Concept
The model is not retrieving. It is predicting.
Each word is chosen from a probability distribution over what usually follows. Nothing is looked up.
Concept
How LLMs Actually Work (No Math)
Understand how large language models predict text, why that explains both their fluency and their mistakes. Free, one sitting.
Start Quick LessonFree to explore · ~3 minutes
Why this matters
Most AI confusion comes from imagining the model as a database that looks things up. It is not. It predicts the next piece of text from patterns, which is why it can write a working function and invent a citation with identical confidence. Once you hold the right mental model, model behaviour stops being mysterious and starts being predictable.
What you'll cover
What the Model Is Actually Doing
- Prediction, not retrieval
- One token at a time
- Probability and temperature
- What "training" actually produced
- Instructions are just more text
Tokens, Context and Attention
- What a token really is
- The context window
- Attention, in one paragraph
- Position matters
- Long context is not free memory
Where Capability Comes From
- Scale and what it bought
- Pretraining vs. post-training
- Reasoning models
- Emergence and its limits
- Why models differ
Why It Gets Things Wrong
- Hallucination as a consequence
- No calibrated self-knowledge
- Training cutoffs and stale facts
- Arithmetic and counting
- Sycophancy and anchoring
Using the Mental Model
- Diagnosing a bad answer
- Predicting what it will be bad at
- Judging a claim about AI
- What this does not explain
- Where to go next
What you'll understand
- Explain next-token prediction in plain language
- Say why fluency and accuracy are unrelated in an LLM
- Predict which tasks a model will be unreliable at
- Stop treating the model as a search engine
A look inside
Three moments from this Quick Lesson
The real thing — not a mockup of it.
How it fits
Context → Prediction → Token → Repeat
The output you read is that loop run a few hundred times, each pass seeing everything written so far.
Apply
Why does a model invent a plausible citation?
Its training data was wrong
A plausible-looking citation is a likely next sequence
It ran out of context
How it works
01
Understand the concept
A plain-language walkthrough of the idea itself, no prior context assumed.
02
See it in practice
A simple diagram or example showing how it actually fits together.
03
Apply what you learned
One quick check that you can recognise it, not just recall it.
Useful for
DevelopersProduct managersFoundersAnyone working with AI
Ready to understand How LLMs Actually Work (No Math)?
Want to go deeper? Explore Context Engineer →
Frequently asked
Do I need maths to understand this?+
No. This Quick Lesson uses no equations. It builds the mental model in plain language.
What will I learn?+
How next-token prediction works, why models hallucinate as a consequence of it, and which tasks that makes them unreliable at.
Is this Quick Lesson free?+
Yes. Quick Lessons are free, short, and do not require a paid plan — sign in only to save your progress.