How a chatbot writes an answer

How does a chatbot write its answers? It's all about predicting the next word. Chatbots are built on something called a large language model. It learned from huge amounts of text, like books, websites and articles. When it writes, it predicts which word should come next, adds it, and does it again, one word at a time.

What comes next?

Let's see. If you read the words, the sky is, which word comes next? The model gives each possible word a score. Blue scores highest, so it's the most likely choice. A word like falling is very unlikely.

Another example

Here's another one. Jollof rice is, what? Delicious, spicy and popular all score well, because they often appear after words like these. Purple almost never does. These numbers are just examples, to show the idea.

New word: Large language model

A large language model is an A I trained on huge amounts of text, to predict the next word. It's the engine inside a chatbot. It's called large because it learned from so much text, and because the model itself is enormous. Here's the important part. The model predicts words that sound likely, and that isn't always the same as true. So sometimes a chatbot makes up facts, names or numbers that sound right but are wrong. This is often called hallucination.

Quick check

Quick check. How does a chatbot choose its words? It predicts a likely next word, adds it, and repeats, one word at a time. Options: It predicts the most likely next word, again and again; A person types every answer; It copies one website exactly; It picks random words.

Recap

Chatbots run on large language models. The models learned from huge amounts of text. They write by predicting the next word, again and again. And likely isn't always true, so check the facts.

I've read this lesson