Latest / Bad Decisions Podcast / #5 AI Updates and Use Cases: GPT4 and Midjourney 5 | Bad Decisions Coffee Break
Transcript
- 0:00Your metier. Whoo.
- 0:05Cheers, Miami guy. Oh, wow.
- 0:08This is good. My goodness gracious, thank God
- 0:11for this drug. That is called coffee, and it's
- 0:14legal. What are we having for you
- 0:17today, man? Today, we have a new packaging.
- 0:20How do you pronounce? I don't even know how to provide
- 0:22sir. Lavazza is Italian.
- 0:23Right? Should be Italian, what's a
- 0:26roast? Or I know it's mild and
- 0:27productive Italy. It reminds me of Fabio actually
- 0:30the way you pronounce it. So this is mild and fruity.
- 0:34Yeah. Yeah.
- 0:34It's probably medium roast, I would say.
- 0:37But it's Yeah, amazing coffee. I'm a fan.
- 0:40Yeah. Welcome.
- 0:41Welcome to another episode of bad decisions.
- 0:43Coffee break. We are down to episode number
- 0:45five. I cannot fucking believe it.
- 0:47We are at episode 5. No.
- 0:49Stop. We didn't miss any weeks.
- 0:51And we are here and another Wednesday, and hopefully,
- 0:53hopefully we're not going to miss any not gonna miss, not
- 0:56gonna miss, not God. If you have like seasons of
- 0:59podcast, I know some people do that actually take a break, and
- 1:01then they continue again. Oh, I have something to tell
- 1:03people. So yet last night, I was
- 1:05figuring out how we can post all of this on spot.
- 1:08Fi Google podcast on Apple podcast and it was asking me
- 1:11what season you already your and I was like, oh my God, we never
- 1:14thought about that. So I just put everything is
- 1:16season 1 and then episode 1, 2, 3 and 4.
- 1:18Yeah, I miss, let's keep it. I think we.
- 1:21Okay, so the thing is, we don't want to plan any Seasons because
- 1:24the goal is we just want to keep doing the podcast.
- 1:28And because we are working all the fucking time.
- 1:30We're super Workaholics. We are not thinking of taking a
- 1:34break anytime soon. So if any time you feel like all
- 1:36right, we've done like hundred episodes.
- 1:38Loads and we need like two weeks of break.
- 1:40Then maybe we'll do it but we don't have any brakes plans.
- 1:43So you can continue to expect to see us every week on Wednesdays
- 1:47or Thursdays when we do upload, and thank you far out for making
- 1:50sure we actually go on Spotify and apple podcast and all this
- 1:53because catalysts are other member of bad decisions Studio.
- 1:56He did mention, he does not enjoy watching the podcast on 4K
- 2:00quality on YouTube because it eats up his data and he listens
- 2:04to it on the way on the way to UNI.
- 2:06And then so we thought A lot of people might be listening to
- 2:10this in their car, when they are stuck in traffic.
- 2:12Yeah, they're on the way to work or maybe you're taking a shit.
- 2:15Yeah. Listening to us because that's
- 2:17what you want to do and you don't want to watch it.
- 2:19Yeah. And I just listened to it.
- 2:20Yeah. So because we have soothing
- 2:22voices. Yeah.
- 2:23So from I think I can promise from end of this week, I'll put
- 2:28the last four episode and starting next week, every
- 2:31episode and post on YouTube, it will be on different platforms.
- 2:34Audio Only spotify's you still have, you still can have the
- 2:37video? Oh yeah.
- 2:38The rest would be only audio. Perfect.
- 2:40So you have both audio and video choices.
- 2:42Now, we have other news that we want to share with you guys and
- 2:45one of the most important ones is we just started a daily Vlog.
- 2:51Yes, you heard it right by God. From our bad decision Studio
- 2:54work which were working on. Metahumans, motion-capture
- 2:57Unreal Engine, blender, 3D VRA are we just decided, you know
- 3:02what, the podcast wasn't enough. We're going to take up the time
- 3:06to create a vlog. Single day because listen, we
- 3:10move to Canada with the hope of making it right.
- 3:13We quit our jobs acquit, everything.
- 3:15We had to come here and make it and we realize the only way we
- 3:20can do that is by sharing all of the stuff that we learn with the
- 3:24world. That's why we started the
- 3:25podcast, but the podcast was only limited to the information
- 3:28about work. So we like in this one hour, we
- 3:31only tell people what do we do during our work time?
- 3:34What do we research about the daily Vlog is like another win.
- 3:38I do to our process into our so life.
- 3:40Yeah, exactly. If you think about the podcast
- 3:42being more professional, even though we're not really a
- 3:44direction, we joke around a lot, you can only imagine what we do
- 3:48on an hourly basis were fucking around all the time.
- 3:51We're very different than your typical like nerdy 3D artist
- 3:56because we are that artists to we are nerdy when it comes to
- 3:59the actual work, but the same time because we're Gamers, we
- 4:03party as well. All of these characters, we feel
- 4:06like is not being shown to the world and we think Interesting
- 4:08to show that side of us in the block plus, plus we are watching
- 4:15motivational speakers and, you know, big businessmen like
- 4:18garyvee every single day to get motivated.
- 4:20And we feel like our journey can be motivating once we make it,
- 4:24because we're putting in so much time and effort, we understand
- 4:28there's so many people who are in the same situation who either
- 4:31don't know how to get to the top.
- 4:34Not that we do know. It's just that I believe that
- 4:37we're going to To make it two hundred percent.
- 4:39Yeah, no doubt. I have absolutely no doubt.
- 4:42If you see the process, you know, when we get there, how did
- 4:45we get there? And how other people also, can
- 4:47do the same thing with no luck involved?
- 4:49Or watch do that? No luck involved.
- 4:51No magic pills. No, no shortcuts, no shortcuts
- 4:55nothing. It's just hard work.
- 4:57Every single minute, every single minute and we want to
- 5:00show the ups and downs. So the whole thing is
- 5:03documented. So there's proof so tomorrow
- 5:05they won't say a daddy's money or yeah.
- 5:08Their family money. We want to show fucking proof
- 5:11when it's all done and I think at the same time, it's going to
- 5:14be another way for us to meet new people, but that's about the
- 5:18day but where are we putting it? So it will be only on Tick-Tock.
- 5:21Yes, we are not going to put it on YouTube, Instagram, or
- 5:25Twitter, which we are actually very active on this social media
- 5:28platforms. To, we decided on Tick Tock
- 5:30because I mean, so far, we are spending so much time just
- 5:33posting our own content. So if you want to post that in
- 5:36all the social media platforms, take a lot of time.
- 5:39We will put it on Tick. Tock, if it turns out to be
- 5:41good, I mean, it will be good. We might consider YouTube at the
- 5:44end. We wouldn't know about the for
- 5:46the first 30 days. We are committed to tick, tock,
- 5:48right? This is yes, exactly.
- 5:50We were gonna do Tick Tock, only just as a test because we
- 5:53haven't seen anybody do daily Vlogs.
- 5:56Like, people do vlogs, but they're just showing a brief,
- 5:59small showing the same all of the same, they're just preparing
- 6:02coffee, and then going to buy food or something like that.
- 6:06You want to show the raw process.
- 6:08How people used to do it like Casey neistat on YouTube but
- 6:11this time on Tick-Tock, if it doesn't work out, we're gonna
- 6:14quickly, we're like ready to switch to YouTube.
- 6:16So testing Tick-Tock for 30 days, that's where you can catch
- 6:20the Vlog. And then if it, if we feel like
- 6:22we need to change, we're going to come and put it on YouTube.
- 6:24But eventually, when we have a bigger team, we're going to do
- 6:26all social media platforms. For now, Tick-Tock only you can
- 6:29catch it at bat X Studio aside from that.
- 6:33I had another piece of news I wanted to talk about but I
- 6:36actually forgot. So don't we should memory.
- 6:39Yeah. What are we?
- 6:39What are we? Just go ahead.
- 6:40If I remember I will talk about it but why don't we just start
- 6:42with a topic of today? Because it is something that you
- 6:45probably clicked on the thumbnail for and it is huge.
- 6:49It's fresh, it's hot and I can't wait to get it started.
- 6:52If I don't want you take it away.
- 6:53So in the past week or in the past month, you have been
- 6:57bombarded with AI news, everybody's talking about it,
- 7:00you will see thousand tools out there and you don't know which
- 7:03one to use. I personally get confused
- 7:05because I'm in it. I use it every day.
- 7:08But then I see everyday people are posting new video editing
- 7:11tool chat chat bot X2 image, saw the video that the guys made an
- 7:16enemy out of, you know, that was crazy Corridor.
- 7:19Crew carry their crew exactly. That was amazing - so in today's
- 7:23episode we are gonna go through two of our most favorite AI tool
- 7:27and how we are going to use. Use it as a practical
- 7:30application in our daily life and how so you can use it in
- 7:34your business at school at work to actually really Helps you
- 7:38work smarter and speed up your process.
- 7:40That's exactly why we use AI in our everyday life.
- 7:44Yeah. So today's Focus would be on
- 7:46open AI G PT. So, so far, it has been GPT 3.5
- 7:51yesterday. Some ultimate on Twitter
- 7:53introduced GPT for which is mind-blowing.
- 7:56It was crazy. We watch the whole demo video
- 7:59and we're going to break it down for you.
- 8:01Also, we are going to talk about our favorite text to images
- 8:04generator, mid-journey. That's right.
- 8:07It's a Huge week for AI artificial intelligence.
- 8:11Because both of these guys actually announced an update.
- 8:14So a lot of people thought it's just going to be a big week for
- 8:16my journey, which is a text to images from generator.
- 8:19But then at the same time GPT came around and GPT for was
- 8:23announced and mid-journey V5 together.
- 8:26Now, they're different companies doing different things, but
- 8:28essentially, it's just a huge advancement in AI all happening
- 8:32in a week. And I was just telling far out,
- 8:34you know, what's crazy, when Google was the search engine
- 8:37that was Was around that say like 10 years ago or 15 years
- 8:40ago or 20 years ago, when people were like, super excited about
- 8:42Google being around. It wasn't like every two weeks
- 8:45or every month. You had like a huge update to
- 8:48Google. This is the crazy thing about AI
- 8:50because these models and data sets and algorithms can get
- 8:54updated not just by devs but on their own as people are using
- 8:59these Technologies communities. And you know exactly that's why
- 9:03these AIS are advancing on a very rapid pace and Oh, just
- 9:09seeing how fast the technology is improving.
- 9:11I cannot fucking imagine what it will be like in like two years.
- 9:17Oh, my fucking god good yesterday.
- 9:19It blew my mind just watching the demo in two years time, a
- 9:23lot of people have to reconsider their workflow and their job.
- 9:28Not not going to say that a I will take your job but you have
- 9:31to reconsider the way you work. And reconsider the way you use
- 9:34these tool because I mean we are going to I'm every I'm not A
- 9:38fucking lie to you, dude, I'm lucky scared.
- 9:40I'm lucky. Okay, I'm let me tell you where
- 9:42I stand with all of this, right? Where I stand with all of this
- 9:46is very simple. I am definitely for AI.
- 9:50I'm definitely for I and I think it should be there because it is
- 9:53going to make our lives easier like how the phone or the social
- 9:57network platforms. Social media platforms, helped
- 10:00us all of these platforms at first were scary.
- 10:03But then they actually help us in the long term.
- 10:06Of course they have disadvantages.
- 10:08Is. But, you know, that's another
- 10:09topic for another day. These AIS do are kind of scary
- 10:14to me because when they're advancing so quickly, I'm
- 10:17literally scared for my job. Like, some of the things that I
- 10:20do that. I spent so long learning.
- 10:22These a eyes will be able to do it very soon, like modeling
- 10:25something in 3D but this is the thing.
- 10:28I generally believe we have to learn how to adapt quicker than
- 10:32the AI is adapting. And as long as we stay creative
- 10:35and we adapt, I don't think it's something to be scared.
- 10:37About it just scary how fast it's growing.
- 10:40And it using AI is like I would say, it's like driving a car.
- 10:45You can give a Ferrari to a person to two different person.
- 10:48You will get 2 different outcome in the way they drive.
- 10:50One can be you know a Formula 1 driver and then he wins all the
- 10:54races, the other one might you know, crash it and hit the ball.
- 10:57Very good example, if AI is a tool, you need to know how to
- 11:01use it. Now GPT is available for
- 11:03everybody. A lot of people are making money
- 11:05out of it and when people don't know how to use it, Journey the
- 11:08same thing, if you go to mid-journey page, you will see
- 11:10amazing photo on the homepage, go and try out and give a prompt
- 11:14yourself. You will get something that is
- 11:15Half Baked and, you know, doesn't even make sense to use
- 11:18that following any sort of way. So again there would be a lot of
- 11:22jobs as like, how do you have now Machinery operator, you
- 11:25would be AI operator, you would give prompts you would know how
- 11:29to use it. You would know how to exactly
- 11:31get what you want out of this large set of data set and
- 11:35models. So I think there would be new
- 11:37job openings. Innings and of course, there
- 11:38will be jobs at be overtaken by a.
- 11:40I absolutely agree with you and I think as we start discussing
- 11:44GPT for I majored in five, we can discuss this further and
- 11:47also as we get to the end of today's episode, but let's just
- 11:51tell everybody what's the flow that we're going to have.
- 11:54So you know, exactly how we're going to go about this, we're
- 11:56going to talk about GP T 4S, this talk about.
- 11:59I want to I want to tell people how we use these tools on our
- 12:02daily life, so they can use it to can also give them the update
- 12:05because of this really fresh GPT is for.
- 12:07Yes. They meet Journey.
- 12:08If only you are a paid member, you could have, you know, you
- 12:11can try it yet. You had even is not even a
- 12:13samples, right? You can see the samples and read
- 12:15the model to train the model better.
- 12:18Paid members can do that right. Hate members can do that.
- 12:21If you want to give your not out yet.
- 12:22Yeah, and then at the end for us, maybe we can explain about
- 12:25how my Journey works and what are the best practices on how to
- 12:27prompt it and all those things. Okay?
- 12:29So upset, let's start with GPT. Yeah.
- 12:32So gbd4. I know you've been watching a
- 12:34lot of videos about it. You did tell me a little bit,
- 12:36but just go ahead. Until everyone cause it's
- 12:39freaking me out. If you haven't watched the demo,
- 12:41fired is going to give you like a quick brief of everything that
- 12:44he saw. So before we before we talk
- 12:47about GPT, for I'm going to talk about.
- 12:48Because so far we have been using GPD 3.5 and what we have
- 12:52been using as an individual, or as an in-studio, we have been
- 12:55using it to finding title for our games.
- 12:58And a ours is a simple as that. So I will explain what RAR will
- 13:01do and I'll tell it, can you give me some catchy titles?
- 13:04And I'm not kidding. This snap lens lot on Winning a
- 13:08our experience title was picked by chargeability 3.5, that's
- 13:12right, the mystery of King to the adventure through time.
- 13:15So I explained what was the AR about it?
- 13:17Gave me 10 options, all ten, very amazing accurate and I
- 13:21chose this. Another thing that we use is we
- 13:24use to answer our emails for work and also to create invoices
- 13:28for our clients. Sometimes you have a big client,
- 13:30you are charging, a big amount of money.
- 13:32You cannot just write one sentence created this.
- 13:35You have to write like four or five centers, so it can.
- 13:38Deliberately tells you plan research collaborative, all
- 13:41those small details that you need.
- 13:43So if you're a company and your invoicing someone that is use it
- 13:46for that. That was the funniest one as
- 13:48well. Because when you do something
- 13:51for a client, let's say you design an image for a client and
- 13:54you're not using AI. Let's say you're actually
- 13:56designing an image per client. When a client asks you for an
- 13:59invoice, just like parrot said, all you think in your head is
- 14:01like okay, maybe I like open Photoshop and I created the
- 14:03image but then when you actually ask GPT to Eight, the invoice.
- 14:09It manages to write it in 8. Steps in full detail that you
- 14:13like, oh shit. I actually did do all of the.
- 14:15Yeah. And then I think about it and
- 14:17you didn't think about it. So it's because it has a wider
- 14:20vocabulary than we do. And so that's why it can create
- 14:23all these amazing things. The invoice think was really
- 14:25cool. Another thing that we use which
- 14:27was really really helpful was asking asking for a game guide.
- 14:31So we were designing a game and we wanted to, you know, explain
- 14:34every step of the game. So I would tell GPT that what is
- 14:37the The game about what are we going to do?
- 14:39So for example going to open the door now the guys are going to
- 14:41drop or I'm going to pick this up, it will give you a game
- 14:44narrative, which is really enticing and you can use it as
- 14:47a, you know, gameplay hint for the audience mostly I use it for
- 14:50emails. So we had to email immigration.
- 14:54You know, when you're emailing all these like especially like
- 14:57fancy government entities, you have to be very formal with the
- 15:00way you asked your questions and we don't know how to before well
- 15:02obviously. So we're texting all the time.
- 15:05Right. We're very informal.
- 15:06So chatty Beatty was used for Email specifically.
- 15:10I did use it to get ideas for some of the videos that we were
- 15:13putting out on social media. Sometimes I could get what I
- 15:17wanted. Sometimes, I couldn't, you
- 15:18really have to learn how to speak to it.
- 15:20I realize you have a much better understanding of how to speak to
- 15:23chat to Beauty. Like, you know, you literally
- 15:25just keep adding. So, when you asked your initial
- 15:27question, then it answers it for you.
- 15:29Didn't you just keep iterating on that monster and going.
- 15:32Yeah, exactly. So that's not how I would do
- 15:34things on my journey because that's like a very different
- 15:36style. That's what I'm used to.
- 15:37But I've seen you use it, I haven't used it on a sea to the
- 15:41best of its capabilities yet I think like you will use GPT for.
- 15:45I promise you and another thing that a lot of people are like me
- 15:49who just started or who don't have any programming background.
- 15:53So I just started letting JavaScript, it's very easy for
- 15:55me to make an error. When writing a piece of code,
- 15:57specially when I'm developing my a, our experiences.
- 16:00Yeah, I can use GPT 3.5 to get, you know, JavaScript snippet.
- 16:06Cause like I can tell Dysfunction for me, or I can
- 16:09even give the function and asked him to explain.
- 16:11So it's great in that way. But enough about 3.5, let's talk
- 16:15about the hot GPT for. So the difference between three
- 16:21point five and four are described as GPT for is more
- 16:25capable. Less bias and is multimodal,
- 16:29what does my multimodal means in the current GPT 3.5?
- 16:33You can only have text conversation.
- 16:36You can ask question, you will answer you.
- 16:37You can ask another question, multimodal means, now we can
- 16:41have image so you can send it at an image, upload an image and
- 16:44ask GPT for can you please describe it or you can upload an
- 16:48image of a clock and you can ask what time is it?
- 16:50Now it understands the image. Okay.
- 16:53Understand. Why would you give it a clock
- 16:58and ask her what time is it may be?
- 17:00Do you have a better example of where would you use?
- 17:02Oh, wait image like, oh, tell me.
- 17:05So I mean, that's one of the example that they During the day
- 17:08of demo, right? But I can tell you easily, you
- 17:10can send an image and you can ask you to describe the image or
- 17:13you can even ask like, like give me an example.
- 17:15If what kind of image would you need explain or describe to you
- 17:19if you're submitting into GPT? I would say, I would say, let me
- 17:23think. What would be a good example of
- 17:26an image? So you can, you can give an
- 17:27image, okay? Off a person that you don't
- 17:29know, okay? Okay.
- 17:30And it will tell you who he is, what they did.
- 17:32You can upload a photo of, of course, you know who Barack
- 17:34Obama and Donald Trump is. But imagine it's a It is the
- 17:38ceremony on LinkedIn. A lot of seos, a lot of CMOS.
- 17:41You are a sales manager, you want to contact every single one
- 17:44of them, but do you think you can find those people?
- 17:46Or they have to be like, super famous, they have to be in the
- 17:49data set, right? It is, it is a train until 2021.
- 17:54So probably if the guy is a CEO of a company who, like a big
- 17:57team and yeah, you would know who he is.
- 17:59Okay, so that's what, but it's good.
- 18:00That's yeah, another I'll give you an example in a few minutes.
- 18:03I will blow your mind. Okay, will blow your mind.
- 18:06So but let's talk about how Three point, five and four are
- 18:09different. So the first example that they
- 18:11showed in the in the demo was to talk about the capabilities.
- 18:16Okay? So it you uploaded the blog
- 18:18post, okay? 3.5 and it asked a, can you
- 18:22summarize it in one sentence that every single letter every
- 18:26single word starts with the letter, a 3.5 gave up it
- 18:30couldn't do it. Okay, so basically imagine you
- 18:32have a paragraph is like 10 paragraphs and you have to
- 18:35summarize it in one sentence, it should make sense.
- 18:37Ants and every single word should start with the same
- 18:40letter. That is crazy.
- 18:41I don't think any human can do it even after spending an hour,
- 18:44okay? GPT for did it.
- 18:46It asks you to start with a it says AI advances, astonishingly,
- 18:51something as sentence that it actually made sense and it asks,
- 18:55now the audience want people were commenting on this
- 18:58Corridor. It was it was a live YouTube
- 19:00video it says Q. I cannot even give ten words
- 19:05that start with Q right now. I can't fucking do.
- 19:07That it's summarized the whole thing with all the words,
- 19:11starting with q quantifiable quality, something like it's, it
- 19:14was crazy. It was really crazy.
- 19:17So that I mean, that was just a demo.
- 19:19That's not something that you can use every day, but that
- 19:21shows the power of how fast it can learn.
- 19:24Remember the article was not that it wasn't trained or the
- 19:27article before, something that you give to the system right
- 19:30there, GPT is train until 2021. And the article was new, was an
- 19:34open-air websites or just uploaded article in the
- 19:37playground. And it asked question.
- 19:39So the second example this is so you're saying this is something
- 19:41that everybody has access to witness.
- 19:43Once again, it's something that only devs can do it in with the
- 19:47APA or know everybody can just like there's an upload button I
- 19:51can upload and then I can just talk about that.
- 19:54Is that how it works? So there is a playground UI
- 19:57front end that you can use it. But unfortunately, pt4, it's
- 20:00only available for GP t plus know that I'm aware that like no
- 20:04infiltrate members. I get that, but like, and pain.
- 20:07A on, not only for now, right eventually will be free for
- 20:11everyone, I think so. But is a twenty dollar per month
- 20:14so I don't know about the playground that you're talking
- 20:16about. Everybody can use it.
- 20:17So it's open. It's not like I don't go to the
- 20:19backend and the front-end that you can copy and paste and use
- 20:23it in. This is not the API.
- 20:24This is like a front end that you can come.
- 20:26Is it, is it like a copy and paste text or can I upload a
- 20:30document like a PDF or a book? I saw that the guy copy-pasted a
- 20:34text and it's up to a certain. There is a word limit to it.
- 20:37For the 25,000. Yeah.
- 20:39For now. Oh no, that was that 25,000
- 20:41would how much words it can generate I read that?
- 20:44Actually it can generate up to 25,000 words, is that crazy - so
- 20:48you can write a 25 thousand word essay your GPT for, so it's
- 20:53crazy. So the second example, it moves
- 20:56toward the direction of how GPT for is more logical and can draw
- 21:01conclusion. So it's not an abort that goes
- 21:04on find information and spit it out to you.
- 21:06If I draw It can draw conclusion.
- 21:09So what it does is we had an article already.
- 21:11So there was another article it asking.
- 21:13Can you find the common theme between these two article?
- 21:16Did he do it? Yes it analyzed both and and it
- 21:19was pretty fast. Anybody tells okay the
- 21:22commenting between these two article is this and then the
- 21:24conversation kept going. He was like no, I don't think
- 21:27that's accurate. Try it again.
- 21:29Okay. So so it's crazy how now I can
- 21:32imagine if it was there back in my unit days how much I could
- 21:35use it. Like, how many essays I Wrote in
- 21:38English that I had to find a theme or I had to find a better
- 21:41way to describe things. Do you think because, okay, this
- 21:45is the truth. Opening eyes trying to be very
- 21:47politically correct, as well. If you've seen, like, the way
- 21:49they're updating all the information and they're talking
- 21:52about like, all these updates, I'm sure that universities or
- 21:57open the items of this doing something, so that students
- 22:00can't just copy an answer. No, no, open a guide themselves.
- 22:04I saw that you can detect, but come on, you can go and change
- 22:07some As in the, I don't think it can detect after that.
- 22:10I mean, when these things, why would you need students to go
- 22:13and write essays. Now, when there is an essay
- 22:14written for you all the time, what is the purpose of learning?
- 22:18The skill of writing, if you can just get it from the air because
- 22:23I was a student, there is no, let's students learn things that
- 22:27are actually beneficial for. You should fucking learn how to
- 22:30use, AI to better, come up with a good research to come up with
- 22:35something that is accurate, that should be the fucking like a
- 22:39hundred example. 3 is exactly what I think now is an actual
- 22:43real world application, okay? And how people is not a One-Stop
- 22:47shop, how it can, you know, have the conversation going on GPT
- 22:50for created this quart pot? So it asks GPT for it.
- 22:55Can you create this Discord but that does this function for me,
- 22:58it went and read all the Discord documentation prior to 2021 and
- 23:03came out with a code. Not only the code, the pseudo
- 23:06code as well means it What is going to do?
- 23:08Okay, I'm going to write a function.
- 23:09First, we need to Define this variable.
- 23:11This is the code s. So you're telling me all the
- 23:14time, people have spent to create this sort Bots.
- 23:17You can now do by just asking GPT for.
- 23:20Yes, you can. And let me tell you how it went
- 23:23up. So, the first response that it
- 23:24came out, the guy tried it, the the person who was demoing,
- 23:28tried it and the error, there was an error for it, it copied
- 23:32and pasted the error to GPT for and then it found there and
- 23:37solve the problem. So it gave another reply and
- 23:40this time the problem was solved and where did they are?
- 23:42Came from? Because of course this court
- 23:44updated their documentation. You have 2021 now is two years
- 23:48so there was additional thing. There was a mass field column
- 23:50that they had to fill in. There was initial things.
- 23:53Now, what the guy did he updated?
- 23:55The dock, the documents in the in GPT for and now the new
- 24:00result, it worked perfectly. How crazy?
- 24:03Yeah. How how sophisticated was the
- 24:05bought? What was the Discord body?
- 24:08It Discord was actually talking to GPT for, so it was a Discord
- 24:11but when you type in that disk or Channel, it would send an API
- 24:15and then get the response. So what was the request that
- 24:17they put for the Discord? But like he said, no do what you
- 24:19want. Did he wander this world about
- 24:21to do to even it would go to the channel.
- 24:22Now you could ask anything from the Discord are connected you to
- 24:26Black. So he did he did you see like
- 24:28him, tell you, writing that text and everything from now on, it
- 24:32happened on this court from this point.
- 24:34So now she came up it uploaded. In this court and said, tell me
- 24:39what's funny about this image and the image was about I think
- 24:42it was a squirrel, holding a camera and taking photo of a not
- 24:46GPT for said, oh, this is funny because girls usually eat the
- 24:50not. They don't take photo of the
- 24:52not, it's crazy. And then one person from the
- 24:55audience gave a photo, probably, the folder was created by
- 24:59opening I dolly or mid-journey, because it was an astronaut's
- 25:03skiing in a Snowy Mountain. So something that you don't see
- 25:05often it perfectly describes. Grab it.
- 25:08He says an astronaut snowing in, you know, know we hold on, I'm
- 25:12not even surprised by that. That is obvious, you can do.
- 25:15But let's go back to this world. Do you know how many people make
- 25:19money from this court Bots, or is it actual job?
- 25:23Of course it's an actual job that like the music bought that
- 25:26you listen to all these are like the street.
- 25:29There's plenty actually there's a lot of paid Bots that big
- 25:33Discord servers. Use to do different things in a
- 25:36server. So you're telling you right now
- 25:39you can go and start making money by making Discord Bots,
- 25:43that do certain things that are useful to people, and start
- 25:45charging people for that. Yeah, but again, you can do it.
- 25:49So, it's again, it goes back to you on how to use GPT, how
- 25:52creative are you in? Your is exactly how creative you
- 25:55can. Can you it?
- 25:56Potentially mix a bunch of different codes and give
- 26:00something valuable to the people that they really want.
- 26:03So it gives you the two. Okay, I'm done with this
- 26:05podcast, I'm going to go crazy. Somebody Make something funny
- 26:08because this is not making me money right now.
- 26:10Come back here. Come back here.
- 26:11I have a better example for you. So after we were done with this
- 26:16good one, it's crazy. So it gave it an image in
- 26:19explained it it, you know, everybody at this point, I think
- 26:23everybody's mind was blown away already and then the last exam
- 26:27is the second. Last example came up, I'm pretty
- 26:30sure. People saw the photos on social
- 26:32media, but I'm going to put this snippet of the video.
- 26:34So there was a notebook, right? Beside the guy.
- 26:37There was a hand-drawn website, imagine a square and writing in
- 26:42a handwritten word, very ugly. Like he wrote the title and
- 26:46every title very ugly. He said my joke website and then
- 26:49he put some buttons and some text uploaded the photo and
- 26:53said, make it a website. It came out with an HTML code
- 26:58and a working website. So he told Chad gbt to use that
- 27:03photo of his hand, drawing of the website and Chachi PT or D4
- 27:08turn that into an actual working website, actual working code for
- 27:12a website. So HTML code, which he runs the
- 27:15HTML code and the website was it?
- 27:16This was crazy. This was absolutely
- 27:20mind-blowing. So you're telling me we don't
- 27:23need to code this code box anymore, we don't need to use
- 27:28wix.com. And, you know, I'm not saying
- 27:31I'm not saying you have to stop using all of this.
- 27:32Again, it gives you the tool. So now, for example, if I want
- 27:36to create a website it it is much easier and you know what
- 27:38I'm gonna do? Let me tell you what I'm saying.
- 27:40Yeah, I know where you're going to go with this.
- 27:42You're gonna you're gonna be, you know, like political courage
- 27:46in the way that they're okay. I understand you're saying that
- 27:49you can still use weeks and all this.
- 27:51But you know, we're talking about the improvements that are
- 27:55being made fast today. Literally I'm telling you, don't
- 27:59think about now I'm talking about in six months time or in a
- 28:03year's time you're telling me that what we have now won't be.
- 28:07To a virgin that you literally give it an image like you
- 28:10photoshop. Forget about hand-drawn, you
- 28:12photoshop, the website that you want your feet, might know, not
- 28:15even figma. Let's say, some people don't
- 28:16even know how to use singer. Photoshop and Summit, everybody
- 28:19uses Photoshop. I Photoshop my website and I
- 28:22give it and I get the code out. Oh, for sure.
- 28:25Yeah. So why do it now available?
- 28:27I would say, I would say there's a lot of things behind it.
- 28:29When you create a website, you have to make it mobile
- 28:31responsive than you have to say, oh, when I click this button,
- 28:34this happened interaction. So those are the things that
- 28:37people. Now you can, if you are a
- 28:38website developer, you can speed up your process by using GPT for
- 28:42to get all the code. But again, you still need to
- 28:44think of the design. The ux design, all those things
- 28:47are still in place again, speeding up the process, makes
- 28:51it more efficient, of course, maybe five years time, you can
- 28:53do audio. Hey, Gee Pete.
- 28:55I don't. Can you five years, dude?
- 28:57There's no way. Five years probably earlier this
- 29:00fucking year. Yeah.
- 29:03Yeah, yeah, true true. This has been going on for a
- 29:08long time, but everything we've heard of, is the best in the
- 29:11past few months. Yeah.
- 29:12Just just just like, no, that's not, that's very accurate.
- 29:16That's very accurate. And then another thing that I
- 29:19think is very crazy about all of this is that he's doing all of
- 29:24this life and everything is super fast, everything he
- 29:27doesn't wait. And then there is the
- 29:29conversation is ongoing, can you change this?
- 29:31Can you change that? So what I'm going to do, I'm
- 29:35going to give all the metal spark documentation, Oh, and
- 29:38lens Studio documentation to GPT for when I have access to it for
- 29:41whoever that doesn't know, these two are the, a our software's
- 29:44that we use for Instagram and Snapchat.
- 29:46Oh, maybe I'll do Unreal Engine 52 and then as she and can get
- 29:51GPT by my side creating dope as a, our VR experiences.
- 29:56And I mean, as long as you have a little bit of a basic coding
- 30:00knowledge, you can then go ahead and add your own stuff to it and
- 30:04just perfect it and have a perfectly running code.
- 30:07Yeah. And you think do you think I can
- 30:09use it for blueprints and underline?
- 30:11100, 200 percent already, use it for blender, python like python
- 30:17using python to run things in blender.
- 30:19It already works. If you haven't tried it out, if
- 30:21you're a 3D artist, go check out some videos of people using chat
- 30:24gbd3, not even for, for blender. Just ask it to create.
- 30:29Let's say, I want an array of 10, cubes in a circle and it
- 30:33will do that for you. So if you were to create that
- 30:35yourself and it would take you time to adjust Adjust it because
- 30:38it's using Code, it's going to be like a perfect circular
- 30:40shape, and it doesn't like a second.
- 30:42So, I know people already using that, but I can imagine we GPT
- 30:45for is going to get more complex and it's going to become smart.
- 30:48Yeah, hundred percent. Now we talk about some exciting
- 30:51stuff. The guy did I would say a thing
- 30:54that a lot of people considered boring so it told GPT that you
- 30:58are tax GPT. So you have to do my taxes, you
- 31:01uploaded all the tax codes and then it gave it a scenario.
- 31:05This guy is married to this guy their income This these are the
- 31:08expenses. What is their tax deduction?
- 31:10It came up with all the tax deduction wasn't accurate?
- 31:13Yeah, it was accurate and the best part is you can ask judge
- 31:16ability to explain it to you. So sometimes you can say, oh,
- 31:20what is my tax deduction? So is thousand-dollar your tax
- 31:23deduction? But how did you get that
- 31:24thousand dollar? It will say that all your income
- 31:26is 20,000. Let's your expenses.
- 31:29You have this much. I don't know, profit from here.
- 31:32Rental properties. It will explain everything.
- 31:34So not only you get the last result you Understand the whole
- 31:38process. And I personally did a lot into
- 31:41all these tags grants and everything.
- 31:44I have to read the documentation over and over again to
- 31:46understand and sometimes something is in chapter 1, and
- 31:49then you have to go to chapter 22, continue is all over the
- 31:52place. This brings everything into one
- 31:55place, summarize it and you're good to go.
- 31:57That's fucking crazy. But sorry, guys, the guy said,
- 32:00nor mean either charge your PT is your Tax Advisor, anything
- 32:03happens. I'm not gonna be responsible for
- 32:05that. Of course you should do.
- 32:07Final check yourself just just that it will help you speed up
- 32:10the process. Okay, I won't question.
- 32:12So these are all the examples that you see.
- 32:14This was all the example I saw in the show and I'm going to put
- 32:16a link of the show so people can watch it for themselves.
- 32:19Okay, so that's GPT for now. Let's move on to mid-journey
- 32:24because mid-journey just talked about V5 now.
- 32:29There's unfortunately, no demo video.
- 32:31Like the one found out watched where they describe all the
- 32:35different possibilities. And I believe that the GPT for
- 32:39release has a lot more aspect to it.
- 32:43Very simply put me Jenna, V5 actually comes up with higher
- 32:46resolution better detail and wider range of styles.
- 32:50Compared to read for. This is something they announce
- 32:52themselves in their discreet Channel but what really matters
- 32:56is also. The human anatomy.
- 32:59This is something that I saw myself.
- 33:01So, when you actually go to the Discord server and in the
- 33:04announcements tab, they have a Teaching model link.
- 33:09And what they want to do before they release V5, they want to
- 33:12make sure people are one hyped up about 35, going to make sure
- 33:16that the model itself that is being trained is accurate.
- 33:20Not just based on their own biased.
- 33:22But based on the actual users and the users is what matters.
- 33:26We are the ones training, the datasets in reality, by choosing
- 33:29the images that we liked. And so they have this training
- 33:33model on the website where it shows you two photos of The same
- 33:38essentially prompt I imagined and also there's another version
- 33:42where it showing you two photos. That is not from the same prompt
- 33:45and your job is to choose the one that you like more and that
- 33:48could be based on the resolution based on the detail based on the
- 33:51realism. But generally speaking even
- 33:53though they said, this is not V5 yet, it's still plain dull and
- 33:58it's not super accurate. Like what V5 is going to be the
- 34:03Improvement is crazy. We have a video popping up.
- 34:07Now and this is showing you did the fucking difference of
- 34:11mid-journey. V5 speaking of human anatomy
- 34:15prior to V5, it's difficult for me Journey.
- 34:18It's challenging to get the human anatomy, right?
- 34:20So sometimes you would end up with six fingers, oh, I've seen
- 34:23that photos all over social media like they were photo of
- 34:26three guys sitting down and they were guy asking what was wrong
- 34:29with the photo. And it was like, what?
- 34:31The guy has fucking six fingers. So, it was like, sitting like
- 34:33this and you would count there were twelve fingers.
- 34:36Yes, sometimes you wouldn't Even notice it.
- 34:38Yeah, that's like an extra. Yeah, or even the face,
- 34:40sometimes, like the eyes wouldn't be symmetrical, you
- 34:44know, you could tell that that was a mid-journey before or V3
- 34:48or whatever. But in this upgraded version
- 34:52more often than not, you're going to get really crazy
- 34:55realistic faces and reminding you.
- 34:59Again, these spaces are not real faces, like when you look at
- 35:02these photos you like oh that's looks like someone I might know
- 35:05but then they don't actually fucking exist.
- 35:07That's Crazy. Crazy.
- 35:08That's a great. There's no one living in this
- 35:11world that looks like that. That's the fucking crazy part.
- 35:14Unless we find out that actually does, I don't know.
- 35:16But according to themselves, these are all a i generated.
- 35:20So they don't look like anybody. You would see in real life and
- 35:24then again, there's other things like the the ABS, you know, the
- 35:28complexity of the muscles all these things are a lot more
- 35:31defined and the same thing goes for Animals as well.
- 35:34There's a lot of animal photos. We saw in.
- 35:37V5 the, the rating of the models.
- 35:39Like the cat's. The dog's the Wolves the for
- 35:43looks a lot more real. They look a lot better.
- 35:46So I'm imagining. It's just the fact that they've
- 35:48trained their model really well. They've upgraded the algorithm.
- 35:53It seems like that's the case here.
- 35:55So, so, but I think it's good to talk about how we actually, I
- 35:58mean, of course it creates amazing images, but how we
- 36:01actually use it and how people can have practical application
- 36:05from my journey. So my journey, Just for
- 36:08everybody. It's a texture image.
- 36:09Prompt application. It runs on this court and you
- 36:12essentially go in there. You type in / imagine and you
- 36:18just give it a prompt and it will turn that prompt into an
- 36:21image based on its data sets. Now, we use it for a lot of
- 36:25different things. You've used it for logo ideas,
- 36:29where we needed to see, for example, for the coffee break.
- 36:31Here, we need a logo. We actually started looking and
- 36:35asking for different designs. We've done that.
- 36:37That we have looked for images for the Snapchat lens autonomy
- 36:40wanted to join a competition and we actually used it.
- 36:43Yeah, actually. So the thumbnail is from MIT
- 36:45Journey. So we started, I think I gave 20
- 36:48prompts to get the final result but final result was amazing.
- 36:52And then also for the game, you I display be used and I said I
- 36:56want an ancient Egyptian script with symbols on it.
- 37:00Of course it was three lines but it gave you and I used it
- 37:02because I suck at UI design and you cannot do it in five
- 37:06seconds. And it did it.
- 37:08Oh my fucking god. So I I use it for other things
- 37:12like thumbnails, sometimes, I don't, I know people on YouTube,
- 37:15actually, use it straight away for their thumbnails.
- 37:19I don't like to do that yet, maybe in V5 the quality is so
- 37:23good that I will, because it's so real.
- 37:25But up until now, even if I want to do a thumbnail, I probably
- 37:28just ask it for the idea and then I will create my own based
- 37:32on the results. Actually, that's honestly, I
- 37:35would say up until V4. That's most of the use, Case for
- 37:37us is getting ideas from it. So let's say, we want to design
- 37:41a flyer or a poster, or even a scene in Unreal, Engine 5, or a
- 37:47scene in blender or even a human.
- 37:49For metahumans is not like we're going to use that photo.
- 37:53Straight up is, we want to get ideas for what that might look
- 37:56like. Let's say I want to do a scene
- 37:57with Joker, right? And I don't know what my sink
- 38:01should look like, or the vibe, or the colors using my journey.
- 38:04I get references, and I'll put in a Like Joker dancing in a
- 38:11red, you know, Tone Room and I'll just write cinematic
- 38:15lighting and all that and it will spit out for images that
- 38:19look like something. I probably have in my head, but
- 38:22I just can't put it down into the software.
- 38:25So this is how I use it, usually to get references.
- 38:28However, I think would be 5 and probably possibly with the
- 38:33upcoming versions after that, it's turning to be so real.
- 38:37The results that you can actually end up using it as it
- 38:40is and not just for references. So for your thumbnails, for your
- 38:44posters, and people won't be able to realize that it's a, i
- 38:48generated that's true. But one thing I'm always say,
- 38:51not not, say disappointed that I really wanted to happen is to
- 38:54having words on the image, I still don't know why they cannot
- 38:58figure it out. Like something as simple as
- 39:02press start. Or let's play the words.
- 39:05It cannot comprehend the words. Putting the image so I hope
- 39:09version 6 it comes with the boards.
- 39:11But an important thing about my journey and all this a, I text
- 39:15with image tools are how you prompted virus is a real Pro.
- 39:19I have, I would be using it in our house but I've seen the real
- 39:25Pro. So I mean it's crazy.
- 39:27They that the words that they use and how it can change the
- 39:31image. So usually when I go, I'll make
- 39:33it very simple. I would say, for example, and
- 39:36ancient Actionscript with symbols on it.
- 39:39But when Paris goes give us the size cinematic color type light,
- 39:43like everything material angle, all those things actually have
- 39:47effect, so it helps for a better image creation, right?
- 39:51100% it. You have to learn how to talk
- 39:53today. I and the same thing goes with
- 39:55you like I cannot talk to GPT like you do because you know how
- 39:59to, you know, keep a conversation going and make sure
- 40:02you get a better answer over time.
- 40:03It's the same thing with mid-journey, the more detail you
- 40:07put Into your prompt. The better your result is going
- 40:09to be and the detail goes down, literally by you usually start
- 40:14by explaining what you want in the image.
- 40:16So let's say you want a woman in the red dress and of course,
- 40:20already I've described the color of the dress that I want so that
- 40:23helps a lot but then after that you go and continue to see what
- 40:28style do you want if you wanted to be in a style of someone art?
- 40:32Let's say Picasso you would write that right in the style of
- 40:35Picasso and then If you want it to be a certain Lighting in that
- 40:40photo, you go ahead and explain that.
- 40:42So if you want cinematic lighting, or if you want it to
- 40:45be completely black and white still in the style of Picasso,
- 40:48but black and white, you would go ahead and describe that.
- 40:50If you want the photo to be a certain aspect ratio, right?
- 40:54Because you want to use the photo as a thumbnail for
- 40:56YouTube, then you go ahead and explain that aspect ratio to The
- 41:00Prompt. If you want it to be as specific
- 41:04if you know, even in chat gbt there's things.
- 41:07Called creative. I believe like you can give it a
- 41:11level of zero to 10, you can give thumbs up and thumbs down
- 41:14and that's what I do. Not that one.
- 41:15I'm talking about the dev version where you actually pay
- 41:18to it. Like, how, how creative I want
- 41:20you to get with the answer? I know what I mean.
- 41:22It is the same thing exists with me Journey so you can explain
- 41:25that in the problem. Like I want you to go super
- 41:28creative or I want you to be not super creative as in like stick
- 41:31to exactly what I'm telling you you know.
- 41:33So you have all these prompts. Now, we found a GitHub Blink
- 41:37that we're going to put into the description.
- 41:40And some guy, I forgot his name. He shout out to him.
- 41:43He went and he took basically a reference of all the different
- 41:47styles, the different colors that you can explore like, lime
- 41:52green, or whatever and literally has photo references of all
- 41:55these prompts and then you can see if I want my photo to look
- 41:59like this. I'm gonna put this style that he
- 42:01put. I want to put this color and it
- 42:03just take all of these and start experimenting.
- 42:06Just try it out. And and create and of course you
- 42:10can also steal other people's prompts.
- 42:12That's something that I do. That's like my biggest hack.
- 42:15Just go to the Newbie channel on this court and look at all these
- 42:19prompts that people are putting out and see the photos that look
- 42:22really good. Take their prom, you probably
- 42:25find something in there prompt that you find.
- 42:27Interesting, use that for yourself.
- 42:29Keep experimenting, keep adding removing modifying, and you'll
- 42:32end up with really great images. We talked about the crazy major,
- 42:37Ernie 5 updates, we talked about GPT for updates.
- 42:41There's something interesting that I searched about yesterday,
- 42:43and I want to share it with you guys.
- 42:45It's essentially telling you guys.
- 42:47How do these texts to image prompts work?
- 42:50Because find out was like, hey, tomorrow, what are going to talk
- 42:53about? Was I'm going to talk about my
- 42:54journey. He was a, maybe you can talk
- 42:56about how my Journey works, and I was like, wait a second.
- 42:59I actually have no fucking idea how it works and I feel like we
- 43:03need to share that with you because we did do the research
- 43:06and it is actually Quite interesting is if up until now
- 43:09you had no idea how these a eyes work.
- 43:11You're going to get a better understanding if you want to
- 43:13talk about any sort of deep learning technology.
- 43:16We have to look at the algorithm that is being used.
- 43:18But before looking at the algorithm, in this case, you
- 43:20want to talk about the data set, the data set is how we train
- 43:24these models to essentially, learn and improve and get better
- 43:28over time. So these data sets are
- 43:31essentially images books, journals websites, social, Yeah.
- 43:37Yeah, exactly. All these things that I've been
- 43:40in the internet for years and have been written and added to
- 43:43by us human beings and the data sets for companies such as
- 43:48opening, I can only imagine how large they are, right?
- 43:52So these data sets are then created, but the problem is, how
- 43:57do you tell? For example, the mid-journey
- 44:00dataset, how do you differentiate between a banana
- 44:03or a balloon? So looking at this video
- 44:05yesterday is very Seeing this whole data set.
- 44:09Space is called the multi-dimensional late in space
- 44:12and how it works is for computer.
- 44:15It doesn't see images like, how we do with doesn't have eyes.
- 44:18So for us, it's very easy to differentiate between a banana
- 44:22and a, and a red balloon. But for computer, it's
- 44:25essentially only seeing numbers, right?
- 44:27We're talking about RGB planes, right?
- 44:30The red green and blue, and these planes all have pixels,
- 44:33and these pixels have numbers from 0 to 255, which Define
- 44:38essentially, what the image ends up looking like and for the
- 44:42computer to Define what a banana looks like it starts adding in
- 44:46dimension. So the First Dimension we can
- 44:48look at here is for example yellowness, right?
- 44:52Between all the bananas that are put into the data set, it can
- 44:56sort of tell, okay? I can see a pattern here.
- 45:00A combination of RGB values that together shows me yellowness,
- 45:04right? That I do not see when I look at
- 45:06the red balloon. Once so then I can tell these
- 45:09are yellow bananas. But then what happens if the
- 45:12balloon now is also yellow. How can you tell the difference?
- 45:15Well, then you add another dimension, you add the second
- 45:18dimension, which could be roundness.
- 45:20And then you find patterns in the data sets that you give to,
- 45:25you know, the the machine and then it can tell.
- 45:27Okay, now I know what roundness sort of looks like, now I can
- 45:31tell the round yellow balloon from the banana, but in today's
- 45:36a I models we have like probably 300 400 parameters that defines
- 45:41this Dimension. Yeah.
- 45:42And we go down to the level of shininess like I dropped
- 45:47roughness. Yeah.
- 45:48And and it's crazy like the more parameters that we have probably
- 45:51the more accurate the results will become.
- 45:53And also you can think about the fact that it makes it more
- 45:56complex so you need more computational power to go
- 45:59through all these Dimensions. Now, what's really crazy is
- 46:02also, I've seen the controversial things on the
- 46:04internet about AI. It's About this lead in Space,
- 46:08the multi-dimensional lead in space because some of these
- 46:11comparisons and these parameters are going to be dark because
- 46:17essentially, if you think about opening idea also had to block
- 46:20some of these parameters. Let's say you want to talk about
- 46:24what happened in World War 2, right?
- 46:26You know I'm even afraid to fucking say it on the podcast
- 46:28because you never know how like YouTube algorithm actually looks
- 46:31for these words. But essentially how Kanye went
- 46:33down. Let's just, you know, I don't
- 46:35even need to say it but you guys know what I'm talking about.
- 46:37So for the AI to be smart enough to tell things apart, there
- 46:41needs to be more parameters and sometimes these parameters are,
- 46:45you know, parameters, that are controversial.
- 46:47So that's why open a.i., every time they update their model,
- 46:51they have to look for some of these parameters or Dimensions
- 46:53that are added, that create unwanted results, results that
- 46:57might be too gory to bloody, or racist or sexist.
- 47:01And so, that's why you see with open eyes are always, updating,
- 47:03the model, same thing with mid-journey.
- 47:05Now, what's really funny We did talk to one of our friend
- 47:09developers. His name is Graham.
- 47:11Shout out to you. He did say there was one
- 47:13developer that he met at a tech meeting that he managed to as
- 47:18highly has all of this and yeah you can tell them maybe what?
- 47:21Well yeah so so he bypassed some of these policy rules and
- 47:24created. I think nudity on a journey that
- 47:29it's crazy to see what the outcomes of those things when we
- 47:32remove all these filters. Yeah, but it's probably they
- 47:35took his account down and I mean, you can always create new
- 47:38accounts, right? But you can imagine there's a
- 47:40lot of blocks and filters between, you know, open eyes
- 47:44data set and you the user. Because if it was completely
- 47:47open, you can essentially search and get anything you want.
- 47:50And I'm sure they don't want that because it's going to go
- 47:52crazy. But that is the, the topic of
- 47:55the the latent space. Let's talk about the actual
- 47:58algorithm. That generates the images.
- 48:01There's two algorithms that we're going to discuss today.
- 48:02The first one being Gann, which stands for generative
- 48:05adversarial Network now, Algorithm uses generators and
- 48:10discriminators. So the generator within this
- 48:12algorithm creates an image and the discriminator, you can think
- 48:16of it. Like a police that is watching
- 48:18every single image that is generated and its defining if
- 48:22the image looks real or fake, and how it defines.
- 48:25That is, based on the data set is like, oh no, this human has
- 48:28like, five eyes. That's not real go and recreate
- 48:31it again, right? So Sailors.
- 48:32I like, you will come to Liza's the generator.
- 48:35Okay, right now, let me sense. Yes, So so you will generate as
- 48:38long as a discriminator approved.
- 48:41The final output. Then only will publish the
- 48:43results. Exactly.
- 48:44And we're not the discriminator. Just when was this every time?
- 48:49The discriminate is like, this looks real, this looks fake
- 48:51District. This looks good.
- 48:52This looks fake. But then, there's some issues
- 48:55with Ganso have you're doing a bit of research, we realized
- 48:57scan is not the most stable algorithm to take.
- 49:01This is, of course, a lot more complex when you go down to the
- 49:03mathematics, and the details were just giving you an overview
- 49:06now, Again is a model that I believe my journey was using,
- 49:10but they have now either completely switch to a diffusion
- 49:14model which were going to discuss now or they're probably
- 49:17done a mix of diffusion and Gano, probably other algorithms,
- 49:21but diffusion, which stable diffusion uses as well, is
- 49:24essentially using gaussian noise and it essentially takes your
- 49:29prompt breaks it down into small parts, that the system can
- 49:32understand. And then he goes and looks for
- 49:34those prompts. In its data set and it generates
- 49:40a preview or beginning image and then it starts using the
- 49:43gaussian noise to add and take away and then predict what the
- 49:48image will look like. Which is why I believe when you
- 49:50go to the journey or stable diffusion, you always see a
- 49:54noisy image, right? And then it's a general percent
- 49:57and it's going up to thirty percent 60 percent, 90 percent.
- 50:00But what's actually happening behind the scene?
- 50:02Is crazy because these sort of generations is essentially for
- 50:07Of noise being added and removed according to our understanding,
- 50:11and it's happening on a very rapid Pace, even though the
- 50:13process is huge, you can imagine going through all those
- 50:16Dimensions that we just talked about.
- 50:18So, this is essentially how the fusion works.
- 50:21So now we have covered GPT and mid-journey.
- 50:25Yeah, I think these are two most practical AI tools that we use
- 50:29on our daily basis. We give you the updates that,
- 50:31you know, it just happened in the past week and we are so
- 50:34excited to try these new things. There are other AI Tools out
- 50:37there which I have a few names that I'm going to tell you guys.
- 50:40If you're going to go on, try it which will help your workflow.
- 50:43One of them is Adobe podcast. So we haven't used it for our
- 50:46podcast, but it apparently enhance the audio quality of
- 50:50your mind. You've seen the videos, right?
- 50:51I've seen like, you can get this quality like with this expensive
- 50:55Mike with a very small more than iPhone.
- 50:58Yes, an iPhone. So what is Adobe podcast?
- 51:01Another interesting one that I saw was called, I think dummy do
- 51:04um, was that about it? When's your podcast to YouTube
- 51:08shorts, it automatically cuts transcripted and and make it
- 51:13ready to publish on YouTube choice.
- 51:15Is it out? Then you can request for a wait
- 51:18list so that that's another one which was already.
- 51:20Yes, I agree. So I just want to try it with
- 51:23another another to which I use for Content writing or blog post
- 51:28is copy Ai and also Jasper Dale. I so Jasper is a half not so
- 51:33usually people use it to write blog posts website content. 10th
- 51:38on social media. I've seen a lot of people using
- 51:40why won't you want to use like GPT for four different models so
- 51:44they will give you different results, Jasper.
- 51:45And copy that AI are really focus on content writing.
- 51:49Okay? So so if there are categories,
- 51:51you want to write blog posts, website content, social media
- 51:54flyer, like like down, drill down to the details.
- 51:58Okay, yeah. And then there was one that you
- 52:01mentioned. Wonder Dynamics, do you want to
- 52:02talk about it? Yes.
- 52:03So, what a dynamic, sorry, I almost forgot about this one or
- 52:06Dynamics. We did mention it in our
- 52:09broadcast channel, we have a broadcast channel, by the way,
- 52:11on Instagram, if you guys haven't checked it out, it's
- 52:13essentially a private group, called secrets of a decision in
- 52:17which we send out all these news that we find out before we
- 52:21talked about on the podcast. So if you want to stay up to
- 52:23date, make sure you join the broadcast channel.
- 52:25Now, I do want to say one thing, Wonder Dynamics is a website we
- 52:29found out about just a couple of days ago.
- 52:32Essentially, it takes a feature film sort of video, Gio's blow a
- 52:37video that you record from The Real World.
- 52:40Let's say I record firehud running on the street and I feed
- 52:43it that video, then it will analyze Farhad as a human in the
- 52:48video. It will analyze his body
- 52:50movement so you will attach an Armature to him.
- 52:53Analyze his body movement analyze, his facial movement,
- 52:57and then it has a list of 3D characters that other artists
- 53:02have created or you can upload your own 3D character and Get
- 53:07and drop it onto Farhad and then you will remove fat heard.
- 53:12It will automatically remove fat from the video and replace him
- 53:16with the 3D character and the 3D character will also have perfect
- 53:21lighting and composition according to your seen, all of
- 53:24this sound fucking unbelievable. Is it a bait listing?
- 53:28Or is it open? It is on a waitlist.
- 53:31So you have to request for Access.
- 53:32It's in closed beta but I have not seen people with videos the
- 53:36video. Oh, is on the website?
- 53:37Look too good to be true, is crazy.
- 53:40The whole concept, like reading their character, like, replacing
- 53:44lighting, all this take hours like these to do.
- 53:48And, and for the AI to be able to all of that cleanly, it
- 53:52doesn't make any sense to me. And honestly, I do not want to
- 53:55trust Wonder Dynamics. However, there's this thing
- 53:59where you go to the team and you see the one of the founders is
- 54:03the actor, the main protagonist of Ready Player.
- 54:06One. That gives it credibility, then
- 54:09on the board of advisors, you have Steven Spielberg, you have
- 54:12loose. So the guy who was behind Marvel
- 54:15Avengers and all that. So, you have this All-Star team
- 54:19that is the board of advisors, and you have two Founders.
- 54:22One of which I know from Ready Player, one, it looks like these
- 54:26guys know what the fuck they're doing.
- 54:28And on top of that on March 21st, 22nd or 23rd, at one of
- 54:33the Nvidia talks, I believe at GDC, these These guys Wonder
- 54:37Dynamics is doing a presentation and we're going to watch, let's
- 54:41watch them and see what like, I'm really Keen to try it.
- 54:43Exactly. So if these guys are going there
- 54:46and this is a team they probably will not release a product that
- 54:50does not work. But the thing is, if this does
- 54:52work, you have no idea like all the limitations and
- 54:57capabilities. All of the limitations that it's
- 54:59bypassing to reach that final output.
- 55:02It means we have so much advancement in CGI and VFX that
- 55:06a lot of the That we're doing right now in blender.
- 55:08And on religion is just extra work that we don't even need to
- 55:11do if the a, I can do it. So, looking forward to wonder
- 55:15Dynamics, go and request for Early Access for 12 Dynamics.
- 55:19If you can, and the moment it comes out, we'll be testing it
- 55:22out. So make sure you stay tuned for
- 55:24that. If you guys have any other tools
- 55:26that you want us to try or analyze, put them down in the
- 55:28comments, will be trying out all the cool things.
- 55:32Like I've tried instruct pics to pigs, what was it?
- 55:35Dream excite. All those different Tools that I
- 55:37posted a video on our social media, you can follow us on
- 55:40Tick-Tock Instagram, Twitter, we will post all of our, you know,
- 55:43small content. But we really are looking for
- 55:46practical use cases, like GPT and me journey in our database
- 55:50who can help and speed up our processes.
- 55:52Yeah, it was great talking about AI.
- 55:54This past week has been full of News full of information and
- 55:58next week we're going to have a little surprise.
- 56:00That's right. Actually.
- 56:01Oh, that was the news I want to talk.
- 56:03Okay, that was the news I want to talk.
- 56:07We have been dreaming of doing this as part of the podcast, but
- 56:11it took us five episodes. Not that long to get it, but
- 56:14Farhad has already requested three guests for the next three
- 56:19episodes. And how we're doing this is
- 56:21we're going to be essentially inviting guests on the podcast.
- 56:24So this podcast is not just going to be about me and Farhad
- 56:27talking about topics. It's a chance for us to talk to
- 56:30other people. They are artists 3D artist
- 56:33business, you know, business owners.
- 56:36Yeah. And then Creators, all, we want
- 56:38to get everybody's opinion. So I think talking to other
- 56:42people and how they made it or what is the process?
- 56:45Like it gives you a better understanding of processes and
- 56:48maybe it can inspire you as well as Inspire us.
- 56:50We're going to be announcing who the guests are on our social
- 56:53media platforms, like, Instagram.
- 56:55So make sure you guys stay tuned, but I think that is it
- 56:59for today. Episode number five, we fucking
- 57:01did it. Thank you so much.
- 57:03Thank you so much. We'll see you on the next
- 57:05episode.