Lesson 1 · VIDEO
What Generative AI Actually Is
Approx. 5 minutes
Learning objectives
- Explain, in plain terms, what generative AI is and how it differs from traditional software.
Audio-first lesson
Listen to What Generative AI Actually Is
The narration follows the approved lesson script. The full transcript remains available below for accessibility and study.
Technical explanation
Large Language Models (LLMs) are neural networks trained via next-token prediction across massive text corpora. Given a sequence of tokens, the model outputs a probability distribution over possible next tokens and samples from it — repeated until the response is complete.
Real-world example
Typing the start of an email and having autocomplete finish your sentence is a tiny version of the same underlying idea, scaled up enormously.
Visual learning aid
1.1.1 — What Generative AI Actually Is (and Isn't)
Comparison graphic (two-column process diagram)
A
Looking something up.
B
Predicting likely words.
Same-looking question. Very different process.
Make the "prediction, not lookup" distinction concrete by contrasting two side-by-side processes.
Accessible description: "Diagram comparing two processes. Left side, labeled 'Looking something up': a query goes into a database and returns an exact stored result. Right side, labeled 'Predicting likely words': a prompt goes into a pattern-based prediction engine, which generates the most likely next words one at a time, shown as the words 'The,' 'Eiffel,' 'Tower' appearing in sequence. Caption: the same-looking question involves very different underlying processes."
Full lesson text used for narration
Learning objectives Explain, in plain terms, what generative AI is and how it differs from traditional software. Technical explanation Large Language Models (LLMs) are neural networks trained via next-token prediction across massive text corpora. Given a sequence of tokens, the model outputs a probability distribution over possible next tokens and samples from it — repeated until the response is complete. Real-world example Typing the start of an email and having autocomplete finish your sentence is a tiny version of the same underlying idea, scaled up enormously. Knowledge check **Q1 (Apply).** What does an LLM predict at each step? - The next most likely token (word/word-piece), based on patterns learned from training data. - Skip the lesson's verification or validation step. - Treat the first result as automatically correct. - Remove human review even when the task is important. *Explanation:* This answer matches the lesson's stated knowledge check. **Q2 (Apply).** Why might the same prompt produce two different answers on two different runs? - Response generation involves sampling from a probability distribution, so outputs can vary run to run. - Skip the lesson's verification or validation step. - Treat the first result as automatically correct. - Remove human review even when the task is important. *Explanation:* This answer matches the lesson's stated knowledge check.