Day 1: Stepping into the World of Generative AI

Day 1 – Generative AI & Transformers

• Learned what AI really is: not magic, but data-driven generation.

• Generative AI = creating something new (not hard-coded). GPT by OpenAI made this mainstream.

• Difference between high-level (user view) and low-level (internal working).

• Transformer model (Google’s “Attention is All You Need”): Input → Transformer → Output.

• Key concepts:

  • Tokens vs Sequences

  • Self-Attention & Multi-Head Attention

  • Vector Embeddings (turning words into numbers)

• Transformer has Training Mode (learns) and Inference Mode (answers).

• Special tokens: <bos> (beginning), <eos> (end).

• Softmax + Temperature → control probability & creativity.

• Assignments:

  • Build a tokenizer

  • Explain GPT to a child

  • Explain embeddings to mom

  • Explain tokenization to a fresher

---

📝 Diary Style (Casual, Human-Like)

Day 1 of AI Class
Today I started learning about Generative AI. First thing I realized: AI doesn’t actually “think”—it just generates based on patterns from data. Generative AI means it creates something new, unlike hard-coded systems. GPT, made by OpenAI, is the one that started this big wave.

We talked about low-level vs high-level views: one is how things work inside, and the other is what we see outside. Then I learned about Transformers—Google’s big idea from “Attention is All You Need.” The main trick is attention, and especially self-attention (words looking at other words in context). Multi-head attention is like looking at the same sentence from different angles.

I also learned about tokens and sequences (breaking text into parts), and vector embeddings (turning words into numbers so machines understand). Transformers have two modes: training (learning) and inference (answering). We even saw special tokens like <bos> and <eos>.

Finally, we discussed softmax and Temperature, which basically control how predictable or creative AI’s answers are.

My tasks: build a tokenizer, explain GPT to a 5-year-old, explain embeddings to my mom, and explain tokenization to a fresher. It was a lot, but super interesting…. #chaicode