Chapter 9: Creating AI

What he had planned was nothing less than creating a chatbot, an artificial intelligence that had once dominated his previous world.

The intelligence of ChatGPT and other AI models had been beyond imagination, capable of performing tasks that once required human minds.

Since it replaced writers that was once thought to be a uniquely human skill was now performed by AI with ease. 

The tone was no longer robotic or dull; eventually, AI could write full novels in minutes.

Writers, painters, and many other artists who once believed that AI could never replace human creativity were proven wrong.

AI had only been given five years to evolve, yet in that short time, tools like Midjourney and DALL·E showed that illustrators could be replaced.

Copywriters, translators, data analysts, animators, voice actors, and even musicians all saw their roles begin to shift or disappear.

The world had advanced so far that even Elon Musk, the CEO of SpaceX, had warned about the accelerating rise of artificial intelligence.

Musk had even built a massive spaceship designed to inhabit Mars, making the world feel like something straight out of a sci-fi novel.

In his previous world, companies no longer needed large teams of employees. 

With the help of AI, an entire company could operate efficiently with only a handful of people or none at all.

The leading giants in artificial intelligence back then were OpenAI, the creators of ChatGPT, followed by others like Google Bard and DeepSeek. These technologies reshaped every aspect of life.

It disrupted entire industries. It redefined the meaning of creativity. Artists were replaced by AI-generated visuals created in seconds, and writers, designers, even coders, found themselves competing with lines of code.

Now, having spent a long time in this new world, he noticed something shocking, not even the most basic chatbots had been developed.

And now, in a world untouched by such technology, he intended to build it from scratch, this time, with no competition.

It was literally a land full of potential, untouched by digital evolution. To him, it was a blank canvas.

So without wasting any more time, it was already around 8:00 pm. He started his work. 

"Since I'm creating my very own version of chatgpt, I will name it Razon Labs" he decided.

With that, he got to work. The first step was project initialization. He created a new folder and named it:

 RAZi_v1.0 – Razon Artificial Intelligence – Zero Interface

Just as GPT stood for Generative Pre-trained Transformers, his system 'RAZi' stood for Responsive Autonomous Zero Interface, developed under the name of Razon Labs, of course.

It was his first step toward introducing true artificial intelligence to a world that had never seen it before.

He opened his terminal, spun up a Phyton virutal environment and began to build its foundation.

If you ask why Jeff chose Phyton, if theirs a language like Java, C++, Rust. Because the answer is simple.

It was the language of AI.

The world's most advance machine learning frameworks, like PyTorch, TensorFlow, Transformer, were all built around Phyton.

It cleans syntax making him code much faster, debug quicker, and focus on logic rather than boilerplate.

And for someone like him, who had already understood how models worked, Phyton let him move at the speed of though.

So he began by isolating the project from the rest of his system using a virtual environment.

It was a dedicated space, separate from global installations, where he could install only the tools and dependencies that RAZi would require to ensure stability, efficiency, and complete control over the development process.

He ran: 

…

Bash

mkdir razi_v1

cd razi_v1

python -m venv env

…

This created a clean environment with no interference from global Python packages, no version conflicts. Just pure control.

…

Bash

env\Scripts\activate

…

Inside this space, he could install libraries, test models, and debug safely.

Next, he created the project structure:

…

Bash

touch app.py # Flask server will go here 

mkdir models # To store model-related code 

mkdir templates # For the future interface 

touch requirements.txt # To track all dependencies 

…

touch app.py — this is where the Flask server would live, the heart of the backend.

 mkdir models — a folder to store all model-related code, from data processing to AI logic.

 mkdir templates — reserved for the future user interface, a space for HTML files and frontend design.

 touch requirements.txt — to track every dependency, ensuring the project could be replicated or deployed anytime, anywhere.

In just a few keystrokes, the skeleton of RAZi was ready. A clean, modular structure that is easy to build upon.

But it's not done yet, but its bone was formed now. 

Inside the razi.core.py, he began scripting out the foundation of his own transformer, a lightweight, attention-based architecture capable of predicting the next word from a given sentence.

"Everyone else is writing if-else rules," he muttered. 

"I'm going to make something that learns what to say on its own."

He initialized tokenization logic, positional embeddings, multi-head self-attention layers, things no one in his world had even written papers on yet.

Back in his previous world, this level of architecture was known as GPT-2, but he gave it a name of his own 'RAZiCore-117'.

A 117 million parameter transformer, built entirely from scratch. No pre-trained base. No borrowed weights. 

Just his own logic, hand-curated datasets, and a series of optimization tricks only someone with godly programming skills could devise.

Unlike OpenAI, he didn't train it on the entire internet. Instead, he fed it carefully.

He used a massive dump of forums, conversations, books, and code he had collected, each file chosen for its quality and contextual richness.

The data had been meticulously cleaned and preprocessed using his own custom scraping and formatting scripts. 

Every duplicate, noisy line, and irrelevant entry was filtered out with precision.

[Hello, I hope you are having a good day] 

"Hmm' now that's fast" Jeff muttered as he cracked his neck.

He then ran his fingers to the keyboard to ask if its functioning correctly. 

[How to build a nether portal?] – He asked.

[…] – RAZi typing.

After a while a reply was sent.

[What's a nether portal?] 

Jeff narrowed his eyes, a hint of doubt creeping in.

He wondered if his program had an error or if the output was just a lucky coincidence, he decided to test it again.

[You know the Nether Portal from Minecraft Mojang] – he asked again.

[…]

[I'm sorry, but there's no such thing as a Nether Portal that is found in my database. However, there is a game called Minecraft, but it's developed by Razon, not Mojang. Here's the source: (Link to his game on Itch.io)]

Reading the text and realizing something, he laughed at his own forgetfulness. 

He had completely forgotten that Minecraft didn't exist in this world yet.

If it weren't for him, there truly would be nothing like it. But this realization answered an important question, his code worked.

But it has its current limitations:

Like memory, it doesn't retain previous messages in a conversation. Every response is generated based solely on the current input. 

So, if he says "hello" followed by "how are you," RAZi won't remember the first message unless Jeff manually provides both.

Next is external knowledge access. It doesn't have access to the internet or real-time data, meaning it cannot search for or retrieve up-to-date information beyond what it was initially programmed with.

It can't Google things, check current events, or fetch updated information. All of its replies come from what Jeff trained it on locally.

Lastly, there's the graphical interface or rather, the lack of one. 

Currently, his AI runs on the terminal Python shell. There's no chatbox, no browser page, and nothing resembling ChatGPT's interface.

Glancing at the time, it was 10:23 p.m. He decided to stop and get some rest, since there was no real urgency in building his own AI.

He had class tomorrow in order to not be tired the next day, so he will not finish it the whole night.

[A day has passed. One spin has been given to the host.]

[Do you want to spin?]

Jeff nodded his head, and the system began to spin. The wheel turned before finally landing on a reward section.

[Congratulations, you have won "2x Earnings."]

[2x Earnings: Earn twice the rewards from any work related to making money.]

Description: The bonus activates every day at 12:00 AM (midnight) and lasts for 24 hours. After the day ends, the bonus resets and will wait to reactivate again the following midnight.]

Seeing this, a smile bloomed on his face, his luck wasn't bad after all. He had expected to win a bit of money, but this reward was even better.

With that, he shut down his laptop, took a quick bathroom break, and went to sleep for the night.

As he drifted into slumber then he's full sleep, elsewhere to another country, where the sun was high in the sky.

Inside a room filled with gaming gear and posters covering the walls, a laid-back gamer sat comfortably in a sleek gaming chair.

"Hey, what kind of game is this?" the guy asked, curiosity in his voice.

For more advanced chapters, please visit my Patreon: patreon.com/Aphelious