The many kinds of AI
AI is a broad landscape, and large language models are just one corner of it. A quick tour of the whole family, and why this curriculum then zooms in hard on the one corner that changed everything.
Before we go deep, a moment to zoom out. "AI" is one of the broadest words in technology, and that's the first thing to get straight. Generative text models get all the headlines, so it's easy to think "AI" means "ChatGPT." It doesn't. AI is a whole landscape, and most of it has been quietly running in products you use every day for years. A short tour so you can place any AI product someone shows you, and then we'll explain why we spend the rest of this section on just one part of it.
The main territories
- Generative AI — creates new content: text, images, audio, video, code. The LLMs we've focused on live here. This is the part that exploded in 2022.
- Predictive / analytical AI — the older, quieter workhorse. Forecasting, scoring, recommendations. Your credit score, a churn prediction, a sales forecast: all this. (We compare it head-to-head with generative AI on the next-door page.)
- Computer vision — recognizing what's in images and video: face unlock on your phone, defect detection on a factory line, "find all photos of dogs."
- Speech — turning speech into text (transcription) and text into speech (the voice your phone talks back in).
- Recommendation engines — predicting what you'll want next. Netflix's home screen, YouTube's "up next," your social feeds. Enormous business value, almost invisible.
The everyday point: AI is already everywhere. Search results, spam filtering, photo tagging, recommendations, these have been AI for years. Generative AI is the loud newcomer, not the whole family.
Narrow vs. general AI
One more distinction worth holding, because it's where a lot of hype lives.
Narrow AI is built to do one thing (or a bounded set of things). Everything that actually exists today is narrow AI, even the impressive general-feeling chatbots are, under the hood, narrow systems doing token prediction very well.
General AI (often "AGI") is the hypothetical idea of a single system that can learn and reason across any domain like a human. It does not exist today, and whether or when it will is a genuinely open, speculative debate. When someone talks about AI "becoming conscious" or "outsmarting humanity," they've quietly jumped from the narrow AI we have to the general AI we don't. Spotting that jump is a real literacy skill.
Where the business value is today: narrow AI doing one job well, not a single all-knowing system. Match the tool to the job and you'll be right far more often than the people chasing one model to rule them all.
So why focus on one corner?
If AI is this broad, why does the rest of this section zoom in almost entirely on large language models? Because they're the corner that changed everything in the last few years. The other territories, vision, speech, recommendations, have been quietly improving for a decade. LLMs are what made AI suddenly feel general, conversational, and useful for almost any task, and they're the engine underneath the products you're here to understand, Agentforce included.
So the move from here is: we've seen the whole map, now we go deep on the part that matters most for your work. Everything that follows, what a model is, how it learns, what it can and can't do, is really about this one remarkable corner.
📝 Practice
List three AI features you used this week without thinking of them as "AI" (autocomplete, a recommendation, photo search, spam filtering all count). For each, name which territory above it belongs to. AI gets a lot less mystical when you notice how much of it you already rely on.