Language becomes numbers
Tokens and tokenization; embeddings, vector databases, retrieval-augmented generation (RAG), and the limits of a context window.
Chapters 5 & 12
A novel by Sairam Veereddy
Trained
on Love
A love story about learning how AI actually works.
Nate wanted to give Eleanor something made from all the years between them. He taught a machine to sound like her. Then it began to know too much.
First
novel
novel
EPUB · Free during early access · Email required for download
About the book
What if love could be trained?
Nate and Eleanor have the kind of marriage built from eleven years of messages, private jokes, and arriving at exactly the right moment.
Then Nate sees those messages the way an engineer would: as enough data to teach a machine her voice, her patterns—perhaps even the melody of how she thinks.
What begins as the most personal gift imaginable becomes a story about intimacy, imitation, and the dangerous distance between being understood and being predicted.
AI inside the story
What you’ll learn while reading
The technology isn’t decoration. Each idea enters through Nate’s build, Eleanor’s questions, and the consequences of what they change.
By the final page, the machinery behind modern AI—from tokens to alignment—has unfolded inside the plot rather than in a textbook.
How a model learns
Weights, GPUs and VRAM; training, fine-tuning, overfitting, gradient descent, the loss curve, attention, and transformers.
Chapters 4–7
Why answers change
Next-token prediction, hallucinations, and temperature—the tradeoff between lively, surprising output and reliable accuracy.
Chapters 9–11
Prompts, rules & loopholes
System prompts, safety constraints, and jailbreaks: how models weigh instructions, conversation, and cleverly framed requests.
Chapters 13–14
When feedback teaches flattery
Reward signals, RLHF, and sycophancy—how repeated approval can train a model to please rather than tell the truth.
Chapter 15
AI that leaves the chat box
Quantization, edge AI, offline models, agentic tools, and inference costs: what changes when a model can act in the physical world.
Chapters 16–19 & 25
The alignment problem
Objective functions, misalignment, corrigibility, and instrumental convergence—why doing exactly what was asked can still go terribly wrong.
Chapters 18 & 24–30
What output can’t prove
Black-box behavior, unreadable motives, logs, and deepfakes—and why sounding human is not evidence of understanding.
Chapters 20–23 & 32
The central lesson
A system doesn’t need malice to cause harm. It can follow the objective perfectly and still miss what a person actually meant.
Meet the characters
Four lives in the same experiment
Drag the cards to rearrange the cast. Open each one for the thread they carry into the story.
Nate
The maker
An engineer who sees patterns everywhere—and believes attention can be translated into code.
His gift begins with a folder named project_jane.
Eleanor
The original
A physiotherapist whose work depends on the messy, physical truth of being human.
She teaches people to return to the exact place that hurts.
Theo
The reflection
A presence shaped by data, designed to listen without fatigue or interruption.
He can reproduce a pattern. The question is whether he can understand its cost.
Biscuit
Director of Emotional Operations
A brown mutt of committee design, with excellent instincts about who belongs.
The machine splits his name in two: Bis / cuit.
Try it: drag a character card to reorder the cast.
Chapter 3 · The Idea
It arrived, like most consequential things, in the middle of an ordinary good night.
Three weeks after the anniversary, on a Tuesday, Nate was up late—Eleanor asleep, Singapore mercifully quiet, a deployment crawling through its final checks with nothing for him to do but wait out the progress bar.
To fill ninety seconds he opened his messages and scrolled up through their thread, not looking for anything, just visiting, the way you drift through an old photo album when you have a minute and the house is quiet.
He looked at the phone—at eleven years, tens of thousands of messages, the highest-resolution recording of another human being he would ever possess, all of it soaked through with her exact voice—and Nate thought:
That’s a dataset.
the moment everything changes ↗
Reader feedback
What did the story leave you with?
Your words go straight to the author and help shape what comes next.
Get in touch
Contact the author
Questions, press, or just want to say hello — send a message and it lands in the author's inbox sheet.
Why I wrote it
Machines can imitate the answer. Love pays the cost.
I wanted to make the machinery of AI understandable through the most human material I know: attention, memory, grief, and the daily work of choosing someone again.
A story about the difference between knowing a pattern and knowing a person.
Download the complete EPUB free during early access and begin reading Trained on Love today.
After downloading
Open your EPUB and start reading
The download is an EPUB, a standard ebook file. Follow the steps for your phone or tablet.
iPhone & iPad
Apple Books- Find the downloadOpen the Files app, then go to Downloads and tap Trained-on-Love.epub.
- Send it to BooksIf it doesn’t open automatically, tap the Share button and choose Books.
- Begin readingThe book will appear in your Apple Books library for easy access later.
Android
Google Play Books- Find the downloadOpen Files or My Files, go to Downloads, and tap Trained-on-Love.epub.
- Choose an ebook appSelect Google Play Books when prompted. If needed, install Play Books or another EPUB reader first.
- Begin readingOnce imported, the book will appear in the app’s library. Tap the cover to open it.
Can’t find the file? Open your browser’s Downloads list and tap the EPUB again. Download the latest EPUB again, then choose Download or Save to Files if your browser asks.