
Wise AI Ambassador · Set 15 of 20
The foundation
Cards for the second degree of the Wise AI Ambassador
Explains how models learn and why they get things wrong, in plain language with no mathematics.
A card for each unit of the degree, with its idea, its exercise and its takeaway, and the gate's card at the end.
- Degree 2 · The foundation1What is artificial intelligence?
Not a mind, but patterns a machine has learned.
ExerciseExplain to someone in your family the difference between a calculator and a language model.
TakeawayCall it by its real name, and you will handle it better.
Fawzoozfawzooz.ai - Degree 2 · The foundation2How a machine learns
A machine learns from its mistakes, millions of times.
ExerciseDraw this cycle by hand in four boxes, without a single equation.
TakeawayWhat a machine learns is only as good as what it is fed.
Fawzoozfawzooz.ai - Degree 2 · The foundation3Language models: how they “think”
A model predicts the next word.
ExerciseAsk a model to finish the same sentence three times, and notice how the answers differ.
TakeawayFluency is no proof of being right.
Fawzoozfawzooz.ai - Degree 2 · The foundation4Tokens, vectors and context
To a machine, words are numbers with a direction.
ExerciseGive a model a long document, then ask it about the beginning and the end, and compare how accurate it is.
TakeawayWhat falls out of the context window falls out of the model's memory.
Fawzoozfawzooz.ai - Degree 2 · The foundation5Beyond text: image and sound
Today a model sees, hears and speaks.
ExercisePhotograph a chart from your work, ask for an explanation, then check every number.
TakeawayEvery new medium is a new door for error too.
Fawzoozfawzooz.ai - Degree 2 · The foundation6The machine's three limits
Know where the machine stops.
ExerciseAsk a model for the source of a precise fact, and check: does the source really exist?
TakeawayConfidence in the tone is not confidence in the fact.
Fawzoozfawzooz.ai - Degree 2 · The foundationGate 2 · The output
A page that explains the machine
What you doWrite one page explaining how a model learns, and at least two ways it gets things wrong, with no equations.
You have passed itDo not move on to the next degree before you pass this gate.When someone who is not a specialist reads it, then explains it back to you in their own words.
Fawzoozfawzooz.ai
