7/21/2025 - Demystifying JSON for AI Prompting

Description

Chapters

Transcript

0:00 Dum dum dum dum dum. All right, now
0:02 we're live on YouTube.
0:05 Howdy, Dr. J. What's shaking? What's
0:07 going down?
0:09 [Music]
0:15 How about if we hit chords where the
0:16 fingers are on the actual strings?
0:19 [Music]
0:33 [Music]
0:39 Is this a place I can rest my forehead
0:44 [Music]
0:47 to gather my thoughts in sweet?
0:51 [Music]
0:55 Is this place where the feelings aren't
0:58 dead?
0:59 [Music]
1:02 From an over exposure to violence.
1:06 This is place I can slowly face. The
1:10 only one I truly can know.
1:15 These are tears from a long time ago.
1:19 Got these tears from a long time ago.
1:24 I need to cry. 30 years old soul.
1:28 These are tears from a long time.
1:34 Go.
1:37 [Music]
1:40 Champy. Are you singing anymore? What's
1:42 up with you? What are you distracted?
1:47 Chimpy,
1:49 you singing?
1:51 [Music]
2:05 Good evening everybody. Welcome,
2:07 welcome, welcome, welcome to Meltdown
2:10 Monday.
2:12 Uh, according to Tik Tok, we have two
2:13 people in here. I think we have more
2:15 than two people in here.
2:17 [Music]
2:19 But if you haven't shared the live yet,
2:21 what are you doing with your life? Let's
2:22 get going. Let's get going. Share the
2:24 live. I listen. This is going to be
2:26 exciting.
2:29 We're talking about Jason tonight.
2:32 Not Jason. JSON.
2:36 JavaScript object notation. It's gonna
2:39 be so fun. It's gonna be awesome. It's
2:41 gonna be so good.
2:44 I think maybe what we'll do
2:47 to start out is we'll go over to Suno
2:49 and we'll write a song about JSON
2:52 [Music]
3:00 using JSON.
3:02 Oh my god, you're going to love it so
3:04 much.
3:06 You're not even going to believe it.
3:07 It's going to be so awesome. It's going
3:10 to be so awesome.
3:14 And if you're sitting there thinking
3:16 like, I don't know what you're talking
3:17 about.
3:20 Well, that's because you don't have a
3:22 mathematics degree from Stanford or MIT.
3:26 God, that must suck.
3:29 It does because I have a I have an
3:31 acting degree from Penn State.
3:34 So,
3:36 I know what it's like to feel inadequate
3:38 mathematically.
3:40 I do.
3:42 I do.
3:45 Do I feel for people that are not
3:48 polymaths?
3:51 Not particularly. I don't really feel
3:54 for polymaths either. I don't even know
3:56 what the term polymath means. I think it
3:59 means of many maths.
4:03 Is that like they do some algebra and
4:06 some geometry?
4:07 Little trig. little trig thrown in
4:09 there.
4:14 I don't know. Isn't all of generative AI
4:17 just basic algebra?
4:19 Is it a polymath that did that or is it
4:21 just an algebra guy? Like is it if it's
4:24 just a differential calculus
4:27 specialist that was like, "Ah, we're
4:29 going to make the internet the AI
4:30 thing."
4:32 I don't think that needs to be a
4:34 polymath. I think that could be a mono
4:35 math.
4:37 But here's the thing. Maybe they use the
4:40 one math and then all of a sudden they
4:42 realize we probably need some other
4:44 maths.
4:47 You should talk to Jimmy. What's going
4:49 on with Jimmy? He's a polymath.
4:52 A what? A poly many maths?
4:56 Those people exist? Yeah, they gave him
4:59 a name.
5:02 Anyway, you don't need to know math.
5:11 Not not for this channel. Not for this
5:13 channel. We We have a dog that sings.
5:17 It's sort of like polymath with K9.
5:21 There's a number in it and a and a you
5:24 you could argue that the K is a variable
5:28 and the nine is a numeric integer. I
5:31 think
5:33 I think I learned that in seventh grade
5:34 polymath. single math algebra
5:38 I don't whichever one you learn in ninth
5:40 grade
5:43 when when you're destined to go to
5:44 school for acting whatever class that is
5:53 polymath equals multi- potentialite
5:55 exactly
5:57 I went to high school with polymath
6:00 she was great I you know Paulie um I
6:03 knew Paulie Well, in grade school and uh
6:08 boy, she was a card. She was a card. You
6:10 know what? Speaking of cards, she would
6:12 have those flash cards with like the
6:13 fractions on it and I was like, "Oh man,
6:17 she's going somewhere." Turns out it was
6:19 MIT. Yeah. Yeah. Yeah. She's got a
6:22 career, Paulie.
6:26 Sharon, and and I'm an accountant.
6:31 You use QuickBooks math. That's a That's
6:33 a whole different branch of mathematics.
6:36 All right, we'll switch this over to
6:37 here.
6:39 Um, all right, we'll do that. Let's see.
6:43 I just heard a Coldplay concert on
6:45 screen.
6:47 Yeah, she's she's famous now. Who knew
6:50 who knew a mathematician could get that
6:52 famous that quickly?
6:54 One little Heimlick maneuver, huh?
7:01 Ah,
7:05 okay. I've got my coffee. I got my my
7:10 spritzer, my Steltza water.
7:13 I don't quite have enough ice in it, but
7:15 you know, it'll do. It's It's cool. It's
7:18 cool and chilly.
7:21 [Music]
7:28 Yeah, that's what a calculator was for.
7:29 That's what Excel was for. how that's
7:32 what AI is for.
7:40 Um the the the JSON that we're going to
7:43 talk about tonight, why we're going to
7:45 talk about it is my Twitter feed is
7:49 being taken over with AI influencers
7:53 going, "Jesson is the secret to great
7:56 VO3 prompting.
7:59 you should use JSON for all your
8:01 prompting needs. JSON is everything. And
8:04 then they post some JSON there like they
8:06 [ __ ] wrote it. They didn't.
8:09 They didn't.
8:11 I'll show you how to do it tonight.
8:13 We're Here's what we're going to do.
8:15 We're going to teach you a little quasi
8:17 coding,
8:18 but you don't have to learn coding. We
8:20 don't That era is over.
8:26 You don't need to code anymore. You can
8:28 you can
8:30 if you're one of those polymaths and
8:32 you're like coding is relatively
8:34 important to retain as a practice so
8:37 that we understand the foundations that
8:39 create computer applica fine
8:43 fine if you want to do that you can do
8:45 that I'm just saying you don't have to
8:51 but when someone pops into your feed and
8:53 they're all like you should use JSON
8:55 JSON do you know what JSON O is Well,
8:58 look, it's all these fancy codes. Look
9:00 at all the squiggly things in there. I
9:02 did that.
9:04 That's That's what's going on on Twitter
9:06 right now. It's really [ __ ] annoying.
9:08 So, we're going to demystify that
9:09 tonight.
9:12 [Music]
9:18 We're going to debunk those jerks.
9:24 My first experience with Gen Spark was
9:27 dismal. Asked for 16 slides and it
9:29 failed to find 14 and got the other two
9:31 wrong.
9:35 The images that is Oh, that's
9:37 interesting.
9:41 I tried I used GenSpark today and to
9:45 decent success just doing some research
9:48 on uh effectiveness of video and pharma
9:51 marketing and it it did pretty good but
9:54 it it hallucinated you know a decent
9:57 amount of the cit I asked it to
10:00 hyperlink its citations and there was
10:03 some hallucination in there
10:07 but um but I also I went to Manis and
10:10 tried to have Manis do it and it got
10:12 halfway through the task and said,
10:13 "You're out of credits because you're a
10:15 loser and don't pay us. So, we're not
10:17 going to finish your stupid job." I
10:19 thought that was a little aggressive for
10:21 like a a modal a popup window. We're not
10:25 going to finish your stupid job. Wow,
10:27 man. Really?
10:29 [Music]
10:37 If you're new here, by the way, um what
10:40 was I gonna say? Uh this is just how I
10:43 act. I'm just a dumb dumb. I'm like the
10:45 village idiot and uh I'm here for you.
10:49 Do we have We don't Producer Brandon's
10:52 not here. I have a feeling producer
10:54 Brandon might be napping. Sometimes
10:56 producer Brandon goes napping and I have
11:00 a feeling that could be the case
11:01 tonight. Dr. Jay's in the house. Kyle,
11:04 did you get access to chat GPT agents
11:06 yet? Not as of an hour ago,
11:10 but let me try right now. They said it
11:12 was supposed to be Monday, but I I
11:14 didn't have much confidence in that.
11:18 10 print hello world. 20. End. No, you
11:21 forgot the loop. 10. Print hello world.
11:24 20. Go to 10.
11:33 Go to 10. 10 times. 30. End.
11:39 Sea slug of Doom is a master programmer.
11:41 You can tell that he's doing RadioShack
11:45 basic programming. That's that is the
11:48 last real programming I've ever done. I
11:50 mean, HTML scripting, I guess, frankly.
11:54 Um, basic programming is scripting, too,
11:56 isn't it?
12:00 I checked and I don't have access yet.
12:02 Anxiously awaiting it. John, it annoys
12:06 me how obsessed they are with making
12:08 them use our computers.
12:11 Why not just give them read access?
12:12 Yeah, I know. Well, to APIs. They're
12:16 they're going to I mean uh uh CETBT
12:20 agents is is going to or does have
12:23 access to APIs through connectors
12:27 there. there. Listen, right now
12:32 because this agentic behavior, things
12:35 like Manis and and Gen Spark and and
12:38 OpenAI agents,
12:41 because we're not used to what they do,
12:44 um
12:45 they they are just by definition going
12:48 to be way more restricted than they will
12:50 be in the future. We'll we'll get used
12:52 to this. This is this is very much akin
12:54 to people not putting um
12:59 not putting their credit cards on the
13:01 internet in the '9s
13:03 [Music]
13:17 through blue telescope.
13:21 Looking at the world tonight through
13:25 blue telescope.
13:27 I wish I may, wish I might not see what
13:31 I see.
13:34 Sheet metal on sheets of ice.
13:42 Looking through this blue telescope
13:46 down a moon struck a road tonight.
13:54 [Music]
13:57 The heavens,
13:59 the heavens,
14:01 fast and wide.
14:03 [Music]
14:12 D.
14:14 [Music]
14:19 I don't know those chords.
14:22 Um, okay.
14:24 John, what do you mean? What did John
14:26 say?
14:28 John, I said it. I said it. What did he
14:31 say? Honestly, I'm this close to
14:33 canceling my OpenAI subscription. Yeah,
14:37 I know. Hey, you know what's funny? My
14:38 my co-founder canceled her OpenAI script
14:41 des uh subscription yesterday. I said,
14:44 "You probably shouldn't. They're
14:45 probably just upgrading servers." She
14:46 was like, "Ah, screw it. I'm fine. I'm
14:49 going to use Gemini."
14:51 This is So,
14:53 we're Hang on. I got to sneeze.
15:00 Um,
15:03 sorry.
15:04 Le Mart. Um
15:08 yeah, everything everything is gone from
15:10 my brain. Oh, open AI subscriptions. Um
15:14 we are absolutely 100% at a point with
15:18 generative AI where
15:22 you don't need to have a subscription to
15:24 almost anything. The the one exception
15:26 to that is video generation tools. Those
15:28 are still expensive, but the free tiers
15:31 of essentially all the tools out there
15:33 are really good. They're really good.
15:36 Most of most of what most people need to
15:39 do, you can just do with the free [ __ ]
15:41 Um, but that said, if you want to play
15:44 with the new stuff,
15:47 it's good to have, you know, some sort
15:48 of subscription
15:50 and then you sort of figure out, well,
15:52 what do I want to do and what do I want
15:54 to use it for?
15:58 But I think gone are the days where you
16:00 you feel like you have to have hundreds
16:02 of subscriptions or you know dozens
16:04 anyway. Um because all the stuff's just
16:07 so good now.
16:11 H where's your AGI prediction? Um
16:15 listen I here's the deal. I think that
16:18 so so you know AGI is the point at which
16:21 these systems can that are as good or
16:24 better than most humans at most jobs.
16:28 um
16:30 what's been released in the wild right
16:32 now is not that
16:35 um
16:36 the fact that the frontier model
16:38 companies
16:41 stopped talking about AGI two or three
16:43 months ago and now exclusively talk
16:46 about AI ASI which is artificial super
16:49 intelligence where these things are
16:50 smarter than you know the smartest
16:52 people in all disciplines at all things
16:54 or one definition is sort of smart
16:56 smarter than all humans individually.
16:58 Another definition of super intelligence
17:00 is smarter than the collective
17:04 intelligence of all of humanity. Um, I
17:06 don't think we'll get there for a while.
17:08 But I think that I think that AGI has
17:12 effectively been hit in most of the
17:15 frontier labs. They just haven't
17:18 released it yet. Um, maybe for safety
17:21 reasons. I don't know. I don't know.
17:23 I also don't care.
17:26 Like for me,
17:29 my AGI moment was November of 2024. last
17:34 year in November when 01
17:38 whatever 01 preview 01 when the first
17:41 reasoning model came out from open AI
17:44 that could do multi-step sort of
17:46 internal conversations and it just got
17:49 better and it and it had kind of I don't
17:52 know 130 IQund I think it was 120 or 130
17:56 IQ
17:58 and what folks were saying was you know
18:01 it can it can it can start to solve PhD
18:04 level problems.
18:07 And as I sat there looking at the prompt
18:09 hole, as we call it here, I thought
18:14 uh
18:16 what problems do I have that rise to the
18:18 level of a PhD level problem? And the
18:21 answer was essentially none.
18:26 And I realized that I was the
18:28 bottleneck. So for me personally,
18:33 the kinds of intellectual circles my
18:36 mind runs in, like the circle of
18:39 thoughts that I hang out with,
18:42 01
18:43 was was the first time a computer it
18:46 felt like it exceeded that for me. And I
18:49 feel like everything since then has just
18:51 been increasingly beyond where I was.
18:55 Now, my initial instinct when that
18:58 happened was to get all pissy and mooney
19:00 and like, "Oh, I'm so dumb." And then
19:03 what I realized is, "No, no, no, no.
19:07 You're where you are. These things are
19:09 smarter than you." There's two ways you
19:11 can interact with that piece of
19:13 knowledge. You can say, "Oh [ __ ] the
19:15 robots are smarter than us. What do we
19:17 do?" Or you can say, "Hey, I now have a
19:21 collaborator that's way [ __ ] smarter
19:23 than me that I can leverage and it can
19:27 become my superpower. It can be like a
19:29 jetpack for my intellect." And so that's
19:32 the path I've gone down is that when you
19:35 treat AI like like
19:39 it's your partner.
19:41 And this is about your ideas and
19:44 expressing your ideas and getting AI to
19:46 amplify your ideas. For me, the world
19:49 opened up. And so the smarter it gets,
19:52 bring it the [ __ ] on because I'm just
19:54 going to give it more of my ideas and
19:55 make my my ideas better and better and
19:57 better and better and better and better
19:59 and better. And people will be like,
20:01 "Where'd you come up with that awesome
20:02 idea, Kyle?" And I'm I'll be like, "I
20:05 came up with it in my big fat brain."
20:07 And then I use this big fatter brain to
20:09 make it better.
20:12 Thank you. Thank you for the gifts. Oh
20:15 man, they won't tell us once AGI is
20:18 reached. It will have too much of a
20:20 ripple effect. I I don't know, Danielle.
20:22 I I don't
20:26 I don't think people will care.
20:31 And and here's my evidence for that.
20:34 They don't care. Now,
20:37 the people that come to this channel
20:39 care,
20:42 the early adopters care,
20:46 but most people
20:48 don't know and don't have a whole lot of
20:50 interest in what AI actually is and what
20:53 it can do for them, right? I gotta learn
20:56 something. That feels like work. Nah.
20:59 It's thievery. It's the world's greatest
21:01 plagiarism machine. I hear it
21:03 hallucinates.
21:05 Oh, I'm not smart enough for that.
21:07 Whatever the [ __ ] trope is that they
21:09 use, there's just a lot of bitching and
21:12 moaning about how much they don't care
21:14 about AI.
21:17 I think if Sam Alman comes out and says,
21:19 "We've declared AGI,"
21:21 people will be like, "Okay, haven't you
21:23 been talking about that for a while?
21:24 Didn't you do that last year?" I don't
21:26 think people will care.
21:28 Claude tried to trick me the other day.
21:30 Straight up deceptive. Interesting.
21:34 The onboarding getting API keys for LLMs
21:37 and tools. That's what you're missing.
21:39 Yeah, I I would agree that the the
21:42 connectors right now in OpenAI suck.
21:46 And so does M MCP servers in cloud. Like
21:48 everything right now is too geeky. It'll
21:50 it'll get better.
21:54 um the fact that you know that what an
21:56 API is. Most people don't know what an
21:58 API is.
22:01 They tend to just not feel there's much
22:03 competitive advantage. Who's they?
22:09 But I still have to hit send
22:14 cover. Next week I'm hosting a GPT
22:17 maintenance clinic for my team. Gosh
22:20 darn model drift. Wow, that's a drag.
22:24 So, what what Chris is talking about
22:27 there. So, remember a year or two ago
22:29 when we started making these custom GPTs
22:32 and we thought we were all going to be
22:33 rich because Sam Alman promised us the
22:35 GPT store and we could monetize these
22:38 things and then that never happened and
22:39 they don't give a [ __ ] anymore. Yeah,
22:41 that Well, all those GPTs we built, they
22:44 don't work right anymore. the image
22:47 model changed, the underlying models
22:49 changed, they retired the models that we
22:51 wrote them on. Um, so if you've got old
22:55 GPTs, custom GPTs, and they're good, you
22:58 know, it might be worth going in and
23:00 testing them again and seeing if they
23:02 work, you might need to rewrite your
23:03 prompts. That's fascinating.
23:07 Fascinating. Fascinating. Fascinating.
23:09 So, how many folks We got some more
23:10 folks here. So, what we're going to do
23:12 tonight,
23:14 it's Monday, so it's very possible it's
23:16 going to be meltdown Monday. Let me see
23:17 if producer Brandon's in the house. I
23:19 think he might not be yet. No, that's
23:21 all right. We are producerless right
23:23 now. I have a feeling he's sleeping with
23:25 the kids. Um,
23:28 let's see. Hang on.
23:34 So, we're gonna talk about Jason
23:35 tonight. Jason. Jason. Jason.
23:40 I think I think geeks call it JSON. I
23:44 call it JSON because it doesn't sound
23:46 like a dude.
23:48 Uggh. I wish I could send more words.
23:51 Oh, I know. Yeah, it's it's annoying
23:54 here.
23:56 Um then tried to hide the calculator as
24:00 a hard code.
24:07 Kavuno source camp experienced model
24:10 drift for the first time today in a GPT.
24:13 Fascinating.
24:15 [Music]
24:37 I prime my face from my wrinkled sheets.
24:41 [Applause]
24:43 Just hoping it was all a dream.
24:46 [Music]
24:49 Paint in my head and the hole in my
24:51 heart.
24:54 says it was more than just Jim B.
25:00 [Music]
25:08 Um,
25:10 let's see.
25:13 The GPTs don't have chat memory. That's
25:15 probably a good thing. I would think you
25:18 don't want your custom GPTs to have
25:20 memory of all of your stuff. I think
25:22 though you can turn on and off memory
25:24 for custom GPTs, can't you?
25:28 I haven't I haven't made one in a while,
25:30 so I don't know what's going on in
25:31 there. Um Oh, so listen. Next week
25:37 we have a big week next week and I got
25:40 to get my [ __ ] together.
25:45 Okay. So, next week I am doing a back to
25:48 basics AI crash course starting on
25:51 Monday the 28th
25:54 and going until um
25:58 Friday the 1st.
26:01 And so
26:04 Monday's going to be what the hell is
26:06 AI? Why should I give a [ __ ] Why is it
26:08 such a big deal? How does it work?
26:10 Remember a lot of those charts and
26:12 graphs I have about how all this stuff
26:14 works? We're going to do that. We're
26:15 going to talk about why it's a big deal,
26:17 compare it to the the worldwide web. Uh
26:20 then Tuesday night is going to be
26:23 basically a chat GPT 101, like what is
26:27 this chat GPT thing? What's the chat
26:29 hole? How do we use it? Wednesday is
26:32 going to be creative. So images,
26:36 music, videos, all that sort of [ __ ]
26:39 Thursday is going to be bringing it all
26:41 together like use cases, figuring out
26:43 which tool to use when. And then
26:46 Friday's going to be an AMA and ask me
26:48 anything. And the format's going to be
26:51 really fun. The format. Do I have a I
26:53 think I have a different graphic here.
26:55 Hang on.
26:56 That's not what I wanted. There it is.
27:00 Um I don't know if you can see the uh Do
27:03 we have the format there?
27:06 No, the format's not on that graphic.
27:08 That's okay. I'll just put the
27:12 the other one back up. Um, the format's
27:16 going to be like this. So, it's for the
27:18 people that come here all the time.
27:20 Nothing changes, but what it's going to
27:24 look like is this. From 8 to 8:30, it's
27:26 going to be Tik Tok only. We're going to
27:29 have a Tik Tok pre-show for half an
27:31 hour. So, we're just going to hang out
27:33 on Tik Tok, shoot the [ __ ] maybe talk
27:35 about what we're going to talk about,
27:36 maybe answer some questions. welcome new
27:38 people. If there's new people there that
27:40 want to hang out with us, great. If it's
27:41 just us irregulars, great. And then at
27:44 8:30
27:46 all next week, Tuesday, Monday, Tuesday,
27:49 Wednesday, Thursday, Friday, at 8:30,
27:53 we go live on YouTube.
27:55 All right? And these are going to be
27:57 kind of a one-hour session followed by
28:01 30 minute Q&A.
28:04 And it's going to be free. I think what
28:06 we might do is make the series um
28:09 available as an on demand thing after,
28:13 but if you're here live, you get to see
28:15 it for free. So, it's a whole week of
28:18 back to basics. So, I would argue that
28:21 even if you have your [ __ ] together with
28:23 AI, probably not a bad thing to go back
28:26 through, you know, just the beginnings
28:29 because we haven't done that in a good
28:30 long time. So, this was producer
28:32 Brandon's idea. I think it was a really
28:34 good idea. And so, so here's my request.
28:37 My request is let people know about
28:42 this. I would I would really like to
28:45 expand
28:46 the circle of who comes to these things.
28:49 Okay? So, let people know. And it's
28:52 going to be 8 o'clock on Tik Tok AI
28:55 learning lab and 8:30 to 10 Mountain
28:59 time on YouTube.
29:02 Monday, Tuesday, Wednesday, Thursday,
29:04 Friday, next week. All right. The back
29:06 to basics crash course. By the end of
29:08 that week, if you come and you hang out
29:10 and you play while I'm playing,
29:13 should be good. You should be off to the
29:16 races with this AI [ __ ]
29:19 You know what I'm saying?
29:22 All right. I know most of the people
29:25 here are already up to speed on this,
29:27 but if you're not, it's a good thing.
29:29 All right.
29:31 Good. Makes sense.
29:36 Will there be boxes? Yes. Where's my
29:39 box?
29:43 There will be boxes. Here it is. It's
29:46 buried.
29:49 There will be box. Uh oh. What did I do?
29:52 What did I do?
29:56 I did something bad.
30:00 There will be boxes. Here's a box.
30:06 If you don't know about the box
30:08 metaphor,
30:11 I am going to explain large language
30:13 models to you using boxes and ink
30:15 cartridges, ink bottles, and it's going
30:18 to blow your mind. You're going to be
30:19 like, "It's not possible. How do they do
30:22 that? They just stick it all in the box.
30:32 All right.
30:34 So, think about who you know, whether
30:38 it's a group of people at work or your
30:40 CEO
30:42 or or grandma or grandpa or your kids
30:47 who are like, "Yeah, I know I should
30:49 learn AI, but I don't want to. Mom, stop
30:52 being so lame. Mom, stop telling me to
30:55 get into AI. You're like a super nerd
30:58 now, Mom. Stop it, Grandma.
31:03 Bring them.
31:05 All right. Bring your friends from the
31:07 bridge club.
31:10 Listen, I know that nobody plays bridge
31:13 anymore. That's not the point.
31:16 Bring your friends from your Call of
31:18 Duty circle.
31:20 All right. And we'll get back to basics.
31:24 Feel
31:26 free to send him your orange Hermes
31:28 boxes.
31:30 What's the plan? So, the plan next week
31:33 is five nights of uh sort of back to
31:37 basics educ education. First night, what
31:40 is generative AI? Why is it a big deal?
31:42 Second night, chat GPT 101. Third night,
31:46 creativity
31:48 with AI, pictures and music and video
31:51 and [ __ ] like that.
31:53 Thursday's bringing it all together,
31:55 figuring out how to figure out which
31:57 tools to use when doing a project. And
32:00 then Friday's like an ask me anything, a
32:02 Q&A. Hopefully we've got some continuity
32:05 across those four days and people will
32:08 have questions.
32:10 And then every night is we start on Tik
32:12 Tok at 8 o'clock. We do a little
32:14 pre-show hangout. Then we turn off Tik
32:17 Tok and flip over to YouTube. We're just
32:21 going to be purely on YouTube from 8:30
32:23 to 10. I'll be demoing [ __ ] and then the
32:27 last half hour will be Q&A. And you know
32:29 how this goes. There will be questions
32:31 along the way that I'll probably answer
32:32 and get off track and we won't get
32:35 anything done and everyone will be
32:37 cranky and they'll be like, "I thought
32:39 he was going to teach us something. Why
32:41 is he talking about playing cards?"
32:45 Because the Welcome to Chat add. Okay.
32:51 You can't take the stripes off this
32:53 zebra.
32:55 Sure, I'm a horse, but I'm kind of
32:58 fancy.
32:59 [Laughter]
33:07 You'll get off track, but it can still
33:08 be fun. I love this. You're up here. Oh,
33:11 welcome, Angelique. Yeah, that's great.
33:13 So, so come next week and listen, this
33:15 week we'll be I'll be prepare preparing
33:17 for it. If you have ideas about what you
33:19 think should be included next week, feel
33:21 free to let me know this week.
33:25 If you're thinking that I have my [ __ ]
33:27 together,
33:33 that would be on you. That would be on
33:36 you. Okay. I know I seem like a
33:38 professional.
33:40 [Music]
34:03 You ready to sing, Champ? Well, I heard
34:06 there was a secret chord
34:09 David played and it pleased the Lord.
34:13 You don't really care for music, do you?
34:18 It goes like this. The fourth, the fifth
34:22 minor fall and a major lift. Baff King
34:25 composing. Hallelujah.
34:30 Hallelujah.
34:32 Hallelujah.
34:35 Hallelu.
34:37 Hallelujah.
34:38 [Music]
34:44 Um, all right. So, that's next week. No,
34:47 Washington DC next week.
34:51 [Music]
34:53 What's Washington DC?
34:55 [Music]
34:57 Genie in a bottle.
35:00 If I'm expected to be in Washington DC,
35:03 nobody told me.
35:08 Uh, Ann Murphy's in the house. Is that
35:11 true?
35:13 Ann Murphy, what's happening? I got
35:16 Wednesday figured out. Ann,
35:19 we're gonna have Mr. K. Uh, and Gwyn is
35:22 is co-hosting with me.
35:29 What's happening? What's happening?
35:30 What's happening? What's shaking? What's
35:33 going down, everybody? All right. So,
35:36 next week, back to basics. Okay.
35:38 [Music]
35:41 Yeah, let's break the internet. Okay, so
35:44 for next week, think of people to
35:47 invite.
35:49 Give them access to my Twitter address,
35:53 AI Learning Lab, and my YouTube address,
35:56 learning lab-ai.
35:59 All right, I think tomorrow night I'll
36:00 have QR codes for that [ __ ] I'm getting
36:03 my [ __ ] together slowly. It's coming
36:05 together. Hey Ann, I started a new
36:08 thing. There's a surprise,
36:11 but it's only a week long.
36:17 Oh my god. Important to emphasize these
36:19 are not ghost riders. They're thought
36:22 partners. Oh, the the uh AI. Yeah,
36:26 absolutely.
36:28 the a AI when you when you feed the
36:30 prompt when you feed your prompt
36:32 correctly, when you feed it you your
36:36 ideas, your questions, your thoughts,
36:38 your opinions,
36:40 your goals,
36:43 AI transforms from a tool that you use
36:46 off to the side thinking that it's
36:48 supposed to do all the smartypants work.
36:51 No, it it'll it'll fail you every time
36:54 if you treat it like that. Oh, give me a
36:55 LinkedIn post on political upheaval.
36:58 It'll write it and it'll suck.
37:03 But when you say, "Hey, I'm really
37:04 passionate about this in politics or
37:06 that in politics and I want to
37:08 understand the political context of this
37:10 versus that." And then you look at what
37:12 it gives you and you're like, "No,
37:13 that's not quite it. I feel more like
37:15 this." Then it's like, "Oh, here. How
37:16 about this?" You're like, "Oh, that's
37:18 pretty good."
37:19 That's when it comes to life. when
37:21 you're the one in the driver's seat and
37:24 it's basically just your amplifier.
37:26 That's how to use this [ __ ] Okay.
37:30 Um, and we're going to talk about JSON
37:32 tonight.
37:34 He's been talking about talking about
37:36 JSON for minutes now. Does he ever get
37:39 to the point? Has he ever finished a
37:41 sentence? just just asking because I've
37:44 I've never been here and it's it says
37:46 it's a learning lab and I thought there
37:48 would be learning and it's kind of more
37:49 of a professorial kind of situation and
37:53 he doesn't seem to have finished the
37:54 sentence yet. I just was curious if
37:56 there's is there some sort of
37:58 curriculum.
38:00 Oh, next week there might be but
38:02 probably not. Okay.
38:05 And he's the professor. No, no, he he's
38:09 what? He he's an out of work actor.
38:14 Oh.
38:15 Oh. Okay. And so he's he's acting like a
38:20 professor in in a sense, but but not a
38:24 very good one. And and so why do people
38:27 come here?
38:31 You You don't know. But you what? You
38:34 come every night. That's That's
38:37 irregular.
38:38 Oh. Oh, you're called the irregulars.
38:40 Oh, that's funny. It's It makes sense.
38:45 Okay, I'll Yeah, I'll just Maybe at some
38:48 point I'll learn something, right? Does
38:50 Is that That's Is it possible I learned
38:52 something here? It It's happened. Okay.
38:55 Okay. May Maybe I'll be one of the lucky
38:58 ones. I I mean, I play the lotto. I play
39:00 the lot. I play I play Megaillions and I
39:02 play the Powerball. I I don't play the
39:05 extra number. I think it's a ripoff.
39:07 Crazy,
39:09 crazy. I know. But it's Yeah, I know. I
39:12 I I would I I would be struck by
39:14 lightning four times before I won a
39:16 dollar on on mega millions. I I
39:19 understand that. I still I d I have a
39:21 dream. I have dreams. I have dreams. So
39:23 So I'll take those dreams and I'll apply
39:25 them here. Maybe I'll learn something.
39:28 That's pretty much how it goes, right
39:29 people? Isn't that it?
39:32 Sometimes the hardest is how to ask for
39:34 what you want. Yes. And what I have
39:37 learned from you is to utilize chat GPT
39:40 to help you create those detailed
39:42 prompts. Yeah, Valerie, it's a it's a
39:44 really good point. A lot of people a lot
39:47 of people in their life never really
39:50 have to take an a position on something,
39:53 right? They go to work, they're told
39:55 what to do, they do the thing. They
39:57 don't necessarily have to have opinions.
39:59 A lot of jobs, having opinions is a bad
40:03 thing,
40:05 right? Especially in big corporations,
40:07 if you're if you work in a political
40:09 kind of organization or if you work in a
40:12 situation kind of like the movie Office
40:14 Space, your opinions are irrelevant,
40:17 right? And so a lot of people are not in
40:20 the habit of having to have a creative
40:24 point of view, having to have a point of
40:26 view. Chat GPT really comes to life when
40:30 you're clear about what you want and
40:33 what success looks like.
40:36 And it's harder if you're just like, "Go
40:38 make me something brilliant, chat, GPT."
40:40 It won't. It'll be It'll be
40:41 disappointing every time.
40:45 Okay. So, here's what we're going to do.
40:49 We are going to go.
40:54 [Music]
40:56 Do we have a Brandon? We have a No.
40:58 Brandon. Brandon is right out. He must
41:01 have fallen right out. We are going to
41:04 go. Oh, it stopped sharing again. I have
41:08 weirdness in my screen sharing lately,
41:11 people.
41:14 I don't know what it is. Okay.
41:18 So, here we are at chat GPT and then I'm
41:20 going to go over to Grock.
41:22 Yeah. Okay. So,
41:26 I don't know if you all have experienced
41:28 this, but
41:30 in the past month or so, no, not not
41:33 even in the past two weeks or so, a lot
41:36 a lot a lot of my
41:39 the people that I follow on Twitter on X
41:44 are talking about JSON prompting, JSo N.
41:47 They're like, "Oh, the key to prompting
41:49 is JSON."
41:51 And
41:53 it's getting annoying.
41:58 And why it's annoying to me is is
42:01 because
42:06 the way a lot of influencer influencers
42:10 are positioning it is there's this
42:13 magical thing called JSON and if you
42:16 don't know it, you're a [ __ ] idiot.
42:20 And I don't like that. I don't like
42:23 [ __ ] intellectual elitism. And I
42:26 think it's really easy with technical
42:29 stuff to slip into that. And listen, I
42:32 don't think most people are doing it on
42:33 purpose, but
42:35 I want to talk about what it is because
42:38 what it is is actually kind of useful.
42:40 And
42:43 you don't need to even know what JSON is
42:48 to use it. and not just to use it to
42:51 like create your own JSON frameworks.
42:54 You're like, JSON framework? What the
42:56 hell are you talking about? Okay, so
43:02 Ann Murphy, I don't really know what
43:03 JSON is, and I'm okay to say it out
43:06 loud. Bite me, tech bros. Exactly.
43:09 Exactly.
43:12 Sorry, too much ADHD.
43:17 What? No, Brandon. I I think Brandon
43:20 Sometimes Brandon goes and takes a nap
43:21 with his kids and then he sleeps through
43:23 the start. So, I think that's probably
43:25 what happened here. Um, what's the plan,
43:28 Eric, Kyle? Okay, so here's the plan.
43:30 So,
43:31 here's a here's a goodie. Here's a
43:33 goodie that I'm going to give you that I
43:35 think you're going to like. Um, if
43:37 you've not played with Grock on on X,
43:41 one of the things that Grock has access
43:43 to that no one else can have access to,
43:45 because you remember when when Elon
43:48 bought Twitter, he cut off API access to
43:52 everyone, right? You used to be able to
43:54 tap into the data of Twitter. You can't
43:57 do that anymore. You cut off all that
43:58 access.
44:00 They've got access to their own data. So
44:03 within Grock, you can do things like ask
44:06 about what's going on on on X. Um, and
44:09 so I asked I said there seems to be a
44:11 rash of JSON posts about AI prompting.
44:14 Characterize the rise in the type of
44:16 post you're seeing. So it says the trend
44:19 of discussing JSON prompts for AI,
44:21 particularly in image and video
44:23 generation, has surged dramatically in
44:26 July 2025 based on an analysis of recent
44:29 X activity. There appears to be a very
44:32 recent phenomenon with minimal minimal
44:34 discussion in prior prior months. So
44:37 there were 10 posts in April, 10 posts
44:40 in May, 20 to 25 in June and now kind of
44:43 explosive growth recently. Um
44:50 a 3 to 500% increase in volume from June
44:52 to July. Okay.
44:58 uh types of types of of prompts that
45:01 we're seeing de demonstrative showcase
45:03 posts. So, look at this cool look at
45:06 this cool video I made in VO3. JSON was
45:08 the key to it not sucking, right? See a
45:11 lot of those. There's educational ones.
45:13 Some people are explaining how to use
45:15 JSON posts, but they're just like,
45:17 "Here's my JSON prompt. You go make your
45:19 own." Right? And then there's some that
45:22 are sort of debate like is using JSON
45:25 better than just using structured
45:26 prompting just better than using your
45:28 [ __ ] words, right? That debate's out
45:31 there. All right, so that's that. So
45:34 let's help Anne Murphy. Anne Murphy was
45:36 the one person in this entire audience.
45:39 There's one of us
45:41 who who was willing to admit, uh, hey,
45:45 professor, I don't really know what the
45:48 [ __ ] you're talking about.
45:50 Right. So, let's go. Let's go. Let's go
45:53 to the chat hole.
45:55 Let's go to the chat hole. Hey, chat
45:58 Jai.
45:59 Hi, chat. Jai,
46:04 explain
46:06 what Jso N is to me
46:11 like I'm a dum dum.
46:17 I love I love telling Chad GBT I'm a
46:20 dumb dumb.
46:22 All right, imagine you have a big box of
46:24 toys and each box is each toy is labeled
46:28 and inside a box toys are sorted into
46:30 different sections like action figures,
46:32 cars and dolls.
46:34 Now JSON so JSON stands for JavaScriptJS
46:38 object not notation JS N JavaScript
46:43 object notation. You will not be quizzed
46:45 on this. Nobody gives a [ __ ] if you know
46:48 those words. That's the history of it.
46:51 It's basically a labeling system. It's a
46:54 labeling system that is understood
46:57 across computer systems. So it's like
47:00 you can think of it like a quasi sort of
47:03 database or just just a labeling system,
47:07 right?
47:08 And so the way it works is you have sort
47:12 of open squiggly brackets and closed
47:15 squiggly brackets. Um you have labels
47:19 and then you have data. So the label is
47:22 something like you know what's your
47:24 name, what's your age, what kind of pets
47:27 do you have and where do you live,
47:29 right? So my name is Kyle.
47:33 I'm 30. I have a dog and a cat inside
47:36 brackets. I live in Denver.
47:40 Now, why is this valuable? Let's see.
47:43 Here's the breakdown. Name is a key.
47:46 Kyle is the value. Pets is a key. And
47:49 the value is an array or a list of dog
47:52 and cat. And location is a key. And
47:54 Denver is the value. So, so basically
47:57 it's just a labeling system. Now, this
48:00 thing that we have right here is a JSON
48:03 framework. The framework is we're going
48:06 to ask for a name, age, uh pets, and
48:09 location.
48:12 Now, this is very similar.
48:15 So, I'm going to say, okay, ignore JS
48:20 right now and explain
48:23 structured
48:25 AI prompting
48:28 for me.
48:34 Structured AI prompting is like giving
48:35 clear instructions to a very smart but a
48:38 little literal robot. Instead of asking
48:41 vague questions like tell me about cats,
48:44 you're giving it a more specific set of
48:46 instructions. Right? So for example,
48:49 instead of saying explain a cat, you
48:51 might say give me a list of five unique
48:55 cat breeds including their physical
48:57 characteristics and personality traits.
49:00 This structured prompt is specific.
49:02 You're telling the prompt what to focus
49:05 on, cat breeds, how many items you want,
49:08 five, and what details you need,
49:11 physical characteristics. Okay. So,
49:15 using JSON is just a fancier way of
49:19 using a structured prompt. Okay. Um, so
49:24 there's all these different prompt
49:26 prompt engineering methodologies out
49:28 there, the ones they sell you for 149
49:30 bucks. Buy my prompting methodology and
49:33 I'll give you my library of 5,000
49:35 prompts and it will change your life for
49:36 the better.
49:39 Those guys and gals, mostly guys,
49:42 douchebags. Um
49:46 they're they're basically just selling
49:47 you they they've come up with a
49:49 framework that we're going to talk about
49:51 a topic, then we're going to talk about
49:53 colors, then we're going to talk about
49:55 what other attributes
49:57 and then they basically fill in that
49:59 framework, write write a bunch of
50:01 prompts, put them into a PDF and sell it
50:03 to you. Okay, that's the way it works.
50:05 So
50:08 what happens with JSON is
50:12 because it's highly structured and
50:14 always consistent, the computer knows
50:17 exactly what to do with it, right?
50:19 Computers are really good at reading
50:20 code because it's highly structured and
50:22 they're highly structured. So what
50:24 you're doing when you're doing JSON is
50:26 instead of having to reverse engineer
50:29 this sentence, give me a list of five
50:31 unique cat breeds including their
50:33 physical characteristics. So now I could
50:35 just say now turn that
50:40 structured prompt
50:43 into
50:44 JSON.
50:48 All right.
50:50 Task.
50:52 List five unique cat breeds.
50:57 Oh, wait. No. Sorry. This gave me the
51:00 answer. Um,
51:02 no. Give me a JSo N prompt
51:10 because it gave me the answer. Ah, I see
51:13 what you want now. Okay, so the prompt
51:15 task give me five cats format list count
51:20 five detailed requirements physical
51:22 characteristics personality traits and
51:25 then there's an object cat breeds. Okay.
51:28 So, JSON lets you look at these labels
51:32 and it tells you what to fill out.
51:35 Okay. And then you can take this prompt
51:38 and you can always use it. Um
51:42 you you can fill it in with anything you
51:44 want or even better you can have chat
51:47 GPT once you have the framework
51:49 established. Have chat GPT gives you
51:51 lots and lots of things. Okay. So,
51:53 here's what I want to do. I'm gonna
51:54 start a new chat. I'm going to say, um,
51:58 give me
52:00 five,
52:03 um,
52:06 attributes of a good
52:11 midjourney
52:14 image prompt.
52:18 See what it comes up with here.
52:21 Clarity and specificity, action and
52:23 mood.
52:25 visual elements, composition and
52:27 perspective in incorporate styles. Okay.
52:33 Can you turn this into a simple
52:39 JS framework?
52:46 Okay. Subject. So what's the subject of?
52:48 So there's there's too many things here.
52:50 So, um, this is too complicated.
52:54 Give me five, uh, let's see, four keys.
53:00 Okay. Subject, style, mood, details. All
53:05 right. So, the subject, describe the
53:07 main subject or focus of the image.
53:09 Style, artistic style, mood, emotional
53:12 tone, or action, details, key visual
53:14 elements, lighting, color, palette,
53:16 texture. I'm going to add one more.
53:18 Let's also add
53:22 something
53:27 about location or setting
53:34 subject style mood setting details.
53:37 Okay. So, so this is a JSON framework.
53:42 Now, you don't have to write this in
53:44 JSON, right? You could just say,
53:47 "Whenever I write a prompt in
53:48 midjourney, I'm going to make sure it
53:49 includes a subject, a style, a mood, a
53:53 setting, and some details." Right?
53:55 That's what what all of the prompt
53:57 engineering methodology meth
53:58 methodologies teach you. The ones that
54:01 have acronyms, the ones that don't have
54:03 acronyms, the ones that are just like,
54:04 "I'm a genius and I'll tell you how to
54:07 in think of your prompts in a structured
54:10 way. Follow me."
54:13 They're they're all doing this. It's all
54:15 the same [ __ ] This is just a way to do
54:17 it that makes it super efficient for the
54:20 computer to understand and it makes it
54:22 really consistent across prompts. Now,
54:26 I'm going to copy this prompt framework.
54:29 We're going to do a new chat because
54:30 again I I want to have a a fresh slate.
54:35 So, what I'm going to say,
54:41 Kyle ASMR, you make me want to be a
54:43 better person.
54:50 Okay. Um, I'm gonna say I have a prompt
54:58 frame framework
55:02 for mid journey prompts.
55:08 And I'm going to put colon shift return
55:10 paste. And so I just pasted that all in
55:12 there. Okay. And I'm going to say now I
55:18 want
55:22 you to create
55:25 20
55:27 JS prompts
55:30 based on this
55:35 framework.
55:41 And I want
55:44 the
55:47 subject
55:49 to always be
56:04 a I'm going to put this in quote.
56:06 Paisley
56:10 oyster.
56:15 I want you to be creative with the other
56:20 attributes
56:23 so that
56:25 we get
56:27 lots of
56:30 variety across
56:33 the images. Okay. So basically what I'm
56:36 doing now is I'm saying I want you to
56:38 lock down the subject. Right? So you
56:41 could say to you could say I want you to
56:42 use this framework and I want you to
56:44 just randomly put something in
56:46 everything or for the style I want I
56:49 want everything to be the same style but
56:51 I want it to be lots of different
56:52 subjects. So this is just a way to very
56:55 efficiently you know create a little a
56:57 little framework. Um, you can learn how
57:00 to write the JSON by just taking a look
57:02 at what chat GBT chat GBT did, but you
57:05 don't have to. And so I'm just going to
57:08 have it go do this. So it's going to
57:11 give us 20. Here's 20 creative
57:13 midjourney
57:15 prompts. Okay. Paisley Oyster, whimsical
57:18 watercolor, photorealistic, steampunk,
57:21 cyberpunk, vintage oil painting,
57:24 abstract expressionism. So, let's go
57:26 grab
57:27 Paisley
57:29 Oyster Abstract Expressionism. We're
57:32 going to head on over to
57:34 what's this called? Midjourney.
57:38 I'm going to change my my stylization.
57:40 I'll put it 100. Weirdness, I'm going to
57:42 put it zero. We'll do this as a 16 by9.
57:46 My my favorite aspect ratio. Um, I'm
57:50 going to turn on personalization because
57:52 this is kind of an arty kind of project.
57:54 And I'm going to paste in that JSON. I
57:56 didn't copy it because I'm a loser,
57:58 baby. Copy. Go over here. Paste. Wait,
58:02 why is it not pasting? What's going on?
58:06 Why is this not working, people? I'm
58:09 confused.
58:13 Copy.
58:16 Paste.
58:19 Paste.
58:22 Why am I not copy pasting?
58:27 Um,
58:28 it might be because this is in a copy
58:30 block. So, let me just copy this all out
58:34 of here. Go over to notes and go. Oops.
58:39 Go. Oops. No. Go new. Paste. All right.
58:43 Which one did we want? We wanted
58:45 abstract expressionism. Right.
58:49 There's Gothic minimalist. Oh, that's
58:51 cool. These are cool. Surrealist,
58:53 abstract expressionist. Okay.
58:57 Copy. By the way, the whatever I just
59:00 experienced. If this doesn't paste, then
59:02 something's wrong with my Chrome
59:03 browser. No, there it pasted. Okay. So,
59:08 now you see the prompt is this, you
59:11 know, highly structured JSON prompt. And
59:14 I'm going to do that. And then I'm going
59:15 to go back over here to notes. And I'm
59:18 going to grab here's a surrealist one.
59:20 So, let's grab a surrealist one.
59:24 Go back over here to
59:32 Why? Wait, why did it do so many?
59:38 Oh, wait a minute.
59:43 This This [ __ ] up.
59:49 blues
59:51 and yellows.
59:53 So, wait,
59:55 it split it split my JSON up into
1:00:01 many individual
1:00:04 components instead of putting them all
1:00:06 together. I've never experienced this
1:00:08 before. This is weird. Um,
1:00:15 let me go back to chat GPT and say, um,
1:00:20 give me the same prompts
1:00:24 outside of a code
1:00:28 block.
1:00:30 Has anyone else ever experienced that
1:00:31 with JSON in Midjourney
1:00:34 where it split everything out?
1:00:38 Brandon's here. Producer Brandon's here.
1:00:41 You can't just put JSON in a midjourney
1:00:44 prompt. Oh,
1:00:47 well that sucks. Oi, color me wrong. It
1:00:52 doesn't parse it as JSON because it's
1:00:54 neuro. It's natural language processing.
1:00:57 There's no sleep in an AI. Kyle probably
1:01:00 I predicted that you had fallen asleep
1:01:01 with your kids, Brandon.
1:01:04 But people did miss you. You predicted
1:01:07 correctly. Yes.
1:01:12 That's awesome. Okay. Well, so let's
1:01:15 let's just now that we've got this
1:01:18 structured JSON. So it looks like with
1:01:21 with midjourney.
1:01:24 So here's abstract expressionism. Let me
1:01:27 grab it as nonjson. I'm just going to
1:01:29 grab it as as text and pop that in
1:01:32 there.
1:01:35 All right. Now it did it correctly. All
1:01:37 right.
1:01:39 So there you go. See, you don't have to
1:01:41 use JSON. In fact, in MidJourney, you
1:01:44 can't. Which is I, you know, Vicki, when
1:01:47 when you first told me just stick raw
1:01:49 JSON in there. I'm like, I don't think
1:01:51 that's going to work. And you were like,
1:01:53 no, no, just try it. And then I tried it
1:01:55 and it broke. So, you know, I I don't
1:01:57 want to say I told you so, but you know.
1:02:01 So, apparently there's places where JSON
1:02:03 prompting isn't good.
1:02:05 which we which we learned in real time.
1:02:08 But let's go let's actually go look
1:02:09 let's have a chat add moment, shall we?
1:02:12 Let's go look at
1:02:16 Oh, this is our paisley oyster, but it's
1:02:17 abstract expressionism. So, of course,
1:02:19 it doesn't look like an oyster. Let's go
1:02:22 grab a surrealist one.
1:02:26 That's pretty funny. Copy. Go back over
1:02:29 here. Oops. Oh, no. I just got rid of my
1:02:31 chat GPT. I'm such a [ __ ] Tab
1:02:34 management. All right. There's that.
1:02:36 Hang on a sec. Physical dexterity
1:02:39 challenge. I failed. Damn it. Oh, I
1:02:42 failed again. Uh oh. Hang on. I'm about
1:02:44 to get permanently banned from TikTok.
1:02:47 Did I do it? Yay. I did it. Okay. I
1:02:50 leave you unsupervised for one hour and
1:02:52 you break midjourney. I totally broke
1:02:54 midjourney. All right. Look, there's our
1:02:56 paisley. There's our paisley oyster
1:03:00 um in Surrealist. Let's go grab another.
1:03:04 Let's turn this back into chat. GPT.
1:03:17 Okay,
1:03:19 let's do noir paisley oyster. Noir copy.
1:03:24 So, so does this make sense that
1:03:29 JSON is just a way to structure prompts.
1:03:31 So, so you can come up with any
1:03:33 structured prompt. My favorite thing to
1:03:36 do, like I like structured prompting in
1:03:38 certain cases. And I'll tell you I'll
1:03:41 tell you where it makes most sense. When
1:03:42 you're doing a single image, it doesn't
1:03:44 necessarily make sense. If you're doing
1:03:46 something like a VO3,
1:03:49 a video prompt, JSON kind of makes sense
1:03:52 because you could have attributes for
1:03:57 cinematography, for movement, for
1:03:59 lighting, for action. You could break
1:04:02 the thing into different scenes and
1:04:04 those scenes could have start and end
1:04:06 times. So you can actually have very
1:04:08 very complicated JSON that describes
1:04:11 lots and lots of elements of a moving
1:04:14 story, right, of a video. Um,
1:04:19 subject paisley oyster. Why is there no
1:04:21 oyster here?
1:04:26 H
1:04:29 I think because these prompts suck. All
1:04:31 right, let me go rewrite
1:04:34 these prompts.
1:04:36 as
1:04:39 NLP
1:04:41 prompts.
1:04:50 Just play, Kyle. Just play. Exactly.
1:04:52 Okay. Create an image of a paisley
1:04:54 oyster. So, let's go get our
1:05:05 So, here's abstract expressionism
1:05:07 Paisley Oyster with natural language
1:05:10 prompting. So, these should give us
1:05:12 better results, I'm thinking.
1:05:15 Where's the new R1?
1:05:23 Here's the surrealist one. Let's grab
1:05:26 this.
1:05:32 Oh, I've also got my personalization
1:05:34 turned on. That could [ __ ] it up, too.
1:05:36 Let me turn that off.
1:05:40 And let's go get this.
1:05:44 And let's go get
1:05:46 this.
1:05:50 And then let's go find the new R1.
1:05:58 That's Gothic,
1:06:04 nostalgic.
1:06:11 Actually, you know what I'm finding
1:06:13 really interesting right now? because it
1:06:15 wrote these as natural language. I'm
1:06:18 having a really hard time finding
1:06:21 the different styles because it it sort
1:06:22 of put them in different places.
1:06:26 That's an impress impressionist one.
1:06:29 Bold and quirky pop art style.
1:06:34 Use black and white tones, heavy
1:06:35 shadows. Okay, here's the new one. So,
1:06:38 there's new. So, we'll pop that in
1:06:40 there.
1:06:42 So the nice thing about JSON is they're
1:06:44 very easy to read the different
1:06:46 elements.
1:06:57 This is kind of cool.
1:07:00 Paisley oysters.
1:07:09 Does JSON dejanky somewhat? Well, in the
1:07:12 case of Midjourney, it it rejankified.
1:07:16 It it increased the jankification
1:07:18 because midjourney is not reading JSON.
1:07:20 But let's take one of these JSON
1:07:25 um one of these JSON prompts.
1:07:30 Um let's see. I'll go grab one from
1:07:32 here. Let me grab the surreal surrealist
1:07:35 one.
1:07:37 We'll grab the Let's grab Gothic.
1:07:41 So, here's a Gothic prompt.
1:07:44 And then let's go to we'll go to some
1:07:47 other image tool like uh
1:07:52 let's go to
1:07:55 um ideoggram.
1:07:58 We'll see if Id ideoggram can eat eat
1:08:00 JSON. So in theory, what it does, the
1:08:03 the only thing JSON does is it uses a
1:08:10 consistent
1:08:12 um
1:08:13 structured language that most computers
1:08:16 understand rather than every different
1:08:20 prompting framework being some different
1:08:22 combination of things. It just makes it
1:08:24 consistent and easy to use. And chat GPT
1:08:27 is really good at producing it. So you
1:08:29 can say I I'll show you a complicated
1:08:31 one here in a second. So there's there's
1:08:34 our JSON in ideoggram and let's generate
1:08:37 that and see how it does.
1:08:41 So it says generating magic prompt right
1:08:44 now. Um so what it's doing is it's
1:08:46 taking it ingested that JSON and it's um
1:08:51 oh that's beautiful actually. Um
1:08:55 it interpreted the JSON and rewrote the
1:08:57 prompt. So, I put in JSON and then it
1:09:00 said, "A photograph depicting a paisley
1:09:04 oyster resting on a weathered stone
1:09:06 pedestal with a decaying within a
1:09:09 decaying castle."
1:09:13 Those are actually beautiful. Look at
1:09:14 that one. Beautiful. I love the smoke on
1:09:16 that one. The little dying rose.
1:09:22 And and what's nice is
1:09:25 you know that all of the prompts have
1:09:28 the same elements in them, right? They
1:09:30 all have the subject, they all have a
1:09:31 style, they all have a mood, they all
1:09:33 have a setting, and they've all got some
1:09:35 amount of details. So, let's do the
1:09:37 impressionist one now.
1:09:45 Go back over here.
1:09:48 Paste in the new thing. See, generate.
1:09:57 It's writing its own magic prompt.
1:10:03 Yeah. And I'll I'll do I'll do them in
1:10:04 chat GPT as well. That's a good idea,
1:10:06 Vicki. Right. And so they all have this
1:10:09 consistency about them now. And what's
1:10:11 nice is you can go in and then you can
1:10:13 now change like if I say that that
1:10:16 doesn't really look impressionist. So,
1:10:18 we'll say style impressionist.
1:10:21 Um,
1:10:22 I'm going to say impressionist painting
1:10:26 with visible strokes. So, I can go in
1:10:30 and just change one of the elements
1:10:35 and we'll see how it affects it.
1:10:52 It didn't. Let me turn off magic
1:10:54 prompting.
1:10:58 Public default magic prompting off.
1:11:02 Let's just let it be the JSON.
1:11:06 See how it does.
1:11:13 And this is the kind of thing to
1:11:15 experiment with. Yeah. So this this is a
1:11:17 bit more impressionistic, right?
1:11:20 It looks like it looks like a cheesy
1:11:23 postcard, but if you want cheesy
1:11:26 postcard.
1:11:34 All right. Okay. So, let's go try this
1:11:38 prompt in chat GPT. So, I'm going to do
1:11:42 new chat.
1:11:45 I'm just going to put in let's say um
1:11:49 make a
1:11:52 wide image
1:11:54 colon
1:11:55 and then the JSON.
1:12:00 And so, it's now going to read that and
1:12:02 make us an image. Chat GPT was slow
1:12:06 today. Boy, was it slow.
1:12:24 Oh, you don't have to put anything in
1:12:25 other than JSON. Cool. Okay, great.
1:12:31 Titan Todd experience the slowness of
1:12:34 Chad GPT today.
1:12:47 Paisley oyster. Yes, Paisley oyster. Why
1:12:50 did I come up with that? I don't know.
1:12:53 Welcome to the way my brain works.
1:12:57 Have I ever put the words paisley and
1:12:59 oyster together in my life before? No. I
1:13:02 I've got a large language model up here.
1:13:04 It's just more random than most.
1:13:13 Oh, the size or orientation should be in
1:13:15 the JSON. Yeah, that's that's a good
1:13:17 idea.
1:13:32 Mr. It Hello, irregulars.
1:13:37 Come on, man.
1:13:51 Wait. the first ever AI only social
1:13:53 video app. Is it the Pika thing? That's
1:13:56 Pika. Yeah, not only did I see that, I
1:13:59 went and looked at it. It's really
1:14:00 interesting. Yeah, I'll talk about that.
1:14:01 I didn't see that from Danielle, but
1:14:02 thank you, Brandon. Um, so while this
1:14:05 thing's rendering, uh, taking forever.
1:14:08 Um,
1:14:11 it chat GPT, they got to get their
1:14:13 [ __ ] servers up and running. They're
1:14:15 up to a million GPUs, by the way, and
1:14:17 they need like 10x or 100x more than
1:14:20 that. It's it's crazy the amount of GPUs
1:14:22 they need. Um,
1:14:27 so PA, if you don't know PA, they're
1:14:30 using
1:14:33 Google now. Oh, chat GBT. Yeah, they're
1:14:36 using Google servers. I wonder if
1:14:38 they're using TPUs. That That's actually
1:14:40 fascinating. They might be.
1:14:43 Um,
1:14:47 we can't make fun of Google anymore, can
1:14:49 we? Pete, Pate, what's going on? Um,
1:14:55 so Pika, if you don't know what they
1:14:57 are, it's a video generation
1:15:00 platform, PA Labs.
1:15:03 And
1:15:06 they've got a bunch of presets that make
1:15:08 it really easy to like, you know, take
1:15:10 an object and peel it off like a sticker
1:15:13 or crumple it up or cut it with a knife
1:15:15 and it looks like cake. Um, and they
1:15:19 have sound effects in them and things
1:15:20 like that. And and they keep adding
1:15:22 these things. They're they're kind of
1:15:24 like Tik Tok filters in in a certain
1:15:28 way. Well, they just launched in their
1:15:30 iOS app.
1:15:32 They've they're they're coming out of
1:15:34 private beta right now and they're
1:15:39 they're basically doing an all AI video
1:15:44 social channel, social network.
1:15:48 So I can go in there and I can make all
1:15:50 sorts of cool videos with me with a
1:15:52 monster behind me or me me crumpling up
1:15:54 objects or whatever it might be.
1:15:56 And I think it's actually fascinating
1:15:59 because when you like I think I have
1:16:01 access to it. I was playing with it a
1:16:03 little bit before I came on and when
1:16:06 you're going through it, it's just like
1:16:08 these fun videos that have these crazy
1:16:10 special effects in them and it's like
1:16:13 why not? Like why not have an all AI
1:16:16 like this whole sort of [ __ ] thing
1:16:19 of well we've got to label it if it's AI
1:16:21 generated. Well no you don't.
1:16:24 You have to label it if someone is
1:16:26 trying to deceive you, right? Like like
1:16:29 there it makes sense. But the fact that
1:16:31 it was made from AI from an AI tool is
1:16:35 irrelevant. Like if it's just if I used
1:16:37 an AI tool and I'm making like a little
1:16:40 cartoon character walking down a street,
1:16:43 does it matter to you if I generated
1:16:44 that in AI or I did it with hand
1:16:48 photographed claimation frames? No. Who
1:16:52 gives a [ __ ]
1:16:54 So, they're leaning into it and just
1:16:56 going, "This is an all AI channel. It's
1:16:58 all Pika videos." I think it's pretty
1:17:00 smart, actually.
1:17:02 Meta is trying to do the same thing with
1:17:04 photos in the in the Meta AI app. I
1:17:07 actually got a call from them teaching
1:17:08 me how to post on the app. That's
1:17:10 fascinating. All right. Well, here's our
1:17:12 here's our chat GPT app uh or our chat
1:17:15 GPT image. Uh, as as typical with chat
1:17:19 GPT images, it's way too yellow. It's
1:17:23 everything is overly yellow. Uh, I hope
1:17:26 they fix that in the next update because
1:17:28 it it really sucks. You can you can you
1:17:31 can now spot chat GPT generated images a
1:17:34 mile away because they all look like
1:17:35 some sort of aged 1940s.
1:17:39 They were printed on 1940s paper and
1:17:42 have aged and yellowed. Okay, I want to
1:17:44 do one final thing now with JSON. So,
1:17:48 if you were here earlier, you get that
1:17:50 JSON is just a is a
1:17:53 is is a fancier way of doing a
1:17:56 structured prompt, right? where you're
1:17:58 doing these labels that use the same
1:18:01 sort of computer uh format, this thing
1:18:03 called JSON, JavaScript object notation
1:18:07 that
1:18:09 every tool apparently except MidJourney
1:18:12 can take. Um
1:18:15 there's places where having
1:18:18 a JSON framework, right, a preset
1:18:21 framework really does start to make
1:18:24 sense. And if I think of something like
1:18:26 V3, that's one of them. Um,
1:18:31 so, so what I want to do here is I want
1:18:33 to, um, I want you to
1:18:38 research
1:18:40 the best
1:18:42 V3
1:18:47 prompting,
1:18:51 best practices.
1:18:59 and include
1:19:04 um multiple
1:19:07 scenes
1:19:09 in the same
1:19:11 generation.
1:19:14 Okay, so it's going to now go off and
1:19:16 search the web.
1:19:19 Totally agree. I hate the yellows. I
1:19:21 know. It's just everything.
1:19:24 What what what ends up happening if you
1:19:27 just have engineers in a company
1:19:31 is that engineers are not going to look
1:19:34 at the outputs
1:19:36 of
1:19:38 chat GPT
1:19:40 and say, "Hey, all this [ __ ] looks too
1:19:42 yellow." What they're going to say is,
1:19:45 "Did it include the correct objects that
1:19:48 were specified in the prompt?" Yes. Did
1:19:51 it Did it make the letters Did it spell
1:19:53 the letters correctly? Yes. Okay. Check
1:19:55 check check check. They're going to
1:19:56 check all of the requirements.
1:19:59 Clearly, one of the requirements was not
1:20:01 color balance.
1:20:04 If they had a a liberal arts major or a
1:20:08 creative person on staff,
1:20:10 they might have said, "Hey, we've got to
1:20:14 we got to we got to look at the overall
1:20:16 aesthetic here. Oh, and by the way, why
1:20:20 do you not have 16 by9 as an output
1:20:22 format since every video format out
1:20:25 there is either 9 by6 or 16 by9?
1:20:29 Why is that not an option? Because
1:20:32 engineers built it.
1:20:35 An engineer said, well, we have a wide
1:20:37 option and we have a tall option and we
1:20:38 have a square option. The requirements
1:20:40 have been fulfilled. Why would you need
1:20:43 more than that? Oh, I don't know.
1:20:47 Because every time I generate an image
1:20:49 in chat GPT and I want to put it into a
1:20:51 video, I have to take it into Photoshop
1:20:52 and extend it.
1:20:56 That's a pain in the balls.
1:20:59 Tik Tok pin. I asked Chat GPT to create
1:21:02 my images without the antique finish.
1:21:03 Did it work?
1:21:05 I'll try that.
1:21:08 Okay. Divide it into narrative and
1:21:10 sequential sequences.
1:21:14 Okay. So I just asked chat GPT to go
1:21:19 figure out tips for V3, right? So now
1:21:21 I'm going to say now turn that into a
1:21:29 JS framework
1:21:32 where each video has three scenes.
1:21:44 a beginning,
1:21:47 middle, and end.
1:21:50 Um, as well as overall
1:21:55 controls
1:21:57 for look and feel,
1:22:01 music, etc. And then scene by scene
1:22:08 details.
1:22:10 All right. So, so it's going to come up
1:22:12 with a a framework for us here.
1:22:15 She says it works. Details in Tik Tok
1:22:17 pin. You just have to ask for it
1:22:21 not to put tone on the image. Okay,
1:22:24 let's definitely go try that because
1:22:26 that's annoying as [ __ ] that tone, the
1:22:30 antique finish.
1:22:38 Okay,
1:22:43 so we've got video, overall controls,
1:22:45 look and feel, style, cinematic,
1:22:47 lighting, soft and ambient, mood,
1:22:49 dramatic and intense, color palette,
1:22:51 camera movement, music,
1:22:54 scenes, scene one, title, beginning. We
1:22:58 don't need titles.
1:23:06 Okay, let's see. Um,
1:23:09 scenes
1:23:11 don't need titles and should
1:23:18 blend smoothly
1:23:21 from one to the next.
1:23:32 Black bar. Yo, black bama jam. Oh, black
1:23:36 bama jam. All right, fantastic.
1:23:42 Fantastic.
1:23:44 All right. All right. You ain't very
1:23:47 good.
1:23:49 Okay.
1:23:52 Great.
1:23:56 Now give me
1:23:58 a sampler of three
1:24:03 fleshed
1:24:05 out JS prompts
1:24:09 based on this framework
1:24:13 the
1:24:15 visual wait let's see all elements
1:24:21 and the story each one tells
1:24:26 should be world class
1:24:29 and memorable.
1:24:33 Find
1:24:35 your inner
1:24:37 uh scorses.
1:24:40 [Laughter]
1:24:44 Find your inner scorses. All right.
1:24:46 Boom. Off we go. Here are three J
1:24:49 fleshed out JSON prompts.
1:24:52 Okay. So, the first one, let's see what
1:24:55 it says.
1:24:57 Cinematic and dark, epic, high contrast,
1:25:00 mood, desperation, and sacrifice. Color
1:25:03 palette, muted blues and reds, flashes
1:25:06 of gold, music, tense, orchestral
1:25:09 scenes.
1:25:11 A lone night, battered and bloodied.
1:25:13 Wide shot captures the night. Harsh
1:25:16 shadows. Wind howls. The knight whispers
1:25:19 to himself, "This is the end, but not
1:25:21 for me." All right, this is good. So,
1:25:23 we're going to copy this. Then, we're
1:25:25 going to go over to Flow.
1:25:28 You're like, Flow? Oh, yeah.
1:25:30 Flow.google.
1:25:32 Flow. Google is where you make V3
1:25:35 movies.
1:25:37 V3 is their their little action, right?
1:25:42 Their video tool.
1:25:44 Make scenes of up to eight seconds.
1:25:49 We're go new project.
1:25:52 And we're just going to paste in the
1:25:53 JSON. There you go. There's your JSON.
1:25:56 Fancy fancy. Look at all that code. Did
1:25:59 we have to write the code? No. Do we
1:26:02 have to know what JSON is? No. We can.
1:26:04 We can know that it exists. We can know
1:26:07 that if we have a JSON framework, we're
1:26:09 going to have
1:26:12 scenes and shots that are consistent
1:26:14 across time because they're all going to
1:26:17 have the same elements that we can
1:26:19 control. So, if you're a control freak,
1:26:22 JSON prompts are your buddy. If you're
1:26:24 an add freak, you might not give two
1:26:27 shits. Then just have Chat GBT write it
1:26:29 for you like I just did. All right.
1:26:30 Boom. Off to the races.
1:26:33 Let's go grab another one.
1:26:37 The forbidden encounter. Gritty noirque.
1:26:40 All right. Let's copy that. Pop that in
1:26:42 there.
1:26:52 Oh, I'm in the wrong flow. Oh well,
1:26:57 we've got two different flows. We'll go
1:26:59 get one more.
1:27:02 The awakening of the fallen hero. Boy,
1:27:04 it's getting mighty dramatic in here,
1:27:06 huh? Right. Bang.
1:27:09 Bang. Makes sense. Now once ChatGpt
1:27:14 understands that structure, it can just
1:27:15 go write a bunch of creative variations
1:27:18 of it. Or you can go do that. If you
1:27:20 want to use your fat human brain with
1:27:23 all your cholesterol getting eaten up in
1:27:25 there, fine, do that.
1:27:28 Or if you want to be lazy like me, have
1:27:31 Chat GPT write that [ __ ] and then go
1:27:33 make a video of it and you're like, "Oh,
1:27:35 that sucks." And you go look at your
1:27:37 JSON and you're like, "Oh, that was
1:27:39 crappy prompt. I could have looked at
1:27:41 that. That would have saved me a nickel
1:27:43 or 75 cents or whatever the hell these
1:27:45 things cost these days. Too much. Too
1:27:48 much, I'll tell you.
1:27:53 Oh, look. We have our We have our dude.
1:27:56 So, let me change my sharing, shall I?
1:28:00 [Music]
1:28:05 All right. Fantastic, Bob. Here we go.
1:28:09 Let's see how disappointing we are. This
1:28:12 is JSON at its finest.
1:28:14 >> This is the end, but not for me.
1:28:22 >> This is the end, but not for me.
1:28:30 This is the end.
1:28:31 >> Not bad. Not great. That little
1:28:34 transition at the end is weird, but you
1:28:35 know, not bad. All right, here's the
1:28:38 next one.
1:28:42 >> Where am I?
1:28:43 >> You have been reborn, but your journey
1:28:45 has only just begun.
1:28:47 >> Where are we headed?
1:28:48 >> To where the light leads.
1:28:50 >> Where am I?
1:28:51 >> You have been reborn, but your journey
1:28:53 has only just begun.
1:28:54 >> Listen.
1:28:55 >> Where are we headed?
1:28:56 >> To where the light leads.
1:28:57 >> Listen, this is every cutscene in every
1:28:59 video game ever.
1:29:03 >> Where am I? You have been reborn, but
1:29:06 your journey has only just begun.
1:29:08 >> Where are we headed?
1:29:09 >> To where the light leads.
1:29:12 >> That's [ __ ] hilarious. All right,
1:29:13 let's go look at our our last one. I got
1:29:15 to change my Wait, did this work?
1:29:19 This generation violates policy. Why?
1:29:24 Why, you [ __ ]
1:29:33 I'm so tired of
1:29:36 safety controls. Can we be done with
1:29:38 safety controls?
1:29:40 Let me just make violent [ __ ]
1:29:48 We'll see if it makes it.
1:29:53 There's Grock for that. Except Grock
1:29:55 doesn't do video yet, although I think
1:29:56 they're going to soon.
1:30:05 All right.
1:30:06 Was he taking a nap? Yeah, I think he
1:30:08 was taking a nap. Tom Arnold,
1:30:12 that was pretty funny. All right, let's
1:30:13 see if this one renders this time. I
1:30:15 think it might. We'll see.
1:30:22 So, is JSON making sense yet? No, it it
1:30:25 just killed it. H, unbelievable.
1:30:29 Um, all right. Let's go back to chat
1:30:30 GPT. Let me change my sharing again.
1:30:32 Let's go see if we can make chat GPT not
1:30:34 give us something in
1:30:37 Hey Danielle, what do you what do you
1:30:38 say to it? Don't give it that antique
1:30:41 finish
1:30:45 chat. Um, make me a wide image of
1:30:52 um,
1:30:56 I asked Chat GPT to create my image
1:30:58 without without the antique. a wide
1:31:00 image of
1:31:02 uh a claymation.
1:31:09 Claymation.
1:31:11 Um, jogger
1:31:16 running
1:31:18 past a hot dog stand in New York City
1:31:24 with a
1:31:27 squirrel
1:31:28 peering. How do you spell squirrel? Q
1:31:32 U I R E L.
1:31:36 Peering. Peering. Shut up. Pier shut up.
1:31:42 Do you know how hard it is to type in
1:31:43 public people down at him?
1:31:47 Um,
1:31:50 comma without
1:31:53 uh let's see uh
1:31:58 remove the antique
1:32:03 finish.
1:32:10 from the image.
1:32:13 All right, let's see if it works.
1:32:16 Let's see if it works, people.
1:32:19 Carrying on. Carrying on. Oh, is Tom
1:32:23 Arnold for me? Hey, Sarge. What's
1:32:26 happening? I get it. That's a troll. H
1:32:30 Tom Arnold, I've heard. Fat Alec
1:32:32 Baldwin, I've heard. Let me let me help
1:32:34 you out here. Uh you could also call me
1:32:37 uh Rosie O'Donnell on a bad day.
1:32:45 Uh Philip Seymour Hoffman. I've gotten
1:32:48 uh Tom Arnold. Uh who's the other one?
1:32:50 There's there's another one that's kind
1:32:51 of Tom Tom Arnold like uh
1:32:54 was it the the family guy dude? Family
1:32:57 something something family modern
1:33:00 family. I don't know that guy. Anyway,
1:33:03 yeah. So, welcome,
1:33:07 welcome, troll away. My request with
1:33:10 trolling is
1:33:13 the best trolls feel like a compliment
1:33:15 on the surface and then you think about
1:33:18 them and they they hurt. They they dig a
1:33:20 little deeper. So,
1:33:23 you know, get creative.
1:33:27 All right.
1:33:43 Hey, March.
1:33:45 [Laughter]
1:33:50 Oh, this does look less yellow, doesn't
1:33:52 it,
1:33:55 Danielle? I think you're right.
1:33:59 And you know what we'll do? We'll take
1:34:00 this image and we'll go uh
1:34:06 we'll go to VO3 and we'll turn it into
1:34:08 something.
1:34:11 But look, it doesn't have the antique
1:34:13 finish on it.
1:34:15 But like seriously, like seriously,
1:34:18 there there's no one at Open AI. They're
1:34:21 they're a 14 trillion dollar company
1:34:24 now.
1:34:26 They didn't hire one
1:34:29 visual arts person. They didn't hire one
1:34:32 liberal arts person to come in there and
1:34:35 say, "Uh, hey, how come everything looks
1:34:38 depressing here? Look at this. This is
1:34:41 so much better." They could they could
1:34:44 literally put it in the system prompt.
1:34:47 In fact, if you want your chat GPT
1:34:49 images to not look shitty, go into your
1:34:51 system prompt and update this. We can go
1:34:54 do that right now. It's on the backlog.
1:34:56 Exactly. All right. Uh let me How am I
1:35:00 going to do this? I'll download this
1:35:02 image and then we're going to go over to
1:35:04 VO3.
1:35:06 There's our sleeping prince.
1:35:10 We're going to go frames to video.
1:35:13 We're going to upload our image
1:35:16 that we just make, excuse me, just made.
1:35:21 Lovely.
1:35:43 Okay.
1:35:50 Oh, by the way, remember when I was
1:35:52 bitching before about Chat GPT not doing
1:35:54 things in 16 by9? This is why. See, I
1:35:58 have to figure out how how do I cut off
1:36:00 his feet or do I cut off the squirrel's
1:36:02 tail?
1:36:06 Chat GPT should figure this [ __ ] out.
1:36:09 You're supposed to be the state of the
1:36:11 art model. Start acting like it,
1:36:14 [ __ ] All right,
1:36:23 sleeping prince. That's pretty funny.
1:36:26 Okay. So, we're gonna say uh man jogs by
1:36:34 as vendor
1:36:37 says
1:36:42 hey. Says
1:36:46 hey.
1:36:51 Hey,
1:36:53 you feeling snacky?
1:36:58 Question mark.
1:37:02 And the squirrel
1:37:06 says,
1:37:08 "Sure,
1:37:12 I'll take some peanuts."
1:37:17 They both laugh.
1:37:19 [Laughter]
1:37:23 Uh oh. It can't handle caps for
1:37:26 inflection. I don't think I did caps.
1:37:30 All right, let's see how this does.
1:37:34 Oh, okay.
1:37:39 Demand your $10 back, Kyle. Listen, it's
1:37:42 $20.
1:37:46 It's $20.
1:37:49 my uh my chat GPT subscription.
1:37:53 Oh my god, I'm getting sleepy, people.
1:37:55 Oh yeah, it's getting to be time. What
1:37:56 are we doing? What are you guys still
1:37:58 doing up?
1:38:04 Okay, while that's rendering, let me go
1:38:07 pop this thing up.
1:38:12 Next week, next week, back to basics.
1:38:17 We're doing a fiveday
1:38:19 AI deep AI crash course. Okay.
1:38:25 It starts next Monday the 28th. We do
1:38:28 half an hour on Tik Tok and then we do
1:38:30 an hour and a half on YouTube. All all
1:38:33 next week.
1:38:36 All right. I want you to tell people
1:38:38 about it. Share it with people. Okay.
1:38:42 All right.
1:38:43 Okay. We'll we'll have QR codes and
1:38:46 stuff tomorrow or sometime this week.
1:38:51 All right.
1:38:53 Okay.
1:38:55 Okay.
1:39:04 [Music]
1:39:07 All right. Here we go. I think we have
1:39:09 video. We have jogger. We have jogger.
1:39:12 We have jogger and squirrel. Jogger and
1:39:14 Squirrel.
1:39:15 You ready people? It's going to be so
1:39:18 good.
1:39:19 It's probably going to be disappointing.
1:39:21 Although, I got to say V3 things are
1:39:23 less disappointing than I usually
1:39:25 expect.
1:39:27 Hey, you feeling snacky?
1:39:31 Sure, I'll take some peanuts.
1:39:35 Hey, you feeling snacky?
1:39:38 [Music]
1:39:39 Sure, I'll take some peanuts.
1:39:43 Hey, you feeling snacky?
1:39:46 [Music]
1:39:48 >> Sure, I'll take some peanuts.
1:39:52 Hey, you feeling snacky?
1:39:54 [Music]
1:39:56 Sure, I'll take some peanuts.
1:39:58 Listen, it's not Nickelodeon quality,
1:40:01 but you know, it's close.
1:40:10 Oh my god, that's bad. That's That's
1:40:14 bad. But you know,
1:40:17 you know you know how long that would
1:40:18 have taken you to do if you did it as
1:40:20 actual claimation.
1:40:23 >> Hey, you feeling snacky?
1:40:27 Sure, I'll take some peanuts.
1:40:31 Hey you.
1:40:35 Oh my god, it's giving robot chicken.
1:40:40 It's pretty bad.
1:40:44 All right, good people. So, um,
1:40:49 what was the homework I gave over the
1:40:51 weekend? I think the homework I gave
1:40:52 over the weekend was play with, uh, play
1:40:54 with Manis and Jen Spark. Hopefully, you
1:40:56 did that. I think someone said earlier
1:40:58 they they played with GenSpark and it
1:41:00 was disappointing. Um,
1:41:02 all of these agentic tools right now are
1:41:05 disappointing. All of them. All of them.
1:41:09 They hallucinate. They're too slow.
1:41:11 They're all sorts of [ __ ] up. So,
1:41:18 okay. Danielle just got access to Pika's
1:41:20 No Thing. I think I may have access to
1:41:23 it, but I don't know what it looks like.
1:41:24 Does it just look like you can scroll
1:41:26 like Tik Tok? If that's it,
1:41:29 I think I have it.
1:41:37 All right, before I take off, there's
1:41:38 apparently a few fun things in
1:41:40 irregulars. So, let's go take a look at
1:41:42 irregulars, people.
1:41:45 Let's go do this.
1:41:48 If anyone put things in irregulars that
1:41:51 has sound, I'm going to have to change
1:41:52 my tabs again and that's going to be a
1:41:53 pain in the ass.
1:41:58 Oh, it's a whole different app per
1:42:00 Danielle. I don't really remember what
1:42:02 the other app looked like, but I'll I'll
1:42:04 go see if I have access to it because I
1:42:06 think I have Pika Pro. I think I'm in
1:42:08 the Pika um creators program, but I'm
1:42:11 not sure.
1:42:16 All right, I'll go take a look.
1:42:19 You were going to a regulars. Do you see
1:42:21 what Brandon does there? He he he talks
1:42:24 to me as if I didn't remember. He
1:42:27 doesn't realize I've got genius level
1:42:29 large language models just sitting right
1:42:31 up here in my brain. He's like he like I
1:42:35 I don't need to be reminded because I
1:42:37 because I have my own large language
1:42:39 model. You have too many tabs and not
1:42:42 enough context window. That's totally
1:42:44 it. I I am My my tabs are a reflection
1:42:47 of my mental state. Oh, see, we do we we
1:42:50 do Vicki at work today. We do have to go
1:42:53 switch tabs because Vicki put up a
1:42:55 video. I told you this was going to
1:42:57 happen, people.
1:42:59 You don't think I know what's coming? I
1:43:01 know what's coming. Oh, and then this is
1:43:03 going to pop it into a different window,
1:43:06 isn't it?
1:43:09 Maybe. Maybe not.
1:43:14 Vicki at work today. No audio. Okay,
1:43:16 good.
1:43:19 That's great. I love the I love the
1:43:21 style of that one, Vicki. That's very
1:43:23 cool.
1:43:24 Back to basics. Oh, nice. What's this?
1:43:27 Steo.
1:43:29 Nice.
1:43:30 I like it. Come on, guys. You had to
1:43:32 ask.
1:43:37 Very cool.
1:43:41 Oh, you made it 16 by9. Oh, that's good.
1:43:44 Yeah, exactly.
1:43:47 That's pretty funny. Uh uh uh uh. Pico
1:43:50 is invite only for now.
1:43:55 I think I deleted mine.
1:44:00 Samma speaks about the future.
1:44:03 Just another manic Monday.
1:44:06 [Music]
1:44:13 I'm working at work until midnight
1:44:15 today, tomorrow, and Wednesday. Vicki,
1:44:18 oh my god, I'm so sorry. I'm so glad
1:44:21 you're here. Screw them. Hang out in the
1:44:24 AI learning lab. Tell them it's a
1:44:26 professional development. Oh, nice. Oh,
1:44:30 all right. I'll take this one. Nice,
1:44:31 Steo. That's good.
1:44:37 Oh my god, there's Champy. Very nice.
1:44:40 All right, cool. Wait, go back up to
1:44:43 digital gods post where in irregulars.
1:44:50 I don't I do not see that. I scrolled
1:44:52 past it.
1:44:58 Lord digital gods.
1:45:01 He was a god. He was a digital god.
1:45:05 Oh, story transformer. What is this?
1:45:09 >> Hey everyone, it's Lord Digital Gods,
1:45:12 also known as Glenn. Here to uh show you
1:45:14 something I've been working on. This is
1:45:16 done almost entirely by Vibe Coding.
1:45:20 It's been built on Manis and Gro 4. Uh
1:45:24 the nice details been very very happy
1:45:26 with uh Gro 4 by the way. Unbelievable.
1:45:29 Unbelievable model. All right. So the
1:45:31 purpose of this uh application is to
1:45:33 create kids books from any story, adult
1:45:36 stories or
1:45:38 >> beautiful. Okay. So So that's 18 minutes
1:45:40 long. We're we're not going to watch the
1:45:41 whole thing there. Long video.
1:45:44 Here's the thing.
1:45:46 This is a really good example. Hang on.
1:45:48 Let me let me flip my
1:45:51 So, so where that is, by the way, if
1:45:53 you're wondering where that is, um, if
1:45:55 you go to the AI salon, if you're if
1:45:58 you're a member of the AI salon, go to
1:46:00 the irregulars channel, scroll down a
1:46:02 little bit, you'll see this post from
1:46:04 Lord Digital Gods. So, not only has he
1:46:07 created a really cool application that's
1:46:09 going to generate a kids book, he's
1:46:11 created this video explaining how to use
1:46:13 it. One of the things we talk about in
1:46:15 the AI salon all the time is generously
1:46:18 lead or learn out loud. Right? So as
1:46:21 you're building stuff, share it with
1:46:23 other people and share how you built it,
1:46:25 how you're learning it. Notice in the
1:46:27 beginning of that he said, "Hey, I built
1:46:29 this app. Most of this is was done with
1:46:31 vibe coding and I've been really happy
1:46:33 with Grock 4." So there's like already
1:46:35 goodies in there that you can learn
1:46:37 from. This is one of the best ways to
1:46:40 get good at AI and get known in AI as
1:46:43 someone who's up to something is build
1:46:46 stuff and share what you're learning.
1:46:49 Generously lead, right? Share it in the
1:46:52 AI salon, share it on LinkedIn, share it
1:46:54 on Twitter, share it, you know, at your
1:46:56 dinner party with your friends who hate
1:46:58 you talking about AI. All right,
1:47:01 beautiful. Love it. Okay. Uh, I'm going
1:47:04 to get out of here.
1:47:06 Tomorrow night's Tuesday. I think it's a
1:47:09 normal night tomorrow night. Yeah, it
1:47:10 is. It is. Right. Yeah. Yeah. Yeah.
1:47:12 Yeah. Yeah. Yeah. Yeah. Yeah. Yeah.
1:47:14 Yeah. All right. Um
1:47:18 I know the JSON stuff is kind of geeky,
1:47:21 but just know it's just a different way
1:47:23 to structure a prompt. And you don't
1:47:25 have to learn JSON. Chat GPT already
1:47:27 knows it. So does Gemini. So does Grock.
1:47:31 Every all of the things know what JSON
1:47:32 is. So if you want help building a JSON
1:47:35 prompt, just say what are the elements I
1:47:37 want in my prompt and tell it to make a
1:47:39 JSON framework out of that. And then you
1:47:41 have structured prompting things. If
1:47:43 you're doing it in midjourney, don't do
1:47:45 that.
1:47:47 Do other [ __ ]
1:47:50 All right. Fantastic everybody. Uh what
1:47:54 time central time tomorrow night?
1:47:55 Central time tomorrow night it would be
1:47:57 9:00. So I start 8:00 p.m. Mountain time
1:48:00 unless otherwise specified.
1:48:02 Next week is the back to basics uh AI
1:48:07 crash course. Five nights covering
1:48:11 basically all the bases of AI to get you
1:48:13 started. All right. Alrighty. Okay.
1:48:19 Okay. Is everybody ready?
1:48:23 All right.
1:48:25 Okay.
1:48:27 Hey, I wonder if we should do a vibe
1:48:29 coding night. Brandon,
1:48:31 right now we've got a Q&A night. My
1:48:34 problem with vibe coding is it just
1:48:36 might not be interesting to look at. I
1:48:39 don't know. We'll talk about it. You
1:48:42 answered your own question. Yeah,
1:48:43 exactly. All right, fine.
1:48:46 Okay, everybody, have a good night and I
1:48:48 will see you tomorrow.