Drag the sliders and watch a tiny AI "brain" turn two numbers into one decision, step by step. No login, nothing saved, no math required.
Green lines mean a connection pushes the decision up, red means it pulls it down — thicker lines carry more influence. Each circle's color shows how "activated" it is. Move the sliders and watch the decision update instantly, in one straight line from input to output.
It's just weighted opinions
Every connection line is how much one signal influences the next step. "Training" a model is nothing more than tuning those influence levels using past examples.
More layers, more nuance
A single layer can only catch obvious patterns. Stacking layers lets the network build up from simple signals to more sophisticated judgment calls.
It's deterministic, not magic
Given the same inputs and the same trained weights, a network always produces the same output. What looks like intuition is really consistent, repeatable arithmetic.
What is a transformer?
The design behind every modern AI chatbot, plain-language explained.
Try Machine Learning Types →
Supervised, unsupervised, and reinforcement learning, hands-on.
A history of AI, 1950–2026
How we got from the earliest ideas to today's AI tools.
Try the Generative AI tool →
See next-word prediction — the core loop behind text generation.
Think of it as a small panel of judges. Each judge looks at the same information, weighs it based on what matters to them, and passes a score forward. The next layer combines those scores into a final decision. "Training" a network just means adjusting how much weight each judge gives to each factor, based on past examples of right answers.
One judge working alone can only spot simple, obvious patterns. Stacking layers lets later judges combine the earlier judges' opinions into something more nuanced — the same way a hiring committee's final call weighs input from several interviewers, not just one.
They're just three different rules for how firmly a judge reacts. This tool labels them plainly: "gentle squeeze" softens any input toward a small range, "balanced pull" does something similar but allows negative reactions, and "sharp cutoff" only reacts to positive signals and ignores everything else. Try each one — the diagram updates instantly.
This tool shows the network already "trained" — you're just watching it make a decision. Real training happens separately: the network guesses, gets told how wrong it was against a known correct answer, and nudges every weight slightly to be less wrong next time. Repeat that millions of times and the weights settle into something useful.
Same basic building block, much bigger and more specialized. Large language models use a specific design called a transformer, with a mechanism called "attention" that lets every word pay attention to every other word. This playground is the foundation those systems are built on, simplified down to something you can watch happen.
Tokenizer Playground
Type anything and watch it split into tokens live — the units LLMs actually read and get billed for.
Context Window Visualizer
See how much of a model's context window your text fills up, and what happens when you run out of room.
FROG in a Bowl Prompt Builder
Fill in Format, Role, Objective, Goal, and Context — get a copy-ready structured prompt in seconds.
Machine Learning Types
Three tiny games show what supervised, unsupervised, and reinforcement learning actually mean.
Generative AI Playground
Watch AI write one word at a time by predicting what's most likely to come next.
RAG Playground
Ask a question, watch notes get retrieved, then see a grounded answer versus a hallucination.
Embedding Map
Click two words and see why similar meanings sit near each other — the idea behind vector search.
Attention Visualizer
Click a word and see which others a toy transformer looks at — including the classic “it” puzzle.
Prompt Injection Lab
Watch a pasted email try to override a support agent, then flip a switch that treats it as data.
