How AI Actually Works

Three videos that build your intuition for what happens inside an LLM — no math required, just watch in order.

1 The Surprisingly Simple Idea Behind Every LLM

Start here. Modern LLM chatbots like ChatGPT, Claude, and Gemini generate responses by predicting the next token, one step at a time. This video makes that idea concrete and visual.

14:34 · Attention Visualized · 30K+ views

💬

Autocomplete on steroids

LLMs choose the most probable next token — that's it.

🎯

Probability, not truth

Output is shaped by patterns in training data, not direct access to facts.

🔄

One word at a time

ChatGPT streams because it literally computes one token per step.

2 The $5 Million Reading Game Behind Every AI

Now that you know LLMs predict the next word, where do they learn those patterns? This video shows the pretraining process: how 15 trillion words get curated, cleaned, and fed to GPUs for months.

15:53 · Attention Visualized · 3.8K views

📚

The patient reader

A model that covers up the next word and guesses — 15 trillion times.

🗑

Data curation

100TB of raw web crawl gets filtered down to 6.8TB of clean training text.

💻

Scale

Thousands of GPUs running for months. No teacher — just text.

3 Self-Attention: How Transformers Actually Work

The deepest dive. Self-attention is the core mechanism that lets every word in a sentence look at every other word simultaneously. This is what makes modern AI different from everything that came before.

13:01 · Visual AI · 19K+ views

🔗

Words talk to words

Every word queries every other word to understand meaning in context.

🧠

Q, K, V

Query, Key, Value — three projections that compute relevance between words.

💡

Why it matters

Self-attention solved the long-range dependency problem that held AI back for years.

4 The Full Picture: From LLM to Agentic Workflow

Now that you understand how LLMs work, see how engineers build real products on top of them. This Stanford course summary covers the full technology stack from base models to multi-agent systems.

Language note: This video is in Chinese (Mandarin). To watch with English subtitles: 1) Click the CC button (bottom-right of the video player) to turn on captions. 2) Click the gear icon (⚙) next to CC. 3) Select Subtitles/CC. 4) Select Auto-translate. 5) Choose English.

27:24 · Gary Chen · 58K+ views · Based on Stanford's "Beyond LLM" course

Topics Covered (English Summary)

🚫

LLM Limitations

No domain knowledge, outdated info, hard to control, struggles with long context.

📝

Prompt Engineering

A core skill for every AI user. Includes two usage styles: delegating whole tasks vs. collaborating step by step.

🔍

RAG

Retrieval-Augmented Generation — give the model access to your own documents.

Fine-Tuning

Train the model on your data to specialize it for specific tasks.

🤖

Agentic Workflow

LLMs that can plan, use tools, and take actions — not just answer questions.

👥

Multi-Agent

Multiple AI agents collaborating on complex tasks, each with a specialized role.

Try It Yourself: Train an AI Model

Now that you understand the theory, try training a real AI model yourself. Google's Teachable Machine lets you build an image, sound, or pose classifier in minutes — no coding required. This short video shows how it works in a classroom setting.

5:50 · STEM Learning By Doing · Advanced / Optional

📷

Train with your data

Collect examples, label them, and watch the model learn the patterns.

🧠

Test and break it

See where the model gets confused — that's how you build intuition for AI limits.

🎯

Try it free

Go to teachablemachine.withgoogle.com and train your own classifier.

🎓
These videos are optional enrichment. You do not need to understand transformer internals to use AI effectively. But knowing what's happening under the hood makes you a sharper, more critical AI user.
← Back to AI Life Skills