How AI Works — Deep Technical Dive - Daftar Isi

Tujuan: Memahami cara kerja AI secara teknis dan mendalam — dari neuron pertama sampai LLM 100B parameter, dengan visualisasi dan analogi Level: Intermediate — Advanced (tidak perlu PhD, tapi perlu ketekunan) Prasyarat: Basic Python, matematika SMA (aljabar, probabilitas dasar) Estimasi Total: 12-16 jam baca

---

Struktur Materi

#	File	Topik	Estimasi
01	01-what-is-ai-really.md	What Is AI, Really? AI vs ML vs DL vs GenAI — Hierarki & Sejarah Singkat	25 menit
02	02-neural-networks-fundamentals.md	Neural Networks dari Nol: Neuron, Layer, Activation, Weight, Bias	45 menit
03	03-training-and-backpropagation.md	Training & Backpropagation: Gradient Descent, Loss Function, Chain Rule	50 menit
04	04-from-perceptron-to-deep-learning.md	Dari Perceptron ke Deep Learning: CNN, RNN, LSTM, dan Kenapa “Deep”	45 menit
05	05-attention-is-all-you-need.md	“Attention Is All You Need”: Transformer Architecture Dibongkar	60 menit
06	06-how-llms-work.md	How LLMs Work: GPT, Claude, Gemini — Pretraining, Tokenization, Inference	55 menit
07	07-embeddings-and-vector-search.md	Embeddings & Vector Search: Representasi Semantik, Similarity, Vector DBs	40 menit
08	08-multimodal-ai.md	Multimodal AI: Vision-Language Models, Text-to-Image, Text-to-Video	40 menit
09	09-reasoning-and-chain-of-thought.md	Reasoning & Chain-of-Thought: o1, o3, DeepSeek-R1, dan Model yang “Berpikir”	45 menit
10	10-mixture-of-experts-and-scaling.md	Mixture of Experts, Scaling Laws, dan Efisiensi: Kenapa Model Bisa Besar Tapi Cepat	35 menit
11	11-ai-safety-and-alignment.md	AI Safety & Alignment: RLHF, Constitutional AI, Red-Teaming, Mechanistic Interpretability	40 menit
12	12-future-architectures.md	Future Architectures: Beyond Transformers — Mamba, JEPA, Liquid Networks	30 menit

---

Peta Konsep

NEURON → LAYER → NETWORK
  ↓
BACKPROPAGATION (cara network belajar)
  ↓
CNN (gambar)  ← →  RNN/LSTM (teks/urutan)
  ↓
TRANSFORMER + ATTENTION (revolusi 2017)
  ↓
LLMs (GPT, Claude, Gemini, Llama)
  ↓
MULTIMODAL (teks + gambar + audio + video)
  ↓
REASONING (o1, R1 — model yang "berpikir")
  ↓
SAFETY & ALIGNMENT (RLHF, DPO)
  ↓
FUTURE (Mamba, JEPA, Liquid Networks)

Sumber Belajar Kunci

• 3Blue1Brown — Neural Networks playlist (YouTube, gratis)
• Jay Alammar — The Illustrated Transformer (blog, gratis)
• Andrej Karpathy — Zero to Hero / nanoGPT (YouTube, gratis)
• Distill.pub — Interactive ML articles (web, gratis)
• Anthropic — Transformer Circuits (blog, gratis)
• HuggingFace — NLP Course (web, gratis)