Slate
Explore
Tracks
History
Pricing
Start session
All tracks
Computer Science
How LLMs actually work
From tokens to transformers to the training loop — what's really happening when a model responds.
7 lessons
~120 min total
First principles
What you'll learn
Explain, end-to-end, what happens when an LLM responds to a prompt
Understand why scale changes everything — not just "more is better"
Separate real LLM capabilities from demos and hype
Progress
0 / 7
Start track
Lessons
1
Text as numbers
Tokens, embeddings, and why "king - man + woman ≈ queen".
3 objectives
2
What "attention" really means
Before the jargon: it's a soft lookup over context.
3 objectives
3
A transformer, layer by layer
Embed, attend, transform, repeat — N times.
3 objectives
4
Training: next-token prediction
One objective, trillions of tokens — and somehow, most of what we call "intelligence".
3 objectives
5
Why scale changes everything
Emergence, scaling laws, and the bitter lesson.
3 objectives
6
RLHF — teaching the model what we want
How you get from a next-token predictor to an assistant.
3 objectives
7
What LLMs can't do (and why)
Hallucinations, reasoning limits, and what context windows really buy you.
3 objectives
Related tracks
Software architecture
Why real systems are shaped the way they are — monoliths, services, databases, caches, queues, and the tradeoffs behind each.
8 lessons
SLATE
How the internet actually works
What happens between typing a URL and the page appearing — packets, DNS, TCP, TLS, the works.
8 lessons
SLATE