AI MODEL (Machine Learning & LLM)

What it is: Artificial Intelligence models, đặc biệt là *Machine Learning* và *Large Language Models (LLMs)*, dùng thống kê và tối ưu số để học từ dữ liệu.

What it’s for: Tự động hoá suy luận, dự đoán, sinh nội dung và hỗ trợ ra quyết định trong hệ thống phần mềm.

—

* Math for AI (Concept-level)

• Linear Algebra (Vector, Matrix)
• Probability & Statistics
• Gradient & Optimization intuition

* Machine Learning Basics

• Supervised Learning
• Unsupervised Learning
• Reinforcement Learning
• Overfitting vs Underfitting

* Neural Network

• Neuron & Weight
• Layer & Depth
• Activation Functions

* Training Mechanics

• Loss Function
• Backpropagation (Concept)
• Gradient Descent

* Tokenization

• Token vs Word

* Embedding

• Vector Space & Similarity

* Attention Mechanism

• Self-Attention
• Query – Key – Value

* Transformer Architecture

• Positional Encoding
• Multi-head Attention
• Decoder-only (GPT-style)

* Pre-training * Fine-tuning * Instruction Tuning * RLHF (Human Feedback)

* Inference Basics

• Training vs Inference
• Latency & Throughput

* Context Window * Model Optimization

• Quantization
• Distillation

* Prompt Structure

• System / User / Assistant

* Sampling Parameters

• Temperature
• Top-p

* Prompt Risks

• Prompt Injection

* Embedding Search

• Vector Database

* RAG (Retrieval Augmented Generation) * Tool / Function Calling * Agent Pattern

* Hallucination * Non-determinism * Security & Data Leakage

* AI Evaluation

• Golden Dataset
• AI Regression Test

* Human-in-the-loop * AI-SDLC

* Explain why model fails * When NOT to use AI * Design AI-resilient systems