What Really Happens When You Press Send in a Chatbot?
Starting with a request to explain why the sky is blue to a ten year old, this explainer follows the words after you press Send. They travel to remote computers, break into tokens represented by numbers, and pass through a neural network that predicts the reply one piece at a time before it streams back to the screen. It makes clear that no person is reading or typing, and that the answer comes from patterns learned in training. It is for a general audience and drawn as sketchnotes on a dotted notebook page.
Technology Sketchnote style
- Voice
- Warren
- Language
- English
- Length
- 0:52
The brief
This is the brief that made the video. Start from it and change what you like.
Make a one minute landscape (16:9) technical explainer for a general audience about what really happens when you press Send in a chatbot. Open with someone typing, Why is the sky blue?, adding, Explain it like I'm ten, and pressing Send. Explain that the prompt travels to remote computers and is broken into tokens, pieces of text represented by numbers that may be a word, part of a word or punctuation. Show the numbers flowing through a neural network that estimates what comes next, predicting one piece and then using the growing answer to predict the next. Show the response streaming back piece by piece, and close by noting that it comes from a model generating text from patterns learned in training, with no person reading or typing. Use a curious, clear American male host and sketchnote visuals on a dotted notebook page.
The script
You type, “Why is the sky blue?” Then add, “Explain it like I’m ten.” You press Send. What happens to those words next?
Your prompt travels to remote computers, where it is broken into tokens: pieces of text represented by numbers. A token might be a whole word, part of a word, or punctuation. The numbers give the model a form it can process.
Then comes the calculation. The prompt moves through a neural network, where many mathematical operations help the model estimate what response should come next. It predicts a piece, then uses the growing answer to predict another. It isn’t a person reading your question or typing a reply.
As the response is generated, it streams back to your screen, piece by piece. That quick appearance can feel like a conversation. But the answer comes from a model generating text from patterns learned during training—not from human understanding.
