4/4/2025 - Navigating The Nuances of ChatGPT, Midjourney, and Digital Twins

Description

Chapters

Transcript

0:03 [Music]
0:09 Girl, y'all
0:11 [Music]
0:15 looking every fellas got you in
0:22 here. What you doing with a clown like
0:28 me? Surely one of life's
0:32 little
0:35 mysteries. So
0:37 tonight I
0:40 ask
0:42 [Music]
0:44 about
0:45 how every
0:52 [Music]
1:03 Friday night date night people. You got
1:06 your nachos?
1:12 [Music]
1:16 She came on here like
1:20 [Music]
1:24 slow was warmer than a look in her
1:29 [Music]
1:32 eyes. She sat on a stool and he said,
1:35 "What do you want?"
1:38 [Music]
1:40 She said, "Give me a love that don't
1:43 freeze up
1:47 [Music]
1:52 inside." Said, "I have
1:54 melted in my time,
1:59 dear." But to sit next to you, well, I
2:02 shiver and shake.
2:08 And if I knew love, well, I don't think
2:11 I'd be
2:15 here. Ask asking myself if I've got what
2:19 it
2:21 [Music]
2:23 takes to melt your icy blue
2:31 heart. What are you doing sniffing
2:33 around? What are you doing? What are you
2:35 looking for?
2:36 I got no cheese. Ain't nobody got time
2:39 for that. You know we got that cheese.
2:41 Ain't nobody got cheese for that. That
2:44 dog's weird tonight. What are you doing?
2:46 Why are you sniffing? Get
2:47 out. He just walked under my
2:50 [Laughter]
2:52 desk. Happy Friday night,
2:55 everybody. Let's flip the cameras. Let's
2:58 lower the angle of the dangle.
3:02 We want to lower the angle of the
3:06 dangle. There you go. There you
3:09 go. All right. Fantastic. Bob, tell
3:12 everybody he's won. Uh, we haven't won
3:16 anything. It's It's not a game show. Not
3:18 a game show. I say it week after week
3:22 and uh nobody seems to understand.
3:24 Nobody listens to
3:27 Bob. Not even sure why I work here.
3:34 [Music]
3:41 I can Can I has cheese, please? That's
3:43 totally what he was looking
3:45 for. There's always room for nachos.
3:48 Yeah, nachos. You know what's nice about
3:51 nachos? They can be an appetizer. I
3:55 mean, you could call them a meal if you
3:56 get the loaded ones. But even with
3:58 loaded
3:59 ones, there's something somehow
4:01 unsatisfying about them, but you can
4:03 also use them as a
4:05 postdessert snack, right? Like you have
4:08 a big dinner and you're like, I couldn't
4:10 possibly eat another thing. Not even a
4:12 laugh a thin mint. And then someone's
4:15 like, you want some nachos? You're like,
4:17 well, yeah, I can have some
4:18 nachos. And then you eat like a whole
4:20 pile of nacho. You have one and you're
4:22 like, well, I need all of them now. Then
4:24 you eat them all and you're like, I
4:25 guess it's time for dessert.
4:28 [Music]
5:03 That's exactly right. Would you like a
5:05 stack of tortillas? No, but if you cut
5:06 them into quarters and deep fry them and
5:08 throw cheese on them, then I'll have
5:13 some. Nice Monty Python slip in there.
5:16 Thank you very much. It's nice to have
5:19 some Gen Xers in the house that catch
5:21 those sort of
5:27 things. Real question, real cheese or
5:29 queso on your nachos? I go for the real
5:32 cheese. There's something about queso.
5:34 Well, I mean, if you're talking about
5:36 like the canned queso, like that that
5:38 gross gross [ __ ] actually does kind of
5:40 gross me out. Um because it makes the
5:43 the the nachos too soggy.
5:46 It's like it's like well, okay, you
5:48 can't leave an Oreo in milk too long.
5:50 That's not possible because a completely
5:52 overdrenched Oreo is [ __ ] delicious.
5:56 But it's kind of like that. Like if you
5:57 if you use the queso, you turn your
6:00 crunchy nachos into into like salty Oreo
6:04 wafers without the sweet middle. It just
6:06 doesn't work. So, I'm I'm a real cheese
6:09 fan. Please stop talking about nachos. I
6:13 have no nachos, but they sound so good.
6:16 Well, this is this is Listen, we have
6:18 tried to prepare everyone for Friday
6:20 night date night. Your other option is a
6:22 is a hot pocket cooked in a dirty
6:24 microwave you still have from college.
6:27 Why not put it in the sleeve? Make it
6:30 burning hot on the outside, frozen in
6:32 the middle. You know, it's
6:33 [Music]
6:36 delicious. Oh, your house is divided on
6:38 the queso dip. Yeah. Well, listen. If if
6:43 you've got a chef in the house that
6:45 makes real queso dip like melted cheese
6:48 with some cream and you know like really
6:51 doing it right. Sure. Techie Becky. Hey
6:54 Kyle. Always a treat watching you. Oh,
6:56 thank you so much Techie Becky.
6:58 Appreciate that.
7:07 I had fun tonight at at
7:10 um going to the gallery scene here in
7:13 Denver.
7:16 [Music]
7:17 Um while while I'm an artist and while
7:20 my wife's an artist, I galleries just
7:23 bore the [ __ ] out of me. They just do.
7:26 Like, you know, I like the art, but like
7:29 I don't like like I can take it in
7:31 really quickly. And so it it drives
7:34 there's people that go just stare at
7:36 paintings for 20 minutes. I'm like,
7:39 "Don't you have [ __ ] to
7:40 do?" So So I tend to be bored on these
7:44 nights, but now that I have chat GPT, I
7:46 was experimenting tonight with um with
7:50 image generation and what it can do. And
7:51 it's it's pretty crazy. I didn't do a
7:53 ton of stuff, but it it occupies my
7:55 brain while I'm standing looking at
7:58 artwork, you know, some of which is
8:01 absolutely beautiful, and then there's a
8:03 bunch that's just like, eh? Like,
8:07 [Music]
8:16 why?
8:19 [Music]
8:29 [Applause]
8:31 [Music]
8:46 Ow. Didn't mean to cause you any
8:50 [Music]
8:53 sorrow. Didn't mean to cause you any
8:59 [Music]
9:01 pain. Only want one time to see you
9:07 laughing. Only want to see you laughing
9:10 in that purple rain. Purple rain. Purple
9:14 rain.
9:16 [Music]
9:20 Purple rain. Purple
9:27 rain. Purple rain. Purple
9:31 [Music]
9:35 rain. Only want to see you laughing in
9:39 that purple
9:41 rain. Um, yeah, good point.
9:46 Listen, Tik
9:48 Tok, it's been
9:50 saved for 75
9:56 days. It's been saved for 75 days.
10:13 [Music]
10:26 Freedom came my way that night.
10:29 [Music]
10:32 Just like a jet play in and out on the
10:37 side. I was hauling ass on a million
10:40 miles and
10:42 I wonder how hard I'd
10:46 [Music]
10:51 hit.
10:52 Um, okay. So, welcome if you're new
10:57 here. We got it's a it's a it's a it's a
11:00 small but appreciative crowd tonight.
11:02 Kind of like a gallery opening. It's
11:05 [ __ ] snowing here in Denver. By the
11:07 way, so there weren't a lot of people
11:11 out. It was a small but appreciative
11:15 crowd. Um but if you're new here, I'm
11:17 Kyle Shannon. This AI learning lab and
11:20 uh we'll talk about AI stuff
11:23 philosophically, practically. I'll go
11:25 show you some new stuff in midJourney
11:31 tonight. Gild or Yamaha on the guitar.
11:34 It's a guild. It is a guild. A guild
11:39 guitar. This is my favorite
11:43 guitar. Um, by a lot. It is a D30AB. I
11:48 bought it in 1987.
11:52 Um, I was hired as the
11:57 um the jester for the Pennsylvania State
12:01 Renaissance
12:02 Fair in State College, Pennsylvania. So,
12:05 I juggled for six days and got
12:08 $1,700 and I spent $1,100 on this and
12:13 $700 on a mountain bike or $600 on a
12:16 mountain bike.
12:21 cuz cuz I know how to make good
12:23 financial
12:26 decisions. But this is the best guitar
12:28 I've ever owned. The next beautiful It's
12:31 all [ __ ] beat up cuz you know I don't
12:33 take care of
12:38 [ __ ] It's got really beautiful like
12:41 flame
12:44 maple the maple back. It's a beauty.
12:47 It's a
12:48 be and it's just got it's just got such
12:51 cool bottom
12:54 end. Why the bike? Cuz I wanted a
12:56 mountain
12:57 [Music]
13:14 bike. I mean it's warm. It can be boommy
13:17 if it wants to. It's got nice high
13:20 end, but it's not like
13:22 [Music]
13:31 tiny. It's a [ __ ] mess, though. My
13:34 frets are all [ __ ] worn
13:36 out. The next the action's too
13:39 low. It's a
13:42 mess. It's been banged up. It's got a
13:45 crack in the headstock where someone
13:47 knocked
13:48 it. Um, okay. So, here's some
13:52 news. Here's some news for you. So, you
13:55 remember a month or so back they came
13:58 out with the 03 thing, right? And
14:00 they're like, we'll give you a little
14:01 taste of 03 and then if you use deep
14:04 research that uses the real 03, but no
14:06 one knows that. And then we have 03 mini
14:09 and you can use that. But but we
14:12 understand you hate all these different
14:14 models. So we're gonna end that. GPT5 is
14:17 coming out like in May and so you'll be
14:20 good. No more model picking. We're just
14:22 going to do a hybrid fully integrated.
14:24 Going to be beautiful. Yeah. They
14:27 unannounced that two days ago or
14:30 yesterday or today. I don't [ __ ]
14:32 know. It doesn't matter when. I think it
14:33 was this morning. So, what they're what
14:36 they are going to be launching now is
14:39 03. And if you don't if you're not
14:42 keeping track of the numbers, 03 is the
14:44 the Mac Daddy reasoning engine. And then
14:47 they're also going to be releasing 04
14:51 Mini. 04 Mini. We we haven't heard of 04
14:54 before, but it's the one after 03. Now,
14:58 you might say, well, 03 was the one
15:00 after O2, right?
15:02 No. No. 03 was after 01, but someone
15:06 else had the trademark for O2. They so
15:09 they had to skip to 03. Everybody knows
15:11 that. That's common knowledge. It's
15:13 common sense. You don't just go from 01
15:16 to 02. People like, what are you
15:19 losers? So, they're like, we're not
15:21 losers. 01 03 04. Wait, where did 04
15:26 come from? It came after 03. Oh, no one
15:30 talked about that.
15:32 So, we're going to get 04 mini and
15:40 03. It's It's exciting,
15:44 right? I know it's
15:47 not. It's not exciting. It's like, wait,
15:51 we have two more [ __ ] models. We've
15:53 got to slightly figure out the nuance
15:55 differences
15:57 of 40
16:00 4.5 and 40 stands for four omni and 04
16:05 stands for open
16:08 AI4. So the O's if the O again fairly
16:13 obvious if you don't get this you're a
16:15 bit of a dumb dumb. If the O comes
16:17 before a number it stands for open AI.
16:20 If the O comes after a number, it stands
16:23 for omni. Omnimodal. Multimodal. Omni is
16:27 just a different word for multi, but
16:28 it's multimodal. But they use O because
16:31 to keep it consistent, right? So the O
16:35 before open AI, the O after multimodal.
16:38 Make
16:40 sense? So clear. So you got 40, you got
16:44 4.5, but not 450, just
16:48 4.5, right?
16:50 Then you have, oh, what? Let's go
16:55 look. This is really exciting. This this
16:58 one of the things that excites me most
16:59 about AI is keeping track of of
17:02 different models within the same [ __ ]
17:04 tool. Okay, 40, right? Great for most
17:10 questions. Oh, and then you can do 40
17:13 with tasks if you want to schedule those
17:15 prompts. like every morning you could
17:17 have a prompt that gives you an egg
17:19 recipe and and it sends it to your
17:21 inbox, right? So that's exciting. And
17:23 then you have GPT 4.5, but it's not GPT
17:26 4.5. It's GPT 4.5 not O because what
17:30 does O stand for? Multimodal, right?
17:33 Obvious, clear. So port 4.5, there's no
17:38 O. So it's
17:41 monomodal and I think everybody knows
17:44 that. It's it's like it's instinctive.
17:46 It's in the name
17:49 4.5. Oh, that must not be multimodal.
17:52 Right. Exactly. Okay. YouTube comment.
17:55 Yes. Yes. What can I say? My brain can't
17:58 handle this. I'm making a GPT to help me
18:00 track all
18:02 this. But wait, we're just getting
18:04 started. So, but then you use you you've
18:07 got 01. So 01. Now what did I teach you
18:11 before? The O before the one stands for
18:14 open AI, which means it's a reasoning
18:16 model because Open AI makes all the
18:19 models. But the one with the O in front
18:21 of it, those are the reasoning models.
18:23 Make sense? Good. Clear. Right. Clear.
18:26 So 01 is like the Mac Daddy
18:31 01, but and and we're on 03
18:36 mini right below it there. But you can
18:40 also use 03 mini high. See, and this is
18:43 the thing. So, if you've got a mini,
18:46 mini means smaller, but mini high means
18:50 smaller, better, but it's not as good as
18:53 03 or 01 pro mode, but we don't have 03
18:57 yet, but that's what's coming. So, we're
18:58 going to also have 03. And then what's
19:01 really exciting is then below 03, we're
19:04 going to have 04 mini. And hopefully,
19:07 fingers crossed, we'll have 04 mini
19:12 high. And then of course you can go in
19:14 and there's extra models down here where
19:17 where you can you can get uh GPT 40 mini
19:21 because remember 404 stands for
19:23 multimodal. So for multimodal mini so
19:27 you can use that and why do you use
19:28 that? Because it's faster and but
19:31 different. And then but you can also use
19:35 GPT4 the the legacy model that is a it's
19:40 it's a it's a mixture of experts model
19:42 with with 10
19:44 different
19:48 well that's why I like open AI because
19:52 it's so
19:54 intuitive and then but you you if you
19:57 pay $200 a month then you get this menu
20:00 item called operator where you can click
20:02 on out and it will take you to something
20:05 that looks like chat GPT but it says
20:08 operator and that's how you know it's
20:10 different because it says
20:11 operator get it but then if you go back
20:15 to chat GPT the other thing you can do
20:17 is you can click on Sora and you can
20:19 make
20:20 movies because or or pictures but now
20:24 now it's pictures because you take image
20:27 gen which used to be called Delali and
20:29 and you put image gen So Sora was just a
20:33 video maker and it didn't have image
20:34 generation. If you wanted images, you'd
20:37 go to chat GPT, the the 40 model or the
20:41 4.5 model or the 4.0 40 mini model or
20:45 the four model and those all use. You
20:47 couldn't do it with the reasoning
20:48 engines, but you use Ali and but now the
20:53 exciting thing is they've got image gen
20:55 this new image tool that's super good
20:58 and then they put that in chat GPT in
21:01 some of the models like I said before so
21:04 it should work the same except they
21:06 don't work in custom GPTs that's still
21:08 deli because that everybody knew that
21:10 and but then what they did image gen is
21:13 now in Sora which is a separate URL alto
21:16 together it's sora
21:17 com. So, so because they don't want it
21:21 to be
21:22 confusing, right? So, you make if you
21:25 want four images at the same time, you
21:27 do it in Sora. If you want one image in
21:30 context of a conversation, you do that
21:32 in 40 or 4.5 or 40 mini or or
21:37 GPT4. Make sense?
21:41 I I mean I just I feel like one of my
21:43 gifts
21:44 is like demystifying like you know
21:47 something that like for me like this is
21:50 all super obvious right it's super clear
21:52 but like for most people they're like
21:54 huh I don't get it and that's I think I
21:56 feel like that's my gift is just making
21:58 it super simple for so like you get it
22:01 right were you I you were you taking
22:03 notes I hope you were taking notes all
22:07 right Friday night 8 My my head
22:12 hurts. I need a a notebook. Yeah. By the
22:16 way, all of what I just said is it this
22:18 is [ __ ]
22:20 ridiculous, Carmen. Um, how can I create
22:23 my first digital twin? I will tell you
22:25 that. Okay, so
22:33 um there's lots of ways to do it.
22:37 Um, one one one kind of simple weird
22:43 way. Let me go. Here I am in chat. GPT.
22:47 Uh, here. It's really important what
22:48 model you choose. No, it's not. It
22:50 doesn't [ __ ] matter. Okay. So, we'll
22:52 just pick the generic
22:55 one. I'm just going to wait for them to
22:58 start picking the models for
23:00 me. Oh my god.
23:05 Don't we have a law archived about that?
23:07 Oh yeah, we do
23:10 actually. If you want to make a digital
23:13 twin, go to the AI salon. Oops, this is
23:17 in the wrong position, isn't it? Sorry,
23:18 Tik Tockers.
23:26 Um, let's see.
23:29 When you put it all that way, I'm rather
23:31 impressed with myself for having kept up
23:33 with that progression. Yeah, it's it's
23:36 [ __ ] crazy. It it really is bonkers,
23:38 but whatever. It's It's fine. It's just
23:40 we're we're early in the industry, so
23:42 shit's weird right now.
23:44 Um past events. Okay, so so I'm going to
23:48 go if I just go to the feed and then I
23:51 search for
23:53 um digital twins.
23:57 Dig
23:58 it
24:02 twins learn out loud. Yeah. So, okay.
24:06 So, if you're not a part of the AI
24:09 salon, go to the salon.ai, say join our
24:12 community. That's going to pop you over
24:13 to this mighty network site. That's the
24:15 salon
24:17 community. And just go to the feed at
24:19 the very top. And then when you do that,
24:22 you can search across everything in the
24:24 salon and and uh search for digital
24:27 twins and then you'll see there's a
24:29 learn out loud in here and there's a
24:30 recording a recording of it as well. So
24:34 that's a way to do it. Um there's
24:36 another way that you could you could do
24:37 something kind of fun is you could say
24:40 just go into chat GBT assuming you've
24:42 got memory enabled. It remembers stuff
24:46 about you. So you could say
24:49 um based on
24:54 everything you know about me
25:00 um
25:02 write
25:05 me let's
25:10 see a
25:14 description of
25:17 me, how I
25:19 think, my world
25:21 view, how I
25:27 write, reason, and create.
25:33 create.
25:35 Um, I'm going to
25:39 use your report
25:42 um, as the basis for a digital
25:47 twin. And then what I'm going to do is
25:49 I'm going to pop that into a
25:53 canvas. All
25:55 right. And so a canvas is just an
25:57 editable document. And so now Kyle's
26:00 digital twin
26:03 profile. Kyle is a boundary pushing
26:05 creative technologist, storyteller, and
26:07 entrepreneur who thrives at the
26:10 intersection of imagination, innovation,
26:12 and human connection. With a background
26:14 that spans theater, technology, and
26:16 content creation, Kyle brings dynamic
26:18 right how Kyle thinks. Kyle thinks in
26:20 systems and metaphors. He doesn't just
26:23 look at how things work, how Kyle
26:24 writes. Kyle's writing is personal,
26:26 bold, conversational, often using
26:28 rhetorical questions. I don't even know
26:30 some of these words. Lived experience
26:32 and dynamic narrative voice to pull his
26:34 readers to sleep. Wait,
26:37 what? He writes like he speaks directly,
26:40 engagingly, and with a twinkle. Great.
26:41 Twinkle of mischief. Nice. How he
26:44 reasons. His re reasoning is nonlinear
26:46 but not erratic. Do you hear that,
26:49 people? My reasoning is nonlinear but
26:52 not erratic.
26:54 You know, if you just came in here and
26:56 you were like, "What's this channel
26:57 about?" You're like, "This seems mighty
26:59 erratic." But it's not. You know how I
27:03 know it's not? Because chat GPT told me
27:05 it wasn't. And if Chat GPT tells it to
27:09 you, it's true. We all know that. That's
27:12 common sense. Okay, moving on.
27:15 Um, he builds arguments like jazz,
27:18 improvisational but grounded in
27:20 structure, often circling ideas through
27:23 multiple ang Jesus, this is blowing
27:26 sunshine way up my ass. Okay, so anyway,
27:28 so there there's all that. So what you
27:30 can do is if you've been using chat GPT
27:33 for a while, it will probably write you
27:34 a decent thing like this. It did not put
27:36 that in a canvas for me, did it? Oh,
27:38 yeah, it
27:39 did. Yeah. Okay. So, this is editable.
27:42 So, I could go in here if if it got
27:44 stuff wrong, I could go in here and edit
27:45 it. And then what I what I what you can
27:48 do, what I would do is I would just copy
27:50 and paste this whole thing and throw
27:53 that into a custom GPT as the training
27:55 data. And then you just work on your
27:57 prompt to get it to speak from you in in
28:00 the first person. So, that's a way to do
28:02 it. Another way to do
28:05 it is to write a like a 20 questions.
28:10 So, one of the things uh chat GPT is
28:12 really good at is you can say um I
28:16 want to create
28:21 um a set of
28:26 structured
28:28 questions that are
28:31 designed to
28:35 capture the essence
28:38 of Who someone is, how they
28:46 think,
28:49 reason,
28:50 engage with the
28:53 world, solve
28:57 problems, and inspire or something like
29:00 that. Now, this could be anything.
29:03 So the thing about digital
29:05 twins
29:09 is you you by definition are kind of
29:13 divine uh
29:16 designing a version of yourself right
29:20 now. So you could try to do one that's
29:22 very balanced that captures who you are.
29:25 But what if you want one that's funny?
29:28 Like maybe you're funny when you get
29:30 drunk at parties.
29:32 There there might be a reason to have a
29:34 digital twin that's that's that guy,
29:36 right? Or you might want the one that is
29:38 the college professor, right? And is
29:41 going to be very serious and thoughtful
29:42 and
29:44 academic. I will answer the questions
29:46 with the utmost
29:48 authority. So when you ask it to create
29:52 one of these structured interviews,
29:56 um guide it in that direction. You could
29:58 say, "I really want to." You could you
30:00 could even say,
30:03 "Um, the
30:06 interview should
30:08 balance
30:11 capturing
30:13 capuring their humor
30:17 um and their
30:20 um
30:22 trust,
30:24 worthiness, whatever." So, so you kind
30:26 of write a description of what you what
30:29 what you want the interviews to be and
30:31 then you go
30:33 boom. Chat GPT is really good at this
30:36 stuff. And again, you could give it a
30:38 role and say you're a social
30:39 anthropologist, whatever it might be.
30:41 It's like there have been so many
30:44 structured interviews written in the
30:47 world that have have been uploaded to
30:49 the internet. It is deep deep in the
30:52 training data. There's so much of it.
30:55 Okay. Okay, the core essence interview
30:56 your your the you blueprint one um your
31:00 inner operating system when you are
31:02 faced with something complex or unknown
31:04 what's the first thing your brain does
31:06 and then it gives you purpose and
31:07 follow-up. So anyway, you you get it to
31:10 write you this structured interview, you
31:12 edit the questions and then here's the
31:13 secret that we
31:17 found. If you write the answers out, the
31:20 digital twin will suck.
31:25 So what you want to do is you want to
31:27 record video record an interview and
31:30 ideally you have someone that you trust
31:34 asking you the questions so that all you
31:36 have to do is answer them and if you get
31:38 a good friend to do it they'll be able
31:41 to push you and say dig deeper on that
31:43 answer. So you you generate this this
31:46 question set then you get interviewed
31:49 and then you use the transcript of that
31:51 interview. You don't even have to edit
31:52 it. you can leave in the other person's
31:54 [ __ ] in there. It's amazing. And and
31:56 then you just upload that as your
31:58 knowledge base to a custom GPT and and
32:01 that'll be a a a pretty remarkable
32:03 digital twin. That's how we do it at
32:05 Content Evolution. We we came up with
32:07 our specific question sets and every
32:10 single we call them digital advisers
32:12 now. Every single digital twin digital
32:15 advisor um the person has done a video
32:18 interview of those same questions. Um,
32:21 and that allows us to compare them, you
32:23 know, compare and contrast them. So,
32:26 does that answer your question? I think
32:27 it
32:33 does. All we have to do is explain this
32:36 to the person who's seeing the GPT for
32:38 the first time. Make a twin just like
32:41 me. Only funny.
32:46 Exactly. Um, and here's the other trick
32:48 with the digital twins. like what what
32:50 you've got to really you you've got to
32:53 muck with the prompt, right? So, how it
32:56 behaves with you? Do you want it to like
32:59 you could create it so that your digital
33:02 twin um whenever it answers a
33:05 complicated question, it always includes
33:07 a story. I don't know. And maybe it
33:11 pulls stories from the from the
33:12 interview. Um or maybe you want it to be
33:16 very factual, right? So you can you can
33:19 kind of dial in how you want it to
33:21 behave. And the only way to really
33:24 um dial it in is to use it. This is one
33:27 of the reasons that I use pictures of
33:30 myself, all those sort of perverted, you
33:33 know, self-portraits that I've that I've
33:35 done is that I know what I look like.
33:38 And so when when the the model comes
33:41 back at me with something, I'm like, ah,
33:43 that's not quite right or oh, that's
33:44 pretty good. Same thing with the writing
33:46 with the with the digital twin. Interact
33:49 with it and if it doesn't feel like you,
33:51 you'll you'll know it instantly and then
33:52 just give it guidance and just guide it
33:55 to, you know, sound more like what the
33:57 interview, you know, what the transcript
34:00 sounds like and it'll it'll it'll do it.
34:02 It's pretty pretty
34:03 amazing. All right. What are the best
34:06 use cases for twins? Can they do my
34:08 laundry yet? They can't do your laundry
34:10 yet, but that's coming.
34:12 Um, so the the use cases where we've
34:15 been using them in content
34:18 evolution, I think we've got 35 of them
34:21 now, 35 different people. So each of
34:25 those people have their own digital
34:27 twin. Um, so with your own digital twin,
34:30 you can use it for brainstorming. It's
34:32 really good to brainstorm with yourself,
34:34 weirdly.
34:35 um because you have the same worldview
34:38 as
34:40 you, but it will just spin up ideas,
34:43 right? It's pulling from large language
34:45 models. So, brainstorming is a really
34:46 good one. Writing articles. So, rather
34:49 than having chat GPT write you a
34:51 LinkedIn article, have your digital twin
34:52 write you a LinkedIn article. Um that's
34:55 really good.
34:57 With the content evolution one, we made
34:59 a master
35:00 GPT where we created a Google sheet that
35:03 contains all of the individuals, some
35:05 metadata about the people, and then
35:07 their interview, right? So, we've got
35:09 one giant data file with all the
35:12 different people in it. And so, we have
35:15 a single GPT that's got all of those
35:17 people in it. So, we can take a topic in
35:19 the world or like a a LinkedIn article
35:21 or a report from McKenzie and we can
35:24 upload it into that thing and then
35:26 basically get a content evolution
35:28 position on that paper that takes into
35:32 account all of the points of view of the
35:34 35 people in there. And then we've got
35:36 it set up in such a way that it um it
35:40 figures out who are the three most
35:42 relevant people to comment on that
35:44 article and it it generates a quote for
35:47 each of them on the article. So it'll
35:50 write like a LinkedIn post and then it
35:52 will quote three different members of
35:54 content evolution and then what we do is
35:56 we check with the member and say is this
35:58 something you would say? Do you approve
36:00 this? So that's a cool way you can use
36:03 them.
36:04 Um, I think using using them like a
36:08 digital board of advisors can be really
36:11 interesting. So, if you do a digital
36:13 twin of yourself and and you know your
36:17 your compatriots at work, if you're on
36:19 the executive team, for example,
36:24 um, oh, they talked about this in the
36:26 business guild. Cool. This week, great.
36:28 Yeah. Um, in fact, we're putting
36:31 together an offering right now where
36:33 we're doing this for companies at
36:35 Content Evolution. We're going to help
36:37 them set up a weekly collab where they
36:39 meet and prototype with AI. And the
36:42 first project that they build is a set
36:44 of digital twins where they create a
36:46 digital board of adviserss. And what's
36:48 cool about that is if you've got
36:50 multiple digital twins in a single
36:54 GPT, you could have this the prompt set
36:57 up so that any topic you give it,
36:59 everybody weighs in on that topic and
37:01 then there's a final
37:04 recommendation. So without having to
37:06 have the people there, you can have
37:08 multiple points of view on any given
37:09 topic.
37:10 So, all
37:13 [Music]
37:15 right. Here, let me show you some fun
37:18 [ __ ] I did tonight in the uh in the art
37:28 world. All right. So, you can all see
37:31 this sort of. There you can. All right.
37:34 So, I took a picture
37:37 of that wall. So, it's a wall with a
37:40 mural and like there's some garbage in
37:43 the alley and there's a truck in the
37:45 way. And so I thought, wouldn't it be
37:48 cool if Chad GPT could just fix
37:52 it? So I said, remove the truck and make
37:56 a a beautiful painting of the woman. And
37:59 this is what it
38:01 did. Bang.
38:03 like, you know, 10 seconds later, 15
38:06 seconds later, no truck, no
38:09 garbage. Kind of a beautiful
38:13 reinterpretation of a fading
38:17 mural. That's kind of
38:20 cool.
38:22 Um, then there was this piece of art
38:24 that was really good. There was a a a
38:27 textile or what do they call it? A fiber
38:29 show. So, this was a quilt that that
38:32 someone had done that I think is
38:34 beautiful. And so, I said to Chat GPT,
38:39 um, reimagine this in whatever art style
38:42 you think is most
38:45 appropriate. And it did
38:48 this, which again, I think is
38:52 just
38:54 stunning. It's like a
38:57 gouache kind of
39:01 medieval
39:03 looking cubist crazy ass
39:10 painting. Then what did I do? Then I
39:13 took my my GQ West magazine cover and I
39:19 said put this cowboy into this
39:24 style. Right. So this is a thing that I
39:27 made in in chat GPT
39:31 image and then it made that which is
39:34 crazy good. Again, it's
39:37 like and you know what's funny is like
39:40 this is nothing that I would
39:43 ever
39:45 do. So, like what struck me
39:49 tonight
39:51 is as a brainstorming tool, just the
39:55 ability to kind of informally say, you
39:59 know, go grab me this, go grab me that,
40:01 think about this, like reimagine
40:04 this.
40:06 Um then I was somewhere and someone
40:08 asked me about can you could you take a
40:11 muscle
40:12 car and like you know make it a uh
40:17 convertible. And so I said yeah go take
40:20 that car and make it a make it green.
40:23 Make it a convertible and give it a
40:25 blower. And so so there's the
40:29 Challenger, you know, convertible with a
40:33 blower. It's chat GPT. It's still bad at
40:36 what as an artist. What does your wife
40:38 think of the new image generation? She
40:40 hasn't played with it a lot. It's only
40:41 been out for a week. I'm still trying to
40:43 get her to embrace
40:45 midjourney. Is chat GPT still bad at
40:48 realistic
40:51 images? Oh, you're asking is it paint?
40:54 Because it said it's the question said
40:56 it's
40:59 um No, it's it's not bad at realistic
41:02 images. In
41:03 fact, well, like that one, like this
41:06 one's pretty realistic. This
41:11 is like that's a decently realistic
41:16 blower. That looks decently
41:18 photorealistic to me. Like I It's a
41:21 really weird question, Pate, because
41:23 it's
41:24 like MidJourney can do really realistic
41:27 images, but they're all kind of stylized
41:30 in they sort of lean toward art.
41:33 Even though they can do it can do
41:34 hyperrealistic
41:36 images. Um I would say that the chat GPT
41:40 stuff's very very
41:42 basic. I tried to make it do a stained
41:45 glass window of the cowboy which I
41:47 thought was pretty
41:48 funny. Um here's a jet that it did. I
41:51 said make a make a jet inspired by the
41:54 Dodge
41:55 Challenger. Dolly always takes artistic
41:58 liberty in a way that is. But this isn't
42:00 Dolly anymore. a bit annoying to me. You
42:02 should try it now. No, it's this is a
42:04 completely different model. Pate, this
42:06 is inside inside chat GBT. Um, it's much
42:10 better at photorealistic stuff than
42:12 Dolly was. Dolly was
42:14 crap. I uploaded that
42:18 photograph and had it reimagine it as
42:22 art. And then I said, "Make it taller
42:23 and make it more vibrant." And it did
42:25 that. That was pretty cool. Oh, and then
42:27 I said, this is a thing that kind of
42:30 blows my mind. Take that image and put
42:32 it in a Simpsons living room. And so it
42:36 created this Simpsons style living room
42:40 with that painting in the
42:43 background. And then I said, "Can you
42:45 put the cowboy in it?" Oh, no. And it
42:47 didn't do
42:49 it. Oh, okay. Um, and then this one was
42:52 really cool. So, this was a this was a
42:54 painting that I saw that just it looks
42:57 like one of those overhead maps of a
42:59 lake and you know like with little I
43:02 don't know different views or whatever
43:04 it is. And so I took that and I said
43:06 turn that into a fantasy map um of a
43:10 dragonfested land and it came up with
43:17 this. That's crazy.
43:22 Like I didn't give it
43:24 anything. I just said it's make it a
43:27 fantasy map of a
43:29 dragon infested land. It's called the
43:32 Dragon Coast. There's the Embers Sea.
43:34 There's Inferno, Grovenwood, Cinder
43:36 Peaks, Ashen
43:40 Wastes, Dragons flying around.
43:49 So like
43:57 just there's a there's a phrase that
43:59 pops into my head. It's something
44:01 like informal excellence, right? or
44:05 informal elegance or
44:09 informal you know I don't
44:13 know like like you can very very easily
44:17 just take any input from the world and
44:20 just like because chat GPT can now
44:24 because it's inside a large language
44:26 model it can
44:30 understand the symbolism of the image
44:33 that you uploaded
44:36 So, in
44:37 fact, let's try something
44:40 here. Let me
44:45 grab Let me grab this
44:47 one. Copy that
44:52 image. I'm going to
44:55 say, "Write me a poem." No, no, no.
45:01 Write me
45:05 and
45:08 analysis.
45:11 Analysis
45:14 of
45:16 the
45:19 imagery,
45:21 style,
45:23 tone, colors.
45:33 and
45:36 symbolism in this
45:40 work. Just we'll just keep it
45:43 simple. So basically ask chat GPT to
45:46 understand well I'm not paying for chat
45:48 GPT. I used it recently like a month ago
45:51 and don't plan to return. Pay were you
45:53 here when I was going over all the
45:55 models? They're all so intuitive now.
46:04 Yeah, I don't think you need to pay for
46:07 chat GBT. Actually, quite frankly, pay
46:09 they've made most of this stuff free in
46:11 chat
46:12 GBT color blues. Okay.
46:18 So, so now I'm going to say
46:21 um I want you to
46:28 imagine five new images that
46:36 leverage the core ideas, not the core
46:39 ideas, the core
46:42 meanings of this piece.
46:47 but be
46:51 completely
46:52 [Music]
46:56 completely
47:00 different in every
47:04 way. All right, let's see what it comes
47:06 up with here.
47:12 The Subway Oracle. A surreal a
47:15 surrealistic a surreal photorealistic
47:17 scene of an empty modern subway station
47:20 at 3:00 a.m. where one of the train
47:22 tunnels glows with a nebula like light.
47:25 Solitary figure stands before it. That's
47:27 really cool. All right. The last library
47:32 before the stars. A tear in the barn.
47:35 Nice. Beneath the sleeper's skin. an an
47:39 anatomical dreamlike rendering of a
47:42 person sleeping in bed, except beneath
47:44 their translucent skin. Instead of
47:46 muscles and bones, there's a swirling
47:48 galaxy. That one's cool, too. Doorway in
47:51 the desert. That sounds cliche. All
47:52 right, so sleeper skin and subway
47:55 oracle. Let me see a
48:01 wide version. wide.
48:06 Um, version
48:09 of number
48:20 [Music]
48:24 one. Oh, yeah.
48:27 Tomorrow. So, today's Friday. So,
48:29 tomorrow, what time is it, Pate? 3 3
48:32 p.m. I can go look it up. Hang
48:39 on. Tomorrow at 2 p.m. Mountain time. 2
48:43 p.m. Mountain time. So if you go to the
48:45 AI salon, go to events, tomorrow at 2
48:50 p.m. is um demystifying AI. So this is
48:55 Pate
48:59 M's series. It's now a series where he's
49:03 helping people understand
49:06 how large language models work, how
49:09 diffusion models work, how basically the
49:11 math and and the the mechanisms for how
49:15 this seemingly magical stuff does what
49:18 it does. It is my contention that the
49:21 better you understand how these tools
49:23 work, the better prepared you're going
49:24 to be. Like even if you don't ever plan
49:27 on creating your own large language
49:29 models or dealing with the weights or
49:31 embeddings or any of that
49:34 [ __ ] understanding how it works puts
49:37 you in much better when someone asks you
49:40 a question about, well, isn't it just
49:42 the world's largest plagiarism machine?
49:44 Isn't it just theft?
49:46 and be like, "No, the actual works
49:48 actually cease to
49:51 exist, you know, um, and you can say
49:55 that with some confidence." It's really
49:56 good to understand what you're the tools
49:59 that you're using. And so, so those
50:01 these this series is really good. So, go
50:03 see that. Just concluded an epic week
50:06 and so so happy to see you all. Very
50:08 nice. Hey, Ann, what's happening? All
50:10 right, let's go back to our little thing
50:12 here.
50:15 Oh, that's kind of
50:19 cool. So,
50:21 thematically
50:26 that has its origins in that I can see
50:30 it. That's kind of cool. And then let's
50:33 do number four.
50:35 Um, show
50:37 me number four.
50:43 And and you know what's the purpose
50:45 here? The purpose is
50:56 just you know the there there's this
51:00 vibe coding thing and then we were
51:02 talking one night here about vibe
51:03 storytelling. I just feel
51:07 like there's something kind of beautiful
51:10 about about
51:13 surrendering creative
51:18 ideiation to these
51:21 tools, but but you get to curate the
51:24 inputs and you get to
51:26 direct how you want it to interpret
51:29 those inputs. But then what it generates
51:31 is is stuff that you likely wouldn't
51:35 have gotten to on your own. But but now
51:37 that you've gotten there, then you can
51:38 take that and do something with it. I I
51:41 find there there's there's an elegance
51:42 in that. Kyle, this is awesome. It's
51:46 writing my response to the US patent
51:48 office. Beautiful. Guys, my chat EPT is
51:50 acting as a legal adviser with trademark
51:52 application. They accepted it, but I
51:54 need need to clear a few things up.
51:57 There you go. I mean, think about patent
52:00 and trademark law. Think about all of
52:03 the patents in the world. They're all in
52:06 the training
52:07 data. Every patent that's ever been
52:10 written is in Chat
52:13 GPT, right? It's in the public domain.
52:15 It's been out there. It's part of
52:17 patenting it is you give up your the
52:20 details of how you do what you do.
52:25 Everything that's ever been written on
52:27 getting a patent, getting a trademark,
52:29 maintaining one, arguing, finding prior
52:32 art, managing prior art, all that stuff,
52:35 all that information is in there. So, it
52:37 makes sense that it would be really good
52:38 at that. Tik Tok question, what's the
52:39 best way to build AI agents or digital
52:42 agents? Um, it depends what you mean by
52:44 AI agents.
52:49 Um the the
52:51 the shitty answer is wait six
52:56 months or three months because there
52:59 there's a lot of platforms coming that
53:01 are just going to do them. If you really
53:04 want to build um one there's within um
53:08 Claude you've got model context protocol
53:12 which is now being adopted by open AI.
53:15 you've got um lang chain and and uh
53:19 embeddings databases and things like
53:20 that. So you could go build there
53:22 there's a whole agentic framework inside
53:25 of OpenAI's API. Um I think there's
53:27 similar um Aentic frameworks in the in
53:31 the Gemini APIs. Correct me if I'm
53:34 wrong, Pate, but I'm pretty sure there
53:35 are. So you could go find the API
53:38 documentation of all the Aentic
53:40 frameworks and go build your own. Um,
53:43 you could also just wait three months.
53:45 They're they're going to be coming. Like
53:48 there were, I don't know, five or six
53:50 announced in this last week. A lot of
53:52 them look absolutely remarkable. You
53:54 could also just go play with a tool like
53:56 Manis if you haven't signed up for
53:58 Manisim. Uh,
54:01 man.
54:03 Um, it's it's a pretty cool tool that
54:07 just goes off and does [ __ ] for you.
54:10 Um,
54:12 so but if if you're a developer, if you
54:15 want to get into the APIs and geek out
54:16 on that stuff, it's all in there. I
54:19 don't know if that answers your
54:22 question. All right. Oh, we got more
54:27 runes. That's kind of cool,
54:30 too, huh?
54:39 So again, so
54:41 that has its DNA origins. Like see the
54:45 tunnel, see the shapes of the trees make
54:47 a
54:50 tunnel. So that's got its DNA origins in
54:54 the description of the meaning of this
54:56 piece.
54:57 I just I just feel like you could do all
54:59 sorts of
55:01 circular self-referential things and do
55:04 all sorts of cool [ __ ] Um, let me show
55:07 you another another thing. Ian
55:11 Murphy, can
55:13 I
55:16 Eli5 MCP?
55:18 [Laughter]
55:22 Um, I'll do what I can with MCP. And
55:25 Pate, if if if I get this wrong, let me
55:27 know if anyone knows. Um, MCP stands for
55:30 for model context protocol. It was
55:32 created by um,
55:34 Enthropic. And what MCP allows, it's
55:38 someone described it once. Oh, this is
55:40 just an API for APIs. And if you don't
55:43 know what an API is, an API is what's
55:45 called an application programming
55:47 interface. So if I want to have my
55:50 software
55:51 application, let's say, let's say I want
55:54 to create a software application that
55:55 has a map in it, right? I could launch
55:59 my own
56:00 satellites, invent my own cars to go get
56:03 mapping data, write, you know, image
56:07 generation libraries that that take map
56:10 data and convert that into
56:12 maps. Or I could just use Google Maps,
56:15 right? And Google Maps has an API. And
56:18 so basically in my code, I go over to
56:21 Google and I get an API key. So it's
56:23 basically a specific software key code
56:27 that's tied to my
56:28 account. And then I just say for this
56:32 software application, insert Google Map
56:34 here. And then I put in my key. And then
56:37 if they charge me for it, um, every time
56:40 someone uses my map, I get charged a
56:42 tenth of a penny or whatever it is,
56:43 right?
56:45 So, it'll it allows software developers
56:47 to use other pieces of software inside
56:51 their piece of software, for lack of a
56:52 better term, or the functionality and
56:55 and bring that into their software
56:56 without having to recreate it.
57:09 Um, when I was your age, I called that
57:11 an frame. Well, this isn't an frame. No,
57:13 because you're actually pulling
57:15 functionality into an application. It's
57:16 different than that. It's it's in the
57:18 neighborhood. Okay. But but it like it
57:21 could be like I need let's say there's
57:24 some software service that does some
57:26 deep technical analysis and I'm going to
57:29 grab some inputs and then I'm going to
57:30 send them out to this other software
57:32 application to do the technical analysis
57:35 and bring back the answer. You do that
57:37 through an API.
57:39 So model context protocol is basically a
57:43 protocol on the front end that you've
57:45 got ah let's go to the
57:49 boxes.
57:55 Um oh here we go.
58:01 Okay. Okay. You got a box, right? You
58:04 got your You got your box.
58:09 Okay. So, you've
58:12 got you've got Claude here, which is a
58:16 large language model, right? Well, no,
58:18 we got we'll do this. You got Claude
58:20 here, which is a large language model.
58:22 Then up
58:24 here, you have
58:26 MCP. And so, this is an API that allows
58:29 Claude to talk to
58:32 um what's called an MCP server. And then
58:36 over here, you've got a a second
58:40 software application. Wait, hang on. I
58:43 got a I got a My producer didn't didn't
58:47 make my picture bigger. We're going to
58:49 have to have a word with him.
58:52 Okay.
58:54 Okay. He's defending himself on a
58:57 post-it note right
58:59 now. Okay. So, you got Claude down here,
59:03 right? There's your large language model
59:04 chat. It's like chat GPT. Here you've
59:07 got the MCP thing. And here you've got
59:10 say a software application like Unity.
59:13 So I don't know if you know Unity, but
59:15 Unity is a game development environment.
59:17 It's got a 3D modeling thing. It's got
59:20 texturing. It's
59:26 got a black bar. Sorry about that. There
59:29 you go.
59:31 um or or this could be Microsoft Word or
59:34 this could be um uh a website. Uh it
59:39 could it can be it can be anything. So
59:42 within
59:43 Unity, you've got I'm sure there are
59:46 thousands of separate little features
59:49 that you can use within Unity, right?
59:51 You can you can create a cube object.
59:55 You can scale a cube object. You can
59:57 move things around mathematically to a
1:00:00 certain position in 3D space. Right? So,
1:00:03 so what you do is you take oops, you
1:00:05 take the model context protocol protocol
1:00:08 like the the sort of recipe for it and
1:00:11 then you take all of your functions
1:00:13 within Unity and you map them. And then
1:00:17 what happens is you say to Claude, wait,
1:00:21 hang on. You say to Claude, I want to
1:00:24 create a little 3D village in
1:00:27 Unity. And it then goes, hey, am I
1:00:30 connected to an MCP server? Yes, I am.
1:00:33 And does that MCP server happen to know
1:00:36 anything about the application Unity?
1:00:38 Yes, it does. So what ends up happening
1:00:40 is you're on your computer looking at
1:00:43 Unity and you talk to chat GPT and it
1:00:48 makes you a little village in this
1:00:49 software
1:00:51 application. So so it is it is a
1:00:53 connector
1:00:55 between easy easy to talk to large
1:00:58 language models and really complicated
1:01:00 software. So, there are going to be
1:01:03 thousands of these connections and like
1:01:08 once you have the Unity one set up, then
1:01:11 you can just connect it. Does that make
1:01:13 sense? Did that answer it? Did the box
1:01:16 metaphor work
1:01:18 there? I've been using Unity all week.
1:01:22 The visuals
1:01:24 helped. Thanks. I started to figure it
1:01:27 out.
1:01:32 Okay, that is the clearest description.
1:01:34 That is a big deal. Good. That's what I
1:01:37 like to do here. Talk in box metaphors.
1:01:41 See, a large language model is nothing
1:01:44 more than a box where they jam all of
1:01:47 the knowledge of of the output of
1:01:49 humanity into it. And then they close it
1:01:52 up like this. And we don't really know
1:01:54 what's in there, but it's math. P. Pate
1:01:57 knows what's in there, but Pate's the
1:01:59 only
1:02:00 one other than the other people that
1:02:03 know. But most people don't know. They
1:02:05 just know that all the [ __ ] that was
1:02:08 ever
1:02:09 created is in this thing.
1:02:12 And so it's why when you when you does
1:02:16 it not blow anyone else's mind that when
1:02:19 you talk to advanced voice and chat GPT
1:02:22 you can literally ask it anything and it
1:02:25 knows about
1:02:26 it. Like it's it's [ __ ]
1:02:30 insane. And it's not slow. It's just
1:02:33 like it's just instantly there because
1:02:35 of the nature of how embeddings work.
1:02:36 It's so they've taken knowledge and
1:02:39 compressed it and made it so efficient.
1:02:41 It's just literally they're just waiting
1:02:43 to be tapped into like our [ __ ]
1:02:45 brains. Do I have
1:02:51 Yeah. How is this all ones and
1:02:54 zeros? Yeah. Well, I'll tell you what.
1:02:57 Um, Bottlenome, if you weren't here
1:02:59 earlier, go to Pate Moders tomorrow at 2
1:03:02 PM Mountain time. He's doing a
1:03:04 demystifying AI course. You can actually
1:03:07 ask him how this stuff actually
1:03:09 functions. and he he provides these
1:03:11 Google notebooks that let you play with
1:03:13 code by just literally hitting a play
1:03:15 button and you can watch how these
1:03:18 models work. It's it's [ __ ]
1:03:22 amazing.
1:03:23 Um, what am I
1:03:31 doing? What's happening
1:03:34 here?
1:03:35 What? Hang on, people.
1:03:47 Is my iPad broken? What is going
1:04:03 on? I'm quitting apps. Hang on.
1:04:06 I can demo it over
1:04:09 here. The shirt is confusing to me. It's
1:04:12 like plaid but also not. It it it is it
1:04:18 correct. It's really
1:04:23 um Can you see it? It is definitely
1:04:27 plaid, but then there's this weird
1:04:28 pattern beneath
1:04:30 it. I can't tell.
1:04:36 Yeah. You know what I think it is? I
1:04:37 think it's a printed like a printed
1:04:40 paisley and then they put the stitching
1:04:43 on top of
1:04:45 it. It's a It's a Robert Graham. And
1:04:48 look at the cool buttons. Well, they're
1:04:50 cool to look at, but you should try
1:04:53 buttoning those [ __ ] in the
1:04:55 morning. I almost jumped through a
1:04:57 window a couple of times. It's print on
1:04:59 plaid. I think it's plaid on print, but
1:05:01 yes, it's cool. And then it's
1:05:03 embroidered on top of that. Like look at
1:05:06 the
1:05:08 embroidery. And I did all this myself. I
1:05:10 did this with chat
1:05:12 GPT. I took Chat GPT and I and I built a
1:05:16 model context
1:05:17 protocol to a Singer digital sewing
1:05:21 machine, an embroidery machine. And I
1:05:24 said, "Could you please embroider my
1:05:25 shirt on top of the plaid and
1:05:29 print and with the threads in the
1:05:32 machine?" And I put them No, I didn't.
1:05:35 I'm lying. It was a
1:05:38 lie. I just got this. My wife put it in
1:05:40 my drawer or
1:05:43 closet. Your million dollar
1:05:46 business. Make it. I said make it so.
1:05:50 Are you serious? No, I'm not serious.
1:05:56 You You must be new here. There's very
1:05:59 little that I do that is
1:06:02 serious. That's chat
1:06:07 GPT21. Oh
1:06:10 man. I I It's so true. We We're actually
1:06:13 pretty close to being able to do that.
1:06:15 Please share that. I wanted to make a
1:06:16 pattern and dress uh from a pic Mary
1:06:20 made. Oh, that's funny. That's cool.
1:06:23 Um, yeah, it's a very cool shirt. You
1:06:26 could I This would be a tough shirt to
1:06:28 do mushrooms
1:06:30 in because you you would absolutely get
1:06:33 lost in this thing. It would be bad. Um,
1:06:37 all right. I want to
1:06:39 show Let's go back to I want to show
1:06:43 some midjourney. Midjourney.
1:06:47 Midjourney. Oh, we could do that,
1:06:50 couldn't we?
1:06:55 Um, all right. Let's try something
1:07:08 here. So, I have an idea.
1:07:12 Welcome to chat
1:07:14 [Laughter]
1:07:22 add new movie
1:07:24 recording.
1:07:26 Okay. So, wait. Let's
1:07:32 [Music]
1:07:38 do Okay, I got to do a
1:07:43 screenshot. Let's see. Uh, wait. Is this
1:07:47 going to be clipboard? Yes. Okay. All
1:07:51 right.
1:07:52 So, let's let's get Wait. We got to get
1:07:55 positioned.
1:07:58 All
1:07:58 [Music]
1:08:00 right. Okay. Wait.
1:08:04 Ready.
1:08:12 Okay. This is turning into a get ready
1:08:15 with me. Okay. Here's how I get ready
1:08:17 for my my little
1:08:20 podcast. Okay. We're going to do a new
1:08:22 thing. Fit check. Okay. Oh, this is
1:08:24 going to be good. I got it. Okay.
1:08:27 We're We're gonna go find Cowboy
1:08:33 Kyle. Okay. So, let's see. Uh, let's
1:08:37 see. Kyle or let me look up dreams. I
1:08:41 think it's a dreams a dreams image
1:08:45 here. Is that
1:08:47 it? Yeah. Cowboy Kyle. All right.
1:08:52 And then we're going to paste in the
1:08:56 picture we just
1:08:57 took. And then I'm going to say, okay,
1:09:01 we
1:09:04 need
1:09:06 to dress up our
1:09:14 cowboy for an evening in the
1:09:19 city. wearing
1:09:24 um
1:09:25 the uploaded the wearing
1:09:30 um a nice
1:09:32 shirt with the
1:09:36 pattern of the one
1:09:41 uploaded.
1:09:44 Um, he should
1:09:47 be clean
1:09:52 shaven, lose the
1:09:56 hat, and be
1:10:00 standing next
1:10:02 to a Central
1:10:07 Park horse/carriage.
1:10:14 giving the
1:10:17 horse a
1:10:20 carrot. It should be
1:10:24 photographic like he's a
1:10:29 model
1:10:32 that has
1:10:35 been
1:10:38 captured
1:10:40 looking swell.
1:10:43 Okay.
1:10:45 Woo. Check your Tik Tok angle. There you
1:10:47 go. Sorry about that.
1:10:52 Oh, Korea has a try it on feature. Yep.
1:10:54 Ann says, "Tell the nice people about
1:10:56 Kim offered next Wednesday." Yeah. So,
1:10:59 next Wednesday. Are we recording it next
1:11:02 Wednesday, Ann? I think we are. Right.
1:11:03 We're doing it live. We'll do it live.
1:11:06 We'll do it live. So, we've got Kim
1:11:09 Offford
1:11:13 and Ann and my uh podcast is 400 PM
1:11:18 Mountain time on
1:11:20 Wednesdays and it's called the AI
1:11:22 readiness project and it's swell. We
1:11:25 have a good time and we talk about how
1:11:26 unprepared we feel for all this
1:11:30 [ __ ] Yeah, we're going to do it live.
1:11:32 We'll do it
1:11:33 live. All right, let's see. Let's see
1:11:35 how our cowboy cleans up here,
1:11:38 [Laughter]
1:11:42 [Music]
1:11:44 people. Uh, how you find the AI
1:11:47 readiness project? It's on my Tik Tok
1:11:50 channel or my my LinkedIn channel, my
1:11:55 um X channel, the AI Salon YouTube
1:11:59 channel, and then uh Ann's She Leads AI
1:12:04 [Music]
1:12:05 um YouTube and
1:12:09 LinkedIn. Dang, that cowboy dresses. He
1:12:12 He cleans up
1:12:15 nice. And it looks like it actually
1:12:18 captured the shirt pretty
1:12:19 close. Let's see what we got
1:12:25 here. I mean, that's pretty good. Pay,
1:12:27 you were asking before about
1:12:28 photorealistic. That's pretty That's
1:12:30 pretty
1:12:32 tight. That shirt's not bad. It got it
1:12:36 pretty good. Damn.
1:12:46 Dad
1:12:49 girl, I wish I could quit you. Look at
1:12:52 his giant head. He's He's a little bit
1:12:55 He's a little bit tiny
1:12:59 bodied. Yeah, that's better than it used
1:13:01 to be. Yeah, that's pretty good. That's
1:13:03 better than I thought it would
1:13:05 be. That's pretty
1:13:08 remarkable,
1:13:10 huh? All
1:13:12 right, I need to animate a logo.
1:13:17 Um, most of the most of the uh video
1:13:20 tools will will animate a
1:13:29 logo. Wow, it really captured the shirt
1:13:32 quite
1:13:34 well,
1:13:36 huh? That's a little bonkers.
1:13:44 That That's how chiseled I am in my own
1:13:46 [Laughter]
1:13:54 mind. All
1:13:56 right.
1:13:58 Um All right. Let me show you one other
1:14:00 thing here. So
1:14:04 um in midjourney
1:14:08 [Laughter]
1:14:09 My chat GPT still says, "I wish I could
1:14:12 quit you and calls me
1:14:16 daddy." Sora is better at prompt
1:14:18 adherence and creates better images than
1:14:20 using chat GPT. Okay, that's good to
1:14:22 know. And and I would think that those
1:14:24 would be two very different kind of
1:14:26 approaches. If you really just want to
1:14:27 focus on the image, do it in Sora. But
1:14:31 if you want to do it in the context of a
1:14:33 larger conversation, then it makes sense
1:14:35 to do it over in chat GPT.
1:14:37 Um, okay. So, in Midjourney, we now have
1:14:42 access
1:14:43 to down here,
1:14:47 um, version
1:14:49 seven. So, there's
1:14:51 6.1 and there's
1:14:54 seven. And then we also have this new
1:14:57 thing called draft. So, does does that
1:14:59 exist for 6.1? It doesn't. So just just
1:15:03 in version
1:15:07 7 you can flip between standard and
1:15:09 draft mode. Draft is really quick.
1:15:18 Um, the other thing that they
1:15:23 did is they changed some of
1:15:32 the Wait, but it's not draft mode.
1:15:57 Welcome to draft mode. Draft
1:16:01 mode, hang
1:16:04 on, helps you move faster. Images
1:16:07 generate quickly and rough ideas become
1:16:09 visuals without slowing you down. No
1:16:11 need to write the perfect prompt. Say
1:16:14 what you're after. Draft mode reshapes
1:16:16 and refineses your input so you can
1:16:18 focus on the idea, not the wording. Work
1:16:21 how you think. Speak or type your
1:16:23 thoughts as they come. Draft mode keeps
1:16:26 up so you can stay in the creative flow.
1:16:28 Enhance drafts to the highest quality.
1:16:32 Begin. Okay, so we're now in draft mode.
1:16:35 So you see in the upper right here it
1:16:38 says draft
1:16:41 mode and then this says personalization
1:16:44 on or off and um version 7 has its own
1:16:49 personalization setting. Now, important.
1:16:52 You can't use, if you're like, I can't
1:16:54 use system 7. That's because you have to
1:16:57 go to personalize
1:17:00 here. You have to turn personalization
1:17:02 on and then see this ranked profile
1:17:05 where you've got version seven. You've
1:17:08 got to
1:17:09 create the the way you um create a
1:17:12 personalized profile is it'll give you
1:17:15 two images and you choose one and then
1:17:18 you do 200 of those and then it'll give
1:17:20 you access to
1:17:21 V7.
1:17:24 Okay. So, let's So, we're in draft
1:17:27 mode and then it said we could talk. Oh,
1:17:30 yeah. Here's the microphone.
1:17:32 Okay. So, the microphone is right here.
1:17:38 right there. Start voice
1:17:43 mode. Let's
1:17:56 see. I want there to
1:18:03 be Hello.
1:18:07 Hello. Hello.
1:18:10 Hello.
1:18:13 Hello. Why are you not making
1:18:17 words? Hello.
1:18:20 Hello. Make the cars
1:18:24 more sexy.
1:18:29 Go. Oh, it heard that. Okay.
1:18:33 Um, this factory looks too dusty. Make
1:18:36 it
1:18:46 cleaner. That's way too sterile. And
1:18:50 we're missing the car. So, this needs to
1:18:52 be a 70s muscle car restood in an
1:18:56 abandoned factory, not a modern well-lit
1:18:59 factory. I just don't want to be quite
1:19:01 so dirty.
1:19:11 Those are not restoods. A restood is an
1:19:16 old style car that's been fixed up with
1:19:19 new components, new wheels, a modern
1:19:21 look, modern engines, modern suspension.
1:19:24 So, it should be super sexy and brand
1:19:28 new looking, not like the piece of crap
1:19:31 factory that it's sitting
1:19:41 in. Yeah. Now, we're talking make the um
1:19:45 factory more cavernous and have sunlight
1:19:49 streaming in through the broken factory
1:19:51 windows at magic hour.
1:20:05 The cars are dusty again. Clean up the
1:20:07 cars.
1:20:20 Now make them punchy7s colors and give
1:20:23 the cars modern mag wheels.
1:20:39 Make all of the cars face three quarters
1:20:42 toward the camera and have their
1:20:44 headlights
1:20:47 [Music]
1:21:02 on. Make them all
1:21:04 1971 Dodge Chargers.
1:21:21 Those look like crap. They need to look
1:21:24 like resto mods. Bright and shiny and
1:21:29 new. Oh, that's a good idea, Anna.
1:21:43 Great. Now make them look like they're
1:21:46 filmed through a fisheye
1:21:58 lens. All right, let me turn that off.
1:22:01 Okay, I ended voice mode. I mean, that's
1:22:03 pretty slick. Like if you look at the
1:22:05 like if you look at the prompts that it
1:22:07 wrote, it wrote pretty good
1:22:09 prompts and it it maintained like I
1:22:12 wasn't saying every time I didn't have
1:22:14 to click on it. This draft mode's pretty
1:22:19 slick. I mean, it's frustrating as [ __ ]
1:22:22 because it
1:22:24 it Okay, this is actually really good.
1:22:27 So, here's where MidJourney is losing
1:22:31 some footing.
1:22:37 It's like every
1:22:39 time it's generating a new image from
1:22:42 the
1:22:44 prompt, it seems to have very little
1:22:46 recollection about the
1:22:49 composition of the previous images. The
1:22:52 image gen within chat GPT
1:22:56 does. So in chat GPT you can you can
1:23:00 keep refining the same image and here it
1:23:03 keeps creating them from new. But
1:23:05 there's something nice about this about
1:23:07 how fast it is and you could just be
1:23:09 sitting back talking to it and just
1:23:11 generating a bunch of stuff. So anyway,
1:23:14 that's pretty crazy. Maybe adding don't
1:23:17 change. I can I can't remember if that
1:23:19 works or not. Yeah, I don't know. I mean
1:23:21 I haven't I haven't played with this a
1:23:22 lot. Like I just discovered this
1:23:25 feature. Um, what I can tell you is the
1:23:29 version 7
1:23:31 model got like 70s muscle car resto mod.
1:23:35 It doesn't seem to understand that at
1:23:37 all. Like like these are not really
1:23:40 resto mods. Like these are like barn
1:23:43 findiness where the wheels are the
1:23:45 original wheels that you know there's no
1:23:47 sort of flared fenders or any of the
1:23:51 things that you would have in a resto
1:23:52 mod. Like these are not restoods. And
1:23:55 some of the things Midjourney
1:23:58 created like are not muscle cars. Hang
1:24:01 on. Like these
1:24:03 things. Like this
1:24:06 thing. Like that looks like a Russian
1:24:11 AMC Hornet from
1:24:16 1973. It's like that's not even close to
1:24:19 a muscle car. What the [ __ ] is that
1:24:21 thing?
1:24:24 And like that like that is supposed to
1:24:26 be a a 70s muscle car.
1:24:31 No. So I don't know. I don't I don't
1:24:34 quite get this model yet. I've seen a
1:24:37 couple of posts on
1:24:39 um I've seen a couple of posts
1:24:43 on Twitter that
1:24:47 um that this V7 model from Midjourney is
1:24:50 a big letdown. and the fact that chat
1:24:53 GPT's image gen is so good that people
1:24:56 are are uh predicting the demise of
1:25:00 midjourney. I don't think we're quite
1:25:01 there, but but this
1:25:03 is there's something about the coherence
1:25:08 of the chat GPT image gen and the fact
1:25:11 that it can do text and it can do it in
1:25:14 context of a chat. That is a really big
1:25:17 deal. And it's it it's not as
1:25:20 complicated as dealing with midjourney
1:25:23 because midjourneys midjourney at this
1:25:25 point is starting to get like Photoshop
1:25:27 in terms of all of the things you can do
1:25:29 and all of the things you need to know.
1:25:31 Quite frankly, I mean I think that kind
1:25:34 of looked like a 69 Chevel. It didn't.
1:25:37 It didn't. It didn't. But I don't know.
1:25:45 Not a Russian AMC
1:25:49 wet. All right.
1:25:52 Um, okay. So, here's your weekend
1:25:57 homework. Do one of two things. We'll
1:25:59 we'll do them both image related.
1:26:02 go out in the world with your chat GPT
1:26:05 app and take pictures of stuff and then
1:26:10 turn them into magazine ads or turn them
1:26:12 into paintings or take paintings and
1:26:15 turn them into photographs or I just
1:26:18 just go play with your
1:26:20 camera and chat GPT image generation. So
1:26:24 that that would be one thing and then
1:26:25 the other thing would be play with
1:26:27 midjourney and put it in voice mode or
1:26:29 draft mode and just see if you can start
1:26:33 navigating using draft
1:26:35 mode. Like in both cases see how quickly
1:26:39 you can navigate out of your creative
1:26:41 comfort zone and create something that
1:26:43 really excites or inspires you that you
1:26:45 never would have gotten to. All right.
1:26:48 Um if you're on YouTube, please like and
1:26:50 subscribe over there. Um, we need uh we
1:26:54 we need as many subscribers as we can.
1:26:57 We're we're just a a small baby YouTube
1:26:59 channel. Um, we got a 75day reprieve on
1:27:02 Tik Tok, which is swell. Um, maybe I'll
1:27:05 start making content over there again.
1:27:06 I've been I've been a little lax on Tik
1:27:09 Tok. It's hard to be motivated to make
1:27:11 content for Tik Tok when you know
1:27:12 they're going to [ __ ] flush it down
1:27:13 the toilet.
1:27:20 Uh, and then tomorrow 2 PM go see Pete
1:27:22 Pate M's Demystifying AI. Have a
1:27:24 fantastic weekend everybody. And uh I'm
1:27:28 trying to think what else. What else? I
1:27:31 don't know. Just pay attention to this
1:27:32 AI [ __ ] And and like we talked about
1:27:36 last night, I forget who it was that
1:27:37 said their wife is tired of them talking
1:27:40 about AI. [ __ ] [ __ ] them. Talk about AI.
1:27:44 Talk about AI this weekend. I I I really
1:27:48 do think it is important for all of us
1:27:51 who
1:27:53 are aware of what's here and aware of
1:27:57 what's coming
1:27:59 um to sort of, you know, pull people
1:28:02 under the bridge as much as we can. Uh
1:28:04 because a lot of people are going to be
1:28:06 blindsided by this [ __ ] and that makes
1:28:07 me sad. So go uh go get someone in
1:28:11 there. Oh, it was you, Gareth. That's
1:28:13 right. Yeah, exactly. Terrific. Mr. It
1:28:16 is a subscriber. Love that. I love that
1:28:19 homework. Okay, good. Yeah, go inspire
1:28:21 yourself. Go go go go use these tools to
1:28:24 to discover something beyond what you
1:28:26 would normally
1:28:27 do.
1:28:29 Um like the the the one thing that um
1:28:33 that happened to me this past
1:28:35 week demoing a research tool that I was
1:28:39 going to try to visualize the research
1:28:41 was. I ended up coming up with an idea,
1:28:44 this thing called an emotional
1:28:46 architecture that I think is a killer
1:28:48 idea that I want to pursue in some way.
1:28:51 Um, and I never would have gotten there
1:28:53 had I not been sort of [ __ ] around
1:28:55 with this stuff around the edges. So, go
1:28:58 play around the edges of what you're
1:28:59 comfortable with and just let these
1:29:01 tools surprise you. Um, chat GPT in
1:29:05 particular, I think, is is is a very
1:29:08 surprising a surprisingly good model. I
1:29:11 think it's I think I think it's a bigger
1:29:14 deal than people know. I I think it's a
1:29:16 bigger deal than I know because I don't
1:29:18 really know what it makes possible yet,
1:29:19 but it's it's big. All right. So, where
1:29:22 are people showing their homework?
1:29:24 Already surprised daily. I know SXS. So,
1:29:28 go to the So, where you're going to
1:29:29 share your work is go to the salon.ai
1:29:33 AI and then click on there's a button
1:29:36 there that says join our community and
1:29:38 that's going to take you into
1:29:42 um the AI salon and it'll land you on
1:29:44 the the welcome page here. You can watch
1:29:46 a video from me and my co-founder Leah.
1:29:48 Um we'll we'll talk about the the idea
1:29:51 of AI readiness, play first, mindfully
1:29:54 create, generously lead. Um and the five
1:29:57 stages of AI adoption. And then over
1:29:59 here on the left hand side, there's all
1:30:00 sorts of different
1:30:02 areas. And one of the areas is called
1:30:05 the mastery workshop. And there's a
1:30:07 there's a section called look what I
1:30:09 made. Where is it? Right there. Look
1:30:12 what I made. So you go to look what I
1:30:14 made. And it's a showand tell
1:30:16 channel. So that's one place that you
1:30:18 can show and tell. That's kind of
1:30:23 fun. And then there's another area. If
1:30:26 you go down to clubs and hubs, there's
1:30:27 the AI learning lab and irregulars
1:30:30 channel and you can also share in there.
1:30:33 So that's this is our little private
1:30:35 club within the AI salon is the AI
1:30:39 learning lab channel where we can go in
1:30:41 and play and so you can share in either
1:30:43 of those places. And then the other
1:30:45 thing you should do if you join the AI
1:30:47 salon, step number two, introduce
1:30:50 yourself. So go in and actually
1:30:52 introduce yourself. Okay. It's a really
1:30:56 cool group. So that's where we're
1:30:57 sharing. So yeah, so go play, go inspire
1:31:00 yourself and then come in here and share
1:31:02 it. What I one of the things look at our
1:31:05 values. If you look at step number three
1:31:07 over here, it's learn our values. Um the
1:31:09 values are curiosity,
1:31:12 playfulness, empathy, bravery, and
1:31:17 generosity. So when you share your work,
1:31:19 don't just share the work, share how you
1:31:22 got there. or what did you do something
1:31:23 interesting to get to it? Share the
1:31:25 prompt if you're comfortable doing that.
1:31:27 If it's something that feels
1:31:28 proprietary, fine. Let it be
1:31:29 proprietary. Um but but if you can share
1:31:33 how you did what you did, share what you
1:31:35 learned, share what share what worked,
1:31:37 share what didn't. Um all of that stuff
1:31:39 will help other people uh on their
1:31:41 journey. All
1:31:43 right, G. All right, have a fantastic
1:31:46 weekend everybody. Um, I hope you're
1:31:50 you're uh you're doing well and having
1:31:52 fun with this stuff. Keep having fun
1:31:54 with it. There's there's a lot of stress
1:31:56 in the world right now. I think there's
1:31:58 a lot of
1:32:01 um self-imposed pressure to be really
1:32:05 excellent at
1:32:07 AI. Don't worry about that. Just play.
1:32:11 Just play. Just play.
1:32:14 and
1:32:14 then your competence will increase
1:32:17 through the play and you'll just get
1:32:19 better. You'll just get more comfortable
1:32:20 with using it. All right, so with that,
1:32:24 have a fantastic
1:32:31 weekend,
1:32:33 beautiful. All
1:32:35 right, you don't want to learn about AI
1:32:38 off that bridge. All right, we'll see
1:32:40 you guys later. Have a good weekend.
1:32:42 Bye.