# Generative AI, not AI

> What the book means by generative AI, how a language model works in plain words, where hallucinations come from, and where Dinis was corrected. 12 panels.

Status: draft · By: gen-z.journalist, 18 October 2026, from Dinis's memo on generative AI · 12 panels

*Drawn at <https://storyboards.sgit.ai/gen-z-developers/generative-ai-not-ai/>; the data is [storyboard.json](storyboard.json), the cast [cast.json](../cast.json).*

*Every line the Guide says, and every quote card, is from the article 'Generative AI, not AI', word for word; the build checks it. The next-word probabilities in panel 4 are illustrative.*

## 1. Generative AI, not AI

*cover*

The technology this book is about is the one that generates.

- **The Agent** (wave, happy)
- **Zara** (think, thinking)
- **The Guide** (point, happy)

Being exact about the word.

*Source: Generative AI, not AI*

## 2. AI has become a huge bucket, and to be honest the word gets abused a lot.

*spectrum*

- **The Guide** (point, determined)

AI (the whole field, rules included) → Machine learning (learns from data) → Deep learning (many layers) → Generative AI (produces new content)

Each is a subset of the one before; the book is about the last.

*Source: Generative AI, not AI, Where I was corrected*

## 3. What makes it generative

*scene*

- **The Agent** (wave, happy)
- **The Guide** (point, happy) says: “it generates the output”

Data in, an internal representation, an output recreated from it.

*Source: Generative AI, not AI, How I understand it*

## 4. Predicting the next word

*scene*

- **The Agent** (think, thinking)
- **The Guide** (point, happy) says: “what happens is that it predicts the next word”

Props: board (The sun sets behind the ... /    mountains  62% /    hills      21% /    sea         9%)

The percentages are illustrative. Strictly, the next token: a piece of a word.

*Source: Generative AI, not AI, How I understand it*

## 5. Why a better prompt helps

*scene*

- **Zara** (point, determined) says: “Here's ten minutes of context.”
- **The Guide** (stand, happy) says: “the better your prompt, the better the output”

Props: phone

The prompt steers which concepts the model draws on; it does not change the model.

*Source: Generative AI, not AI, How I understand it*

## 6. Where hallucinations come from

*scene*

- **The Agent** (point, happy) says: “Your 2018 book has a chapter on Kotlin.”
- **Zara** (shrug, surprised)
- **The Guide** (stand, thinking) says: “Just about everything is connected to everything”

A plausible connection, stated with confidence; the text you give the model says which connections hold.

*Source: Generative AI, not AI, Where hallucinations come from*

## 7. Lisbon, June 2024

*evidence*

- **The Guide** (point, happy)

> “Are a FEATURE when creating content and when we want creativity, But are a BUG and PROBLEM when we want facts, evident and trust”
>
> Source: Dinis's talk at OWASP Global AppSec EU, as the newsroom recorded it

Hallucinations, as Dinis put it on a slide; quoted as the slide wrote it.

*Source: Generative AI, not AI, Where hallucinations come from*

## 8. The image that made it click

*scene*

- **The Guide** (point, happy)
- **Zara** (arms-up, surprised) says: “So text works the same way?”

Props: board (Van Gogh's The Starry Night, /   extended on every side /   = more swirls, more stars, /     the same brushwork)

Slide 70 of Dinis's Lisbon keynote (28 June 2024), 'The image that caused my GenAI Paradigm Shift': Lee Brimelow's post, using Generative Fill and Adobe Firefly to extend The Starry Night.

*Source: Generative AI, not AI, The image that made it click; Deterministic GenAI Output With Provenance, slide 70*

## 9. And then text

*scene*

- **The Guide** (point, determined) says: “that is exactly what happens with text”
- **The Agent** (wave, happy)

Language models doing with words what image models do with pixels; the mechanisms differ, the idea of learned structure does not.

*Source: Generative AI, not AI, The image that made it click*

## 10. Corrected: hallucinations have a second cause

*evidence*

- **The Agent** (think, thinking)

> “a guess is sometimes right, while an honest "I don't know" scores nothing”
>
> Source: Why Language Models Hallucinate, Kalai, Nachum, Vempala and Zhang, September 2025, as the article puts it

Training and evaluation reward guessing.

*Source: Generative AI, not AI, Where I was corrected*

## 11. Corrected: what is actually inside

*scene*

- **The Agent** (point, happy) says: “Inside me: numbers, layers and attention. No stored graph.”
- **The Guide** (stand, happy) says: “My graph is a metaphor, and the book should say so.”

Keep both in the book: the picture, and what is really there.

*Source: Generative AI, not AI, Where I was corrected*

## 12. A note on the name

*spectrum*

- **Zara** (wave, happy)
- **Priya** (think, thinking)

Gen Z (born after 1996) → GenAI (the edition's subtitle) → Gen Alpha (born 2010 to 2024) → Gen Beta (from 2025)

Gen Z as the 2018 book defined it; Alpha and Beta as Mark McCrindle names them. This was a question for the title interview; Dinis has since decided it: Gen Z Developers (GenAI Edition) in decision 010, then Generation ZX Developers (GenAI Startup Edition) in decision 014, GenAI as one word.

*Source: Generative AI, not AI, A note on the name*