AI Basics: Learn These Tips First

[00:00:00]

Matt: All righty, everybody. This is not an edition of the Web News. This is a whole brand new episode of the HTML All the Things podcast. In this episode, we're gonna be talking about AI basics. Learn these tips first. So what this is, is it's designed to be an evergreen episode, so hopefully, and, you know, AI changes things, the tech world changes, but hopefully you'll be able to listen to this in the future.

No guarantees, of course. We can't predict the future. Listen to this in the future and still get some helpful tips if you wanna learn some AI basics, and I wanna be clear that we're gonna be touching on some, some code stuff of course, but this is primarily, like, an AI basics episode. So think of it as someone who still uses sort of the traditional Google search.

They haven't really touched AI at all. They haven't talked to Copilot. They haven't, you know, talked to Gemini or anything like that, ChatGPT. This is what this episode is designed... This is who this episode is designed for, and we're hoping this will be helpful into the future

Mikhail: I just jump in actually for the first [00:01:00] time ever in your, in your hook? I, I think the way that I wanna kind of frame this episode to our regular listeners who are maybe the developers and maybe have already played around with AI quite a bit, I want this episode to be something that you can recommend to your spouse or your friend, and, uh, that, that's maybe, like, AI hesitant or just hasn't had time to look at AI. And I want this episode to stand in that regard of, like, "Hey, I, I don't have time to, you know, help you with AI. Here's ... Listen to Matt and Mike. They'll get you started at least," right? So that, that's the mindset that I want everyone to come in with. So listen to this episode. That'd be great. Give us some feedback on it, but really we would love it if you would just, like, pass us off to the people that y- need the help in the basics of AI

Matt: It's a good way to put it, honestly, because, like, the, the entire point of this is absolutely just to sort of get you spun up. And then again, also no guarantees, but we do have an episode planned for, like, AI basics for developers, AI basics for small business owners, things like that, or people working in small business.[00:02:00]

We have those episodes kind of planned currently. We'll see how, you know, things go of course. But th- this is, like, a good jumping off point. But If this sounds interesting, interesting to you and you wanna support the show, you can go and check us out on that Patreon, leave a review or rating on your podcast app.

Join us in our Discord server, or share this with your friends. And if you are someone who actually wants to learn how to code, and you wanna learn how to code using some courses, you can use Scrimba's interactive media player code editor, and also get up to 20% off their Scrimba Pro plan. That link will be in the show description, as well as in the show notes, with full details on how it works in the show notes on htmlallthethings.com.

Again, 20% off that Scr- that Scrimba Pro plan, up to 20% off. Go and check that out. And so now that the, the preamble's over, the hook is over, let's get into the meat and potatoes of this episode. Uh, I do also wanna say that we have some pretty beefy show notes that are on htmlallthethings.com if you wanna kind of either follow along or you wanna refer back if you are a complete beginner and you're learning, and you [00:03:00] wanna, you know, refer back and, and, and read through some of the things that we've done without having to scan through the episode.

There's, uh, again, some pretty beefy show notes on htmlallthethings.com for this episode. So I, I wanna, I wanna just kind of have an intro for this episode, which is, like, right for the beginners, and I think that as a blanket statement, as a very generalized statement, A, the best way to learn AI is to actually use it.

Now, that doesn't mean that you, you know, should stop listening to this episode and then, like, run over and start messing around with ChatGPT or whatever. But in my opinion, you know, the best way to learn AI is to actually familiarize yourself with how to use it and see how these tools work. The, the reason why I say that is because there are...

Every single time I talk to somebody, I talk to Mike, I talk to a colleague, I talk to a friend who's not in tech at all, I talk to my girlfriend who's, who's, you know, certainly not in tech and doesn't even have a computer for that matter. They... Ev- anyone I talk to, every time the conversation goes to AI, I find out that they're [00:04:00] prompting AI, they're using AI in a unique way.

"Oh, I'm using it to schedule this thing," or, "I'm using it to manage this," or, "I'm using it to manage my, uh, my, my schedule for, for work and, and my meal prep," or, "I'm using it for this," or, "I'm using it for that." And those are some just scratching the sort of surface. There's some really niche use cases where it's like this person, you know, wants to bike to work, but they have to, like, correlate it with the weather or something, right?

And you can... A lot of data points, and people will use AI for stuff like that. And so I would say try to get exposure to a lot of the different AI services available. But don't run rampant and just sort of run and, and, like, try every single one, uh, every day. You're gonna get overwhelmed. But some, some kind of keystone ones or some main ones, at least right now, are ChatGPT.

You got your Claude, Copilot, um, Gemini as well, and then others as you come across them. If, if you look at them and then you're like, "Hey, I, I wouldn't mind trying it," many of them will have a free tier, maybe a free trial or a [00:05:00] reasonable Sort of like a bronze paid plan that you could maybe pay for one month and then, you know, try it out if you're, if you, if you'd like and, and, and learn that tool.

But in addition, I think that you should also be not only just trying different AI services, I think you should be trying AI in different places. So try it in the web app. So, like, ChatGPT, for example, is, has, uh, various form factors and many others will too, but try the web app, try the desktop app, try mobile apps, um, try it built-in.

What I mean by that is built into existing software. So, like, Copilot is built into Word. Try it in there because it's working in a different context. See what it could do for your Word, Word document. But also, in addition, try to experiment with different ways of interacting. Sometimes these form factors are gonna have different ways to interact, and you're al- you're already gonna be interacting with it differently depending on the, the, again, the form factor, the web app or the mobile app or the, the [00:06:00] built-in.

But I think you should also try things like not only texting back and forth with it. I think you should be trying t-traditional text chat, as well as trying the different voice modes, the advanced voice mode or natural voice mode or whatever tool you're using. They call it all different things. I think that you should be trying by interacting with it differently.

So, like, for example, if something's gone, uh, awry on your computer, some people will try to text it out, and they'll try to explain the whole thing over text. Try to take a screenshot, assuming it's not an, uh, a very secure thing you're doing. Try to take a screenshot, and if you're comfortable with, you know, whatever's in the screenshot, upload that to ChatGPT and, like, ask it, right?

"Hey," you know, "what do you, what do you think about this?" Or if you have, like, a PDF, let's say it's a, a, an event coming up, you're going to, like, a Christmas bazaar or something like that, and you're like, "Man, I got a lot of events coming up at Christmas," maybe you run a vendor... maybe you're one of the vendors at, like, a, at local, local Christmas bazaars, [00:07:00] "Hey," like, you know, "this is a big deal for me."

You could maybe feed ChatGPT those PDFs which are gonna be the flyers to those Christmas bazaars, and it can maybe organize an itinerary for you or a schedule, or you can start planning things out, things like that. The point is, is to not just rely on the text chat. It's, "Hey, what can I do with files that I see?

What can I do with images that I have?" Things like that. And then I would say after you've kind of deal- dealt with all of this, I know I'm throwing a lot at you, mess with maybe some agentic stuff for more complex tasks. 'Cause agents can kind of I guess a really beginner way to describe it is they can run off and do something for you.

So you might be like, like for example, I, uh, didn't have a, a plugin installed on one of my versions of VS Code. It was on one of my laptops, and I was just like, "Ah, shoot, I forget the name of it." So I told it, I was like, "Ah, you know, I, I usually use this plugin." And it was like, "Oh, it's this." So I was like, "Yep, that's it."

I was like, "Can you just handle that for me?" It goes in, installs it, does whatever, and it's like, "Hey, make sure you restart VS Code. Go into the settings and just like check this thing," 'cause it's a [00:08:00] security thing that it can't do. Go in and check this thing, and away you go. So that's an, a very beginner, like big agent task that it could do.

Like, it's just installing a plugin for me. But the idea there is, is like you're not just having a text or verbal conversation with the AI. I think a lot of people get to that level where they're just kind of having a conversation. They don't really use the file sending. They don't really go beyond it just being kind of like a knowledgeable person that they're talking to, if that makes sense.

Mikhail: I- I've told this story a few times now, but, um, I'll tell it again here since it's the beginner episode. Essentially, when I was helping my dad with something on his computer, pulled up or ChatGPT, and I started, like, having it analyze logs, having it do things on the computer, and he was like, "What are you doing?"

Like, "What, w- what is this?" Like, "When I use ChatGPT, I just ask it like a Google Search question." And I'm like, "No, like you treat these agents as if they were agents." Like, you don't have to anthro- [00:09:00] amorphitha- am- morphitize them, whatever, make them pretend that they're human or anything like that. But what you can do is you can kind of pseudo pretend that they're like these little robotic agents and almost talk to it like you would an assistant. That's kind of, at least for a beginner, that to me is the best way to think about it, especially once you start getting to the agentic stuff. But even before the agentic stuff, like when you're just asking it to do things, you can talk to it like an assistant and like in a, you know, search for flights kind of deal.

You're not, you're not saying, uh, you know, uh, "What is the, like, you know, the, the best fl- flight for X and Y?" You would say, "Hey, can you find me the best flights based on this, this, and this?" And then, you know, a- ask me questions along the way. Like, treat it like you were to ramp up an assistant that you would have, right?

And the more you kinda give that context to it, and the more you feel like it can do more for you, you'll start to naturally start to ask it different things that you wouldn't ever ask a Google search, and then it'll start to [00:10:00] click. Before you using these things, and I think that's why it's really good that you put that at the s- top here is like, just use it. Before you start using it, you really do kind of box these in as a Google search, and I really want people to break out of that box a little bit because even though it can do Google searches, who cares? Google has been doing that for us forever. This is a different set of tools. This can act on your behalf. This can do more complex... Like, it can do Google searches for you, multiple Google searches, 10, 100, 1,000 to find the right set of information for you. It can parse information. Um, there's just an unlimited amount of functionality that you can unlock as you start to chip away at the understanding that it has and the limitations that it has, right?

Like, you do need to understand both sides of it, but that's hopefully gonna help you and w- like, and it helped my dad to see me doing that because the two weeks after I showed him how I use ChatGPT, [00:11:00] he came back to me and he's like, "That's how I've been using it, and it's increased like 2x my workflow," like for, for work and stuff like that, for support tickets and all that.

So if you're, if you're out there listening and you already know this stuff, a really good way to teach someone is just showing them how you use it right now.

Matt: I think, I think that, like, you're, you're 100% right in that too, because it's like your dad, at the very least, got started, which is kind of like what I was talking about,

Mikhail: Mm-hmm.

Matt: then there... You... He had enough sort of meat and potatoes, if you will, of a conversation with you where you could be like, "Hey, hey, hey, let's take what you're doing," and, like, kind of steer it, and that's, that's what's gonna happen here.

And I... And it, it, it highlights an important point here is that there's a heck of a lot of AI offerings right now. I mean, we listed four or five there. There's a heck of a lot of, like, form factors, your desktop apps, your mobile apps, and then there's ways to interact with it. But the thing is, is you don't have to learn it all at once, and, and that's important.

Like, it's important that your dad started with just a bit of texting back and forth. Then he was able to, like, "Oh, yeah, I understand where to go," or, "I know where to reach for Copilot," or whatever he was using, [00:12:00] ChatGPT. "I, I know where to reach for it. I know, like, that the prompt is there," and he's familiar with it.

But then when he gets in those conversations, like how I was saying, I, you know, I'll converse with people in or outside of tech, then someone like you can be like, "Hey, I actually use it for this," and you guided him. But even if there was someone there, like let's say it wasn't you, Mike, and y- and someone was there, and they mentioned that use case to him, that would probably stick with him regardless.

And it would be, "Oh, I, I can do that?" And that might stick with him to the point where he would start experimenting himself. The point here is, is that we're still human. We can't just, like, learn everything all at once. I keep mentioning AI bottleneck. You're not gonna learn every single mode, form factor, offering, us- u- use case, utility, prompting method all at once.

Not gonna happen, like period, not gonna happen. Stop trying to do that, period. It's not gonna work for you. What you really should do is exactly kind of like what Mike said. It's like more organic. "Hey, I'm gonna play with this a bit. Oh, I'm getting some utility out of this. [00:13:00] Okay, now I'm gonna have a conversation with somebody else."

Or maybe if you don't have someone to converse with, you could go online and look at how people are using it on Reddit, how people are using it. Like, you... There's tons of articles out there. Uh, you could watch even the, uh, presentations. If there's an AI presentation, they talk about the new use cases, right?

And then you could be like, "Oh, I could mess with imagery, i- images," or, "I could mess with document manipulation," or whatever, right? Again, it gives you ideas, and then you kind of... There's no standard yet, I think is the point right now. There is no sort of AI standard. There's some things being standardized here or there, right?

Maybe some agentic coding stuff or whatever. There's some frameworks in place and things. But for the most part, it's like I'm gonna use AI differently than Mike's gonna use AI, than, like, another team member would use AI if we were all on the same team. There isn't like a, "This is the way to use it," right?

So my, my favorite method, and Mike, you can correct me if I'm wrong here, but my favorite method to, like, cut through the noise and start learning obviously is to start tinkering, like we me- m- like we mentioned. [00:14:00] But once I'm kind of past the tinkering stage and I'm like, "Okay, like I kinda get it," right?

Even... It could even be just a couple of hours into using AI, and not necessarily consecutively sitting there texting for two hours. Could be over a week. You kind of play with it for, for a couple hours. Then I say, if you want to start using this thing, pick a goal and complete it, here's the important thing, with one AI tool.

One AI tool. So you don't go in and go, uh, let's say you have a... And I know people hate, hate the AI-generated posters, but I'm just gonna use it as an example. Like, a small business wants to have, like, a, an event, and they're like, "Hey, like I'm gonna have a, a flea market. I'm gonna have a flea market on Saturday.

I wanna have a poster made up." Don't go in and be like, "Oh my God, I have to like go to Claude or go to Copilot and then go to Gemini and ChatGPT and like try all these different things." In my opinion, try to choose a goal and complete it with one AI tool. What that allows you to do is you go in there, let's say it's ChatGPT, and you say, "Okay, I wanna do the [00:15:00] complete task with this, with this service.

So we've chosen ChatGPT. I want you to like, you know, brainstorm what, like what you want on that poster. I want you to like n- maybe research the audience that's involved. I want you to go in and like let's start maybe designing some stuff." Learn how to, like, let's say the first result that you get for a poster, like the actual image isn't what you want.

Learn how to edit the image. You will learn if... And this is less of a worry nowaday- nowadays, but in one singular conversation, if the bot starts to mess up the context a bunch, like if you're really getting into it, then you go, "Okay, hang on. Let me take my files out, my, my various versions of my poster, start a new chat, dump it in."

The point is, is you're learning this stuff, right? You're learning, oh, this thing could potentially lose context. If it doesn't give me the results I need, it's not a bad, necessarily a bad tool. I can change it. I can go in and I can modify it. How do I change it? How do I prompt a change? Oh, [00:16:00] shoot, I told it to like change it to dark mode and it, and it went way too far and took all the bright colors out.

Okay, now I know I need to kind of prompt around that. And the point is that you'll naturally start to learn things. How much context does this bot need? What does it understand? What does it misunderstand, right? What features are useful?

Mikhail: since this is a beginner episode, Matt, can you explain context a little bit? Like what context means in terms of AI?

Matt: That's a good point. So, uh, let's just say, for example, uh, Mike and I are planning, uh, this podcast is a, is a good example. Mike and I are planning this, this, this podcast, and so we go, "Okay, uh, we wanna have, um, a beginner episode on AI." So there's a bunch of stuff we could, we could talk about, and Mike and I can be in the same chat.

That's another kind of advanced feature, but Mike and I can be in the same chat. So him and I are both prompting ChatGPT. The reason why I wanted to bring us both into this example is because we're gonna be prompting it like crazy. "Hey, like, should we talk about this? Should we [00:17:00] talk about that? Let's get some numbers."

We can tell it to research things like, "Hey, how many... What's the percentage of people that haven't tried AI yet? What's the percentage of people that have? What's the percentage of people that only use sort of the consumer free models like Copilot or Gemini or things? Or what's the percentage of people that only use AI overviews?"

And th- that's just the tip of the iceberg. We could really, really do a lot of research if we, if this episode was stat-based. And then the thing is, is at that point, you're... You've prompted... You could have prompted 200, 300 times in that chat back and forth, back and forth. There's been some mistakes, right?

The bot gives you the wrong stat. You're like, "No, no, no, not that stat, this stat," and you're, like, going back and forth, back and forth. You'll start to notice that the bot will start to lose it a little bit, where it starts to, like, kind of forget what we're talking about. And so it'll be like, "Yeah, AI is used all the time."

And I'm like, "No, no, no, I didn't ask about if AI is used all the time. I asked you if, like, AI is used by this, like, this demographic, or, like, if it's used in this country a bunch or something like that." And it'll be like, "Yeah, AI is used in the US all the time." And it's like, "No, no, I asked about the UK."

[00:18:00] It'll start to mess things up. It'll, like, forget little bits of its memory. Like, so for example, if Mike and I have, like, set up all these stats, and then we're like, "Okay, now let's start to shape the episode, and we're gonna s- like generate a PDF," which is not what we do, but let's just say we generate...

we're gonna generate a PDF show note. It might mess up the stats. It'll, like, forget. It'll be like, "Oh yeah, the UK uses it all the time. Uh, UK uses a lot of AI." And it's like, "Whoa, whoa, whoa. We got you to, to research and generate, like, eight statistics. What the heck?" Once it starts to do that, sometimes then you have to go into a new chat.

It's kinda like having a, a conversation with somebody, and it's just gone on too long, and you're like, naturally your mind's gonna start to wander. Heck, if this episode is too long, for example, you might start to, like, forget elements. Now, thankfully, you can replay it, but the thing is, is if this was a live chat, you start to, like, forget those things, right?

Like 40 minutes ago, somebody was talking about s- the stats of, like, AI usage in the UK, and then right now it's like, [00:19:00] like, I forgot that we even had that conversation, right? Like, your mind starts to wander. So that's what context means with AI. I don't know if you have a, a better way to describe that, um, or a better example maybe, but

Mikhail: Yeah, I th- I think that's a pretty down-to-earth way to do it, like, uh, for someone that's just first getting into this. Essentially, it's just like whatever information you give the AI, it gets added to its context, and each AI has a limited amount of context it could keep in its memory essentially, in that, in that bubble, the context bubble let's say. as soon as you get to a point where you're almost full, the context is almost full, you'll s- you'll also notice a degrading performance. So you'll notice that the AI is just acting weirdly and it's making more mistakes because it just has to... Every, every time you prompt it, every time you ask it something, it has to go through everything you've ever, you've ever mentioned and decide what is relevant and what isn't. And more there is, the more likely it can make a mistake there. [00:20:00] So the, the rule of thumb is to like, to keep conversations as, you know, short as possible. But lately, and if you're listening to this in the future, that context stuff has s- slowly been getting resolved where these apps especially are really good at being like, "Oh, this is too much.

I'm compacting it. I'm gonna put it into a m- Like, I'm gonna put all of this massive context into just one small little file." And it can make mistakes there too, but that allows it to be a lot more accurate as you go. So it's-- I would say for a beginner, don't worry about it too much when you're going into it.

Don't worry about overloading your context. Don't worry about that too much. As you advance, you might start noticing where you like the conversation to be and like in terms of the amount that you, y- you've talked to it, and you'll naturally pick up on that and go in and start new chats. So for now, just ignore it. Talk to it as if, as if it was a person with a long memory, and see how that works for you

Matt: Especially in 2026. I mean, when we started a few years ago, the original Copilot [00:21:00] was, like, limited to, like, 12 messages. It was like a hard limit, 12 messages per chat, 15 messages per chat. It went down to eight or, like, four at one point. It went back up and down as they were, like, messing with models and features and things like that, and performance, they were kind of dialing it in.

But now, especially if you're having sort of a, what I call, like, a simpler chat where you're not bringing a lot of points in and stuff, it, it doesn't really come up, and then usually your issue is kind of resolved. Like, if you're trying to set up an itinerary, your itinerary has been, like, set, solved. It's in a PDF.

They made an... And they made an image for you, and you're, you're done with that chat anyway. You're not gonna be going back 'cause, you know, that you're gonna do the itinerary, and then that time is gone. So you know, it's, it... Th-then that chat's gone. That context is gone anyway. But it, it, it's still absolutely something to bring up because you might notice, like, oh, like, in this case, like, ChatGPT is kind of getting stupid here, or, like, something's really messing up here, and it's like, oh, it might just be...

It... I could really see it, like, especially for a beginner, actually, I just realized something, is, like, somebody... There's definitely somebody out there that is probably trying to keep all of their chats in one chat,

Mikhail: Yeah,

Matt: like [00:22:00] texting a person where it's like, "Hey, do all my recipes, but also this is my job, and I also want to run, and I want an exercise plan, and I want..."

And then it, like, the bot's like, "What are we doing?" And it's just like a person, where if you talked to me all day, you know, starting at the beginning of the day talking about, like, meal prep, and then at the end of the day, we're talking about quantum physics, it'd be like, "What are we talking about? Like, what did we have for breakfast, or what are we making?

Or did we skip breakfast?" You know? It's just like a, like you said, Mike, like a person with a long memory or a big memory.

I think all of this, like, as you start to use these bots and as it... as we've kind of discussed so far, you're gonna start to realize that prompting, meaning the actual things that you're writing or saying or providing to the bot, is the new Google search. So we have an episode from years ago, I don't know if I'll include a link 'cause it's so old now, but th- if you, if you're a fan of, if you're a fan of, uh...

Like, back in the day, if you were a fan of, like, being really efficient at Google, it was a skill. Like, you would know [00:23:00] how to search, and you would... If you wanted to research something, you knew to, like, kinda have a big zoomed-out search, and you would, like, word things a certain way and then kinda, like, zoom in slowly and whatever.

You wouldn't want your search to be too specific 'cause it would cut out results you would want, but you wouldn't want it to be too broad because then you would have too, too, too many results where it's like, whoa, whoa, whoa, like, I'm not even going to that country, or I'm not driving that type of car, or I'm not even into that sport.

Like, what's going on here? So Google search was a skill. Prompting's kinda just like that. Again, it's just a new type of thing where you're, you know, you're talking to the bot, and it's a skill that needs to be sort of learned and practiced, and everyone's gonna have their own, you know, u- unique flair, right?

Overall, though, to kinda boil it down, over time, you're, you're naturally gonna learn, you know, which keywords matter, telling it to do this and specifically not do this if you don't want something touched, for example, in an image. Or if you're like, "I like [00:24:00] this stat," or, "I like this, uh, section of the chart, don't touch it.

Do another chart," or, like, "Mess with this other column," stuff like that. You know, setting up some keywords, like, keywords or key phrases to say to get the result you want. Um, you'll also naturally learn how specific you need to be, and you'll also learn that, in my case, I, I'm noticing that I need to be less specific.

The context around it really helps, especially with newer models. I'm able to just sort of, like, fire off a lesser detailed quote than I, I was a year ago and get the same result, if not better, with the newer models. You'll also learn how to rephrase a bad search. So I've absolutely prompted something, and then it goes off and it's doing something else, and I'm like, "Whoa, whoa, whoa.

Okay, shoot." Like, I missed a keyword in there or I missed a key piece of context. Let me, you know, stop this and, like, you know, re- resteer it. You'll also recognize quirks. Um, so I gave it what I knew yesterday was a good prompt. I was just looking at some, [00:25:00] some numbers, and it, its response was yes. And I, like, hadn't asked it...

Like, I gave it this huge thing and was like, "Hey," like, "what do you think about, like, th- this stat, this stat, this stat?" And I'm like, "We'll move this in this table," and it was like, "Yes." And so I was like, "Can you elaborate?" And then it goes, "Oh, sorry, that yes wasn't a real response." And then it, like, went off and, like, actually responded.

And that was Astra. That's the latest model right now. Um, so it still has those little moments, but the point there is I knew my prompt was good 'cause of the experience, and I knew instead of, like, redoing all my prompt and, like, kind of worrying about the prompt, I just, like, went in and was like, "Hey." Gave it, like, gave it, like, a little, like, push and like, "Hey, man," like, "answer my, my, my, my prompt."

Um, you'll also have... You'll also start to recognize useful results. So you'll start to r- like, go through things, and you'll start to realize, like, okay, like the, the things that it's giving me is useful, or let's say it's giving you things that aren't useful. So for example, uh, sometimes I'll, like, go through and I'll, I'll [00:26:00] need, like, a, a summary of how something works, and it'll just start just piling information in, in paragraph form.

And it's like, whoa, man, if I wanted to read the manual, I would've read the manual. Stop. So I'll literally come in and be like, "Hey, hey, hey, point form, please. I just need this level of knowledge. I'm just trying to understand the context around this technology," or whatever it is. And then it goes, "Oh, okay, sorry."

And it goes it. So you'll, you'll understand, like, what's useful, how to steer that usefulness, and things like that

Mikhail: I, I think, I, I think I'm actually gonna bring up something that we're gonna-- we were supposed to talk about later in this part, that's kind of the verification point. Because a Go- prompting is the new Google search. The problem with that mentality is you're-- you, you typically always trusted a Google search, and the, the reality with these-- with this stuff is, is that if it's something that's important, you really have to verify it.

Matt: A very good point, yes

Mikhail: And verification might not be the same as it was [00:27:00] a couple years ago where you had to manually go in and check everything. Uh, verification now is a follow-up prompt it to provide sources, right? So, "Hey, where'd you get this information?" That's something you can ask a fo- in a follow-up prompt, right? And it can tell you, "Oh, shit, I just made this up." Like, that, that's

Matt: Y- 100%.

Mikhail: to me before. Like, the-- it, it's, it's very rare nowadays to be, to be fai- fair, like an As- like a, a top-of-the-line model like Astra or whatever is out there in the future is to just make stuff up, but it still happens. and if it doesn't have sources, you can always ask it for sources, and then the next thing is, is just ask.

Like, if it's something that you need to validate, tell it that. Tell it, "Hey, this is super important." Like, put emphasis on stuff that you think is gonna be important, and it will usually pr- like, it'll tell it to actually go deeper. And if you tell it to, like, in your initial prompt, if you tell it to, "Hey, make sure to validate this and give me sources," it will do that, and then it'll probably get back information.

It might be [00:28:00] slower, so your, y- the... And it might cost you more usage in t- depending on what plan you're on. But if it's something important, again, like later on in this episode w- we have, or in the show notes, there's this thing where it's a pharmacy versus an icecr- ice cream parlor,

Matt: But we, we can, and we can bring that up now. Like, it's a very important thing to bring up

Mikhail: it, I think it makes perfect sense because, like, if you're looking for medication and you're looking to, to clarify if the medication is correct or whatever, where you would put this, this e-emphasis on the prompt because you want it to be slow. You want it to give you the sources.

You want it to verify. And so yes, it'll be more expensive and slow, but it'll get there.

Matt: And, and you wanna verify the sources too. Like, if it gives you, "Hey, this is where I got it," you wanna read the link yourself

Mikhail: Yeah, exactly. Check the link and see where it, where it had it there. And you can even ask it, like, if you don't want-- If it's a link to, like, a paper and there's, like, a million different lines in there, you can ask it to clarify where in that paper the, it got the information from, and you can go in and verify yourself.

So don't [00:29:00] underestimate the importance of the, the follow-up prompt. Don't underestimate the fact that you can give these agents or, or L- uh, you know, L- uh, whatever, AI's instructions in the prompt to do something more deeply. It sounds weird because, again, in a Google search you would never be like, "Hey, do this deeper."

Like, that just doesn't make any

Matt: No, no

Mikhail: But th- again, if you go back to this as being an assistant level, it's the same thing. If you're talking to an assistant, you would tell it what's more important to you. Is it, you know, getting the store hours of an ice cream parlor? No. Like, you would be like, "Just get me those store hours."

If they're wrong, whoop-whoop-dee-doo. Like, you, you went to the place and it was closed.

Matt: You, you missed out on a vanilla cone

Mikhail: Who cares? But if, if you get the medication wrong, then, you know, that's really, really bad, so you better, you better tell the assistant that, "Hey, if you get this wrong, I'm gonna die." Like

Matt: like, well, and, and the, the specific example I had with the pharmacy was like something, [00:30:00] something like, "Hey, uh, I'm gonna check the pharmacy hours, and I need my medication today." Like, I need to pick up my prescription today. So if you told an AI, "Well, when's it open?" And it was like 9:00 to 5:00, you'd be like, "Hey, I better confirm that."

Whether you call it or you check the official website of the pharmacy or whatever, 'cause you need your medication today. But you don't need an ice cream cone. Like, you could do without the ice cream cone is the idea. So like another... You mentioned the assistant. Um, a thing, a, a, a prime example would be something like if you asked your assistant, like a, a physical person, if you ask your assistant, "Hey, like, do you remember which, which weekend is the long weekend?"

They may be like, "Ah, I think it's in a few weeks, in a couple weeks." Very casual answer. But if you're trying to schedule a flight for that weekend, you would sche- you would prompt your assistant differently. Not talking AI, talking the person. You say, "Hey, I'm booking..." You tell the, you know, you give it the important context.

"I'm booking a flight for the long weekend. Can you, like, look up the flight and tell me exactly when the long weekend is and confirm on a calendar? Or, like, give me a reference on a calendar to [00:31:00] make sure, because I need to be at this place at this day and time." Much different prompt than a casual, "Ah, in a couple weeks

Mikhail: That's correct, yeah. And I, I think just learning that concept as you're going, if you, if th- if that's the only thing you take away from this episode, I guarantee you the results you're gonna get from these agents is going to be way, way better than what you had before. It, it, it's asking it that extra little bit for

Matt: Yes

Mikhail: and giving it the context of why you need the information is also really important.

Like, that's a good point, Matt. Like, all of this stuff going to provide better results. I can guarantee it. It's a weird... I think it's hard to get over that hump of like these are machines, like this isn't a thing. Why would it need the conte- Like, it should just give me the right information anyway. It's not the case. Like, it will decide how deep it goes based on what you ask it

Matt: Well, the other thing too is, like, you have to also remember that when we- w- if we're talking with a friend, and I don't mean an AI, I mean a real friend or a per- [00:32:00] a person, a lot of the stuff that when you have a longer conversation especially, a lot of the stuff is wrong that you, that you share back and forth 'cause it, a lot of it's casual.

Like, I'll talk about video games with my friends 'cause I'm in a video game podcast, Tail and Patch Podcast. Go check it out. Um, but I'm in a video game podcast, and we'll have, like, in-depth conversations about video game history or, you know, whatever, and, like, a lot of the conversation will be like, "Oh, I think that was published by Ubisoft."

And if you really, like, you know, wanted to Google it and confirm it and whatever, maybe it wasn't published by Ubisoft. But it didn't matter in the conver- in the conversation. But, like, if there was something where, hey, we're writing out a huge report, and we wanna, like, have this accurate on our website, it's like, "Hey, someone go and make sure that that was actually published by Ubisoft."

Different prompt, different context, different importance level. Um, I, I think it's a, a good time to also mention this just straight up. Uh, be careful with certain bits of information you give AI as well. I think this is a good time to m- b- mention this. Things such as medical information, financial decisions, legal [00:33:00] information, travel requirements, important deadlines, large purchases, s- and security.

Basically, anything where incorrect information could have serious consequences or if you're unsure or uncomfortable with sharing it with an AI, know that the information could not be encrypted. Uh, look at the security policy of the AI service you are using. Uh, for example, I'm not getting m- me personally, I'm not getting medical information from AI.

I'm gonna go to my doctor still. Uh, whether that makes me a dinosaur or not, cool. Still gonna just call my doctor up. I did the same thing during COVID, called my doctor and asked questions 'cause the news was a flurry, and I was like, "I'm just gonna call my doctor." Like, that's just what I'm gonna do. So you don't have to, like, surrender everything over to this AI, and also, again, be very cautious with important decisions, important information, confidential information, et cetera.

We've be-- and I mentioned, s- mentioned several examples there. I, I, that's literally scripted. It's right in the show notes, but I, [00:34:00] like, I think that's an important almost, like, disclaimer for people, uh, to... Don't overuse AI just because it's like, "Well, I, I think my doctor said take two of these tablets, but I'll just ask..."

Like, no, stop. Call your doctor again and confirm if you're suppo- how much you're supposed to take. Don't-

Mikhail: good warning. It's honestly a very good warning, and especially if you're on these free services, you better believe that they're gonna be capturing whatever information you're giving them, right? Like, there are some caveats when you're paying the monthly subscription where, like, you have to trust them on this because you, you can't verify.

But the reality is that they, in their terms of service, they don't take your information. Like, that's what they say. Um, so take that with the grain of salt that it is. They say they don't. But if it's free, believe they do to a certain degree. So yeah, the, everything that Matt's saying is very, very correct. Be very careful. You can be a little bit less careful. I still wouldn't do the medical stuff, for sure. you can be a little bit less careful on the paid [00:35:00] tiers because, again, they have a little bit of stipulation there. Check the paid tier yourself to

Matt: I was gonna say check the paid tiers and your comfort level though, like

Mikhail: Exactly.

But, like, you can be a little bit more lax, in my opinion, with the paid tiers. But on a free tier, holy crap, they're gonna take it

Matt: Well, it's just like, it's just it like Google and, and, and, and the internet still works on advertising, at least right now. So it's the same type of thing where like people have your information. Like that's just, that's just the way it is. Like, if I start looking up cameras right now, Amazon's gonna be like, "Hey man, we got cameras on sale."

Be like, "Oh yeah. You do, do you?" Just so happens I've never messaged... You know, Ma- Matt, I've never messaged, I never, never told you about all our cameras, and today, today's the day we got, we got some cameras on sale. It's like, oh man, maybe I should buy a camera from Amazon. Interesting. Interesting. I think ultimately, um, one big thing about AI being wrong, too, another good thing to mention, is that AI is confidently wrong.

I've had it say stuff and it'll be... And this is a classic in [00:36:00] IT. It'll be like, I just experienced this actually. It'll be like, "Hey, download, uh, version 4.7." Look around, can't find it. I'm like, "What?" Look around, can't find it. The update button's not working. So you're like, "Man, is my update button broken or something going on here?"

Go look around, look around, look around. Hang on, hold up. 4.7 isn't even out in beta yet. What's going on here? And you could tell the AI that, you could be like, "Hey, I don't see no 4.7." And it's like, "Oh yeah, there isn't a 4.7." Oh. Okay. Fantastic. I've had Gemini kind of give me lip where like I told it like I wanted to code something up, and I just couldn't remember the CSS property, and it kind of, this is years ago now, but it kind of got out of hand and it gave me this, like a totally different way to do it in CSS.

And I was like, "Hey, like, I don't want to do it in this other way. I want to do it like I just told you." And it's like, "Yeah, but my way will still work." And it wouldn't give me the other way. I was like, "Oh. Fantastic." So there's that as well. Like, like AI can be like weirdly [00:37:00] social, I guess. A- maybe antisocial there, and it's, and it can definitely be confidently wrong.

Yeah, yeah, I'm sure there's a 4.7. No, there isn't. Oh yeah, I guess there isn't. Like, thanks for gaslighting me, bro. Like, thanks a lot.

Mikhail: And, and again, you, you've mentioned this, I mentioned this, I just wanna mention it one more time. If this is an assistant, if that's the mentality that you have, this is not a Google search, this is not a computer, this is an assistant, then you have that sa- have that same hesitance to believing whatever they say.

Matt: Yes

Mikhail: you g- you gr- like, the thing is, is that you naturally grow skeptical certain things and confident on certain things with certain people,

Matt: It's true

Mikhail: people are really good at getting you the facts of, uh, y- some, some trivia thing because that's what they love to do, whatever. So you tr- you start trusting them on that. And it's the same thing with these agents, where you start to... The more you use them, you start to get confident on things because they start to reply back with the right answers that you've verified, [00:38:00] and you gain that confidence, and you should double-check it every once in a while, but you gain that confidence.

And then certain things you're not confident on. You're just not confident on, and are just bad at certain things. have that same mentality that this is an assistant. They can be wrong. This is not a deterministic computer system. This is not a right or wrong answer. They can be wrong.

Matt: Yes

Mikhail: mentality, and if you have that mentality, you can get a lot out of them, um, and you can still have successful conversat- like, you know, outputs and stuff like that. You can still use them for most things. Uh, but if you have the mentality that, "Hey, these are computer systems, and they're wrong one time," and that's it, you lose all trust in them, the wrong mental- that's the wrong path.

And I've, I've seen a lot of people go down that path of like, "Hey, I asked it this question. It gave me the wrong answer, so it's useless."

Matt: And, and I get that mentality 'cause our computer systems for the most part have been

Mikhail: like, yes

Matt: you know, for, for the lar- largely through history. I know there's been some, [00:39:00] even like the original versions of something like Google Assistant would sometimes give you an amazing response. You say the same prompt and then it, like, tells you, like, something completely wrong or gives you a bad response the next day.

So it's like a little, it's like almost like that was the beginning of, like, indeterministic in a way.

Mikhail: Mm-hmm. Exactly

Matt: I, I think ultimately too with the prompts, and like correct me if I'm wrong, Mike, what you really wanna do with your prompts is you don't wanna necessarily like literally practice. You're not like, you know, going in and being like, "I'm gonna do seven prompts tonight," or something like that. But you're gonna like naturally put in, put in the reps.

And I think, I think what you slowly but surely are aiming for is you're aiming for less prompts overall. You don't wanna have five prompts to get your event poster made. You want ideally one, and one might not work. You might need more than one, but you- you're kinda like shooting for that, if that makes sense.

So just as you go through and, and as you start getting... [00:40:00] As you start like, you know, really kind of like jiving with the AI and you're going back and forth and you're, you're learning what it, what it's gonna reply with, pay attention to the responses that it gives you and, and kind of ask yourself some questions.

Like what, what did it get right? What did it misunderstand? What did it get outright wrong? Uh, did it get something not necessarily wrong, but something I didn't expect because I didn't provide a guardrail or a piece of information that it needed to, you know, act appropriately, uh, and research the right thing or what have you?

And then also pay attention to what additional prompts you use to then fix that. Because then you'll start to realize like, okay, I'm, you know, new conversation, it's a new day, it's a whole new topic. You know what? I better put those guardrails up, or I better tell it what color I want this thing, or I better tell it that I've attached a file for it to look so that it's not just taking my text, 'cause my text is explaining context of an image and it, you know, they go hand in hand.[00:41:00]

And so thing- things like that you'll start to understand like, okay, like this is, this is how I, this is how I prompt, this is the results I'm getting, this is how I'm becoming more efficient. But you don't need to like go through the reps or anything like that and, and just, just use those follow-up prompts and l- use your, the natural reps you're getting as practice, uh, to like slowly but surely lessen the amount of prompts that you need

Is there anything else that's, like, really standing out to you, Mike, that, like, you think that, like, a beginner, uh, would need to know? Like, we've covered, like, don't overwhelm, we've covered some prompting, um, uh, all, all that kind of stuff. Is there anything maybe, maybe we should touch on when doing the old way is better?

Like, when you should just Google search, when you should just reach for a calculator. It's like, don't just ask an AI what two plus two is. It's like, just use a, use a freaking calculator. You know what I mean? Um, is there anything, like, that you think we should specifically touch on? There's more in the show notes, like I've mentioned before, but [00:42:00] in terms of the episode itself, just to not overwhelm

Mikhail: Uh, yeah, I don't-- I think the next step for you, like, and we'll have, again, we'll ha- we'll probably have an episode on next steps, honestly. But if I was gonna, like, you've learned to prompt, you've gotten some good information of it, I would start to play around with these apps, like the ChatGPT application on your computer that is able to use your computer for certain things, Like it can technically, you know, write an email for you, it can look at your Amazon account and tell you something about your Amazon account if you need it. Like, it can do things on your behalf, behalf. That's a really cool next step, right? I'm not gonna get into how to do that deeply, but, like, there's plenty of articles online just knowing that, hey, these things can do that for you.

They can look at what your, what your computer is seeing and answer questions for you. They can act on your behalf, like actually move your mouse and do things [00:43:00] for you. So just know, know that. They, they can do that, and once you're comfortable with the prompting, once you understand what the agents, what these, uh, like, you know, these, you know, ChatGPTs and Geminis of the world are capable of, maybe the next step is asking it to do something a little bit more complex, right?

You know, maybe creating a, a wish list for you or something like that. I don't know. But I wouldn't go too, too far as you're going through this. And again, I, I think you're right. Let's not overwhelm them too much. Um, if you want, the, we can talk about some things that, uh, shouldn't use AI for. Like off the top of my head, I'll, I'll, I'll,

Matt: Yeah, for sure

Mikhail: couple right now where, for example, if you need, hours, like operating hours on a, uh, for a restaurant. I would still go to Google for that and just, you know, especially if it's something local,

Matt: I've made that mistake

Mikhail: Yep, like the-- it's just gonna be faster and better and more efficient to do it that way for the most part, unless it's like something super [00:44:00] complex and you have to get them to call it or something, but even then, like it's, just go to Google, get the phone number.

So very like direct location-based information would be Google. Abs-abstract information like, "Hey, design an itinerary for me," that's AI. just trying to think if there's anything else. Finding a website, like finding a, website that you know the name of. Like if you know, if you know, you know, uh, this is a stupid example.

Best Buy, uh, has a website y- but you don't know the .ca or .com or whatever. Like a lot of times I would just type that into Google search and it would show me the, the website and I

Matt: Yeah

Mikhail: the website. Um, probably don't need AI for that, honestly.

Matt: I would say if you don't have the context for the prompt, if it's something that you're so outside the wheelhouse on, a traditional Google search will generally give you, like, the startup guide, the beginner's guide. Because if you don't even know what to prompt, [00:45:00] like sometimes you're so beginner, but like sometimes you know enough.

Like you could say something like, uh, I don't... Like I know there's a, it's called the TFSA in Canada, Tax-Free Savings Account. I know there's a TFSA in Canada, but I don't know anything about it. Can you tell me? It's financial information that you are absolutely gonna want it to, like, you're gonna- just gonna get your- you're just gonna get context on it, and then you're gonna, like, be like, "Okay," like, "I have, I, I have a prompt," whatever.

You're gonna ask it, like we said, to, like, give it, give you its sources so you understand how to use it, and then you're gonna go and you're gonna check with a professional before you start using a freaking tax vehicle thing. You're gonna go check with a professional and all that. But the point is that the AI got you started.

But I think that if you are, like, absolutely brand new to, like, the Canadian financial system, you're not even gonna know what to ask. So to me, it'd be like, um, maybe you're new, you just moved to Canada. I moved to Canada, I don't know all the accounts, like what's checking savings, like, what is all this? If you're so far [00:46:00] removed, it's like I wouldn't prompt the AI 'cause it's gonna be like, "Oh, yeah," like, "Come on now, it's just, you know, checking and savings."

And you could, like, prompt it to give you more and more beginner information. But I, I would just prefer to go to Google, do my traditional Google search, get an introduction to banking from a couple of, like, trustworthy sources, and then I could potentially prompt the AI with for additional information.

And again, if it's financial stuff, so definitely check with a professional before you do anything. Uh, absolutely. Um, I think another, I think another actually key point that is in the show notes that we should cover to sort of maybe conclude the episode is you do not need to use AI for everything, which we're already covering, but what I mean is, is in a min-max way.

Like, some people are min-maxers, and they're gonna l- they're gonna be like, "Oh, my God," like the, you know, the, "The possibilities are endless," like, "This is crazy," and you're gonna try to, like, blow up. And the point of this episode is to like, hey, like, get the, get the reps in naturally and learn how to use the AI properly.

Don't necessarily worry about min-maxing, like, every single aspect of your life. Like if you're like, "I'm [00:47:00] gonna be able to automate every single task. I'm gonna build all these workflows. I'm gonna connect all my accounts together. I'm gonna have all these agents running, and then when I come home, I'm gonna have all these agents that have, like, sixteen things ready for me and I'm gonna come...

You know, I'm gonna optimize every single inefficiency so that even though this is an indeterministic technology, I'm gonna do all this prompting to make it deterministic." And like, some people, like, talk at that speed, which I do sometimes. But some people talk at that speed, they get super fired up.

They're all ready, right? The problem is, and I've mentioned this, I think, in this episode, is your brain becomes the bottleneck. Like, you still only have one brain, so if you have, like, 18... If you had 18 employees that only reported to you, you're the bottleneck. If you had 18 AI agents that only report to you, you are the bottleneck.

There's a reason why companies have a management structure that filters, filters, filters, filters until the CEO gets involved. If the CEO of a, of an app that, and let's say it's a larger company, and the app is down for an hour, the CEO does not need to know that the app was down for an hour. The tech team's got it.

The support team's got it. Whatever. IT's got it. Th- they don't need to know [00:48:00] that, right? But if it's down for two days, then it's like, okay, it's been filtered, filtered, escalated, escalated, escalated, it goes up. The, the point is, is that you can't be the CEO, quote unquote, of all these freaking agents. Like, don't try to min-max, don't try to overwhelm yourself

Mikhail: Agreed. I, I, I think, I think that's a good, that's a good ending point, honestly. Um, it's-- it-- for me personally, I'm, I'm better, I'm best at, like, managing two to three agents at a time personally. But th-this is getting into the weeds a little bit. Like, this is what we're gonna be talking about most likely in the advanced or developer episode. get into the weeds of, hey, this, this is my workflow for still having good context. 'Cause like you said, I can't do 18 things at a time. Like, it can do it, yes. But what's the output? Like, the output's gonna be probably pretty trash if I just trust it to do everything on its own. People can [00:49:00] argue with, about that maybe, but, like, I, I'm personally of the mind of I like to focus on things, and I think focusing on one to three at most is enough to still be super productive and get good output. That's where I want to be. I wanna be productive, but I wanna get good output. If my productivity goes up but my output goes down, like the quality, am I doing here? What's the point?

Matt: You want, you wanna be Mike plus. You don't wanna be Mike that controls a company, but the company is literally just a bunch of AIs that are behind you. Like, it's, it's ridiculous. Like, it

Mikhail: I don't wanna be, you know, not me. Like I, I, I don't want the output to not be me.

Matt: Yes

Mikhail: it to, I want it to still be me. So that's, that's where I'm drawing the line. Again, people are different and that's fine, but like that's... It's a good point, Matt

Matt: Yeah, I think, I think that's, that's it for the, uh, introductory episode. There is more in the show notes, including, uh, a brief section which we've already kind of touched on with the safety part and stuff, but just [00:50:00] know when AI needs to search the web. Like, 'cause the AI will sometimes just give you a response from its own internal database, and you'll be like, "Hang on."

Like the hours, for example, that Mike was saying. You might be like, "Hang on, like that, those hours don't sound right." You'd be like, "Yo, go check, go check the internet real quick." Like, it's not just your internal... You know, don't just check your internal dirtabit- database kind of th- kind of thing. But, uh, we've already mentioned, you know, that in passing where it's like, "Hey, make sure it, it gives you the sources that it's using," and, and, and stuff like that.

But I think that's it. That's AI basics. If there's anything we missed or anything that, uh, you think you want covered, or maybe it'd be super interesting to know if this e- episode's gonna become evergreen or if it's gonna become, like, super dated because, uh, you know, AI is rapidly changing, going through the paces right now.

Uh, but I mean, that will likely slow at some point if tr- historical trends hold up. So we'll, uh, we'll see. But anyway, if you enjoyed this episode, please leave us a comment, um, uh, or let us know if you wanna know something else and l- leave us a comment, maybe, uh, DM me on Instagram or something like that.

Um, but [00:51:00] no, we'd, uh, we'd love to hear from you about, uh, this episode or your initial journeys with AI. If you'd like to support episodes like this, you can do so, and that is on Patreon at patreon.com/htmlallthethings. And these $3 tier patrons did just

that. Tim from The Web Hacker on thewebhacker.com, Jason from Geek Life Radio via geekliferadio.com, Garrett Segall, Level Up Financial Planning via www.levelupfinancialplanning.com, Magnus from Yes Web via yesweb.se, Syntaxified from the HTML All the Things Discord server, and Stacy Mostiller from the website swoon- swoonworthydesigns.com.

And remember that we have a contributing author, and he is Michael LaRocca. He is the author of Self-Taught: The X Generation blog at selftaughttxg.com. And if you wanna learn how to code, you can get some courses in you by going to Scrimba, and you can get up to 20% off their Scrimba Pro plan. Use their interactive media player code editor to pause the lesson, play with the code live, and then press play to continue the lesson right where it left off.

Again, go check out that link in the show notes, as well as in the show description. The full details on how the [00:52:00] link works is in the show notes on htmlallthethings.com. That's it. Feel, feel free to leave a comment or review on the platform you're listening to this on, and we're signing off