Thinking Like ChatGPT - Universal Prompting Mental Models for Any AI

1. Introduction: Have you ever pasted one of those so called viral prompts and gotten mid results? Today, I'll show you why some prompts work and others flop and how to think like AI, so prompts, work across Chachi PT, clawed, gemini, ama, and more. Learn on, apply everywhere. This class is the thinking layer. It ties together all the practical prompting and applied AI work. Once you understand this, every other project you tackle with AI will make more sense. The tone, here's a line from Andrew Karpathy. The hottest new programming language is English. And to make this course practical, meet our avatar, Chris, he's a marketer slash creator who struggles with prompts. In each lesson, we'll see Chris's before and after, so you can watch him level up alongside you. And before we dive in, if you'd like to go further after this class, I have an extensive beginner friendly off Skillshare courses on covering everything from content creation to ecommerce and agents. Check those out if you want to build on what you learn here. And as always, feel free to reach out if you have any questions. Ready to dive in? Great. Let's jump 2. Class Project: Okay for our class project, here's what I would like to work on. Pick one of Chris's before prompts or choose one of your own real prompts. Rewrite it using at least two of the mental models you learn in this course. Run both the before and after prompts in an AI tool of your choice and compare the results. Share your before versus after in the project gallery. It's simple but powerful. You'll see for yourself how these mental models transform your results. Keep Chris's examples in mind. His journey will mirror yours. All right. Now, let's get started with our first lesson, a brief history of LLM. 3. Lesson 1 — A Brief History of LLMs: Et's zoom out for a second. How did we even get here? Why are we suddenly talking about prompting as a skill? Back in the 1950s, Alan Turing asked the question, Can machines think? He came up with the Turing test. If a computer could chat so well that you couldn't tell if it was a machine, maybe it could be called intelligent. Fast forward to the 1980s through the early 2000s, and AI was basically if this, then that. These were called expert systems. They were useful in narrow cases, but brittle. The situation and the system fell apart. Came 2017. Google researchers published a paper with a now famous line. Attention is all you need. This introduced the transformer architecture, a breakthrough that lets models understand relationships between words across entire sentences and paragraphs. It was like giving AI a new brain. From there, progress snowballed. Google released B ERT or Bert. Open AI released GPT too, then GPT, trained on Internet scale data. These models weren't searching. They were generating, predicting the next word again. And by late 2022, Chat GPT wrapped this in a friendly chat interface, and suddenly millions of people discovered that they could talk to AI, and prompting became the steering wheel. Chris, our fictional avatar, used to think AI was just like Google Search. But after learning this history, he realizes AI doesn't fetch facts. It generates patterns. That's why how you frame the input matters so much. Let's jump into the next lesson, how AI thinks. 4. Lesson 2 — How AI Thinks (Plain English): Large language models don't think like humans. They don't understand the world. They predict the next word, then the next, then the next. Think of your phones autocomplete, only trained on billions of examples. Sam Altman, CEO of Open AI explained this theory. Large language models are not databases are facts. They are pattern completers. Means when you prompt them, you're not querying a search engine, you're setting up a pattern for them to complete. Our character, Chris, before he typed write me a viral TikTok script. He got random generic fluff. But then after he now writes, You are a TikTok strategist. Your task generate a 30 to 45 second script for a portable brander aimed at UK students. Step one, hitch three hook options. Step two, outline the script structure, and step three, draft the full script in a playful. The result is a sharper structured and much closer to what he needs. Makes sense? Alright. Now let's jump into the next lesson. 5. Lesson 3 — Hierarchy Beats Chaos: Hierarchy beats chaos. The first mental model hierarchy is always better than chaos. When you dump everything into one messy prompt, the AI has no idea what to prioritize. Anthropi says it best. Be clear and direct. Break tasks into sequential steps. Google says the same. Assign a role, set a goal, and give instructions step by step. Our character, Chris, before wrote like this, help me write a blogpost about Tik Tok Shop, give me an SEO title, a full draft, make it fun, at a conclusion. The result a wall of text, off tone and completely unfocused. After now he writes like this. You are an SEO blog editor. Here's your goal. Draft a 1,200 word post on TikTok shop for beginners. Step one suggests five SEO titles. Step two, outline the H one and H twos and step three, draft only the intro and then stop. The AI doesn't drown anymore. It works in stages like a real editor. The takeaway, think roll goal, steps. That's hierarchy. Okay, let's move on to the next lesson. 6. Lesson 4 — Broad → Narrow → Examples: The second mental model funnel the AI's thinking. If you stay broad, you'll get fluff. If you go narrow too quickly, you kill creativity. The trick is broad first, narrow, second, then lock it in with examples. Andre calls examples the Rosetta Stone of prompting. Show the model what good looks like, and you will follow. Before knowing this, here's how Chris typed his prompts. Make me a funny tweet about coffee. Sometimes you got funny, sometimes you got cringe, no con but after realizing this, Chris now writes task, write five witty tweets about morning coffee struggles. Step one, brainstorm six relatable scenarios. Step two, pick the two funniest step three. Write the tweets in this style. My Coffee and I are in a committed relationship. Alarm is the third wheel. Poor over more like poor over thinking. This funnel makes the output sharp and reliable, and the takeaway, don't just type the wish. Funnel it broad, narrow, and then examples showing what TPT or whatever model you're using, what good looks like. Does that make sense? Perfect. Now let's move on to the next lesson. 7. Lesson 5 — Context Stacking: Welcome to Lesson five Context stacking. The third mental model, think of AI like a new colleague. If you explain everything from scratch each morning, you waste time. If you onboard them once, they become more powerful. Fe fee, put it best. AI is like a new colleague. The more you onboard it with context, the more useful it becomes. Our man, Chris, before realizing this was prompting like this. Write me a promotional email for my course. He got generic text with the wrong tone completely. Now that he realizes this model, here's how he prompts. Brand voice, conversational UK spelling, short sentences, audience, TikTok shop sellers, style examples, two or three sample lines, formatting rules, headings, bullets, and one takeaway per section. Then he asks the following. Using the saved context, draft a 150 word promo email with a subject line, short hook, three bullet benefits, and a call to action. Now, every output sounds like his brand. The takeaway onboard your AI once, reuse that context over and over again. Okay? Sounds good. Let's jump over to the next lesson. 8. Lesson 6 - Iteration Is The Key : Okay, welcome to Lesson six. Iteration is the workflow. This is now the fourth mental model. Too many people treat AI like a vending machine. They type once, they take the first draft and walk away. But AI works best in dialogue. Reid Hoffman once said, working with AI is a dialogue, not a command. Our main man, Chris, before he realized this, would prompt his LLM the following way. Write me a business plan for my Tik Tok show. He got a generic plan, side, and then he quit and gave up. After he realized he now breaks his prompt into two stages, Step A, draft the problem and solution only. Step B, pause and ask if you'd like to refine. Step C, propose three revenue models with pros and cons. Step D, combine everything into a one page plan. The output is sharper because he's steering each step. Having now applied this mental model, here's an example of what the output from Chachi PT might sound. UK E Commerce brands on Shopify stores generating 30 120,000 per month often struggle with TikTok at performance. Their in house teams are overstretched and freelance creators can be inconsistent. This leads to wasted at span, poor CTR, and campaigns that stall before scaling. Brands lose momentum because they can't keep up with TikTok's fast paced trends and short shelf life content cycles. Solution. Our UGC Studio delivers ad ready TikTok videos in under 48 hours. Each piece is designed with proven hook formats and tailored to TikTok shop products. By offering consistent creative output, quick turnaround, and knowledge of TikTok shop policies, we help brands lower CPA, increase CTR and refresh ads before fatigue sets in ChatPT with an ask. Would you like to refine or continue? But in all honesty, just that initial draft speaks exactly to what Chris was asking for. The takeaway, Don't expect magic. Build it section by section, refining as you go, right? Hope that makes sense. Let's jump into the next lesson. 9. Lesson 7 — Transferable Thinking Across Models: You're welcome to Lesson seven Transferable thinking across models. So here's the big question. Does all of this only work in hat GPT? The answer? No, these mental models are transferable. Demi Hasais, CEO of Deep Mind, put it simply the future isn't one AI model. It's an ecosystem of them. Before realizing this, our character, Chris, would have thought he'd have to re learn prompting for every tool, but afterwards, he now realizes that he can use the same framework anywhere, whatever tool he's using. His prompt looks like this. You are a TikTok ad strategist. Your goal is to create a 62nd AD script for a collagen face mask. Aimed at UK university students. Step one, write three hooks. Step two, outline the ADA structure. Step three, draft the script in a playful tone, constraints UK spelling under 160 words. Examples, late nights, bed diet, meet your new skin BFF. Claude might lean, thoughtful. Gemini might create neat outlines. Chachi PT might balance tone and structure, but the framework works everywhere. Takeaway, dont learn hacks, learn habits that travel. Okay? Are you with me so far? I hope all of that makes sense. Now, let's jump on to the next section. 10. Wrap-Up: So let's recap. We started with a quick history of how LLMs came to be. We learned that AI doesn't think it predicts. We explored our four mental models, hierarchy, funnels, context, and iteration. We saw that these apply across any model you'll use. Our man, Chris, went from typing vague one liners and getting junk to writing structured, reliable prompts that work everywhere. Jensen Huang, CEO of Invidia, sums it up. AI amplifies human potential, but only if we learn how to think with it. That's what you've learned in this class. Hopefully, this course is part of my wider Skillshare library of AI Learning. Each one builds on the others, but they also stand alone. Feel free to explore those once you have digested this course. Thank you for learning with me. Any questions you've got, feel free to reach out. Go ahead and put these mental models into practice and share your project so we can all learn from each other. I'll catch you in the next one.