Tuesday Morning — AI & Machine Learning Vocabulary
Good morning! Today we dive into the language of AI and machine learning — the terms your teammates use in standups, Slack threads, and pull request reviews. Understanding these words helps you participate confidently in technical discussions and explain complex concepts to non-technical stakeholders.
Word of the Day: hallucinate
IPA: /həˈluːsɪneɪt/ Vietnamese: ảo giác / khi AI bịa ra thông tin không có thật
When an AI model hallucinates, it confidently generates information that is factually incorrect, fabricated, or entirely made up — often without any indication that it is wrong. This is one of the most critical challenges in deploying LLMs in production.
Cambridge Dictionary: hallucinate Hear it spoken: YouGlish — hallucinate
Examples in Professional Context
1. In a PR review:
“The AI assistant hallucinated a method name —
getUserTokens()doesn’t exist in our SDK. Always verify before merging AI-suggested code.”
2. In a team meeting:
“We noticed the chatbot was hallucinating product prices during last week’s demo. We’ve since grounded it with our live inventory database.”
3. Explaining to a client:
“LLMs can hallucinate — meaning they may state incorrect facts with full confidence. That’s why we built a retrieval layer to cross-check every response against verified data sources.”
Vocabulary Table: AI/ML Phrases for Tech Professionals
| Term | Pronunciation | Vietnamese | Example Sentence |
|---|---|---|---|
| inference | /ˈɪnfərəns/ | chạy model để lấy kết quả | The model runs inference on the input data in under 200ms. |
| fine-tune | /faɪn tjuːn/ | tinh chỉnh model trên dữ liệu riêng | We fine-tuned the model on our customer support data to improve domain accuracy. |
| prompt engineering | /prɒmpt ˌendʒɪˈnɪərɪŋ/ | kỹ thuật viết prompt hiệu quả | She specializes in prompt engineering for legal document summarization. |
| RAG (Retrieval-Augmented Generation) | /ræɡ/ | tăng cường LLM bằng dữ liệu thực tế | We use RAG to ground the chatbot in our internal knowledge base. |
| grounding | /ˈɡraʊndɪŋ/ | neo LLM vào dữ liệu thực để tránh hallucinate | Grounding the model in verified sources reduced hallucinations by 80%. |
Pronunciation Guide: “hallucinate”
Full breakdown: hə — luː — sɪ — neɪt
| Syllable | Sound | Tip |
|---|---|---|
| hə | schwa /ə/ | Soft, unstressed — like the “a” in “about” |
| LUː | stressed, long “oo” | This is the stressed syllable — say it louder and longer |
| sɪ | short “i” | Like “sin” without the “n” |
| neɪt | rhymes with “late” | Clear and crisp at the end |
Stress pattern: hə-LU-si-nate (stress on 2nd syllable)
Practice sentence:
“The model hallucinated a reference that doesn’t exist in the documentation.”
Say it three times at normal speed. Pay attention to the long “oo” in the second syllable.
Exercises
Exercise 1: Fill in the Blank
Choose the correct word: inference, fine-tune, prompt engineering, RAG, grounding, hallucinate
- “The base model kept making up citations, so we added a ________ layer to pull real documents before generating.”
- “Our GPU cluster handles 10,000 ________ requests per second at peak traffic.”
- “Instead of retraining from scratch, we decided to ________ the existing model on our proprietary dataset.”
- ”__________ is the practice of designing inputs that reliably guide a model toward accurate, useful outputs.”
- ”________ the LLM with real-time data prevented it from ________ outdated statistics.”
Show Answers
- RAG — Retrieval-Augmented Generation fetches real documents to reduce hallucinations
- inference — running the model to produce outputs
- fine-tune — adapt a pre-trained model to a specific domain
- Prompt engineering — the skill of crafting effective prompts
- Grounding / hallucinating — grounding prevents the model from fabricating outdated info
Exercise 2: Translate to English
Translate these Vietnamese sentences into natural professional English.
- “Chúng tôi đã tinh chỉnh model GPT trên dữ liệu khách hàng nội bộ để cải thiện độ chính xác.”
- “Model này bịa ra tên hàm không tồn tại, vì vậy chúng tôi cần xác minh mọi đề xuất code trước khi merge.”
- “Kỹ thuật viết prompt tốt có thể giảm đáng kể tình trạng AI bịa thông tin.”
Show Answers
- “We fine-tuned the GPT model on our internal customer data to improve accuracy.”
- “The model hallucinated a function name that doesn’t exist, so we need to verify all code suggestions before merging.”
- “Good prompt engineering can significantly reduce the likelihood of the model hallucinating information.”
Idiom of the Day: “garbage in, garbage out”
Vietnamese: Đưa rác vào thì nhận rác ra — chất lượng đầu ra phụ thuộc hoàn toàn vào chất lượng đầu vào.
This classic computing principle is especially relevant in ML: the quality of your model’s output depends entirely on the quality of your training data, your prompts, and your inputs.
Example 1 — Data quality:
“The model’s predictions are unreliable because the training labels were inconsistent. Garbage in, garbage out — we need to clean the dataset before retraining.”
Example 2 — Prompt quality:
“If you give the LLM a vague, ambiguous prompt, don’t be surprised by a vague, useless response. Garbage in, garbage out applies to prompting just as much as it does to training data.”
Recommended Watching
Build your AI English vocabulary through authentic content:
1. Andrej Karpathy — YouTube Former Tesla AI Director and OpenAI co-founder. His explanations of deep learning, LLMs, and neural networks are technically precise yet remarkably accessible. Watch his “Let’s build GPT” series to hear how top AI engineers discuss these concepts. Search: “Andrej Karpathy YouTube”
2. 3Blue1Brown Visual, mathematics-driven explanations of neural networks and machine learning. Excellent for understanding why these models behave the way they do. His “Neural Networks” series uses beautiful animations to make abstract concepts concrete. Search: “3Blue1Brown neural networks”
3. Two Minute Papers Short, energetic summaries of the latest AI research papers. Great for keeping up with the field and hearing the vocabulary used when discussing cutting-edge AI — hallucination mitigation, RAG architectures, multimodal models, and more. Search: “Two Minute Papers YouTube”
Tip: Watch with English subtitles first, then without. Pause and repeat phrases that use today’s vocabulary.
Morning Challenge
Take 5 minutes and write 3 sentences explaining what “hallucination” means to a non-technical manager who has never heard the term.
Your goal: Clear, jargon-free explanation that builds trust and sets realistic expectations about AI tools.
Things to cover:
- What it is (in plain language)
- Why it happens
- How your team handles it
Example opener: “When we say an AI model ‘hallucinates,’ we mean it…”
Share your sentences in your team Slack, or save them for your next presentation about your AI product. Writing for a non-technical audience is one of the most valuable communication skills a tech professional can develop.
Daily English for Tech Professionals — Tuesday Morning Session Next session: Tuesday Noon — Communication phrases for async remote teams