Guess and Adjust A song about how AI learns
Sing it back
Can you finish the lyrics?
Eight quick questions on “Guess and Adjust”. No music, just your memory. Then share your score and challenge a friend.
How AI learns in one paragraph
AI systems built with machine learning learn by finding patterns in large numbers of examples, rather than by following rules a programmer wrote out by hand. Many of today’s systems are neural networks: layers of simple artificial neurons, each of which multiplies its inputs by adjustable numbers called weights, adds them up and passes the result on to the next layer. In training, the network makes a prediction, a loss function measures how wrong it was, and backpropagation works out how each weight contributed to the error so gradient descent can nudge every weight a little in the direction that reduces it, repeated over a huge number of examples. Once training is finished the model is used as it is; large language models, the systems behind chatbots, generate text by repeatedly predicting the next token, a word or piece of a word. Because they learn from data, models can be confidently wrong and can repeat biases in the data they were trained on.
Key facts from the song
- Machine learning finds patterns in examples instead of following hand-written rules.
- A neural network is made of layers of artificial neurons; each neuron multiplies its inputs by weights, adds them up and passes the result on.
- Training repeats one loop: make a prediction, measure the error with a loss function, then adjust the weights using backpropagation and gradient descent.
- Large language models generate text by predicting the next token, a word or part of a word, one after another.
- Models can be confidently wrong, and they can reproduce biases present in their training data.
Where it fits in your course
Lyrics
Verse 1
No one writes the rules by hand, the machine learns from examples instead
Show it many labeled photos, and it picks out the patterns it’s been fed
A neural network stacks up layers of artificial neurons, row on row
Each multiplies its inputs by weights, adds them up, and lets the signal go
Chorus
Make a guess, measure the error, nudge the weights a little bit
Do it again a million times until the predictions fit
The loss is the error, and gradient descent walks it downhill, slow and steady
Backpropagation tells each weight which way to move until the model’s ready
Verse 2
Training is where learning happens, then the trained model’s put to use
Large language models predict the next token, then the next, on through
It learns from data’s patterns, so it can sound sure and still be wrong
And the biases in that data can get carried right along
Chorus
Make a guess, measure the error, nudge the weights a little bit
Do it again a million times until the predictions fit
The loss is the error, and gradient descent walks it downhill, slow and steady
Backpropagation tells each weight which way to move until the model’s ready
Outro
Learning patterns from examples, one small weight change at a time
Guess, measure, and adjust, then do it all again down the line
How AI learns flashcards
Say the answer out loud, then tap the card to check it. Or test yourself quick-fire.
1How is machine learning different from traditional programming?
It learns patterns from examples instead of following rules written by hand.
2What does an artificial neuron do?
It multiplies its inputs by weights, adds them up and passes the result on to the next layer.
3What are the three steps of the training loop?
Make a prediction, measure the error with a loss function, and adjust the weights (backpropagation and gradient descent).
4What does a large language model predict?
The next token, a word or piece of a word, one after another.
Questions about how AI learns
Still wondering about something? Ask us and we’ll reply within two working days.
How does AI learn?
Most modern AI uses machine learning: instead of being given rules, a model is shown many examples and adjusts its internal settings, called weights, until its predictions match the examples better. It repeats this guess, measure the error, adjust loop a huge number of times.
What is a neural network in simple terms?
A neural network is a mathematical model made of layers of simple units called artificial neurons. Each one multiplies its inputs by weights, adds them up and passes a signal to the next layer; training sets the weights.
What is gradient descent?
Gradient descent is the method used to reduce a model’s error during training. It calculates which direction each weight should change to make the loss smaller and moves it a small step that way, over and over; backpropagation is how those directions are calculated efficiently.
How do large language models work?
A large language model is a neural network trained on large amounts of text to predict the next token, a word or piece of a word. It writes a reply by predicting one token, adding it to the text and predicting again, which is why it can sound fluent while still getting facts wrong.