| prompt engineering | kỹ thuật viết lệnh AI | Tối ưu hóa đầu vào cho AI |
| grounding | gắn AI vào dữ liệu thực | RAG, factual accuracy |
| context window | cửa sổ ngữ cảnh | Giới hạn bộ nhớ của AI |
| garbage in, garbage out | rác vào rác ra | Nhấn mạnh tầm quan trọng của data quality |
🕐 Noon session complete! Spend 5 minutes tonight reviewing these terms and use one in a real conversation or Slack message tomorrow.
🌅 Morning session covered daily routines — evening session will focus on AI in writing and documentation.