🤖 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/
VietnameseẢo giác (khi AI tạo ra thông tin sai nhưng nghe có vẻ đúng)
Part of speechVerb

When an AI model hallucinates, it generates information that sounds confident and plausible — but is factually incorrect or completely made up.

📖 Example Sentences

  1. “The chatbot hallucinated a fake research paper with a real-sounding author and title.”
  2. “We need to validate outputs carefully — large language models can hallucinate code that doesn’t compile.”
  3. “The AI hallucinated an entire API endpoint that doesn’t exist in our documentation.”

🔗 References


📚 AI Vocabulary Table

PhraseIPAVietnamese MeaningExample
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

WordSyllablesStressAudio Tip
hal·lu·ci·nat·ed5 syllableshal-LU-ci-na-tedStress on 2nd syllable
ground·ing2 syllablesGROUND-ingShort, sharp first syllable
en·gi·neer·ing4 syllablesen-gi-NEER-ingStress on 3rd syllable

🎧 Audio References


✏️ 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

  1. “The AI gave a completely wrong answer — it seemed to __________ the entire statistic.”
  2. “Our solution uses retrieval-augmented generation for __________ the responses in real data.”
  3. “After __________ on 10,000 support tickets, the model handles customer queries much better.”
  4. “We spent a week on __________ to get consistent, structured outputs from the LLM.”
  5. “The document is 200 pages — does it fit in the model’s __________?”
✅ Check your answers
  1. hallucinate — AI tạo ra thông tin sai
  2. grounding — Kết nối AI với dữ liệu thực
  3. fine-tuning — Tinh chỉnh mô hình trên dữ liệu cụ thể
  4. prompt engineering — Kỹ thuật thiết kế câu lệnh
  5. 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:

  1. “Mô hình đã tạo ra một đoạn code không tồn tại — đây là hiện tượng ảo giác của AI.”
  2. “Chúng tôi đang tinh chỉnh mô hình trên dữ liệu nội bộ của công ty để cải thiện độ chính xác.”
  3. “Kỹ thuật viết câu lệnh tốt có thể giúp giảm đáng kể hiện tượng ảo giác.”
✅ Sample Translations
  1. “The model generated non-existent code — this is a classic case of AI hallucination.”
  2. “We’re fine-tuning the model on internal company data to improve accuracy.”
  3. “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”

MeaningDữ 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
RegisterProfessional / Tech

Usage Examples

  1. “We spent months cleaning training data because, well — garbage in, garbage out. The model quality depends entirely on what you feed it.”
  2. “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:

  1. Ask it a very specific, obscure question: “What was the exact sales revenue of [small local company] in Q3 2019?”
  2. Notice how it responds — does it admit uncertainty, or does it sound confident with a made-up answer?
  3. Try this prompt engineering fix: Add “If you don’t know, say ‘I don’t have reliable information about this.’”
  4. Compare the two responses.

Share what you find — did the AI hallucinate? Did your prompt engineering reduce it?


📊 Today’s Vocabulary Summary

TermVietnameseUse It When…
hallucinateảo giác AIAI tạo thông tin sai
inferencechạy/suy luận mô hìnhNói về tốc độ/chi phí AI
fine-tuningtinh chỉnh mô hìnhCustomize AI cho domain cụ thể
prompt engineeringkỹ thuật viết lệnh AITối ưu hóa đầu vào cho AI
groundinggắn AI vào dữ liệu thựcRAG, factual accuracy
context windowcửa sổ ngữ cảnhGiới hạn bộ nhớ của AI
garbage in, garbage outrác vào rác raNhấn 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.

Export for reading

Comments