OK, so imagine for a second, you go out and you buy a Ferrari. It's sitting in your driveway. Italian leather, the engine is purring. It's capable of these incredible speeds. Right. But every single time you get in it, you never shift out of first gear. That's a painful image. You're just putting around the neighborhood at 10 miles an hour. But I mean, according to the data we're looking at today, that is exactly what is happening with artificial intelligence. Is that right?
Over 90 % of paying users, so people actually dropping $20 a month, have never even opened the settings menu. Wow. Or driving a supercar in first gear. I have to admit, when I saw that stat, I felt a little called out. I'm definitely guilty of just opening the window and typing, assuming the defaults are good enough. We all are, I think. Welcome to the deep dive. Today, we're analyzing a guide called Mastering the Interface. But this isn't just a tech review,
is it? It's more a conversation about how we interact with intelligence itself. It is, yeah. And the mission for this deep dive is really moving from being a casual user to a power user. Right. And the most interesting part, I think, from the source material is that you don't need to learn to code. You just need to know where the buttons are. which is fortunate because my coding skills are nonexistent. So looking at our roadmap, we're going to move through this
logically. We'll start with the foundation. So the privacy and memory settings that I think most people are just afraid to touch. Then we'll move to the art of the prompt. How to structure your language so the machine actually understands you. Right. And then we'll cover the productivity tools that turn it from a toy into a real workspace. And finish up with the advanced stuff. Vision, web browsing, and these new agent capabilities. It's about moving from a passive consumer to
an architect of the tool. Exactly. So let's unpack this. We start with the foundation. The source material calls this fixing the foundation before you build the house. And it starts with something that I think generates a lot of anxiety. Privacy. Right. There's this lingering question. people have, like, is the AI watching me? Is it reading my diary? Yeah. And the reality, according to this guide, is that unless you change a very specific setting, OpenAI is using your chats
to train its future models. So if I'm uploading a confidential business plan or, I don't know, venting about a relationship. It's potentially going right into the mix. Yeah. to make the model smarter for everyone else. See, I think the reason people don't touch this stuff is fear. They think if they mess with the settings, they'll, like, break the bot. Of course. But the guide is pretty specific here. It's actually really simple. What
does it say? It's under data controls. You go to settings, then data controls, and you just turn off the toggle that says, improve the model for everyone. Just a switch. A simple switch. But there has to be a trade -off, right? If I turn that off, does the tool get worse for me? Not immediately. No. But you're essentially opting out of that collective learning process. You
kind of become a ghost to this system. For most people, though, especially in a business context, that trade -off is absolutely worth it for the security. That makes sense. Secure the perimeter first. But then, once you've done that, you have to look at how the AI remembers you. This memory feature. I've had moments where it brings up something I said three weeks ago, and it feels a little, I don't know, a little black mirror. It can be jarring, yeah. But the guide points
out this isn't magic. It's just a database. And you have control over it. What do you mean? You can actually go into the settings under personalization and view the specific list of things it knows about you. So I can see a list that says, like, user likes sci -fi novels or user is a vegetarian. Exactly. And if there's something wrong, let's say it thinks you're a lawyer because you asked one legal question three months ago. Right. You just hit the little trash can icon next to it
and delete it. You can even wipe the whole memory if you want a fresh start. That's useful because, I mean, I still wrestle with... what I call prompt drift. I'll be working on a podcast script and suddenly it starts giving me advice like I'm a Python developer because of something I asked yesterday. Right. It feels like I'm constantly wrestling the steering wheel back. That's the context pollution problem. Yeah. And managing that memory list is exactly how you clear the
air. Okay. So that leads us to the about me section or custom instructions. Now I've seen this in the menu, but I usually just ignore it. Why does the source material put so much weight on this? Because it solves the problem of repetition. Without this, every single time you start a new chat, you have to establish context from scratch. It's like having to introduce yourself to a friend every time you meet them. Hi, I'm Dave. I like
podcasts. Exactly. Custom instructions lets you do that introduction one time, and it sticks. There are two boxes. Box one is, what should chat GPT know about you? Okay. And box two is, how would you like chat GPT to respond? The guide had a really clever example of this. Something about a junior English teacher in Vietnam. Yes. In their instructions, they specified that they teach 10 -year -olds. And crucially, in that second box, they ask for short sentences, no
academic jargon, and please... Use bullet points. And once that's saved, every answer is automatically tailored for a 10 -year -old. You don't have to keep typing. Explain this simply. It just does it. It creates a consistent persona. It stops the drift. There's one more feature in this section, temporary chat, the so -called incognito mode. Right. Be honest. Does anyone actually use this? Or is it just for when you have, like, embarrassing medical questions you
don't want in your history? Well, the medical questions are a totally valid use case. But it's also for those one -off queries. You don't want polluting your memory. Like what? If you're asking for a recipe for a party, you don't necessarily want the AI to remember you love gluten -free cupcakes forever. You toggle on temporary chat, the screen turns gray, and nothing is saved.
It's the burn after reading feature. OK, so looking at these foundational settings, memory, custom instructions, it does raise a question for me. If we automate our context so heavily, do we lose the ability to be flexible in how we ask things? That is the tension. If you hard code your instructions too strictly, you might just create an echo chamber where the AI only talks to you in one specific way. So, rigidity versus convenience. Rigidity versus convenience, okay.
So, we've built the foundation, the settings are locked in, now we have to actually talk to the thing. Right. Section two of our source material is the art of the prompt. And this is where most people struggle. They blame the tool when they get a generic answer. But usually the problem isn't the AI, it's the input. The source compares it to learning to ride a bike. It suggests there's a perfect prompt structure, four parts. Roll, context, task, and format. If you miss one of
those, the bike wobbles. Okay, let's role play this because I want to see if it actually makes a difference. Let's say I own a coffee shop. The bad way would be me just typing what... Write a post about coffee. Exactly. And the AI will give you something so incredibly bland. Coffee is great. Come drink it. It's useless. OK, so fix it for me. How do I use the four parts? OK, so roll. Act as a social media expert. Got it. Context. I own a small coffee shop in Hanoi that
specializes in coconut coffee. OK, specific. Task. Write a Facebook post for our summer launch. And then format. Give it an exciting tone, use three emojis, and keep it under 100 words. Okay, wow. I can see how that would produce a wildly different result. The AI knows exactly what success looks like. You're not making a guess anymore, you're giving it a map. And for listeners who really want to geek out on this, the source mentions Dr. Jules White's course on Coursera. Oh, yeah.
He frames this perfectly. He says language is now a programming interface. You're coding with English? You're writing code just without all the syntax errors. So does the structure of the prompt then matter more than the creativity of the idea behind it? I would argue that in this medium, yeah. The format is what unlocks the function. You can have a brilliant idea, but if you can't structure the request, the intelligence can't manifest it for you. Format essentially
unlocking the function. That's a good takeaway. Okay, let's move to section three, daily productivity tools. This is about turning the AI from a fun toy into a real workspace. And the first tool here is called projects. The best analogy for it is smart folders. This one resonates with me because I hate context switching. Oh, yeah. I'll be asking the AI to help me write a serious
email to a lawyer. But five minutes earlier, I was asking it to generate names for, like, a D &D character, and the tone just bleeds over. It's so annoying. That is exactly what Project Solves. You create a folder, say, Legal Work. You can upload your previous contracts or your company's style guides into that project. Oh. Now, every chat inside that folder is walled off. It knows the context is law, not... Dungeons
and Dragons. And speaking of finding things, the source mentions the new search feature, which seems obvious, but it wasn't there for a long time. Windly. The little magnifying glass icon, it turns your chat history from this messy endless scroll into a searchable library. You just type meeting or recipe and it pulls it right up. It turns your past interactions into a database you can actually use. Now, I want to talk about voice mode. The source describes this as talking
like a human. I've used this while cooking, and honestly, the thing that surprised me most was the interruptions. It's startling, isn't it? You can just cut it off mid -sentence and it adjusts, it handles laughter, it understands cadence. It totally changes the feeling of it. It's not typing at a computer anymore. It feels like a presence in the room. It creates a new modality, brainstorming while driving, practicing a language while you're doing dishes. It frees
the intelligence from the screen. Another productivity feature mentioned is file uploads. The scenario in the source is a classic one. Your boss sends you a 50 page PDF report. Right. You upload it and just say, summarize this. And seconds later, you have the top five points in your action items. But here's my skepticism. Do you trust it? I mean, if I upload a dense legal contract or a complex medical study, am I really going to trust a three second summary? That is the critical
question. The source material frames this as a massive time saver. And it is. But there is a danger. We are trading date comprehension for efficiency. We're outsourcing the reading. Right. And if you outsource the reading, do you lose the nuance? I think it's amazing for triage just knowing if a document is worth your time to read deeply. OK. But I wouldn't let it sign the contract for me. That's a fair distinction. Triage versus
decision making. But just, I mean, pause for a second and imagine the scaling implications. even if it's just for triage. You handed a 50 page dense report and seconds later, you know it's in it. It just changes the velocity of information flow inside a company. It definitely speeds up the treadmill, but does it make us better runners? That's the question. Okay, we're back. We've covered the settings, the prompts, and the daily tools. Now let's unpack the really futuristic
stuff. Section 4 covers advanced vision and what the source calls the doer era. Right. We are moving beyond just text. The source discusses multimodal prompts. This just means the AI can see. And the examples here are so practical. Taking a photo of your open fridge to get a recipe. Or taking a photo of a broken bicycle part because you have no idea what it's called. You just ask, how do I fix this? It bridges that gap between the physical world and digital information. Then
there's canvas mode. This one feels less like a chat and more like, I don't know, like Google Docs. It's a dedicated workspace for writing and coding. So instead of the back and forth chat bubbles, it opens a document window on the side. So you can work next to it. Exactly. You can highlight a paragraph and just say, make this shorter. Or highlight a sentence and say, check grammar. It becomes a side by side writing partner. And we can't forget web browsing. Browse
with Bing. That one's crucial for relevance. Without this, the AI is stuck in the past. Its training data has a cutoff. With browsing, it can check stock prices or current news. It connects the brain to the live internet. Now the final frontier in our source material is customization and agents. This feels like the biggest shift of all. It is. We are moving from the AI being a talker to a doer. Can we make that concrete? Because agent is such a buzzword. What does an
agent actually look like in 2026? Okay, imagine this. You don't just ask the AI how should I plan a dinner date. Right. You tell it, plan a dinner date for me and my wife this Friday. The agent then checks your calendar for when you're both free. It goes to OpenTable and finds a reservation at an Italian place you've liked before. Then it drafts the calendar invite for your wife. It doesn't just give you advice. It executes the task. It connects to the apps. To
Google Drive, your calendar, your email. Exactly. It's editing the files, not just talking about them. So what does this all mean for us then? If the AI starts doing rather than just talking, Does my role change? Fundamentally, yes. You shift from being the creator, the one actually typing the email, to being the manager. You're just supervising the results. So I become the editor -in -chief of my own life. That's the ideal, yeah. But it requires a different skill
set. You have to be really good at spotting errors and really good at giving clear, concise instructions. So to recap the big ideas here, we started with the importance of settings, privacy, memory to stop that drift. We move to projects that hold our context, canvas for co -writing, and finally these agents that actually do the work. And if we look for the thread connecting all of this, it's customization. The tool is only as powerful as the settings you actually decide to turn on.
That hidden menu isn't just about features. It's about moving from a passive user to an active architect. So here is the challenge for you listening right now. The source material is very clear on this. Don't try to use all 17 features we talked about today. You'll just get overwhelmed. Just do one thing. Go to settings, go to personalization, and open custom instructions. Spend five minutes, just five minutes, telling the AI who you are and how you want it to speak to you. It is such
a high leverage five minutes. Once you do that, every single future conversation you have with it improves. Stopped driving the Ferrari in first gear open the menu. Thanks for diving in with us today. Always a pleasure. See you next time