Welcome to "How AI Thinks"!
Hi there! I'm Allie, your AI guide. I'm going to show you how AI language models like me work!
AI doesn't actually "think" like humans do. We learn from patterns in text data and make predictions about what words might come next.
Ready to explore how I work? Click on "Visual Tokens" to get started!
Visual Tokens
AI models like me break down text into small pieces called "tokens". These can be words, parts of words, or even single letters!
Tokens:
When AI "reads" text, it sees these tokens, not the whole words. Each token gets a number, and AI learns which tokens usually come after others.
For example, after "I love" the tokens for "pizza", "ice cream", or "playing" might be likely to come next.
Sentence Builder
Let's play a game! Can you guess what word might come next in these sentences? Then see what the AI would predict!
The cat sat on the
AI's top predictions:
- mat (45%)
- floor (20%)
- chair (15%)
- windowsill (10%)
- table (5%)
Training vs Predicting
AI models have two main phases: training and predicting. Let's try them out!
Training Phase
Help train the AI by feeding it examples:
Prediction Phase
Now test what the AI has learned:
Train the AI first, then select a topic to see a prediction!
How This Works:
1. Training Phase: Drag example sentences to the AI brain. The AI learns patterns from these examples.
2. Prediction Phase: Once trained, the AI can generate new sentences based on what it learned about your chosen topic.
Real AI models work similarly but with millions of examples and much more complex patterns!
Model vs Human
Can you tell the difference between text written by a human and text generated by an AI? Let's find out!
Who wrote this?
The ocean waves crashed against the shore, creating a soothing rhythm that lulled me into a peaceful state of mind.
Build-a-Bot
Let's see how different training data changes how an AI responds!
Choose what to train your AI on:
Glossary
Here are some important terms to help you understand how AI works:
AI (Artificial Intelligence)
Computer systems that can perform tasks that normally require human intelligence.
Large Language Model (LLM)
A type of AI that has been trained on lots of text to understand and generate human language.
Token
Small pieces of text (words, parts of words, or characters) that AI uses to process language.
Training
The process where AI learns patterns from lots of examples.
Prediction
When AI guesses what might come next based on what it has learned.
Dataset
The collection of text that an AI learns from during training.