Tuesday Evening: AI Vocabulary Review & Listening Comprehension Tips
Welcome back! Tonight we wrap up a full day of AI English vocabulary. You’ve learned debug this morning and hallucinate at noon — now we add the final piece, review everything, and build your listening skills for the AI world.
Word of the Day: inference
IPA pronunciation: /ˈɪnfərəns/
How to say it: IN-fuh-runss — stress on the first syllable. The middle syllable is very short and reduced.
Vietnamese meaning:
- General meaning: suy luận (drawing a conclusion from evidence)
- AI/tech meaning: chạy model AI để tạo ra kết quả dự đoán (running a trained model to produce predictions)
In AI, inference = using a trained model to get output. Training is when the model learns; inference is when the model works.
References:
3 Examples in AI Context
- “The server handles thousands of inference requests per second from users around the world.”
- “We optimized the model for faster inference without losing too much accuracy.”
- “During inference, the language model predicts the next token based on all previous context.”
Vocabulary Table: Today’s AI Terms Review
| Phrase | Vietnamese | Example Sentence |
|---|---|---|
| hallucinate | bịa đặt thông tin (nói sai sự thật một cách tự tin) | “The model hallucinated a research paper that doesn’t exist.” |
| inference | chạy model AI / suy luận | ”Inference speed is critical for real-time applications.” |
| embedding | nhúng (biểu diễn dữ liệu dưới dạng vector số) | “The embedding layer converts words into numerical vectors.” |
| fine-tune | tinh chỉnh (huấn luyện thêm một model đã có sẵn) | “We fine-tuned GPT-4 on our customer support data.” |
| prompt engineering | kỹ thuật viết câu lệnh cho AI | ”Good prompt engineering can dramatically improve model output quality.” |
Pronunciation Practice: Stress Patterns in AI Jargon
Speaking AI terms confidently is about knowing where the stress falls. English speakers notice immediately when stress is wrong — even if the word sounds otherwise correct.
Breaking Down “artificial intelligence”
ar-ti-FI-cial in-TEL-li-gence
- ar-ti-FI-cial → stress on the third syllable: fi
- in-TEL-li-gence → stress on the second syllable: tel
Say it in rhythm: ar-ti-FI-cial in-TEL-li-gence (tap your finger on the stressed syllables)
Stress Map for Today’s Words
| Word | Stress Pattern | Stressed Syllable |
|---|---|---|
| IN-fer-ence | ● ○ ○ | IN |
| hal-LU-ci-nate | ○ ● ○ ○ | LU |
| em-BED-ding | ○ ● ○ | BED |
| FINE-tune | ● ○ | FINE |
| prompt en-gi-NEER-ing | ○ ○ ○ ● ○ | NEER |
Rhythm Tips for AI Terms
- Compound nouns get stress on the first word: PROMPT engineering, FINE-tuning, MACHINE learning
- Words ending in -tion/-sion: stress falls on the syllable before the ending — optimiza-TION, representa-TION
- When in doubt, listen: Use YouGlish to hear native speakers say any term in real context
Practice Sentence (say it 3 times, fast):
“Good PROMPT engineering reduces model HALlucination during INference.”
Exercise 1: Fill in the Blank
Choose the correct word: hallucinate / inference / embedding / fine-tune / prompt
- The AI model tends to __________ when asked about very recent events it wasn’t trained on.
- We need to __________ the base model on medical data before deploying it in a hospital.
- The __________ layer converts each word into a 768-dimensional vector.
- The chatbot’s __________ time must be under 200 milliseconds to feel responsive.
- Write a clear __________ that tells the model exactly what format you want the output in.
Show Answers
- hallucinate — “The AI model tends to hallucinate when asked about very recent events it wasn’t trained on.”
- fine-tune — “We need to fine-tune the base model on medical data before deploying it in a hospital.”
- embedding — “The embedding layer converts each word into a 768-dimensional vector.”
- inference — “The chatbot’s inference time must be under 200 milliseconds to feel responsive.”
- prompt — “Write a clear prompt that tells the model exactly what format you want the output in.”
Exercise 2: Translate to English
Translate these Vietnamese AI-related sentences into natural English. Use today’s vocabulary where possible.
- “Model này bị lỗi vì nó hay bịa đặt thông tin không chính xác.”
- “Chúng tôi cần tinh chỉnh model trên dữ liệu tiếng Việt để cải thiện độ chính xác.”
- “Tốc độ chạy model quá chậm, cần tối ưu hóa hơn.”
- “Kỹ thuật viết câu lệnh tốt giúp AI cho ra kết quả tốt hơn.”
- “Vector nhúng giúp model hiểu được ý nghĩa của từ.”
Show Answers
- “This model has a problem because it frequently hallucinates inaccurate information.”
- “We need to fine-tune the model on Vietnamese data to improve accuracy.”
- “The inference speed is too slow — we need to optimize it further.”
- “Good prompt engineering helps AI produce better results.”
- “Embeddings help the model understand the meaning of words.”
(Note: Your answers may vary slightly — focus on using the correct vocabulary naturally.)
Idiom of the Day: “feed the algorithm”
Vietnamese meaning: “cung cấp dữ liệu/nội dung cho thuật toán” — nhưng thường dùng với nghĩa rộng hơn là làm theo những gì hệ thống AI/mạng xã hội muốn để đạt kết quả tốt hơn.
In everyday English, “feed the algorithm” means to create or interact with content in a way that the platform’s recommendation system will reward — getting your content shown to more people.
Usage Examples
- “I post three times a day on LinkedIn just to feed the algorithm — it’s exhausting but my reach doubled.”
- “The app basically trains users to feed the algorithm with engagement data, whether they realize it or not.”
Pro tip: You’ll hear this idiom constantly in conversations about social media, content creation, and AI product design. It’s the kind of phrase that immediately signals you understand how modern tech works.
Speaking Challenge: 60-Second Explanation
Task: Set a 60-second timer. Speak out loud (to yourself, a mirror, or record your phone) and explain how a large language model works — using only today’s vocabulary.
Words to use: inference, embedding, hallucinate, fine-tune, prompt engineering, artificial intelligence
Starter sentence to get you going:
“A large language model works by first converting words into embeddings…”
What to cover:
- How the model converts text (embeddings)
- What happens when you send a prompt (inference)
- What can go wrong (hallucinate)
- How models are customized (fine-tune)
- How users control output (prompt engineering)
Don’t worry about being perfect. The goal is to get comfortable speaking these words out loud in connected speech. Replay your recording and notice: did you stress the right syllables?
Listening Tips: Understanding AI & Tech Podcasts in English
Tech podcasts move fast, use dense jargon, and speakers often talk at 160+ words per minute. Here are three strategies that actually work:
Tip 1: Use Key Term Repetition as Anchors
Technical podcasts repeat core terms constantly. When you hear a word you know — like inference, model, training, deployment — it’s an anchor. Even if surrounding words are unclear, anchors tell you what topic is being discussed. Train yourself to catch anchors first, then fill in context around them.
Practice: While listening, make a quick tally: every time you hear a term you recognize, put a mark. After 5 minutes, you’ll see your comprehension is higher than you thought.
Tip 2: Adjust Playback Speed Strategically
Don’t just slow everything down — that changes rhythm and makes normal-speed speech harder later. Instead:
- First listen at 1.0x speed — get the main idea, don’t stop for unknown words
- Second listen at 0.8x on confusing sections only
- Third listen at 1.25x to build tolerance for fast speech
Most podcast apps (Overcast, Pocket Casts, even YouTube) let you adjust speed per segment.
Tip 3: Use Context Clues from Speaker Behavior
Expert speakers in AI podcasts follow predictable patterns:
- They define new terms right after introducing them (“…what we call inference — that is, running the model to get output…”)
- They give examples immediately after abstract claims
- They repeat key ideas in different words
When you miss a term, don’t panic — wait 10–15 seconds. The speaker will almost certainly explain it, restate it, or give an example that makes it clear.
Evening Challenge: 5 Minutes of Real AI Content
Tonight, watch 5 minutes of an AI-related English video and write down 3 words or phrases you didn’t know before.
Recommended channels:
- Andrej Karpathy — youtube.com/@AndrejKarpathy — deep explanations of how language models actually work, clear and precise English
- Two Minute Papers — youtube.com/@TwoMinutePapers — fast-paced summaries of AI research papers, great for vocabulary density
What to note:
- The 3 new words/phrases
- The context they were used in (one sentence each)
- Your best guess at the meaning before looking it up
This habit — watching 5 minutes of real AI content daily — will accelerate your vocabulary faster than any flashcard app.
Day Summary: Tuesday AI Vocabulary
You’ve now covered three sessions of AI English today:
| Session | Word | Category |
|---|---|---|
| Morning | debug | Technical / Engineering |
| Noon | hallucinate | AI Behavior |
| Evening | inference | AI Operations |
Plus the supporting vocabulary: embedding, fine-tune, prompt engineering, artificial intelligence.
That’s 7 AI terms in one day. Use them in a real sentence tomorrow — write them in a Slack message, say them in a meeting, or just think in English when you open an AI tool.
See you tomorrow morning for a new topic. Good night!