AI Playground

Don't just read about AI. Play with it.

Four tiny experiments that show how AI really works. Everything runs in your browser.

How AI reads your words

Language models don't see letters or whole words. They chop text into small pieces called tokens. Type anything and watch it get split.

0tokens
0characters
0chars per token

This is a simplified tokenizer for learning. Real ones (like GPT's) work the same way: common words stay whole, rare words get split into parts.

Guess the next word

At its heart, a language model does one thing: predict the next word. This mini-model learned from a few sentences about AI. Pick a starting word, then let it write.

Click "Write a sentence"…

Low creativity = always picks the likeliest word (safe, repetitive). High creativity = takes risks (surprising, sometimes silly). Real chatbots have this exact dial, called temperature. Click a bar to add that word.

Teach a machine to sort

This is machine learning in one picture. Click to drop green points, Shift+click (or switch below) for red. The model learns a line that separates them, live.

Class AClass B
–accuracy
0training steps

Try this: put green points in a circle with red around them. The straight-line model fails. Switch to "Curvy" and watch it figure it out. That's why bigger models can learn harder things.

Can AI read the mood?

Sentiment analysis guesses whether text is positive or negative. Companies use it on millions of reviews. Type a review and see which words push the score.

NegativeNeutralPositive

This one uses a word list, the old-school way. Modern AI understands context too, so it can tell "not bad" is actually positive. Try typing "not good" and see how our simple model handles it.