Mohamed Bedier

Most people think AI started with ChatGPT. But we've been using AI since the 90s — the difference is that older AI was narrowly focused, while today's AI can understand, create, and converse.

Traditional AI relies on pre-defined patterns and data. It's exceptional at analysis and execution, but has zero creativity. Think Google Search, YouTube recommendations, or spam filters — all pattern matchers.

Generative AI doesn't just recognize patterns — it creates. Given what it learned from billions of text examples, it generates entirely new content: essays, code, images, strategies. Tools like ChatGPT, Gemini, and GitHub Copilot all fall here.

The real question isn't which is better — it's knowing which one you're using and for what. Traditional AI excels at classification and prediction. Generative AI excels at creation and explanation.

💬 What AI tool do you use daily and couldn't live without?

Here's what no one tells you: AI doesn't "think." It plays a game. Specifically, it plays a next-word prediction game at massive scale.

When you write "I'm going to ____", an LLM assigns probabilities to every possible next word based on billions of training examples — and picks the highest-probability fit for the context. That's the entire trick. It's sequential prediction, done extraordinarily well.

Under the hood, a Neural Network does the heavy lifting. Like the human brain has neurons, AI has nodes — organized in three layers: Input (raw data goes in), Hidden layers (patterns are extracted), and Output (the result comes out). The more hidden layers, the deeper the learning — which is where "Deep Learning" gets its name.

Real-world examples: Image Recognition (Google Lens, Face ID), Self-Driving Cars (analyzing camera + radar feeds in real time), and AI match analysis in sports (tracking passes, predicting outcomes).

🤖 The magic isn't mysterious — it's statistics at a mind-bending scale.

Training GPT-4 on your personal laptop would take thousands of years. That's not an exaggeration — it's why companies spend $100M+ to do it in 3 months.

The staggering numbers: 20,000 NVIDIA A100 GPUs (each priced like a new car), running for 3 straight months, consuming enough electricity to power a football stadium for 9 years.

The training process has 3 phases: Pre-training (the model reads the entire internet and learns to predict the next word), Fine-tuning (it learns to be helpful and polite through Q&A scenarios), and RLHF (humans rank its answers and teach it which responses feel most natural — that's what makes ChatGPT feel like a conversation).

But even after all this, the model still hallucinates facts, carries bias from training data, has a knowledge cutoff, and has no real moral compass — it's pure math.

🚀 Extraordinary power. Still needs a human in the loop.

I applied to the Digital Egypt Pioneers Initiative (DEPI) knowing I wasn't technically qualified. I'd only been studying for two weeks — no grades, no transcripts, nothing to submit. I applied anyway.

Somehow, I got accepted in the first round. I passed the English test. Then came the Technical Exam — I stared at it, it stared back at me, and neither of us understood the other. I was convinced I'd been rejected.

Days of refreshing the portal. Then: Waiting List. Then, out of nowhere — Accepted. I went to register and the website crashed. Errors everywhere. The registration window was closing in hours. Forty-six minutes on hold with customer service — and I finally got my slot manually assigned.

Day one, I showed up full of energy. They told me my data wasn't in the system. I couldn't enter the class. More calls, more waiting. Eventually, I got added to the group.

I'm now officially enrolled in the Data Analyst Specialist track — by Google, under Egypt's Ministry of Communications. And every obstacle along the way is now part of the story.

وَأَن لَّيْسَ لِلْإِنسَانِ إِلَّا مَا سَعَىٰ