LLM Foundations
FreeWhat is actually happening inside a language model
Start at the beginning: how text becomes numbers, how attention finds meaning, and why a model that never sees a letter can still finish your sentence. Free, start to finish.
Beginner–Intermediate88 min11 topics
0 of 11 topics
First Principles
Tokens & Meaning
- Tokens & TokenizationBeginnerAsk a model how many r's are in 'strawberry' and it can get it wrong. It's not stupid — it literally cannot see the letters. Why?7 min
- Embeddings & Vector SpaceBeginnerA computer can't compare two words. But it can compare two numbers — so what if meaning had coordinates?8 min
The Engine
- The Attention MechanismIntermediate'The trophy didn't fit in the suitcase because it was too big.' What is 'it'? You knew instantly. Teaching a machine that took thirty years.9 min
- Transformers, End to EndIntermediateThe most sophisticated machine humans have built does exactly one thing: guess the next word. So where does everything else come from?10 min
- Context Windows & Why They MatterBeginnerWhy does a model with a 128k context still 'forget' what you told it four messages ago?8 min
Controlling Output
- Temperature, Top-p & SamplingBeginnerAsk a model the exact same question twice and you get two different answers. Nothing about the model changed between them — so where does the difference actually come from?7 min
- Why Models HallucinateIntermediateA model has no truthful mode and no lying mode — it runs identical machinery either way. So what is actually different about the sentences it gets wrong?8 min
How They're Made
- Pretraining vs Fine-tuningIntermediateA company fine-tunes a model on its internal docs, and it still gets the policies wrong. The training run worked exactly as designed. So what did it actually change?8 min
- RLHF & Alignment, SimplyIntermediateNobody ever wrote down the rule that says be helpful. Humans just kept pointing at the better of two answers — and somehow that was enough. It's also exactly where sycophancy comes from.9 min
- Prompting FundamentalsBeginnerMost prompting advice is folklore that happens to work. Once you know what the model genuinely does with your words, you can stop memorising tricks and start deriving them.8 min