🔤 1. Everything is tokens

LLMs don't see sentences. They see token IDs.
This is why:

• context length matters
• long prompts cost more
• models lose track of earlier text

🧠 2. Attention

Attention lets each token inspect every other token and decide relevance.

Gives you:

• reasoning
• relationships
• instruction following

Also causes:

• hallucinations (wrong patterns reinforced)

🪜 3. Transformer layers

Layers refine meaning:

• lower → syntax
• middle → facts
• upper → reasoning

🧩 4. Why LLMs hallucinate

Because they predict, they don’t verify.

Agents fix this with:

• tools
• retrieval
• planning loops

📦 5. Why small models often win

Tools + retrieval > raw model size.