Creating a Cognitive AI from Scratch: A 10-Year Journey

🧠 AI didn’t start with ChatGPT. It started over 80 years ago… From simple logic → to machines that learn → to AI that creates. Here’s the journey 👇 1943 → First neural network concept 1956 → AI becomes a field 1980s → Businesses start using AI 1997 → AI beats humans in chess 2012 → Deep learning breakthrough 2020 → AI starts generating content 2023+ → AI creates text, images, videos 🚀 — 💡 What changed? AI evolved from: → Following rules → Learning patterns → Creating content — 🚨 What this means for businesses: We are no longer in the automation era. We are in the **content & attention era.** If your business is not creating content, someone using AI will outperform you. — ✨ Create. Captivate. Convert. 🌐 c3ads.in #AI #GenerativeAI #AIHistory #DigitalMarketing #C3Ads

70 years of AI in one timeline Most people think AI started with ChatGPT. It actually started in 1950 — when a 38-year-old mathematician asked a question that changed everything: "Can machines think?" That was Alan Turing. And the answer took 70 years to arrive. Here's the journey — the milestones that actually matter: 1950 — The Turing Test Alan Turing proposes a test: if a machine can hold a conversation indistinguishable from a human, it has demonstrated intelligence. We're still debating whether it's been passed. 1956 — The word "AI" is born At a Dartmouth summer conference, John McCarthy coins the term "Artificial Intelligence." A field is officially created. 1966 — First chatbot ELIZA is built at MIT. It mimics a therapist using simple pattern matching. People genuinely believe it understands them. Sound familiar? 1997 — Machine beats human champion Deep Blue defeats Garry Kasparov at chess. The world panics. AI turns out to be very good at chess — and nothing else. 2012 — The deep learning explosion AlexNet wins an image recognition competition by a margin so large, every AI lab on earth pivots to deep learning overnight. This is the moment that made GPT possible. 2017 — The transformer is invented 8 Google engineers publish "Attention Is All You Need." This architecture becomes the backbone of every major AI model you use today. 2022 — ChatGPT launches 100 million users in 60 days. The fastest product adoption in history. AI enters the mainstream conversation permanently. 2024 — The agentic era begins AI stops just answering questions and starts taking actions. Agents browse the web, write code, send emails, and execute tasks. The next chapter is being written right now. 70 years. Three winters. Three summers. One unstoppable direction. 📌 Day 2/60 — Learning Agentic AI in public. Follow for the full journey. #AIHistory #ArtificialIntelligence #MachineLearning #DeepLearning #ChatGPT #TuringTest #LearningInPublic

Choosing the right AI is not about following trends. It is about understanding what actually works for your needs. Some tools are built for real-time insights. Others are designed for deep thinking, coding, or structured problem-solving. The real advantage comes when you align the tool with your workflow. If you want versatility and reliability, ChatGPT stands out. If your focus is deep reasoning and long-context analysis, Claude is a strong choice. If you need fast, real-time information from live sources, Grok can be useful. The smartest professionals are not choosing one AI. They are learning how to use each one strategically. The question is not which AI is best. The question is which AI fits your workflow best. #AI

Why does every AI model love em dashes so much? 🤔 You've noticed it too, right? You ask ChatGPT, Claude, or Gemini something. And the answer comes back full of — Em dashes. Everywhere. Here's the actual reason: It comes down to training data. AI models learn by consuming billions of pieces of human-written text — books, articles, blogs, academic papers, Reddit threads. The writers of that content? Disproportionately educated, professional, literary types. And those writers love em dashes. So the model learns: "sophisticated writing = em dashes." And it copies the pattern. Obsessively. There's a second reason, too. Em dashes are structurally useful for a next-token predictor. An LLM generates text one token at a time, always asking: "What word comes next?" The em dash is a cheat code. It lets the model pivot mid-sentence — add emphasis — extend a thought — without needing to commit to a new sentence structure. It's the AI equivalent of "uhh" or "like" in human speech. A filler that buys time while the next token is being predicted. And the third reason: Reinforcement learning from human feedback (RLHF). When humans rated AI outputs during training, they probably gave higher scores to responses that felt polished and well-structured. Em dashes feel literary. Thoughtful. Considered. So the model learned: em dashes = good rating. And doubled down. The result? A writing tic baked into almost every major LLM. So next time you see — this — in an AI response, you're not imagining it. You're watching a statistical pattern learned from millions of writers who thought em dashes made them sound smarter. They did. Until every AI started doing it too. Do you edit out the em dashes? Or have you just accepted them at this point? 👇 #AI #LLM #GenerativeAI #MachineLearning #ChatGPT #Claude #PromptEngineering #AIEngineering #BuildInPublic #TechHumor #ArtificialIntelligence #SoftwareEngineering

🔍 Cosine Similarity — the math secretly running every GenAI app you use When ChatGPT "finds the right context", when Notion AI "searches your notes", or when a RAG pipeline retrieves the perfect document — cosine similarity is doing the heavy lifting. Words and sentences are converted into vectors (lists of numbers) by embedding models. Cosine similarity then measures the angle between two vectors — not their size. A simple example: "Machine learning is powerful" → [3, 5, 1] "Deep learning advances AI" → [4, 5, 2] "Pizza is delicious" → [0, 1, 8] sim(ML, Deep learning) = 0.98 ← very similar ✅ sim(ML, Pizza) = 0.12 ← unrelated ❌ Why not just use word matching? Because meaning ≠ keywords. "AI is transforming healthcare" and "ML is revolutionizing medicine" share zero words — but their embeddings point in nearly the same direction. Cosine similarity catches that. 📌 Where you'll see it in GenAI: → RAG pipelines — finding the right docs to feed your LLM → Semantic search — returning results by intent, not keywords → Recommendation engines — "users who liked X also liked Y" → Deduplication — detecting near-duplicate content at scale The elegance: a single number between -1 and 1, derived from a simple dot product, is powering billion-dollar AI products. If you're building with LLMs and haven't dug into embeddings + cosine similarity yet — start there. It'll change how you think about search and retrieval. #GenerativeAI #MachineLearning #RAG #Embeddings #AIEngineering #LLM #DataScience

🧠 What is a Large Language Model (LLM) — and how does it actually work? You've heard of ChatGPT, Gemini, Claude. But what's actually happening under the hood? Let me break it down simply. 👇 🔹 What is an LLM? A Large Language Model is an AI system trained on massive amounts of text data. Books, websites, code, articles billions of words. The goal? Learn the patterns of human language well enough to generate, understand, and respond to text. 🔹 How does it learn? It doesn't memorize text. Instead, it learns relationships between words. Given a sequence of words, it learns to predict what comes next over and over, billions of times, until it gets very good at it. That's it. Predict the next word. But do it at massive scale and something remarkable happens the model starts to "understand" context, meaning, and even reasoning. 🔹 What makes it "large"? The amount of data it's trained on → The number of parameters (think of these as the model's memory GPT-4 has an estimated 1.7 trillion of them) 🔹 The magic ingredient — Transformers LLMs are built on an architecture called the Transformer. It uses a mechanism called "attention" allowing the model to focus on the most relevant parts of a sentence when generating a response, just like how you'd re-read the important parts of a question before answering. 🔹 So when you type a prompt... Your text gets broken into tokens (chunks of words) → The model processes them through layers of attention and math → It predicts the most likely next token, then the next, then the next → Until it forms a complete, coherent response It's not thinking. It's not conscious. But it's extraordinarily good at pattern recognition at scale. 🔹 Why does this matter? LLMs are becoming the backbone of modern AI applications — coding assistants, customer support, search, healthcare, education. Understanding how they work puts you ahead of most people using them. Found this helpful? ♻️ Repost to share it with your network. What questions do you have about LLMs? Drop them below 👇 #AI #LLM #MachineLearning #ArtificialIntelligence #DeepLearning #Tech #ChatGPT #GenerativeAI #DataScience #LearnAI

Most people are using AI every day but can’t actually explain what AI is. Let’s fix that quickly. Because, No, ChatGPT is not AI. Claude is not AI. DeepSeek is not AI. Stay with me. ➖➖➖➖➖➖➖➖➖➖➖ What you’re actually interacting with is a product. A software interface. Now, in the backend, we have what we call a model. And that model is part of a bigger field called Artificial Intelligence (AI). ➖➖➖➖➖➖➖➖ Now what is AI. Let’s break it down simply: AI is Artificial Intelligence, yes. A field in computing and the goal is to build systems that can perform tasks that typically require human intelligence. ➖➖➖➖➖➖➖➖ We also hear Machine Learning (ML) This, is a subset of AI. Before AI came to be, we had traditional programming where we hardcoded rules, using If statements, etc. But with ML, we train models to learn patterns from data. Same with Deep Learning (DL), still a subset of AI, but this is under ML (more advanced and deep). DL uses neural networks (layers of computation) to learn complex patterns. This is where things like LLMs come in. ➖➖➖➖➖➖➖➖➖ LLMs (Large Language Models) are the actual models behind tools like ChatGPT, DeepSeek, etc. They are trained on massiveeeeeee text data from the internet to understand and generate language. ➖➖➖➖➖➖➖➖➖ So if I ask you what ChatGPT is? I expect, you to be able to say Chatgpt is a product/interface built on top of an LLM. same with DeepSeek. Not, Chatgpt is AI. You’re not talking to “AI” directly. You’re interacting with a system built using AI models. We'll talk about the roles in AI and data in coming posts. ➖➖➖➖➖➖➖➖➖➖➖ If this is your first time seeing my content: My name is Dominion. I build and solve real-world problems using data and AI If you’re interested in: • Data Analytics • Machine Learning / AI • Building real systems Do well to follow me.

🚀 Generative AI Series #6 ChatGPT. Gemini. Claude. You use them every day — but do you know what's actually powering them? The answer is — Large Language Models (LLMs). Let me break it down simply. 👇 🧠 What is an LLM? A Large Language Model is an AI system trained on massive amounts of text data — books, websites, articles, code — to understand and generate human-like language. Think of it as a system that has read billions of pages and learned patterns of how language works — and uses that knowledge to predict and generate text. ⚙️ How does an LLM work — simply? 1️⃣ Training Fed billions of words from the internet, books and code. Learns patterns, grammar, facts and reasoning. 2️⃣ Tokenization Breaks text into small pieces called "tokens". Every word you type is converted into numbers the model understands. 3️⃣ Prediction LLM predicts the next most likely word — one token at a time. That's how it "generates" a response. 4️⃣ Response Tokens are converted back to text — and you see the final human-like answer on your screen. 🌍 Popular LLMs you should know: 🟢 GPT-4o — OpenAI · Powers ChatGPT 🔵 Gemini — Google DeepMind 🟣 Claude — Anthropic 🟠 LLaMA 3 — Meta · Open Source ⚡ Mistral — Mistral AI · Lightweight & Fast 🔴 DeepSeek — DeepSeek AI · Cost Efficient ⚡ In simple words: LLM is not magic. It is pattern recognition at a massive scale. It doesn't "think" like humans — It predicts what comes next, incredibly well. Personal note: I have personally worked with LLaMA 2, Gemini Pro & Mistral — fine-tuning them on custom datasets using LoRA & QLoRA techniques. Understanding how LLMs work under the hood completely changed how I build AI applications. 🎯 💬 Question for you: Which LLM do you use the most in your daily work — and why? Drop it in the comments 👇 🔔 Follow Sheikh Ebada for weekly posts on Generative AI, Agentic AI & Data Science. Next Post #7: What is Transformer Architecture? #LLM #LargeLanguageModels #GenerativeAI #ChatGPT #Claude #Gemini #ArtificialIntelligence #MachineLearning #DataScience #TechTrends #LinkedInCreator

Everyone says AI. Almost nobody uses the word correctly. Day 2 of the LLM series. Let's fix that. Think of it like Russian nesting dolls. The biggest doll is AI. Artificial Intelligence. Any machine doing something that looks smart. Your spam filter is AI. A chess engine from 1997 is AI. A thermostat is... arguably AI. Inside that doll is ML. Machine Learning. Instead of programming rules by hand, you let the machine find patterns in data. You show it 10,000 cat photos and it learns what a cat looks like. Nobody told it "cats have whiskers." It figured that out. Inside ML sits DL. Deep Learning. ML but with many layers of pattern recognition stacked on top of each other. Like going from sketch to shading to full portrait... each layer adds detail. And inside DL... the smallest, most specific doll... is the LLM. A Deep Learning model trained specifically on language. Text in, text out. Not all AI is ML. Not all ML is DL. Not all DL is an LLM. But every LLM is all four. Most people skip this and go straight to prompting ChatGPT. That's like driving a car without knowing what an engine is. You'll get places... but you won't know why it breaks.