Module 1 of StartCloud's AI at Work: Foundations learning pathway, in five short units with a knowledge check: how AI tools generate answers by prediction rather than lookup, the AI already in everyday software, what these tools are genuinely strong and weirdly unreliable at, and the assistant mental model that makes them useful at work.
What AI at Work Actually Is (Without the Hype)
It predicts, it does not look things up
Before you use any of these tools well, there is one idea worth getting straight, and it only takes a couple of minutes. Tools like ChatGPT and Copilot do not look up answers. They predict them.
During training, these systems read a staggering amount of text: websites, books, articles, forum posts, more than any human could get through in a thousand lifetimes. From all that reading, they learned patterns in how words follow other words. When you ask a question, the tool is not consulting a database of facts. It is writing the answer that is most likely to come next, one chunk of words at a time, based on everything it has read.
You ask a question. The tool searches a giant library of checked facts, finds the right entry, and reads it back to you. Like a search engine, only politer.
You ask a question. The tool writes the most likely answer, word by word, based on patterns from everything it read during training. There is no library being checked.
A likely answer and a correct answer are usually the same thing, which is why these tools feel so clever. But not always. And when they differ, the tool writes the wrong answer in exactly the same fluent, self-assured tone as the right one.
Fluent is not the same as right. Once you know the smooth writing comes from prediction rather than fact-checking, a lot of AI behaviour that seems baffling suddenly makes sense. It also makes you a much harder person to fool, which is where this whole pathway is heading.