TL;DR
- This glossary explains the 50 most-used AI terms in plain English with Malaysian business context.
- Organized by category: Foundational AI Concepts, Generative AI Tools, Technical Terms, Cost & Performance, Compliance & Safety.
- Every term includes a one-line definition + Malaysian example.
- Use this when reading vendor pitches, evaluating AI tools, or training your team.
- Bookmark this page — AI terminology changes fast and most articles assume you already know the basics.
[IMAGE 1 — Featured: AI glossary book with laptop showing AI tools]
Intro
AI for Malaysian business owners comes with jargon — LLM, GPT, RAG, prompt engineering, hallucinations, fine-tuning, context windows, tokens. Most articles assume you already know these terms. This glossary explains the 50 most-used AI concepts in plain English with examples from Malaysian businesses, real RM costs where relevant, and what each term means for your buying decisions. Bookmark this page. The next time a vendor pitches you on "our proprietary LLM-based solution with RAG capabilities," you'll know exactly what they mean and whether it's worth the money.
Foundational AI Concepts (10 terms)
The 10 foundational AI concepts every Malaysian business owner should understand are AI, machine learning, neural network, deep learning, generative AI, LLM, GPT, transformer, training data, and inference.
1. AI (Artificial Intelligence). Software that performs tasks usually requiring human intelligence — reading, writing, recognising images, decision-making. Example: ChatGPT writing your Shopee listing is AI.
2. Machine Learning (ML). A type of AI where software learns patterns from data rather than being explicitly programmed. Example: Meta Advantage+ learning which audiences buy your products.
3. Neural Network. A mathematical model loosely inspired by the human brain, used in most modern AI systems. You won't interact with neural networks directly — they power the tools you use.
4. Deep Learning. A subset of machine learning using neural networks with many layers. Powers ChatGPT, Gemini, Claude, image generation, voice recognition.
5. Generative AI. AI that creates new content — text, images, video, audio — rather than just classifying or predicting. ChatGPT, Claude, DALL-E, and Gemini are generative AI.
6. LLM (Large Language Model). An AI trained on massive amounts of text to understand and generate human language. GPT-5, Claude 4.7, Gemini 2.0 are all LLMs. When someone says "an LLM-powered tool", they mean it uses one of these.
7. GPT (Generative Pre-trained Transformer). The technical name for OpenAI's family of LLMs. GPT-5 is the latest as of 2026. ChatGPT runs on GPT models.
8. Transformer. The neural network architecture introduced in 2017 that powers nearly all modern LLMs. You don't need to understand the math, just that it's the foundation.
9. Training Data. The text, images, or other data an AI learns from. GPT-5 was trained on roughly the entire public internet plus licensed datasets.
10. Inference. When an AI uses what it learned to respond to your input. Every time you send a message to ChatGPT, it runs inference. Inference costs are why ChatGPT Plus charges RM95/month.
Generative AI Tools (12 terms)
The 12 generative AI tools and concepts most relevant to Malaysian businesses in 2026 are ChatGPT, Claude, Gemini, DALL-E, Midjourney, Sora, Veo, Llama, prompt, prompt engineering, system prompt, and few-shot prompting.
11. ChatGPT. OpenAI's flagship AI assistant. Free tier available. Plus tier RM95/month. Most familiar to Malaysian business owners.
12. Claude. Anthropic's AI assistant. Free tier available at claude.ai. Pro tier RM95/month. Preferred by many writers and editors for quality.
13. Gemini. Google's AI assistant. Free at gemini.google.com. Advanced tier RM95/month. Best for Google Workspace users.
14. DALL-E. OpenAI's image generation model, built into ChatGPT Plus. Generates business images, marketing visuals, illustrations.
15. Midjourney. Independent image generation service known for artistic quality. Costs USD 10–60/month (~RM45–RM280).
16. Sora. OpenAI's video generation model. Released to general availability in 2025. Available in ChatGPT Pro tier.
17. Veo. Google's video generation model. Available via Gemini Advanced.
18. Llama. Meta's open-source LLM family. You can self-host Llama models. Free to use but requires technical setup.
19. Prompt. What you type into an AI. "Write me a Shopee listing for a women's blouse, RM89, comfortable cotton, sizes S-XL" is a prompt.
20. Prompt Engineering. The skill of writing effective prompts. Good prompt engineering can 5x the quality of AI output without changing the underlying tool.
21. System Prompt. Instructions given to an AI before the conversation starts, often defining its personality or constraints. Used in custom AI assistants.
22. Few-shot Prompting. Including 2–5 examples in your prompt to guide the AI's output style. Example: "Here are 3 examples of our brand voice. Now write a 4th in the same style: [prompt]."
Technical AI Terms (10 terms)
The 10 technical AI terms Malaysian business owners encounter most often are token, context window, fine-tuning, embedding, vector database, RAG, agent, API, model, and hallucination.
23. Token. The unit AI uses to measure text. Roughly 1 token = 0.75 English words. Pricing for AI APIs is usually per token (e.g., USD 0.003 per 1,000 input tokens).
24. Context Window. How much text an AI can read at once. ChatGPT Plus: 200,000 tokens (~150,000 words). Claude Pro: 1 million tokens (~750,000 words — an entire textbook).
25. Fine-tuning. Training an AI on your specific data to make it better at your use case. Costs ~USD 1,000–10,000 typically. Most Malaysian SMEs don't need this.
26. Embedding. A numerical representation of text meaning, used for AI to compare similarity between concepts. Powers semantic search and RAG.
27. Vector Database. A database that stores embeddings, allowing AI to search by meaning rather than keyword. Examples: Pinecone, Weaviate, Supabase Vector.
28. RAG (Retrieval-Augmented Generation). Combining an LLM with your own data so it answers based on your documents. Example: a chatbot that answers based on your product manuals.
29. Agent. An AI that takes actions on your behalf — not just chats, but actually does things like book meetings, send emails, query databases. Most cutting-edge AI work in 2026 involves agents.
30. API (Application Programming Interface). How software systems talk to each other. When you connect Claude to Make.com, you're using Claude's API.
31. Model. A specific AI system. "GPT-5" is a model. "Claude Opus 4.7" is a model. "Gemini 2.0 Pro" is a model.
32. Hallucination. When AI makes up false information that sounds plausible. Example: inventing a non-existent law citation. Always verify AI outputs, especially facts and numbers.
Cost and Performance Terms (8 terms)
The 8 cost and performance AI terms Malaysian business owners need to understand are inference cost, token pricing, latency, throughput, accuracy, hallucination rate, response time, and benchmark.
33. Inference Cost. The cost to run AI on a single request. Powers the per-usage pricing of AI APIs.
34. Token Pricing. Most AI APIs price per token. Example (2026): Claude Sonnet input is ~USD 3 per million tokens. Important for calculating costs at scale.
35. Latency. How long the AI takes to respond. Lower is better for customer-facing applications. ChatGPT typical: 1–3 seconds. Older models: 5–10 seconds.
36. Throughput. How many requests an AI can handle per second. Relevant for high-traffic applications.
37. Accuracy. How often the AI is correct. No AI is 100% accurate. Top models score 85–95% on common benchmarks.
38. Hallucination Rate. How often an AI makes things up. Better models hallucinate less. Always test with your specific use case.
39. Response Time. End-to-end time from your prompt to receiving output. Includes latency plus network time.
40. Benchmark. A standard test used to compare AI models. Common benchmarks: MMLU (general knowledge), HumanEval (coding), GSM8K (math).
Compliance and Safety Terms (10 terms)
The 10 compliance and safety AI terms most relevant to Malaysian businesses are PDPA, data residency, training opt-out, bias, alignment, jailbreak, watermarking, deepfake, content moderation, and audit trail.
41. PDPA (Personal Data Protection Act 2010). Malaysian law governing how personal data is collected, stored, and used. Applies to any AI tool processing customer or employee data.
42. Data Residency. Where your data is physically stored. Enterprise AI plans often offer data residency in Asia or specific countries.
43. Training Opt-out. Most paid AI plans (ChatGPT Plus, Claude Pro, Gemini Advanced) don't use your data to train future models. Free tiers may.
44. Bias. When AI systems produce systematically unfair results, often reflecting bias in training data. Test your AI outputs for bias, especially in hiring and lending decisions.
45. Alignment. Making AI systems behave according to human values and intentions. Modern AI tools are aligned to refuse harmful requests.
46. Jailbreak. Bypassing an AI's safety guardrails. Don't try this for production use — it produces unreliable output and may violate terms of service.
47. Watermarking. Embedding invisible markers in AI-generated content to identify it as AI-made. Important for transparency.
48. Deepfake. AI-generated fake video or audio of real people. Ethical and legal risks. Don't create deepfakes of clients or customers.
49. Content Moderation. AI systems that detect inappropriate content (spam, NSFW, violence). Important for user-generated content platforms.
50. Audit Trail. Records of who used the AI for what. Required for PDPA compliance and many enterprise use cases.
How to Use This Glossary
The 4 best ways to use this AI glossary are: read it once end-to-end for a foundation, bookmark it for vendor pitches, share it with your team for training, and refer back when reading AI articles or contracts.
1. Read once for foundation. Spend 30 minutes reading from top to bottom. You'll have working understanding of AI vocabulary.
2. Use during vendor pitches. When an AI vendor uses jargon, look up the term mid-conversation (or after). You'll spot vendors who use jargon to obscure rather than clarify.
3. Train your team. Share this glossary as part of onboarding for any role that touches AI tools.
4. Reference when reading AI articles or contracts. AI contracts often include technical terms (model, token, fine-tuning, embedding). Knowing what they mean prevents costly misunderstandings.
Frequently Asked Questions
What is the most important AI term for Malaysian business owners to know?
The most important AI term for Malaysian business owners to know is LLM (Large Language Model) — the technology behind ChatGPT, Claude, and Gemini. Every AI tool you'll evaluate as an SME in 2026 is either an LLM directly or built on top of one.
What's the difference between AI and machine learning?
AI is the broader category of software performing intelligent tasks. Machine learning is a specific technique within AI where software learns patterns from data rather than being explicitly programmed. All modern generative AI uses machine learning.
What does "RAG" mean in AI?
RAG stands for Retrieval-Augmented Generation — combining an LLM (like Claude) with your own data (product manuals, FAQs, knowledge base) so the AI answers based on your specific information rather than general training data.
What is an "AI agent"?
An AI agent is a system that takes actions on your behalf, not just chats. Example: an agent that reads your inbox, identifies sales leads, drafts replies, and schedules follow-up meetings — without you doing each step. Most cutting-edge AI work in 2026 involves agents.
What's a "context window" and why does it matter?
Context window is how much text an AI can read and consider at once. Claude Pro's 1 million token window means it can read an entire 750-page document in one go. ChatGPT Plus's 200K token window is smaller but still handles most business needs.
Is "AI hallucination" really common?
Yes. All AI tools occasionally generate plausible-sounding but false information — citations that don't exist, statistics that are wrong, or facts that are made up. Always verify AI outputs for important decisions.
Do I need to know these terms to use AI?
You don't need to know technical AI terms to use ChatGPT, Claude, or Gemini effectively. But knowing 10–15 key terms (LLM, prompt, context window, hallucination, training data) helps you evaluate tools, read vendor pitches, and avoid getting oversold.
Ready to apply AI knowledge in your business?
ONSET, the AI division of Marketing Lancers, helps Malaysian businesses move from understanding AI terms to actually using AI tools — with practical workflows, real RM cost guidance, and the 50% MDEC SME Digitalisation Grant.
- Free 30-minute AI strategy call — we'll demystify the jargon for your specific business. Book a call →
- Free AI readiness audit — send your current process, get a 5-page report. Request your audit →
For the bigger picture, read our complete 2026 guide to AI for SMEs in Malaysia.
Last updated: May 2026 · Written by the Marketing Lancers team
