Lesson 1 of 5 · Fundamentals
AI predicts. It doesn't look things up.
AI chat tools write by predicting the most likely next words, based on the text they learned from. Likely isn't the same as true, especially on topics few people write about. Test it on something you know well.
Try this today
Ask an AI chat tool about something you know really well: your hometown, your job, your hobby. Count what it gets wrong, and notice how sure it sounded.
Read the full lesson
What it is
Behind every AI chat tool is a model: a program trained on a huge amount of written text to predict which words usually come next, then tuned to follow instructions. Think of your phone's autocomplete, scaled up enormously.
Why it matters
Because it predicts, it can sound smooth and sure while being wrong. Where lots has been written, like a common recipe, answers tend to be steady. Where little has been written, like a local rule or a small town, answers get shakier but sound just as confident.
Key points
- Type "Happy" on your phone and autocomplete may offer "birthday". It isn't checking a calendar; it's going by patterns. AI chat tools do the same thing at an enormous scale.
- The model learned from a huge amount of text, its training data, and was then tuned to follow instructions.
- Predicting isn't looking up. Some tools can also search the web and show their sources, but the writing part is still prediction, so open those sources too.
- Lots of text on a topic means steadier answers. Little text means shakier answers, delivered just as smoothly.
- Building a model takes huge amounts of computing power, electricity and water to keep the machines cool.
- Chatbots aren't new: one called ELIZA held simple typed conversations back in 1966. What's new is how natural today's tools sound.
Words to know
- AI chat tool
- A program you type or talk to that writes back in everyday language.
- model
- The trained program behind an AI chat tool that predicts the next words.
- train
- To teach an AI by feeding it lots of examples, sometimes including people's chats.
- autocomplete
- Your phone suggesting the next word while you type.
- training data
- All the text an AI learned from before you ever typed to it.
- chatbot
- A program you can have a typed or spoken back-and-forth with.
Sources
- Models like these are trained to predict the next word, then often tuned to follow instructions. Large language model - Wikipedia (opens in a new tab)
- Some AI tools can search the web and show direct citations for what they found. Claude web search now available globally on all plans (Anthropic) (opens in a new tab)
- Training AI models uses large amounts of water, including for cooling. Making AI Less "Thirsty": Uncovering and Addressing the Secret Water Footprint of AI Models (Li et al., arXiv:2304.03271) (opens in a new tab)
- ELIZA, an early chatbot from MIT, dates to 1966. ELIZA - Wikipedia (opens in a new tab)
Prefer to talk it through?
JdAI is our voice guide. Talk or type your way through this lesson, at your own pace.
Try it with JdAI