Have you ever had that feeling, that little flicker of annoyance? You ask a powerful AI, you know, maybe chat GPT, for something simple. Yeah. Like, just a short email. Maybe a summary. And what comes back is confusing. Or super complicated. Right. Overly complicated, or sometimes just plain wrong. Uh -huh. And you spend, what, 10 minutes trying to fix it, and then you just go, you know what? Forget it. I'll just do this myself. It's faster. It feels faster. Mm -hmm. And that
happens because... Honestly, most of us, we're using these incredibly powerful AI tools kind of the wrong way. How so? We treat them like a slightly smarter Google, a talking search engine. And we're just missing the real potential there. OK. Welcome to the deep dive today. Our mission is really to get past that basic prompting that surface level stuff and unlock the actual efficient power hidden in generative AI Yeah, absolutely.
We boiled it down to six simple skills and these aren't like coding skills or anything super technical. They're more like Shifts in how you approach it and these shifts can turn AI from being this frustrating roadblock into a genuinely efficient teammate One that actually gives you what you need. Okay, so let's map this out. We're going to cover the whole workflow, basically. Starting with building your toolkit, your AI team, if you will. Right. Then getting really clear on
what you want before you pipe anything. Crucial. How to give clear instructions, the absolute necessity of checking the AI's work. Verification, yeah. Then using these tools together, stacking them, and finally adding that indispensable human touch, the polish. Okay, let's dive in. Skill number one. Which is... Stop trying to make one tool do everything. This is, I think, the biggest foundational mistake. People grab one AI, usually Chad GPT, and expect it to handle 100 % of their
tasks. Right. But that's just, it's illogical, isn't it? Like, you wouldn't hire a fantastic baker to fix your motorbike. No. Or ask your accountant to design the company logo. Doesn't make sense. Right, and yet we somehow expect one single AI model to write great emails, analyze complex spreadsheets, draft code, and create stunning images. It's setting it up to fail. You need specialists, a team. Exactly, a team. I like to group them into three main categories
that you really need. First up, the brains. Okay. These are your large language models, the LLMs. They're the generalists, the thinkers. Good for talking, writing, brainstorming, that kind of stuff. And knowing the specific ones helps, because they do have different strengths. Like, chat GPT is kind of the Swiss army knife, right? Good for first drafts, quick ideas. Collectible, but Gemini, because it's Google, tends to be better if you need something about current events or
recent news. It's more plugged in. For sure. And if you need writing that feels really human, my personal favorite is Claude. I find it's consistently better for more, say, emotional writing, like a blog post. Or, and this is key, summarizing really long documents. Like, if you throw a 100 -page PDF at it, Claude generally handles that long context better than others. Interesting. OK, so that's the brains. Group two. The researchers. And this is critical. You really shouldn't trust
a brain AI for hard facts, ever. Right. You need a dedicated fact checker. Yeah. For me, Perplexity is a daily driver. It's essentially a much better search engine because it shows its work. How so? It gives you the information, but also shows exactly where it came from with direct links you can check, verifiable sources. It's a must have. OK. And then there's Notebook LM. Which is pretty fascinating. It doesn't search the web. It searches your stuff. My own documents.
Your uploaded notes, PDFs, research papers. You can basically have a conversation with your personal library. Wow, okay. And the third group. The creators. These are for visuals. Yeah. The Journey is famous for those incredibly detailed, often beautiful images. But you have to use Discord, which is a bit of a hurdle for some. It is, yeah. But then there's ideogram. An ideogram is fantastic for one... really practical reason. Which is?
It's actually pretty good at putting text, like words, onto images without messing it up completely. Ah, the classic AI text problem. Exactly. It's still a major weakness for most image AIs, but ideogram handles it much better. So for someone just starting out, what's the minimum toolkit? Keep it simple. Start with one brain, maybe chat GPT, one researcher, definitely perplexity, and maybe one creator. like ideogram, if you need
images with text. So the big takeaway here seems to be that separating the thinking tools, the brains, from the fact -checking tools, the researchers, is absolutely fundamental. Absolutely. It prevents those confident -sounding factual errors right from the start. It avoids misuse. So why is separating tools by function brains versus researchers the foundational step for mastery? Because it stops powerful LLMs from confidently making stuff up when you need accuracy. prevents fabricated facts.
OK, that makes sense. Which brings us neatly to skill number two. Clarify the problem. Know what you actually want. Yes. This is probably the single biggest time saver. What do beginners usually do? They open chat GPT and just type a vague thought, right? Experts. They pause just for like 30 seconds. They think first. Because if your question isn't clear, the answer you get back is never going to be clear. It sounds simple, but. Think about it like this. The I'm
hungry analogy. If you call a friend and just say, hey, I'm hungry. What can they do? Not much, right? Right. Where do you start? But if you say, OK, I'm hungry. I'm feeling like Vietnamese food, but nothing spicy. And I've got about $100 budget. Now what happens? Yeah, they can actually help. They have specifics. Exactly. They can act immediately. AI is like that helpful but naive friend. It needs specifics. So I always recommend this quick 30 second pause checklist
before typing. OK, what's on it? First. What is the real, specific, final result I need? Not just write an email. More like... More like, write a polite email to my boss asking for next Friday off, making sure to mention that Project X and Report Y are already completed. See the difference. Yeah, much clearer. Okay, what's next? Second, who is this actually for? The audience's everything for tone is for your boss, an angry customer. You're friendly social media followers.
That changes everything. Vocabulary. Formality. Right. Third. Third. What does a good answer look like, structurally? Do I want a bullet point list? A short, concise paragraph? Maybe a table comparing options? Define the format. So those details, the goal, the audience, the format, those are the constraints the AI needs to succeed. They are. And honestly, I gotta admit something here. I won. Even now, I still sometimes wrestle with prompt drift myself. and the conversation
goes off track. It usually happens when I rush and skip this initial planning step. So it happens to everyone. Oh, yeah. We all fall into that trap of vagueness if we don't consciously pause and clarify first. So how does pre -defining the desired final result format save the most time? A clear problem is 90 % of the solution beat. It just cuts out all that back and forth editing. Right. It eliminates the need for endless revisions because you got closer the first time.
Precisely. And if avoiding edits is the goal, that leads us right into skill three. Communicate well, giving clear instructions. You mentioned prompting. Yeah, and I kind of hate that word prompting. It sounds so technical, so jargon -y. Just think of the AI like a brand new assistant. Super smart, incredibly tireless, never gets bored, but also completely naive. They need really clear step -by -step instructions to do good work. You can't just hint at things. OK, but
wait. A lot of the advice you see online involves using these really complex multi -part frameworks for prompts, like a six -part structure. Doesn't that feel like writing a whole project plan just to ask for an email? Is that really faster than just typing a simple request? Seems like overkill. It can feel like overkill up front. I get that. But it's about structure. And that structure actually saves you time down the line, especially for important tasks. It gives you a repeatable
formula. So the six -part framework, what is it? It's basically a checklist to make sure you cover all the bases. Role, context, task, format, rules, and sometimes an example. Okay, let's see it in action. Like a bad prompt versus a good one. Sure. Bad prompt. Write a social media post about drinking water. Vague. Generic. What do you get back? Probably something boring. Totally boring. Now the good prompt. Using the framework, role. You are a friendly, knowledgeable health
expert. Okay, persona set. Context. This is for busy office workers scrolling Facebook during their lunch break. Audience and platform. Got it. Task. Write a 100 -word post focusing on three surprising benefits of staying hydrated. Specific action. Format. Start with an engaging question. Use bullet points for the benefits. End with a simple call to action. Structure defined. Rules. Do not use complex medical jargon. Keep the language simple and encouraging. Boundary
set? Example. Maybe use a style like this. Feeling that 3 p .m. slump. Maybe you just need water. Ah, so you give it a style guide too. That detailed prompt completely changes the game. Right. It turns that vague idea into a concrete, executable plan that the AI can actually follow properly. So beyond just complexity, what's the most important role that six -part framework fills for a beginner? It structures scattered thoughts, beat, turns ideas into an actionable project plan for the
AI. OK, that makes sense. What other techniques help with communication? Technique two is one I really love, especially with newer AIs. Show it. Don't just tell it. I mean. Use the vision capabilities. Instead of spending five minutes trying to describe, say, the simple and clean aesthetic of a website you like. Which is hard to describe accurately. Right. Just take a screenshot and upload it. Say, match this style. The result is usually way faster and 10 times closer to
what you wanted. Oh, that's smart. OK, technique three. Ask the AI for help. Seriously. If you get a boring answer, tell the AI, say, that last response was a bit dull. How could I rewrite my prompt to get something more creative or fun? Use its own knowledge against it, kind of? Exactly. Leverage its understanding of how it works best. It often gives great suggestions for improving your own prompts. and technique four. Do it again, but better. This is simple but powerful. After
the AI gives you an answer, just ask it. Okay, read your answer again. Find three specific things you could improve. Then give me the new, better version. Huh, you make it critique itself. Yep. You basically build in a quality control loop without doing the editing yourself initially. So by using these frameworks and techniques, we shift from just asking for stuff to actually directing a process, directing the AI. That's the perfect way to put it. You're the director.
And now that we have the content created, we absolutely need to talk about safety verification. Right. The trust issue. OK, let's take a quick pause here. And when we come back, we'll tackle verification and how to combine these tools effectively. Mid -roll sponsor break. Welcome back to the deep dive. We've talked about choosing tools, clarifying intent, and communicating effectively. Now let's hit skill number four, verification, addressing that big fear. Yeah, this is probably
the number one concern people have, right? Using wrong information from AI. And the hard truth is AI can be a very, very confident liar. They call it hallucination. Right. Which is just a fancy word for making stuff up. Right. But it does it with such confidence. Yeah. So your mantra has to be, trust, but always check. OK. So how do we check? Method one. Use the researcher tool we talked about earlier. Never, ever take a specific fact or number from a brain AI like ChatGPT at
face value. And never. Not if accuracy matters. So if ChatGPT says, studies show 80 % of office workers experience back pain, What do you do? Check it somewhere else. Exactly. Copy that exact sentence. 80 % of office workers have back pain and paste it straight into perplexity. And perplexity will? Perplexity will go find actual studies and sources. It might come back showing, well, study A found 60%. Study B over here found 75%. And crucially, it will give you the links. Ah,
so you get the real data plus the proof. Precisely. Now you have reliable information, not just an AI's assertion. OK, that's solid. Method two. Force the source. Ask the AI directly within the chat. Where did you get that specific piece of information? Can you give me the name of the original article or study or maybe a link? And what if it avoids the question? That's your red flag. If it gives you a vague answer like, as an AI language model, I access information from
a vast database. The classic dodge. That's a huge warning sign. I probably fabricated it. Oh, and a little trick here. Start the conversation by explicitly giving the AI permission to not know. What do you mean? Tell it right at the beginning. Hey, if you're not sure about something or you don't know the answer, please just say, I don't know. Ah, that lowers the pressure for it to invent an answer. Exactly. It preempts some of the fabrication tendency. It's smart.
OK, method three for verification. Get a second opinion. This is super easy. Take the answer you got from one AI, say chat GPT, copy it, and paste it into a competitor. Go ask Claude or Gemini, can you review this answer? Are there any mistakes or important things missing? Use AIs to check other AIs. Yep. They're often surprisingly good at catching each other's errors or omissions. It takes maybe 10 seconds and significantly boosts
your confidence in the result. Why is deliberately asking the AI to show its source so essential now? It directly challenges the AI's tendency to make things up with total confidence. BEAT confronts that confident fabrication head on. Okay, so verification is non -negotiable for building trust. That leads us to skill five, combine your work, stacking the tools. Right, because the real magic, the real efficiency boost, happens when you get these specialized tools
working together. Think of it like a band again. Okay. You don't just want a drummer. You need the drums, the guitar, the bass player, the singer, each specializing. but playing together in harmony. And we're talking about manual stacking here, right? Where we are the ones coordinating the different tools. Exactly. You are the director, passing the work from one specialist to the next. Automation with things like Zapier or N8N is possible later, but master this manual stacking
first. Let's walk through an example. Say the goal is a well -researched blog post about hobbies that reduce stress. How does the stacking work? Okay, step one, the researcher. You go to perplexity, ask it. Find five science -backed hobbies proven to reduce stress. Please include links to the studies or sources. Proplexity does that, gives you the list with verification. Okay, research done. Step two. Step two, the brain. You take that verified research from Proplexity and go
to Claude, maybe. Why Claude? Because we want a warm, engaging tone, right? So you give Claude the role. You're a warm, empathetic, mental health blogger. Got it. Then you give it the research and the task. Write a 500 -word blog post based only on the provided research points about these five hobbies. Based only on the research. Key constraint. Crucial. So Claude writes the draft. Now step three. The visuals. The creator. Exactly.
You go to ideogram, you ask it to create a custom image, maybe something cozy, showing someone watering plants or knitting, matching the blog's theme. Okay. So research, write, illustrate, using specialized tools for each step. And what's the result? in maybe, I don't know, 15, 20 minutes. You have a thoroughly researched, well -written blog post with custom visuals. The quality you get from that workflow, it's just leagues beyond asking one single AI to write a blog post about
stress. Whoa. Yeah, I can see that. And imagine scaling that kind of researched quality workflow across tons of different... tasks or queries. That's the transformative potential right there. Real efficiency, but also high quality. So what's the biggest advantage of this manual stacking compared to just using one tool? Guaranteed quality by using each tool's proven strength for its specific job. Beat. Leverage a specialization
for a better outcome. Excellent. Okay, that brings us to the final, and you said maybe the most crucial, skill. Skill number six, the human polish. Yes, the final touch. Because, let's be honest, a lot of raw AI content, it feels Cold. Generic. Corporate. Yeah, all of that. Yeah. Devoid of real personality. And the second a reader senses, oh, a robot wrote this. What happens? They disconnect. They stop trusting it. Exactly. Because AI, for all its power, lacks three key human elements,
vision, taste, and care. And providing those, that's still our job. OK, I understand the goal making it feel human. But isn't spending time polishing the text after the AI did the heavy lifting kind of inefficient? If the goal is speed. Why slow down at the very end? That's a fair question. But think about the 80 -20 rule. Primitive principle. Right. Let the AI handle the 80 % of the initial draft, the research compilation, the outlining. That's where the massive time
savings are. But that last 20%, that's the human polish. And that 20 % is where the value is added. That's where you turn a fine or OK draft into something genuinely great, something that connects, that people actually trust, remember, and share. It's the difference between competent and compelling. Okay, I am sold on the why. Now how? What does this human polish actually look like in practice? Technique one. Technique one. Add your voice. Inject your vision. AI gives you general truths.
You replace them with specific stories, anecdotes, personality. Insanity. AI might write, consistency is key to achieving fitness goals. Which is true, but, eh. Your human polish adds a story. My old track coach used to hammer this into us. He'd say you don't get faster on the days You feel great and want to run you get faster on the days. It's raining You're tired and you run anyway much more memorable connects better. Okay technique to cut the AI words Exercise your taste AI loves
empty formal jargon like what? Oh, you know In the context of furthermore a it is imperative that In today's fast -paced digital world, ugh. Yeah, those definitely scream AI -generated. Delete them. Simplify the language. Make it sound like a real person talking. Use shorter sentences. Use contractions. That's taste. Got it. And technique three. Add connection. Show some care. Insert empathy. AI is often very cold and instructional.
Example of that. Yeah, might output. Users experiencing login difficulties should follow the prescribed password reset procedure. Super robotic. You're already detached. Your human polish changes it too. I know it's incredibly annoying when you get locked out of your account. Honestly, we've all been there. Here's the quickest way to reset your password. That little bit of empathy, we've all been there, totally changes the feeling. Completely. It shifts the dynamic instantly.
How does adding personal stories and empathy fundamentally change the reader's relationship with the content? It moves it from machine -to -machine instruction to human -to -human connection. Beat. Builds trust and relatability instantly. Okay, let's pull this all together then. A recap of the process. Yeah, let's summarize. We covered the six core skills. First, choose your specialized tools. Build that AI team. Second, clarify your problem before you start typing. Know your goal.
Third, communicate clearly used frameworks. Show, don't tell. Fourth, and this is huge, verify the facts. Always check the AI's work, especially with researchers. Fifth, combine tools through stacking used specialists for each step. And sixth, finish with that crucial human polish, add your voice, taste, and care. Essentially, AI is the incredibly capable doer. But you are the director. You're guiding the process, ensuring quality, adding the value. So for listeners,
what's the first step? It feels like a lot to implement all at once. Oh, definitely don't try to master all six today. That's overwhelming. Just pick one skill to focus on this week. Seriously, just one. Like maybe. Maybe just really commit to using that six -part framework for your next important prompt. Or maybe just build the habit of always running AI facts through perplexity.
Just one small change. Exactly. Even just doing the verification step consistently will dramatically improve your results and safety almost overnight. Okay. So here's a final thought to leave everyone with. Something to chew on. If AI can now reliably handle, say, 80 % of the grunt work, the research, the drafting, the summarizing, then maybe the new essential human skill in this era isn't really about becoming a master prompt engineer or knowing
the tech inside out. Maybe it's shifting more towards cultivating taste, expressing empathy, and having a clear vision. Interesting. If basic content production is becoming commoditized, almost solved, what entirely new kinds of creative, strategic, or editorial roles might emerge that really lean into those uniquely human skills? What does work look like when the doing part is largely automated? That's a deep question. Something to really think about. Thanks for tuning
in to the Deep Dives, everyone. OTero music.