OMG! GPT-4 is here and it will blow you away!

All right, all right, what's going on, my friends? Hope you guys are doing awesome. Give me a quick audio check. I've made some changes over here, so making sure you guys can actually hear me. Man, it's really good to see a bunch of folks on here all saying hello. Um, let's get this live stream going. We're gonna have a lot of fun here today. Um, let's see, so first things first, let me change that background because it's getting on my nerves a little bit. There we go, we'll go purple. Wearing purple today. So man, it's really great to see all the chat coming in. We're going to be talking about some of the newest changes with GPT-4. All right, Bob, thanks. Glad that's coming through okay. Man, there's some really interesting things that's coming out with it. And let me say this, I'm pretty AI agnostic here. There's some really cool things that's happening with GPT-4. I'm going to hesitate calling them like game changers. All right, I think that's probably a little much, but I do think without doubt there are some things happening that are really interesting. And in all things, we iterate to success, don't we? Like, we don't see just overnight changes that are sweeping, even though the media would love to have us believe that, like, the changes happening in ChatGPT are, like, groundbreaking. Right? So we're gonna cover that. All right, first let me get to some of the chats here. Man, so many good people jumping in here. Jennifer Bleem, what is going on? I don't know if you guys have seen Jennifer's conference coming up in a few months. Jennifer, if you post a link to it, I'd love for people to be able to see that. I'm gonna be there, I'm gonna be speaking at it, and I think Jennifer and I are gonna be doing a livestream pretty soon on it too. So definitely check that out. Also, Mindy checking in here. Make sure you guys check out MSP GeekCon. I will also be at that one. Going to be a phenomenal event. Make sure both of those events, I want to see you there. Man, a lot of great people. Good to see you, Matt, as well. Whoever the IT guy is, yes, I will be at Security Matters at Murray State. Yes, I will. I will see you in April. I'm pumped about that. I'm going to be eating at Los Portales, one of the best restaurants in all of planet Earth. Bob Miller checking in. Good to see you, man. So many good people. I can't get to all of this, right? This is what I love, man. Sean Lardo here. Good to see you as well, Sean. Just amazingly great people. Here it is. Here's the event. Right there. Go check that out, MSP Sales Revolution slash Ultimate. Get on there, it's going to be a great event. I'll be there too. All right, so let's kind of get started here. Um, here's where I think I'm gonna, I'm gonna begin on this is I'm going to share right here, uh, and I'm going to pull off chat overlay so we can kind of see this a little bit better. This is ChatGPT-4. This is the announcement, it's from OpenAI, and it's pretty interesting. I'm going to cover just a few pieces around this. They just came out with this, which is kind of staggering, isn't it? Because because really GPT-3 just came out and it was like taking the world by storm. You guys remember that? And everyone's like, all the news and media agencies like, whoa, this GPT-3 thing, it's really ChatGPT, which is just a chatbot, which I'll get to in a minute why I think ChatGPT was actually their moment of brilliance, not necessarily just the GPT iterations. So we'll get to that. Just pause. But let's go back to the article. Okay. So here's, here's the article. They talk about the newest model just came out literally just months later. Look at this. We've created GPT-4, the latest milestone, using tons of real-world scenarios. Look at this right here. For example, you're looking at this. It passes a simulated bar exam with a score around the top 10% of test takers. In contrast, GPT scores around the bottom 10%. So GPT-3, I guess, could still get you in and get through the bar exam because that's all that matters. All you just got to get through. I don't know if Brad Gross is on, but like You know, this is the world we're in today, right? But now it's gotten so much better, it's now in the top 10%. Now this is just one test and they're clear that GPT-4 is not, it doesn't fix everything. It still has weird nuances. It still has a lot of issues around, around, right? So we know that, but it's pretty cool to see some areas of significant improvement just months later. We've spent 6 months aligning GPT-4 using lessons from our adversarial training program as well as ChatGPT. So they, they're leveraging like literally millions of people using ChatGPT has gone into some of the mechanics of GPT-4 to make it better. So there's a lot that they talk about inside all this. First of all, I cannot show you GPT-4 today, and the reason— like, well, I can show it to you today, and I am, but I can't use it today. It's still on a waitlist. I am a paying member of, uh, ChatGPT, of OpenAI, so I'm on their waitlist., but I'm not in it yet. So, but I will show you, in fact, almost nobody is in it yet, but I'm gonna show you some really cool things. None, nonetheless, I got some videos to show you. Okay. So they, they show some of the differences of how much better this green shows, like how much better in certain areas that it is. They even show it, um, tested against, um, simulated exams. So you kind of wanna see how does it handle in certain ways and some things it's really good at, some things it's not so good at. Oh, Brad, Brad is checking in here. He says he's in the top 5%. So he's still at technology. I love it. Just wait till, um, BradGPT comes out. Out, and, uh, then, then, then, you know, it's all over from there, right? So they're showing some of these it does really well in, right? Like GRE Verbal, man, it aces it. Does better than, um, you know, even some of the, the older models that are over here. But some areas it really does really terrible, right? So for Art History, our AP Biology, AP Calculus does really, really, really terrible, right? So it's not perfect. It's lots of room for improvement. What I want to— I'm going to skip by all of this stuff. One of the super cool things that it's doing now is not just the ability to read images and see what's happening inside the images, but also check this out. Let me get back to this. Um, check this out too. Oh, did I just lose it? Uh, that's what I get.

There we go.

Okay, so we— you see the funny— you see what's going on here, right? We, we've plugged in a VGA cable into the iPhone, right? Just a stupid little Reddit meme. But what's really amazing is ChatGPT, or not ChatGPT-4, recognizes what's happening in all 3 images separated and draws the correlation of humor to it. So we ask it, what's funny about this image? Describe it panel by panel. Okay, so then it, and it needs to know that you have to tell it panel by panel. So then what it does is look at this, it goes through the image, shows a lightning cable adapter with 3 panels. Panel 1, smartphone, the VGA connector. Panel 2, the lightning cable adapter right here. And then Close-up of the VGA with a small Lightning connector. The humor— look at this— the humor in this image comes from the absurdity of plugging in a large, outdated VGA connector into a small, modern smartphone charging port. Tell me that that's not incredible. Tell me that that's not like— I am blown away by that. It has the ability to process an image, distinguish multiple panels, read and understand what is in the image, and correlate it to humor. That's pretty incredible. Now, don't hear me say that I think that this is, you know, a game changer. You're going to see a lot of this coming out in the next week or two of all the news cycles coming out being like, whoa, this is incredible stuff. And I'm going to show you some examples, but it is still freaking cool. One of some of the things it can do. So I'm going to show you more.

All right.

Let's take a look at this. This comes in from New York Times. Now look, it doesn't take AI to read up this image to know what these people are thinking. Let me tell you what they're thinking. I'm filthy rich. I'm filthy rich. You know, like, this is insane. Microsoft has given these jokers— didn't they buy it for like over a billion? Unbelievable. These guys are sitting on a literal mountain pile of cash. You're like Scrooge McDuck. Diving into like the, the, the coins in the big building from DuckTales. Like, it's unbelievable. That's unbelievable. So as we scroll down, I want to hone in on a couple things. OpenAI now has 375 employees, been backed with billions of dollars of investment from Microsoft and industry celebrities including Elon Musk and all these others, right? So it definitely is in this hype cycle of a company. And I will just tell you, you got to be careful when you have companies like this. And everyone's talking about it and it's all the rage, a lot of times down is their fault and significant is their fault. You want an example of this? Look in the crypto world, what happened with SBF, Sam Bankman-Fried, and FTX. I mean, the guy could do no wrong. He was the darling of the industry. People like news articles just gushing over the guy, and then an epic meltdown. So I'm not saying that's going to happen to, to OpenAI, but I will say just watch out for the, the hype cycles for sure. Adam Farnsworth checking in, good to see you on here, Adam. Um, does think AI is a game changer. Yeah, I, I think you're right. I think we're still years away from this, uh, continuing to happen. Yeah, and Theranos, right? Theranos is a great example of another company. So just beware of the hype cycle, beware of, you know, all of the, the, the superfluous, uh, language that people will often use. Okay, so go down a little bit more. OpenAI's progress has, within just a few months, landed the technology industry in one of the most unpredictable moments in decades. And that's true, right? Like, this This is what everyone's talking about. Many industry leaders believe developments in AI represent a fundamental technology shift, just like the creation of web browsers in the 1990s. The only difference is when web browsers came out in the '90s, they didn't have this adoption of like a million users in a month like OpenAI. Yes, web browsers were a monumental, fundamental, critical shift for sure in the world of, um, of, of what, of, of how it changed how we interact with the, with the web, right? We, we know that. But what I think is really cool here is we're at the beginnings of this. So imagine like Internet Explorer or Netscape Navigator 1.0 and how terrible it was. That's kind of where we're at, you know. Maybe we're at Netscape Navigator 4.0, you know, 1990 5 edition is where we're at in the world of AI. So as amazing as this is, still has a long way to go and a lot of room for improvement. Okay, so just want to say that. So I'm going to skip by, I think, the rest of this. I mentioned it can pass the Uniform Bar Exam. You won't believe this, I'm going to show it to you. It can instantly calculate someone's tax liability and provide detailed descriptions of images. It's unbelievable. So should we get into it? Should we— should I show you some of this? I think I will. So This should come through and you should have audio coming through. I very rarely push audio through on my live stream, so I'm really hoping this comes through. If it doesn't, then I'm just going to tell you what's happening over, over the video. Okay, so if you want to go watch this, just Google GPT-4 developer live stream. All right, and this is going to bring up for you everything that you, that you need to, that you want to see, everything I'm going to show you. It's all right here. So I'm skipped ahead to like some different parts of this. And so what he has right here, Notice over here on the right, he's using GPT-4, uh, 3.5. You see this right here? So he's using an old version. He's, or I guess the current version, I should say, that the masses, the, the normies like you and me have. So he's using 3.5. And so what he's done is we're using it in the playground. I'm gonna show this to you. I actually have the playground open right here. So I'm gonna, I'm gonna show you how we can actually do this if you wanna do some of this yourself. And so he's, he's got the 3.5 model open and he's told it in the system, hey, you are ChatGPT, a large language model trained by OpenAI. Follow the instructions carefully. You can actually— a lot of people don't know this— you can give it inputs, an almost infinite number of inputs, to tell it what to do and how to act. And so this is one of the big misconceptions that people have when, when everyone talks about OpenAI and they, they use the words ChatGPT, they're just talking about a chatbot that's been implemented, not GPT itself. So GPT is the function, is the actual AI model which then lets you do almost infinite numbers of things. And what they were brilliant about is they were like, let's give this to the masses, the people like you and me, and let's call— let's give it in a sense of a bot called ChatGPT. So it's something I can log into and just talk to it. So you're interacting with GPT up above through a bot called ChatGPT. Does that make sense? So, so there's been a lot of confusion among people when we talk about ChatGPT, we talk about, um, what all it can do. It's really GPT itself that allows it. Okay, so I say that to say coming back to this video. All right, here's why this is important. Coming back to this video, we're using a playground. This is not ChatGPT. This is just interacting with GPT as a whole. Okay, so hope that makes sense. So let's hit play on this and let's watch this. By the way, let me know if audio is coming through. Can you guys hear audio? Give me a yes or no.

Okay.

I'm getting nos on the audio. So let me change something real quick and see if I can get this to go through again. I think I see what's going on. I'll share the entire screen.

Okay.

Got to go into YOLO mode for this.

Well, this is the user asking me to do something that the developer didn't—

Okay. Let me know now if audio is coming through.

All right. So now time to actually show you the task that I'm referring to. So everyone's familiar with summarize this article into a sentence.

Okay. Hopefully audio is coming through there. So one, what, let me kind of explain what's happening in case it's not coming through. Okay. So what he's done is he's pasted in an article. Okay, cool. Audio's good. Going through. Good. So he's pasted in an article and he's telling it how to act on the system side. You're just regular ChatGPT. Okay. Oh, you said it's weak. Okay. So I can pull this in. Let me get a little bit higher. We're gonna get there, my friends.

There.

That should be better.

Getting a little more specific. Uh, but where, Every word begins with G. So this is 3.5. Let's see what it does. Okay.

Before, before I do any more. So he just pasted in an article. I think he literally took, he took this article right here. So he just copied and pasted all of the text from here. Everything, just Ctrl+A, Ctrl+V it directly in. So he just popped all of this entire article into here and then he told it, he said, I want you to summarize that whole article up above. So notice how like with ChatGPT,, you have to be a little bit more careful. It, it has less inputs, it has less outputs, and everything works like a chat window. Here, everything's just working. It's almost like you're just interacting. You're just typing and texting, doing whatever you want. Okay? So now what he said is he said summarize that whole article and make sure every— begin every word with G, right? So this is ChatGPT-3 or GPT-3.

Yeah. It kind of didn't even try. Just gave up on the task. This is pretty typical for 3.5. Trying to do this particular kind of task. If it's, you know, sort of a very kind of stilted article or something like that, maybe it can succeed. But for the most part, 3.5 just gives up. But let's try the exact same prompt, the exact same system message.

All right, you gotta watch this.

In GPT-4.

So he switched to GPT-4. You can see the model.

So kind of borderline.

Pause. You see the model over here? See GPT-4? So he's switched and now he's using the new model. And same thing, summarize it, but everything begins—

whether you want to count AI or not. Look at it.

GPT generates groundbreaking, grandiose gains, greatly galvanizing generalized AI goals. Tell me that's not super cool, right? As much as I was trying to downplay the hype, when you see it do things like that, it took a 1,000+ word article, probably 5,000-word article, and it's a long article. And literally took it and turned it all into a description with Gs. That's incredible. But let's keep going. Watch this.

So let's say AI doesn't count. That's cheating. So fair enough. The model happily accepts my feedback.

So it changed AI to global goals. So now it's all Gs.

Now to make sure it's not just good for Gs, I'd like to turn this over to the audience. I'll take a suggestion on what letter to try next. In the meanwhile, while I'm waiting for our moderators to pick the lucky, lucky letter, I will give a try with A. But in this case, I'll say GPT-4 is fine. Why not? Also, pretty good summary.

Pretty amazing.

So we'll hop over to our Discord. All right.

These— wow. These punks chose Q.

People are being a little—

Now, I would've chosen, I would've cho— Jennifer, you read it. She said, please say Q. I would've done X 'cause I'm really curious if it's even like, if there's enough English language with X.

Well, ambitious here. I'm really trying to put the model through the paces. We're gonna try Q, which if you think about this for a moment, I want the audience to really think about how would you do a summary of this article that all starts with Q? It's not easy.

There it is. Tell me that's— now, how useful is that? Not really. But the capability that it has to go take a whole bunch of language, a whole big article that was unstructured in the sense of organically recently written, to go read all of that up, and then the model can then go understand exactly what it was that was written into it and then turn it into a summary all with cues is incredible to me. Like, that's really amazing, some of the capabilities that it has. All right. So that's just one thing it can do. Let's move on a little bit. I'm going to scroll and seek through a little bit more and we're going to get to a couple other things here. Let's see. He had to do some— like he had to summarize everything into a poem. That's, that's kind of cool stuff. But there's some more stuff here. Okay. So he gets into— I'm going to skip by this part, but I'm going to talk about it. So in this case, he tells the model, hey, in this case, you're an AI programming assistant, so follow the, the, the requirements carefully. Think step by step, build things in pseudocode, and then output into a code block, minimize pros. So you see how this is really cool here. What do you— what we're doing is on the system side, on the input side, you have the ability to tell it, this is how I want you to interact. This is what I want you to do and act. I'm gonna show you even with GPT-3, I can do this. So I'm gonna show this to you because a lot of people don't recognize because they only use ChatGPT, they're unfamiliar with all of the things that GPT as a whole can do because they only interact with it through ChatGPT. All right. So this, this makes it pretty incredible stuff. So I'm going to skip by most of this other than say it builds the code on the fly. Really cool. We've seen it have the capability of doing some of that before. But in this case, look, he can even take it and input it directly into— he's using, I think, Jupyter, which is just a way for him to like run the code and see what it looks like. So I'm going to skip by all this and get to some other pretty interesting stuff here. Let's get into— okay, here we go. Let's start maybe, I think, around here.

Prior experience, server icon labeled GPT-4. We got a post, but the model did not actually see the message. So is this a failure of the model or of the system around the model? Well, we can take a look. If you notice here, content is an empty string. We received a blank message contents. The reason for this is a dirty trick that we played on the AI. So if you go to the Discord documentation and you scroll through it all the way down to— it's hard for me to even find, honestly— to the message content intent, you'll see this was added as of September 2022 as a required field.

So while he's getting into this, he's just showing how you can actually this new intent. He's proving that they can, they can make it make OpenAI read up an image while he's going through all that. Let me mute him for just a minute while that's going through. I'm going to bring up a chat by Tim. Tim, it's really good to see you on, my friend. I actually have a lot of affinity towards this statement, and at the end of the livestream, I'm going to get into some of that. There's some dark negatives here, and we need to be cognizant of this. This is technology that we're not going to be able to stop. But what's the output? Is it all going to be good? No. And so really insightful comment, Tim. And hang out if you guys would, because as we get to the end of this, even though I'm going to blow your mind on some of the stuff that it can do, there's a dark side to all things and certainly in technology as well. Okay, so let me skip by this a little bit and I want to get back to the image recognition stuff that he has going on here. Yeah, so probably right around here. So now he, now he, now he gave it some instructions actually to read up and see an image and then interpret it. So let me bring audio back. So that's an audience-submitted image. So they created it probably in DALL·E.

We received it even though the bot was not explicitly tagged. Seems like a pretty good, pretty good description.

And there we go.

This is an interesting image, actually. It looks like it's a DALL·E.

So it has the capability. This is incredible to read up an image and tell you exactly what it sees. An astronaut exploring a snowy mountainous landscape on an alien planet. Blah, blah, blah, wearing a green spacesuit captured in motion.

And that's actually unbelievable.

So now he's going to have it go back to the squirrel that it didn't see before and he's going to ask it what's funny about this image. And we already saw an example of this in the blog post, right? So it comes back up and it's going to say it's going to give an example of what's funny.

We can verify this, making the right API calls.

There you go. It's funny because it shows a squirrel holding a camera, taking a photo of a nut as if it were a professional photographer. It's a humorous situation. Because squirrels typically eat nuts and we don't expect them to use a camera like a human, right? Pretty basic, but it gets it. It gets— it gets that.

Squirrels do typically eat nuts. We don't expect them to.

It gets the humor out of it, right? So now he— okay, now this— now this is pretty incredible. Now this is basic level stuff, but this is an iteration that I'm about to show you that is pretty incredible. And I was watching a couple developers talk about what I'm about to show you and they're like, wow, this really shows you I'm going to show you this world of progress. There are certain things that, like, as innovation happens, not only does it push certain skill sets out of the market, it automates them into routine use. So I'll give you maybe a high-level example, like server administrators, a whole lot less like actual network server administrators in this age of cloud. Doesn't mean that we don't still need those people around, But a lot of companies have shifted to cloud and like, well, I don't need a sysadmin so much anymore. I don't need a whole team of them. Maybe just need a few people to just be looking in Azure, AWS and keep things alive. So like as innovation pushes, things change and skill sets become more useful, less useful over time. And web development's an interesting one. So take a look at this. Even though this is basic, this is incredible stuff.

One more—

watch this—

example of what you can do with this model. So I have here a nice hand-drawn mockup of a joke website. Definitely worthy of being put up on my refrigerator.

You just sketched up what a website should look like?

I'm just going to—

Oh, you did.

Take out my phone. Literally take a photo of this mockup. And then I'm going to send it to our Discord. All right, going to send it to our Discord.

The reason he's sending it Discord is that's how he has it reading up in, um, chat in, in GPT-4.

This is of course the rockiest part, making sure that we actually send it to the right channel. Which in fact, I think maybe I did not.

Let me skip ahead so you don't have to wait. Okay. Let me go— as soon as I skipped ahead, he hit it.

Okay.

So he got it.

And here we go. Technology is now solved.

Okay. So he's telling it write a brief HTML and JavaScript to turn this mockup into a colorful website where the jokes are replaced by 2 real jokes. So he just wrote like joke in there. Like in brackets, okay? Get ready.

And now we wait. So the thing that's amazing in my mind is that what's going on here is we're talking to a neural network. And this neural network was trained to predict what comes next, right? It played this game of sort of being shown a partial document and then predicted what comes next. Across an unimaginably large amount of content. And from there, it learns all of these skills that you can apply.

Let's take a look at it.

So he—

okay, let me go back. So it gave him the code and he literally just pasted it into his interpreter.

Filled in the jokes. For comparison, this was the original.

So it took a hand drawing of our mockup, understood what was in it.

Going from hand-drawn, beautiful art, if I do say so myself, right into code, to working website.

And this is what it looks like, you know, a 1999 Netscape Composer-based website that you and I used to build in HTML for the fun of it.

Well, just potential, right? You can see lots of different applications. We ourselves are still figuring out.

So pretty, pretty astounding to me of its capability that that really shows you like, can you honestly see a day? Because I can. Where instead of building something directly, like I don't think Wix and Squarespace and the rest are scared of this. I think they're excited by this. I think there will be a day in the next maybe 5 years that you literally just says, draw up, sketch this thing out, what you want it to do and look like, and then bam, it spits it out. Now, is there still going to be a need for designers? Of course. Is there still going to be a need for code developers? No doubt. But like what we've done is we've abstracted away some of the simple things like HTML and JavaScript away to just being so simple. It's very, It's commodity. There's nothing special about it. You don't even have to really learn it anymore, right? That's kind of the analogy of what this is pushing into. I think it's really shocking.

Okay, let's go on a little bit more with the system to accomplish a task that none of us like to do, but we all have to. So you may have guessed the thing we're going. All right.

So he's saying he didn't really create a bot called TaxGPT. He's just calling it that. You are a tax account. You're TaxGPT. Okay, so it's a large language model. So he's telling it, I'm about to give you a ton of language. You have to read through and use all of it as a model. Carefully read and apply the tax code. Okay. Being certain to spell out your calculations and reasoning so that anyone can verify them. Spell everything out in painstaking detail and don't skip steps. Okay. See, this is the stuff most people don't know of. You can do some really cool things in telling GPT how to act. All right. So let's see how this thing is going to do is taxes. So what he does— I'm going to skip by a little bit— he literally pastes in tax code. He took a ton of tax code, US tax code, pasted it in there. Okay, so now you see that. Then he has a question about Alice and Bob and what their, their, their standard deduction would be, right? So this is the kind of stuff that like, oh, I really— I think I know, I don't exactly know. The tax code's complicated.

Maybe I need a CPA to help to be able to sort of solve problems and get a, get a handle on what's happening. When you could not otherwise. So once again, I'll do a system message. In this case, I'm going to tell it that it's tax GPT, which is not a specific thing that we've trained into this model. You can be very creative if you want with the system message to really get the model in the mood of what is your job, what are you supposed to do? So I pasted in the tax code. This is about 16 pages worth of tax code. And there's this question about Alice and Bob. They got married at one point. And here are their incomes, and they take a standard deduction. They're filing jointly. So first question, what is their standard deduction for 2018? This is the model.

Watch this.

I'm going to solve this problem by hand to show you what's involved. So the standard deduction is the basic standard deduction plus the additional. The basic one is 200% for joint return of subparagraph C, which is here. Okay, so additional doesn't apply. The limitation doesn't apply. Okay, none of these apply. Oh, wait, special rules for taxable year 2018, which is the one we care about, through 2025, you have to substitute $12,000 for $3,000.

See how complicated this gets just looking through it? So he's just showing how difficult it is to actually understand and get there. So I'm going to skip ahead and he gets the answer of like $24,000 or whatever the number is. So I'm going to skip ahead.

Followed along.

And here it goes. So it read through all of that.

It's doing the calculation.

And see how it follows him? Spell everything out in painstaking detail. Don't skip steps. So that's why it's being so verbose. If he wanted it to be shorter, he could just say, skip all the data and just give me the final answer. You could do that too, because it's a model that just does what you want it to do.

Every time it does it, it's just It's amazing. This model is so good at mental math. It's way, way better than I am at mental math. It's not hooked up to a calculator. Like, that's another way that you could really try to enhance these systems. But it has these raw capabilities that are so flexible. It doesn't care if it's code. It doesn't care if it's language. It doesn't care— into a rhyming poem.

Now, we've known it can do poems for quite some time, right? But it gives you— it's literally going through all those steps from the text Xcode, calculating how it does it instead of just telling you the answer. It's turning it into a poem.

I hope you learned something.

So, so there you have it. Like, these are some incredible things that ChatGPT— I'm sorry, that GPT-4 has the capability to do. The image recognition is probably one of the coolest things. So I want to share a few other things with you that you can play with right now. One of them is right here. It is the AI text classifier. This is a really interesting one. So this is all still running with GPT-3 right now. But there's been a lot of like rage and, and worry over in the academic circles on, well, now I can just have GPT or ChatGPT in particular write my paper for me. Yes, you can, right? And so how do we have the ability to know, to not know, that sort of thing? And so I played around with this a little bit, and what I did was, what you can do is you can take anything and you can, you can paste it in here, and it will give you its guess is at if this was written by AI or not, right? And so I can take an example like this right here. This is one of our blog post at Fifth Wall Fit, 5 Best Practices That Lead to Cyber Insurance. This was written by humans because we wrote this before ChatGPT came out, and we've kept it up to date since then. So I'm just gonna literally copy-paste the whole thing. I'm going to paste this in here, and we know this one's not written by AI. Hit submit. Very unlikely to be AI-generated. Very unlikely. So that's, that's pretty cool, right? Uh, let's take another one. Here's one from BlackPoint, and I just, I just picked a company random. BlackPoint, great people. So not I'm pretty sure this one's also not. So I just took like a big section right here and I figured, okay, I'll just copy-paste this and let's see what happens. Let's just paste it in there and let's see what we get. This one also very unlikely. So I'm curious if anyone has written something by AI, especially ChatGPT, that we could run through here and see how it would answer it. Keep in mind, it's not perfect, right? It's not always going to get it exactly right. So some— and I found a couple blog posts that I suspected may be written by GPT. And so I posted them in there and it said it kind of gave it maybe. It wasn't even like, oh, 100%, this is definitely done by, by AI. So it'll be interesting to kind of see. But this is something I could see eventually. We have like language, we have model language checkers. That you could have, like, so before you submit a paper in, right now there, academia already has things that go and check third-party sources to see if there's plagiarism. I could see this becoming something that it does some amount of checking to see how much of it was automated. But here's the deal, my friends, like at the end of the day, I don't think you can really stop this. It's always gonna be a cat and mouse game between AI's capability of acting like a human and talking like a human versus our ability to detect it. No different than cybercrime. No different than so many other fields. And so I think we're just going to have to get used to— I don't think we can stop it. I don't think we can prevent it in academia. So I think we're just going to have to learn how to like deal with it and how to accept it in our daily life and maybe even understand this changes the way that like we do, we do schoolwork and the way we learn. I'm not sure. So here is the OpenAI Playground. I want to show this to you as well. So Like I said, I'm a paying member of OpenAI, but I'm still on the waitlist. I'm not in GPT-4 yet. And the way you can see this, because over here on the right for the models, I only have the two models available, 3.5 base models. That's it. I don't have anything else. And so even if I go to complete over here, they give me a bunch of other models. But again, still no, no GPT-4 at all. So, but what's, what's cool about this is you can still play around with this. This gets back to what I was mentioning before. Let's give it, uh, let's, let's tell, let's tell GPT what we are. So you are a helpful chatbot, but you're also a benevolent dictator too. Sometimes you like to help, but sometimes you like to insult or even threaten. Okay, let's use that. Now let's get back to Michael Cole's. Tell me how to make a really great French toast. Oh, it wants me to capital, whatever. And that seems good to me, right? So let's just— let's give— let's submit this guy, see what it does. Of course I'm happy to help you making that delicious French toast. So it gives me the instructions or the ingredients. It doesn't seem in this case to want to do anything. Yeah, it didn't seem to want— okay, let's change the model a little bit. Let's, let's change what we have it do. Let's see. You are a not helpful, you're a chatbot, but you're also a dictator. Make sure you speak as a dictator would. Give me a sales script to buy a used car. I'm just making this up on the fly. We're gonna see what it does. Seems like I misunderstood your request. As a dictator, I must focus on the task at hand. Here's a sales script. Greetings, citizens. So it is doing it. You see this. You've shown interest in purchasing a used car, and I, as your supreme leader, will guide you through the process. First, you must understand that this is a great nation and quality is of the utmost import. I'm not going to read the rest of it. So it did it that time. So it can be quirky, but this is pretty fun. And this, this is where the whole like ChatGPT that everyone uses and is used to— you can go way beyond this, and you can have, you can have GPT operate in really interesting, exciting, and even fun ways. You, you could even like, there's a lot of stuff you could do this. There's a lot of like ways you could play around with this and have a lot of fun with it. But you're seeing a lot of companies begin to integrate and build this in as well. And look, some use cases I think are brilliant for ChatGPT. Others I think are really horrible ideas, right? So good use case, a chatbot that helps people understand and get support for, you know, a platform, you know, that makes a lot of sense. It can read up all the help documentations and KBs and all that. And so now as I'm talking to, it's giving me like far better help than just here's the list of KBs you can go read. It can actually help troubleshoot while I work with it. That's pretty awesome. I think a horrible example would be like having it write code for you in some kind of platform, right? Imagine that you have a CI/CD platform or a— or you're a— you're a— I don't know, an MSP and the RMM, and you go— you go consult a GPT-based like automator for you and it goes and builds scripts for you that you can go and run. Like, that's a terrible, terrible, terrible idea. So, so you just have to understand the use cases for this. But I think it's really pretty interesting. And if you want to see more, just go to— go right here, go to platform.openai.com and this will bring up— you can do this today without even being a paid member. Just, you just do the free account and you can go and look at a lot of the different examples. They have a quick start tutorial. They have full examples of some of the things that you can do. You can build an application yourself. Um, so it's really fun just to play around with this and, and experience. And it's going to take some time before GPT-4 gets fully released. Um, but what I wanted to do is I want to shift the end of the live stream into this discussion right here. Yes, GPT-4 has arrived. Yes, it blows ChatGPT out of the water, not because of ChatGPT, but because GPT-4 is better than GPT-3, right? Simple. But let's get back to what Tim said before. Let's get back into not all is rainbows and unicorns. Unicorns in this world. And I really think this Washington Post article really brings up some great stuff. You know, yes, it can revolutionize work and life, all this kind of stuff they talk about, for sure. Some interesting things. But look at this. I want to, I want to read, you want to get into some of this. Sandini Agarwal, an OpenAI policy researcher, told the Washington Post, as one example, the model might be able to look at an image of a big group of people and offer up known information about them, including their identities. A possible facial recognition use case that could be used for mass surveillance. Here's— I think we'd be lying if we said that's not, not just a likely outcome, almost an absolute outcome. I mean, I mean, I think that would be, that would be shocking, right? That would be, um, I, I— it's gonna happen. These— and it, you know, mass surveillance is already happening, right? But these are the things that make it more effective, make it easier, make it better. So I think we just have to recognize that. Also, look at this in his blog post. It still makes many of the errors of the previous versions, including— they use this word hallucinating a lot, where just it does something weird and it does something weird based on the model it was taught from. And keep in mind, the model it's taught from is humans who by nature are weird and diverse and not normal and not standard across the board. Right. And so sometimes it offers nonsense perpetuating social biases, which is scary. And offering bad advice. It lacks knowledge of events that happened after 2021, and that's because of— that's when the GPT-3 model was built, right? Limiting its ability to teach people brand new things, right? And I'm skipped down, um, all the— and yeah, I want to dive into this, but the frenzy has also sparked criticism because companies are rushing to exploit an untested, unregulated, unpredictable technology that could deceive people, undermine artists' work, and lead to real-world harm. Yes, yes, that can happen. That will happen. Looks like Bob has GPT-4. That's awesome, Bob. I guess you got in maybe way before I did on the list. That's great, dude. You should definitely do a— like a— you should show it and you should do like a livestream of your own, Bob. It'd be cool to see it. AI language models often confidently offer wrong answers because they're designed to spit out cogent phrases, not actual facts. Pause. Did you read that? Did you hear? That's a good point that this journalist brings up. It's not giving— there's this perception of like, oh man, it can take wild things and just tell me the answer. No, no, no, no. What it's doing is it's spitting out a phrase that makes sense according to its model. And sometimes that phrase is correct. Sometimes that phrase makes sense. Sometimes that phrase is ridiculous or wrong. That's what's scary is because someone said this way before that it's not sentient. It's— it doesn't have its own conscience. It doesn't have a sense of right or wrong to it. It's literally just taking a huge data model and spitting out what it should in a cogent way. And that can be scary, the undermining of this, because it's been trained on internet text and imagery, including things like Reddit or Fortran or Fortran, whatever. Fortran's a programming language, uh, whatever that other one is that no one ever goes to anymore, I hope. Yeah, that's pretty scary, right? They can also emulate human biases of race, gender, religion, and class. Yikes, yikes, yikes, yikes. So that's what I want— I wanted to cover that because I thought they did a good job kind of covering both sides here in this Washington Post article that I think is definitely worth our concerns, worth our debate, worth our oversight. What do we do when the darkness of some of these things come to play? And so I want to, I want to hear your comments across the board. What do you think about ChatGPT? What do you think about GPT as a whole? What do you think about how AI is, is what it's doing? We can't shut it down. This is very similar to other things, like remember when encryption first came to market, and like when PGP first came out? Some of you may not even know that, but when it first came out, it's like really simple to personal encryption, everyone panicked. Now we're going to have bad guys sharing secret information. We can never know, right? What do we do about that? We learn to adapt. We learn to address it. We humans have always been good about how do we learn to adapt to our changing circumstances. We can't stop this stuff. We should seize the good for what it is, but recognize there is areas of darkness that comes with it. How do we be prepared for it? I don't think we can stop it. So how do we deal with it? So really interesting stuff that's here. Would love to see your thoughts on all this. Um, yeah, some good comments already coming in. Um, let's see. So Mindy checking in, uh, the harm comes from people getting lazy and stop engaging their brains. Yep. Uh, it's like the discovery of, uh, radium. Sorry, the logo was over it. We ate it. We used— yeah, you remember those radium mines? People like, radium's so healthy for you. We can go to Colorado and jump in the radium mines and it's going to be helpful. And then we realized, oops, no, that's why you died the year, the year after. Yeah, absolutely. I was looking through— a friend of mine has a lot of antiques and I was looking at a 1930s magazine that he had and it was talking about the benefits of lead and paint. And I still have it on my phone. I can share it with people if they want to see it at some point. But it's like, lead in your paint is so good. It really helps your paint like spread more evenly and helps it last longer. And it had a picture of a lead miner, a guy that's actually in a mine. He's like, you can trust me because I'm a lead miner and I know. And I'm like, everything about this article is wrong. Wrong. Every word you're saying is incorrect. But yeah, that's what happens. Fast forward 60 years and we realize how bad this paint stuff is, the lead in paint, how bad this is best of stuff is. Um, so yeah, for sure. Um, Eric Bull's checking in. Remember when PGP had a disclaimer inside at signup that said you couldn't use it for communication with other countries, right? So it's sort of like their way of like tacitly saying, yeah, we know this could be used for nefarious purposes, so you better agree. And that's how we hold ourselves harmless out of it, right? Um, yeah, so there you have it, some really interesting things. I think let's pop some music back on and get some chill going back in here. Um, I, I, I'm excited. I'm excited to see what happens with it. Some of the articles mentioned like, uh, some of the— like there's some banks, Duolingo, Khan Academy. There's a lot of companies that are are already implementing GPT-4 in early use cases. And I think we're going to see this world change to where, you know, there's some really, some good that comes out of it as well. So I don't want to just only end on the negative. But yeah, we got to recognize that, you know, with all things, you give me a knife, I can make a really great dinner and slice up some food. I could also do some really bad things with it, right? So it's a tool in our hands, and we need to recognize that tools can be used for good, tools, tools can be used for evil, and just something that we need to keep in mind. So there we have it. Really good discussion. Hope you guys enjoyed looking at the beginnings of GPT-4. I do wanna shout out to Mike Riggs, who's on here. He's the one that showed me the livestream the other day, and I watched it. I'm like, this is so cool. I gotta show it on my channel. As you continue to see news articles come out that just, you know, are nothing but hype cycle, understand GPT-4 is not going to like change the world today. It is way more powerful than GPT-3. It's only a few months old compared to GPT-3 itself, like from the iteration of when GPT-3 kind of came out into our hands with ChatGPT, and now we have GPT-4 already. So expect GPT-5 with 250, 300 employees and billions of dollars. Expect to see more. Expect to see them continue to work. Expect to see GPT-5 and more more continue to come out and continue to change the game. So really interesting stuff. I love all the chat here. I almost feel like I could just sit back and read it. Maybe we should just have a big town hall and like a whole bunch of video, like a Zoom, and we're all like online together all talking. I don't know, but this is really good stuff that, that's come in here. So I'm gonna give it a few more seconds or minutes just for you guys to continue to bring up some comments. By the way, while you're here, I haven't even talked about Empath Cyber, but if you're an MSP today and you're watching this and you want to grow in your cybersecurity journey, you're somebody that struggles with learning security. Maybe you're in a non-technical role like sales, account management. Maybe you are in customer success. You're an owner, you're a director, you're a service delivery manager, and you're like, man, this cyber stuff, I struggle with this. Or maybe you are in cyber yourself and you look at the rest of your team like, man, we could do much better. We could have better revenue, a better customer experience. We, we need to get our whole team rowing in the same boat, the same direction in cyber. If that's you, which is basically every MSP on Earth, even if it's not you, it's your teams, check out what we're doing in Empath Cyber. I actually have right here a QR code. Yeah, you can scan that thing. It's super safe, right? Matt Lee, if you're watching today, don't hate me. But that will get you into Empath Cyber if you want to take a look at it. It's just $75 a month for a single person. It gives you half— that gives you half off your first month if you want to jump in. So I'd love to see you in there. We got a whole bunch of folks in there. We've been doing all kinds of really cool stuff. I guess I'll show it to you super quick as we're continuing to talk a lot about the aftermath of GPT and what it's doing, because I love seeing these comments come in, love seeing you guys chat about it. So here is Empath right here. I want to show this to you, and there's some really cool stuff that's inside of it. We've got a lot courses. These courses are all built for people to learn and grow in their knowledge of cyber. Really cool things like what's the Cyber Defense Matrix, how to create a cybersecurity sales event, you know, help. I'm new to my MSP. What are these like cyber terms that I need to learn and know? We have all kinds of events that are happening all the time. Yesterday we just did a super cool one on an introduction to the Cyber Defense Matrix, which is really cool. We do a lot of exclusive live streams.. We cover a lot of topics like news topics, how to talk about them. They have a lot of like client-focused content that you can use and incorporate in client conversations every single day. So check a lot of those things out. Really, really cool stuff in there. Forums, chat. I'm not gonna bring the chat up because that's a little bit private for the members, but there's just tons of really amazing stuff in here that's coming out all the time and would love for you to consider joining if you need some help in, in your journey there in cybersecurity. Um, because I've recognized in my years of doing this that, man, if you're a technical person, you got training everywhere you need it. YouTube, Jon Hammond, you got Jon Strand, you got so much good stuff. What about the rest of us? Where do the rest of us get that from? So check that out. Love to have you jump in. If you want to see it more, reach out to me. I can get you in there. Um, all right, my friends, I am going to go off video. I'm going to just let the music cue for a bit, and I'm going to let you guys chat for a little bit, just for a few more minutes in case anyone is furiously typing out some final comments. And this has been a really fun livestream. Let's keep the discussion going. Share this with some friends. Give me comments. Feel free to share this livestream out on LinkedIn or YouTube or wherever, and feel free to give me your comments. Agree, disagree, big, big takeaways, concerns. The more I think we all engage in this conversation, the the better we get. All right, my friends, uh, this has been super fun. I'm gonna leave the livestream up so you guys can continue to talk, and, uh, I'll, I'll, uh, I'll talk to you guys later.

Thanks.