8/28/2025 - The Power of the Creative Generalist in the Age of AI

Description

Chapters

Transcript

0:13 [Music]
0:16 Every time I say it now, get that off
0:19 your mind.
0:22 Every time I see your mouth,
0:25 I hear that smile.
0:29 Early misty morning light heard the
0:33 engine turning the old fall outside.
0:40 [Music]
0:42 You would even me
0:45 again today.
0:48 You will convince me
0:51 again today.
0:54 You're leaving this hot tail looking for
0:58 someone else's cold and ring.
1:01 [Music]
1:03 should have said
1:05 so long, Suzan.
1:09 [Music]
1:11 Hush now. Don't you cry.
1:14 [Music]
1:16 So long as she was head,
1:21 [Music]
1:22 don't you cry for me.
1:25 [Music]
1:31 Sharing jeans and cigarettes and keeping
1:33 warm out on the road.
1:36 Chasing down a lifestyle out on Highway
1:40 24.
1:42 New York states a rolling breeze and the
1:45 sunshine with a blue sky falling
1:48 chill on September dreaming.
1:55 You believe in me
1:58 again today.
2:01 You will convince me
2:04 again today.
2:07 Leaving this hard time looking for
2:10 someone else's golden ring.
2:15 to say so long Suz.
2:21 [Music]
2:49 Oh, that champ Shannon. That Champ
2:51 Shannon. He's singing so good.
2:55 [Music]
3:27 [Music]
3:31 Good morning.
3:38 Yay.
3:42 Woohoo!
3:46 [Music]
3:53 [Music]
4:07 Hey,
4:10 [Music]
4:17 hey, hey.
4:19 [Music]
4:31 Heat. Heat.
4:32 [Music]
4:38 [Music]
4:46 [Music]
5:12 All right, it is a Th Thursday th
5:16 Thursday.
5:21 Is it Thursday?
5:23 Get it? Like you're thirsty.
5:26 It's Thursday.
5:30 I feel like I'm so clever sometimes.
5:38 [Music]
5:54 Encourage your viewers to share your
5:56 live. Hey viewers, I encourage you to
5:59 share my live.
6:04 I wouldn't say that if Tik Tok hadn't
6:06 suggested it.
6:08 They really did encourage. See Danielle
6:11 sharing away. She's been sharing away
6:13 the whole time. All you other slackers,
6:15 you're on there like, "Where was I don't
6:16 know where the share button is?
6:19 I ain't seen the share button. Is there
6:21 a share button somewhere?
6:24 My eyes aren't so good anymore,
6:26 you know. But I'm not judging you. I'm
6:29 not a little just a little judgment.
6:33 Just small pico amounts of judgment.
6:37 Nano. Nano banana amounts of judgment.
6:43 [Music]
7:04 [Music]
7:14 Oh, good lordy good people. God lordy
7:18 good people.
7:21 [Music]
7:30 The boys were singing shingling
7:33 the summer that we met.
7:38 You were 10 and 17.
7:41 How could I forget?
7:45 All the stars from near and far were
7:48 watching from above.
7:52 too little
7:54 da fall in love. I don't know what those
7:58 words were.
8:00 The way we danced was not a dance but
8:04 more a long embrace.
8:07 We held on to each other and floated
8:11 there in space.
8:14 I was shy to kiss you while the whole
8:18 wide world could see. So shingle said
8:23 everything to me
8:28 and all the poor old old folks. They
8:32 thought we'd lost our minds.
8:35 They could not make heads or tails of
8:38 the young folks. Funny rhymes.
8:42 You and I learned all the words and we
8:45 always shang along
8:48 to old shaming dong ding. Shamboling
8:54 dang.
8:55 [Music]
9:11 That's no Jesse Winchester number.
9:17 Do I know any songs from the dog days of
9:19 summer? I do not. I need to learn some
9:22 new songs,
9:24 as anyone who watches this channel
9:26 regularly knows.
9:29 I will occasionally remember one like
9:31 that one.
9:36 came on him like slow moving cold
9:40 fronts.
9:44 His beer was warmer than a look in her
9:47 eyes.
9:50 [Music]
9:53 Sat on a stool and he said, "What do you
9:55 want?"
9:57 [Music]
9:59 She said, "Give me a love that don't
10:02 freeze up inside.
10:04 [Music]
10:06 [Applause]
10:07 [Music]
10:12 He said, "I have melted some hearts in
10:15 my time, dear.
10:19 But to sit next to you, well, I shiver
10:22 and shake.
10:27 And if I knew love, well, I don't think
10:30 I'd be here.
10:34 Asking myself if I've got what it takes.
10:40 [Music]
10:41 A two. A two. Your icy blue heart.
10:50 Should I start?
10:53 Turn what's been frozen for years
11:00 into a river of tears.
11:05 [Music]
11:14 Oh man. All right, let's get going.
11:17 Let's do something. Let's do something
11:18 tonight. I know. We'll do Nano Banana.
11:22 Okay. I just made a video. So, I think I
11:25 think I'm in some sort of Tik Tok [ __ ]
11:27 list. I don't I don't know what's going
11:28 on. I I guess they've got some new
11:30 algorithm and they don't like my [ __ ]
11:32 Um,
11:34 but I put out a video tonight called I'm
11:36 an idiot.
11:39 Maybe they didn't like me calling myself
11:41 an idiot, but I called you idiots, too.
11:44 Um,
11:47 one of the things that's been striking
11:48 me the past couple of nights, few
11:51 nights, three nights, four nights,
11:54 is as powerful as the new Gemini image
11:58 generation model is, I don't know how to
12:00 use it.
12:03 Like, I'm getting really inconsistent
12:05 results. Sometimes I get results that
12:08 are like mindblowingly good. Sometimes I
12:10 get results that are like it didn't even
12:13 try to give me what I thought it would
12:15 give me.
12:17 Um, I feel similarly with chat GPT5
12:20 right now. Still, I feel like a clueless
12:22 newbie. Feel like an idiot. And I don't
12:25 listen, I idiot might be a harsh word. I
12:28 don't know where your uh sensitivity is
12:30 on the triggering list with the word
12:32 idiot, but but uh but um you know, just
12:37 feeling like a dumb dumb, feeling out of
12:39 it, like not connected, like don't have
12:42 a handle on it. And the point I made in
12:44 the video was if if every time a new
12:47 model gets released, you're like, "Oh
12:49 god, I just I just don't know what I'm
12:52 doing." It's not you. Like that's just I
12:55 think it's the natural
12:58 it's the natural shape of things as as
13:01 especially as we're living in an
13:03 exponential technology development
13:05 curve. These tools are going to get
13:08 better and better faster and faster and
13:10 our ability to keep up with how to use
13:12 them is just it's going to be bad. So
13:15 give yourself some grace. But I figured
13:17 we'd play with that some more.
13:20 We love anyway. Mr. He loves us anyway.
13:23 Even though we're clueless.
13:25 Even though we're clueless. Dude, give
13:28 yourself a break. How many irons do you
13:29 have in the fire
13:31 right now? About 462.
13:35 I hear you. I'm overwhelmed with the
13:37 onslaught of AIS.
13:39 Hi. Sensitivity is high. Yeah, exactly.
13:42 Sounds like midjourney. I know. Like
13:45 even midjourney, just keeping up with
13:47 midjourney. just keeping up with what
13:50 can it do and how do you actually use
13:52 that here. Let me show you another let
13:55 me show you something right now. Am I
13:57 sharing this tab? Share this tab
13:59 instead.
14:03 That's not it. That's not it. That's it.
14:05 Okay.
14:07 Wait, what did I do? Oh. Oh, producer
14:11 Brandon and I were were dual clicking.
14:14 Um,
14:17 so this is Korea's new real time video
14:20 model which you can get on the wait list
14:22 for
14:24 um,
14:31 but do you remember do you remember a
14:34 year and a half ago when Korea Realtime
14:36 Image came out and I was making realtime
14:39 images and then it made a painting of
14:41 boobies and I got perma banned from Tik
14:43 Tok.
14:45 Um, this is like that only worse cuz
14:48 this is video.
14:52 What do you want, champ? You want to get
14:53 up on the bed?
15:17 So,
15:19 so a couple of thoughts on this.
15:23 My first thought is, well, what the [ __ ]
15:24 would you use this for? Right?
15:29 a thing that you could use it for, which
15:31 would actually be pretty cool,
15:33 is real-time performances.
15:37 So, imagine like a theater improv group.
15:40 Oh, great. Another wait list that I'll
15:42 forget I'm even on. So, I don't check
15:44 until 3 months after find out I was let
15:46 in. I know. That's totally what it is.
15:49 You know what's funny, Tom? I was just
15:50 thinking tonight,
15:52 um,
15:54 I went and looked at my at my Discord,
15:57 my list of Discord servers where I've
16:00 got like the official Discord server of
16:03 the thing. I've got some creative
16:04 partner program versions. I've got
16:08 I don't know
16:10 30 different discord servers for all for
16:13 stuff I'm interested in or am in the
16:17 creative partner program of it
16:21 absolutely reminds me of when I was
16:23 following NFT projects like it's like oh
16:27 my god and and they're just relentless
16:31 and I and I feel like it's the same way
16:33 with these things like they just keep
16:34 you know releasing new things. So, you
16:38 know, anyway,
16:40 give all yourselves some grace. I'll
16:42 give myself some grace.
16:45 30 lightweight. Yeah, I know it. It's
16:47 probably more than that, Vicki, but I'm
16:49 trying to not seem like a total
16:51 sociopath.
16:55 Um,
16:58 you know what? Let's um
17:01 let's cruise through.
17:07 Yeah.
17:13 Let's just cruise through X. I'll be the
17:16 sociopath here. That's just your Discord
17:18 horde. Um let's just cruise through X.
17:21 So, so one of the things people often
17:22 ask me is how do I keep up with keep up
17:25 with stuff? Um, one is just really like
17:29 who who I'm who I'm following. So, one
17:32 of the ways that I'll do this is if I
17:34 find someone I like, like if I like this
17:37 guy Mark uh Cretchman,
17:41 one of the things I'll do, like he's got
17:43 14,000 followers, but he's only
17:45 following 493 people. Let me make this a
17:48 little bigger.
17:50 He's only following 493 people. So you
17:53 can click on that and you can now go see
17:55 the people that he follows.
17:58 And
18:00 sometimes you'll find people in here
18:02 that you're like, "Oh, I should follow
18:03 them." So So one of the ways I tweak my
18:06 feed is is to follow people
18:10 who who I like what they do and I want
18:13 to find out who they're following,
18:14 especially if their ratio is really low
18:16 like that. Like I think my ratio is
18:18 stupid because I followed a bunch of
18:21 people dur during the NFT days.
18:24 Um
18:27 yeah, my my my ratio is I've got 3,500
18:30 followers. I'm following 2,000. That's
18:33 useless. Like my list my list is
18:35 useless.
18:37 Um although I suppose if you look at my
18:40 most recent ones, my most recent I don't
18:42 know 200, those are probably decent
18:45 ones.
18:47 Um, OpenAI today did drop um a new
18:52 version of speech to speech for
18:53 developers. Um, producer Brandon also
18:57 posted on the Irregulars channel in the
19:00 AI salon um that his his voice in chatbt
19:04 is glitching a lot. It sounds like it's
19:06 had helium. It's all crackly and and
19:09 sounds like compression artifacts. Um,
19:12 I've had that when I've had low
19:13 bandwidth, but he said he's on really
19:15 high bandwidth. So, this could be a
19:17 thing where OpenAI is throttling access
19:19 because they have too many too many
19:21 concurrent users. I don't know. That's
19:23 total speculation.
19:26 Um, I'm starting to see a lot of um,
19:32 you know those annoying videos like I I
19:35 I've been noticing some recently that
19:37 are like um, moments before disaster 10
19:41 wild examples and it's video of like the
19:44 the one that I've seen is a it's like a
19:47 mall with a giant fish tank and the the
19:49 wall falls out of the fish tank and
19:51 smashes a bunch of people. It's total AI
19:54 [ __ ]
19:55 So, I'm but I'm starting to see a lot of
19:57 those, you know, shitty shitty accounts
20:01 get even shittier. Uh, so that's a thing
20:04 I'm noticing on on X right now.
20:07 Um,
20:09 CES
20:11 is is coming. People are talking about
20:13 that again. I don't know why planes are
20:16 crashing.
20:25 [Music]
20:26 Oh, here's a fake a fake doll. This
20:29 isn't creepy.
20:32 Good luck sleeping.
20:40 Oh my god.
20:43 >> Hey, Kyle.
20:44 >> Yes.
20:45 >> Just a quick question for the X
20:47 uneducated. You've got a tab there that
20:50 says generative AI and another one that
20:52 says AI stuff. How did you create those?
20:55 Because I've tried and not been
20:57 successful. And I could Google it, but
20:59 you're right here. So,
21:02 >> I I don't remember. I think
21:06 I think
21:10 I go to
21:15 >> I thought they were lists, but they're
21:17 definitely not lists.
21:18 >> They're lists. They are lists. your
21:22 lists.
21:26 Let me just make sure. Yeah, here's my
21:27 generative AI list. See how
21:29 >> you pinned that?
21:30 >> I've pinned it.
21:32 Okay, so I definitely know how to do
21:34 this now. Thank you, Brandon, for for
21:37 reminding me to remind myself how to do
21:39 this. So, you go to your three dots and
21:43 then you go to lists and then you can
21:45 make a list. And basically when you make
21:47 a list, you just add a bunch of people
21:49 that you're following or you can like
21:52 like uh Robert Scoble, if you follow
21:55 him, he's got amazing lists of like
21:58 2,000 people. So you can just go grab
22:00 their lists and add them to your lists.
22:03 So that's that's a way that I'll do
22:04 this. Like AI artists, this is this is
22:08 not a list I made,
22:10 but this AI artist is the one that has
22:13 18 members in it. I made the one that
22:15 has 3,500 members in it. I didn't. That
22:17 might be Robert Scobble's list.
22:20 Um, yeah. So, here, AI consultants by
22:23 Robert Scoble. And I've got that pin.
22:26 And so, then when I'm when I'm here, I
22:29 can just like one of What's weird about
22:31 the for you page lately is that it seems
22:35 to have during certain hours of the day,
22:38 it's getting different. Normally, what
22:40 I'll get is mostly AI stuff.
22:44 But it seems like at 3:00 or 4 in the
22:46 afternoon, I get like it's just all
22:48 videos and it's all videos of stupid
22:51 [ __ ] Like stupid [ __ ] on Tik Tok.
22:54 People falling, tripping, you know,
22:56 smacking their head on garage doors,
22:59 burping after they take the Sprite
23:01 challenge, you know, like Tik Tok.
23:05 And so if it gets like that, then I'll
23:07 flip over to one of these lists.
23:10 Um,
23:13 yeah. So, that's what's that's what's
23:14 there.
23:16 And then the other thing I do is,
23:19 um, a lot of times you'll see headlines
23:22 that kind of
23:25 catch you because they're cleverly
23:27 written.
23:32 Let's see if I can find one.
23:37 Grock code was officially released.
23:58 But anyway, I don't I can't find one
24:00 now. Um,
24:03 just forward it to Manis. Like, follow
24:06 the tools. Like, if you like Manis or
24:09 GenSpark, follow the companies that make
24:11 them or their official pages.
24:14 That's how I'll find out there's
24:16 announcements. Um, pay attention to
24:18 people that you like.
24:21 And then the thing I do is if I if
24:23 someone I like posts something about
24:26 here's a thing coming up or here's a
24:28 rumor, then I'll just keep scrolling and
24:31 if if I don't see it come up again, then
24:33 I just assume that it's [ __ ] And if
24:37 but if I see something show up four or
24:39 five times from four or five different
24:40 people, then I have a pretty good sense
24:42 of what's happening.
24:45 Uh, one other thing while we're on X,
24:48 uh, we have, um, a DM from me to you for
24:52 Grock jumping into the vibe coding space
24:55 for free.
25:01 >> Strong model, but it's a strong model.
25:03 Do is there a place where you can
25:05 actually just go play with it?
25:07 >> Uh, here say show the expanded out.
25:15 because I noticed it was on
25:18 like some vibe coding tools like wind
25:21 wind whatever wind surf or wind sock or
25:24 whatever the [ __ ] that thing is.
25:26 >> Yeah, but it but free uh is always a
25:32 it's on co-pilot. I'm not sure how you
25:34 get to it in Copilot, but
25:36 >> yeah, cursor co-pilot. Yeah, it's it
25:39 it's living. I mean, if if you use
25:42 cursor or copilot or things like that,
25:44 then I would go play with it. It's a new
25:46 coding model from Grock, it looks good,
25:48 but I didn't see anything where it's
25:50 like easy to just go play with it. Um
25:53 like like clawed or or lovable, uh
25:56 things like that. So, um it might be
25:59 really good. So, okay, cool. Beautiful.
26:11 There's a lot of Nano Banana
26:14 uh prompts. People are liking it.
26:19 It ain't perfect. Nano Banana,
26:23 if you don't know what Nano Banana is,
26:25 it's Google's new image generation tool
26:28 in Gemini 2.5 Flash.
26:32 There's another weird thing like you've
26:34 got 2.5 Pro which has Imagine in it and
26:38 then you've got 2.5 Flash which which is
26:40 like a smaller less powerful version of
26:43 2.5 Pro but it's got a more powerful
26:46 image generator in it. How are you
26:48 supposed to keep track of that?
26:51 You're you you can't you're not you
26:53 can't it's not you you're not the idiot.
27:01 I'll tell you what, here's a billion
27:02 dollar opportunity. Make an AI to make
27:05 sense of all this stuff to just track it
27:07 to just be able to just go um you know I
27:11 need something that does this and it's
27:12 always got the latest one and you don't
27:14 have to link to it. It's just right
27:15 there. Here you go. Start prompting. Go.
27:19 That'd be nice. Someone go build that.
27:25 I've been working on visual poems.
27:37 If AI became unavailable today, would
27:39 you still continue building?
27:45 Interesting.
28:01 XI is cooking clever and creative ways.
28:05 Oh, aren't they doing something new with
28:07 uh
28:10 Oh,
28:12 there was a post from Notebook LM where
28:15 they're going to put um
28:18 they're going to put the new image
28:20 generation model into Notebook LM in
28:23 their video generation tool. So, you'll
28:25 be able to prompt their video generation
28:27 tool to do to do things
28:31 in a more predictable way with probably
28:34 better image generation.
28:37 I hate it when I'm led in and pay the
28:39 sub for 6 months and then do absolutely
28:42 nothing with the tool. That's
28:45 archetypal. That's that that's
28:53 I would say that 90% of the tools that I
28:56 pay for if I were to divide the monthly
28:59 fee by the number of prompts that I give
29:01 it in a month on average,
29:04 it's probably about 50% of the monthly
29:07 fee. So So I'm paying $10 a prompt for
29:11 most of the tools that I that that I
29:13 use.
29:19 I have completed not one single thing in
29:22 manners. Well, okay. So, here's the
29:24 thing. So, I said this last night, but
29:27 I'm going to say it again because I I
29:29 want to find a way to work this into
29:34 into my more organically into my
29:36 thinking. I saw a LinkedIn post
29:39 yesterday that said 10,000 prompts is
29:41 the new 10,000 hours.
29:44 And basically meaning like if you want
29:46 to get good at a tool, you just got to
29:49 [ __ ] use it, right? Play first. You
29:51 got to play with it. You got to play.
29:53 You got to push it. You got to put in
29:55 your 10,000 prompts. And
29:59 I think
30:00 the expectation is, oh, I should be able
30:03 to do that for lots of different tools.
30:05 I think what's happening is the tools
30:07 are getting sophisticated enough that
30:08 it's like it's kind of like pick a lane.
30:11 Pick a lane. Pick a couple of tools.
30:14 put in your 10,000 prompts and just just
30:17 get good at those things. I think that's
30:20 where it's where it's going because
30:22 otherwise you just end up with like like
30:24 like the like chat GPT at this point I
30:29 would say if you're starting from a
30:31 standing start with chat GPT
30:34 and you really want to master it, you
30:36 really want to understand all of the
30:39 settings and all of the personalities
30:42 and and custom prompting and custom GPTs
30:45 and codecs and uh canvas
30:49 and deep research and deep thinking, all
30:53 that [ __ ]
30:55 image generation, voice.
31:00 Like, if you did it eight hours a day
31:02 for a month,
31:04 you might
31:06 feel like you've got some decent mastery
31:08 of it, but it's probably going to be
31:11 more like three.
31:13 So, just kind of where we are right now.
31:15 The tools are a bit more sophisticated.
31:18 Um,
31:20 [Music]
31:28 do I have a good recommendation for a
31:29 HubSpot AI alternative? I don't, Mark, I
31:33 haven't done I haven't t done a ton of
31:35 pro programmatic marketing. Um, it's
31:38 something I want to get into. Um, Vicki
31:41 in here is a huge fan of Blot from
31:45 Sabrina Romangh. Um, I don't know that
31:49 it's apples to apples to HubSpot, but I
31:51 know it's an AI native tool. Um, it it's
31:55 not cheap, nor is HubSpot. Um,
32:00 yeah, I don't know. I I just I don't
32:03 have a good answer for you there,
32:04 unfortunately. Um, I just don't do
32:06 enough of it.
32:10 Yeah, Vicki says she loves Blot.
32:14 I get loads of use out of pseudorite.
32:18 I don't know what pseudorite is. Is that
32:21 an LLM?
32:31 All right. Um, I think that's true.
32:34 10,000 rule. My prompts are really
32:35 getting next level. Oh, that's cool.
32:38 GPT5 hates us. No, GPT5 doesn't hate us.
32:42 GPT5
32:45 thinks we hate it
32:47 because we haven't taken the time to
32:49 learn it.
32:51 You know what it is with GPT5? It's like
32:54 we went out on one date with it and
32:56 we're we're expecting it to it's like on
32:58 the second date we're like you want to
33:00 get married
33:02 and chat GBD5 is like whoa whoa whoa.
33:06 Why don't you learn to prompt me first?
33:08 You know, I've got different models and
33:10 I've got this routing model that will
33:12 make some choices for you. Why don't you
33:14 learn what's good about that before we
33:16 get hitched?
33:23 I've been using Brievo, but it's falling
33:26 short.
33:28 I mean, you know what's funny, Mark, on
33:30 on the on the marketing automation stuff
33:32 is
33:34 I I've never even heard of Brevo. Um
33:39 SEO is totally [ __ ] right now and
33:42 changing. You've got to you've got to do
33:44 AIO, right? So AI optimization. So how
33:47 do you get into search into you know
33:49 chat results? Um so that whole world's
33:52 changing. The marketing world is so
33:56 fragmented. Like if you just think about
33:58 short form video, Tik Tok, Instagram,
34:02 YouTube shorts, things like that. Like
34:04 where do you put your attention?
34:06 And you know the answer is like you put
34:08 your attention on all of them you dummy.
34:10 You just what you're going to do is
34:12 you're going to make one piece of
34:13 content and you're going to leverage
34:15 that content across multiple channels.
34:19 And that's what Vickiy's doing with
34:21 Blot. But it's like just getting your
34:24 head around where you're going to put
34:26 all this [ __ ] and what you're going to
34:28 put there and what your voice is. It's
34:30 it's a lot. And then and then you have
34:33 to create the content to fill that
34:35 machine,
34:38 right? It's this big circular [ __ ]
34:40 nightmare. Just moved from a sauna to
34:43 ClickUp. Yeah. You're you're in a world
34:45 that I just I haven't I used to run one
34:48 of the biggest agencies in the world,
34:49 but we were more of a creative agency
34:51 than a media agency. And then I I got
34:54 out of that in 2002. So basically from
34:57 2010 on from all the social media stuff
34:59 on I I'm kind of clueless with with what
35:03 to do with this stuff now.
35:05 Vicki, it's not quite HubSpot but fills
35:08 some of that capacity. There you go.
35:12 That's been amazing having all the AI
35:14 integrations. I really just need a CRM
35:17 and an email tool. Really? Yeah. Yeah.
35:20 So, um,
35:25 good question. Keep coming back here and
35:29 when you figure it out, let us know.
35:30 Join the AI salon if you haven't.
35:32 Brandon, if you would pop the uh the
35:34 address for the AI salon up on screen.
35:36 Yeah, I've dropped my blot sub, too. I
35:39 just don't have anything that I want to
35:41 say that badly or that much. I know. I
35:44 I'm I'm the same way right now. Like
35:46 even though I put a fair amount of
35:48 content out there, I'm like, "Wait, what
35:49 am I going to talk about? What do I want
35:50 to talk about? What's important right
35:52 now?" And I don't know. I'm I'm I'm
35:55 still in this weird place where I don't
35:57 I don't have a clear point of view right
35:59 now. There you go. So, the AI salon, if
36:02 you haven't joined the AI salon, please
36:04 go do that. The salon.ai. If you go to
36:06 community.thesalon.ai,
36:08 that'll take you straight into our
36:10 community website. Sign up there.
36:12 Introduce yourself. Step two is
36:14 introduce yourself. just go introduce
36:16 yourself. Um, I need an AI that paints
36:20 my 3D models for me in Bamboo Studio.
36:25 I can't believe someone hasn't come up
36:27 with that by now.
36:30 That seems vibe codable, Vicki. Come on.
36:34 That Come on. You got to be able to vibe
36:36 code that. That's all just that's all
36:39 just geometry and and pixel colors.
36:44 and normal maps. I don't know enough
36:46 about 3D. I know they got these things
36:48 called normal maps and shaders.
36:51 [Laughter]
36:54 Oh, man. All right.
36:57 Is he reading this chat or is he just
36:59 plugging his other stream?
37:02 Uh, I was not reading this chat. What's
37:05 your question?
37:07 Oh, here we go. Do you feel AI has an
37:10 honest case
37:13 to replicate
37:15 human output like full replacement?
37:21 Does it have an honest case? I don't
37:22 quite understand the question.
37:26 What's your what's your name? Aha.
37:29 Ahaz a sh a sha a shush.
37:35 Um,
37:38 do do wait, how did you how did you
37:40 phrase the question?
37:42 Do I feel AI has an honest case
37:46 to replicate human output? Oh, do do I
37:50 feel that AI is going to get good enough
37:54 that it's going to be indistinguishable
37:56 from humans? Is that it? Is AI output a
37:59 replacement for human work? Oh, yes. I
38:04 not only do I think it has an honest
38:06 shot, I think
38:10 I think what's going to happen is this.
38:12 Right now, all of the AIS
38:16 are still bad enough.
38:19 Like Sam Alman talked about GPT4 as
38:22 being slightly embarrassing.
38:26 like he knows
38:28 that it hallucinating that much and it
38:31 being that
38:35 imprecise
38:37 is not acceptable, right? And he called
38:39 it slightly embarrassing. GPT5 is
38:42 better,
38:44 but
38:46 one of the things you'll notice about
38:47 GPT5 is that there's no clear
38:51 there's no clear consensus on is it
38:53 good, bad, or indifferent because all
38:56 sorts of different people are using it.
38:58 People that are using it for high-end
39:00 thinking stuff are saying it's
39:02 absolutely remarkable. people like me
39:05 that are living down in more mundane
39:08 things like, you know, marketing and
39:10 communications and storytelling.
39:12 It's just different. It's not better.
39:14 It's just different. And I don't know
39:15 how to prompt it yet. And so,
39:19 so all of the tools right now are for
39:23 the most part non-automated. The ones
39:25 that are automated are janky as [ __ ]
39:28 But I think what's going to happen is
39:30 this. I think what's going to happen is
39:33 someone's going to crack the code.
39:36 Yes.
39:38 >> Yeah. So there's I want to call it it's
39:40 an important distinction here that you
39:43 know for
39:44 you know the answer not to be an
39:46 immediate no on replacing humans. It it
39:50 will replace most human work in terms of
39:53 the work as we know it today. But I I
39:55 think if I if I know you well enough to
39:58 know that that's not going to replace
40:00 human drive and innovation and I want to
40:03 make sure we get to that point as well.
40:05 >> Yeah. Yeah. Yeah. No, I No, like I don't
40:08 No, I don't think it's going to I I'm
40:10 talking about human human jobs. Um will
40:14 will it replace human jobs? I think it
40:17 will it will replace human jobs of
40:19 today. I think it will enable new jobs
40:23 tomorrow that we can't even imagine
40:25 right now. That's what new technology
40:27 always does. It automates the things
40:30 that we know now. We can't imagine what
40:32 the future would be like without those
40:35 jobs. And the fear is always, oh well,
40:39 what are humans just going to sit around
40:40 and rot?
40:43 No, we're humans. We we seek to make
40:46 meaning. We seek to connect. We seek to
40:48 make a difference. So, we're still going
40:51 to do that. And I think I think what
40:53 happens is what AI does is it it
40:57 automates it replicates the tasks that
41:00 we do today.
41:03 But the tasks are not the important
41:05 part. The important part is what's the
41:07 idea like what's the idea of the
41:08 business that you work for? You know,
41:11 what's the idea of how you want to um
41:14 impact a customer? I think the human
41:16 role is going to elevate into well what
41:20 difference do I want to make in the
41:21 world and then the AIs are going to do
41:23 all the tasks.
41:25 So I think that's that's what shifts and
41:27 so I think our roles elevate or
41:31 you could choose not to live in that
41:33 more elevated highinking world and you
41:37 could just say well [ __ ] all the [ __ ]
41:39 all the tasks are taken care of the
41:43 thing I'm really excited about is LED
41:47 macra
41:49 and so I'm going to master LED macra and
41:53 at some point that's going to become a
41:54 job like stuff that we just like that's
41:57 not a job.
41:59 If anyone from 1895 saw what we do for a
42:02 living right now, they'd go like that's
42:04 not work. Work involves muscles and
42:07 grain.
42:08 [Laughter]
42:13 The tech intakes language and outputs
42:15 the average of that language, nothing
42:16 more. That's not true. That's not true.
42:21 That's not true. That is a trope from if
42:25 you look at how
42:28 a large language model
42:31 predicts the next possible token,
42:35 then what you say is true,
42:38 right? If you're only looking at it at
42:39 the token level,
42:42 right? You're saying we take all this
42:43 information, we embed it into
42:45 thousand-dimensional mathematical space,
42:47 we put all these tokens in in this
42:50 weirdass math space, and when you type a
42:53 prompt, it creates a probability of
42:55 what's the next most probable token. And
42:58 and you could argue that that's some
43:00 amalgamation of all the [ __ ]
43:04 But when you start to do things like
43:06 create chain of thought like the the
43:08 reasoning models, the thinking models
43:11 that are not just doing next token
43:13 prediction and if you start taking the
43:16 agents, the autonomous agents and then
43:18 if you also start taking
43:21 human contribution
43:23 where so like like one of the tropes of
43:26 AI and and the the one that you just
43:28 said is that it just gives you an
43:31 average answer. If you give it a shitty
43:34 prompt, it will give you a shitty
43:36 answer. So if you say, "Write me a
43:38 LinkedIn post about marketing," you will
43:41 get the amalgamation of all the LinkedIn
43:43 articles and every other article of
43:46 marketing, and it will be boring ass
43:49 repetitive dril.
43:52 But if you spend 15 minutes going,
43:54 "Okay, I've got a point of view on
43:56 marketing and I've got some ideas here
43:58 and I'm going to put my ideas into the
44:00 prompt and I'm going to see what it
44:01 gives me back and then I'm going to go,
44:02 "Ooh, I like that idea. I don't like
44:04 that idea." I I'm still actively
44:07 involved in the curation and refinement
44:11 of what the model's doing. So, it's not
44:13 a oneshot thing,
44:16 right? I think that's where you where
44:18 you're like what I would encourage you
44:20 to do is like use these things. Chain of
44:24 thought is still average. It's not
44:26 because it's not because you're you're
44:29 missing the human component here.
44:32 If if what you're asking is might an AI
44:35 be so good at some point in the future
44:38 that we don't need the human involved in
44:40 it, it might be.
44:42 But to say that you can't get high
44:45 quality original content out of large
44:48 language models is just not true. It's
44:52 just not true.
44:56 Um
45:00 it's like that for using AI. Let's see
45:03 want to write. I work with AI daily. I
45:07 use it constantly
45:09 as a staff dev. It does. Okay. But
45:14 it's not original.
45:17 Let Let me ask you something.
45:21 What is an original thought
45:24 out of a human?
45:28 How do our brains work?
45:31 How our brains work is we get some
45:34 inputs and those inputs go into what is
45:37 effectively a large language model.
45:40 that's had a bud a bunch of information
45:42 encoded in it, right? The original
45:46 source material ceases to exist, but
45:48 we've got, you know, synapses and
45:50 neurons that are firing off with
45:52 chemicals.
45:54 And when an input comes in, how do we
45:57 have an original thought? Well, we have
45:59 an original thought like this. We go,
46:00 "Huh, that kind of reminds me of Andy
46:03 Warhol. Oh, and that makes me think of
46:05 Tom Robbins, that author. And oh, that
46:08 thing my mom told me when I was eight.
46:11 And and we mash together those disperate
46:14 thoughts in this crazy straw of a
46:17 [ __ ] brain we have, and we go, "Huh,
46:20 here's a new idea."
46:23 Which is some weirdass combination of
46:26 the thoughts that got put in our head
46:27 over our lifetime. How is that [ __ ]
46:30 different than what a large language
46:31 model does?
46:34 If it's not generating original
46:36 thoughts, then you're not prompting it
46:37 in a way that's taking inputs from from,
46:42 you know, in congruous things and
46:44 mashing them together and putting them
46:46 out in such a way that you go, "Oh, that
46:48 actually seems interesting and
46:50 different." Two nights ago in here, I
46:53 was I was doing something. We were
46:55 talking about writing kids stories, and
46:57 just on a lark, I said, "I don't want a
47:00 kids story like every other kid story. I
47:02 want something that makes them, you
47:04 know, question thought and I I want
47:06 something that shakes up the
47:08 kindergarten. And what Chad GPT came up
47:11 with was really interesting and it was
47:14 new and novel from where I sat. Now, was
47:17 it mathematically new and novel? I don't
47:19 know. Maybe not.
47:22 But I don't see how
47:25 how our brains create an original
47:27 thought. I don't see how that's
47:28 different than a wellprompted large
47:30 language model.
47:34 So for what it's worth, okay, you're
47:38 talking about context size then it
47:40 cannot
47:42 replicate a human buffer size. I I
47:45 agree. I said at the beginning of this
47:48 these tools suck right now. The context
47:51 windows too small. The memory is too
47:53 imprecise. The hallucinations are not
47:55 predictable.
47:57 or or or manageable.
48:00 Although it looks like five has gotten
48:02 that much better and it looks like memor
48:05 is about to get much better. But I agree
48:08 with you. Like yes, human brains are
48:11 better than large language models right
48:13 now. But to say that a large language
48:15 model is not capable of generating an
48:17 original thought, I just don't agree
48:18 with it.
48:20 Um a million tokens cannot speak to
48:23 human experience. I'm not talking about
48:25 speaking to human experience.
48:31 It doesn't have the attention head space
48:33 to differentiate actual content.
48:36 No, but we do. But we do.
48:41 Here's what you're missing.
48:44 We're using these tools.
48:51 If you want the large language model to
48:53 suck less, then find more creative ways
48:57 to prompt it. Find different ways to
48:58 come come at things. Find different ways
49:01 to synthesize the world so that these
49:04 things are reflecting better answers
49:06 back to you.
49:08 But anyway,
49:10 like I I I think you and I are actually
49:12 agreeing. I get what you're saying and I
49:15 actually like like it. Not going to lie.
49:17 But but but here's here's my point. I'm
49:19 I'm not saying that you're wrong. I
49:20 think you're actually right
49:23 that from a
49:28 from a definitional place where large
49:31 language models are right now is like
49:34 child's play to the sophistication of
49:36 our brain. Absolutely agree.
49:40 My point is the tropes of things like it
49:43 can only do predictive outputs and it
49:45 can't it's only going to give you the
49:47 average of things is just not true. if
49:49 you've got a human collaborating with
49:52 that machine because then what h so okay
49:56 so so I'm writing a book right now
49:57 called feed your prompt and and the
49:59 whole thesis of the book is that you if
50:01 you feed your prompt you what the large
50:04 language model does is amplify your
50:06 ideas
50:08 and I think the the the the sort of area
50:12 that you're treading on is I think one
50:14 of the mistakes that people make and
50:16 it's it's one of the things that the
50:19 Silicon Valley bros have have
50:22 precipitated this idea that the AI is
50:25 the genius, right? They're like the AI
50:28 is going to have AGI and it's going to
50:30 be smarter than all of us. And so
50:31 they're saying the AI is the genius. And
50:33 so if you treat AI like this thing where
50:36 you give it a prompt and it's going to
50:38 give you a genius answer, it's horrible
50:40 at that. It doesn't give you a genius
50:42 answer. It gives you a shitty answer.
50:46 But if you shift the framework to say,
50:49 "No, no, no, no. The AI is not here to
50:52 replace me. The AI is here to augment
50:54 me. It's here to amplify my ideas."
50:57 Everything changes
50:59 because now you take what it's good at,
51:02 and you let it amplify your stuff, and
51:06 then you can take your brain, which is
51:07 better than it at some things, and you
51:10 can go, "Ooh, that's a good idea. Let me
51:12 take that one. Put that in my crazy
51:13 straw. spit it back out to it. See how
51:16 it amplifies that. See how it amplifies
51:18 that. And so within an hour, you can do
51:20 in an hour that might what might what
51:23 might have taken you 3 months without
51:27 something like AI, but it's a very
51:28 different mindset. So
51:31 anyway, good question, good banter back
51:35 and forth. Thank you, bro. Truth, Lisa.
51:38 It's happening.
51:40 Um, all right.
51:43 I misjudged this and I do respect this
51:46 te take actually. Thank you for
51:47 entertaining it. Thanks for the qu. Like
51:49 seriously, I love these conversations
51:52 because for one thing it's like like
51:55 when Brandon said, "How do you make
51:56 those things across the top?" I'm like,
51:58 "Uh, I don't know."
52:00 The whole purpose of this channel, the
52:03 whole purpose of this channel
52:05 is for all of us to be in this
52:09 conversation. You're like, "Well, what
52:11 is this conversation?" I don't know.
52:13 It's it's it's like it's questions like
52:16 that saying, well, wait, is it theft? Is
52:19 it copyright violation? Can you
52:20 copyright something? Can it have
52:22 original thought? Like it it is so
52:26 important for us to to start to get our
52:28 head around what these things are, what
52:31 they make possible, what they can't do
52:33 right now, but the trajectory that
52:36 they're on. And that if they can't do
52:38 this right now, are they going to be
52:40 able to do it 6 months from now or 3
52:42 years from now? And if they can do it 6
52:44 months from now, well, we better start
52:46 [ __ ] preparing for that, right? And
52:49 that like, and the only way you can know
52:51 that is like I've been on this planet
52:54 for 60 years.
52:57 The only thing I know is when you're in
52:59 a a burgeoning um primordial soup stage
53:04 of a technological innovation,
53:08 the only way you can learn it is to be
53:10 in the soup. And that's that's what this
53:12 channel's about. Tik Tok question. Um
53:16 McKenzie Sim, how much coding should I
53:18 learn? I'm learning Python.
53:22 Uh might try uh wolf language. Okay,
53:26 great question.
53:28 Not a simple answer.
53:31 Do you technically need to learn coding
53:34 right now? You don't. I would argue that
53:37 learning philosophy is probably as
53:40 valuable, if not more valuable, than
53:42 learning coding.
53:44 And I' I've got lots of reasons for
53:45 that. Um, but that said,
53:49 people who are competent with coding,
53:52 who also understand vibe coding and a
53:55 lot of other things that AI makes
53:56 possible are going to be way more
53:58 valuable than just people that can code.
54:02 So, and and it also has a lot to do with
54:04 what you want to do. If you want to go
54:06 work for Meta or Open AAI or Google, um,
54:11 yeah, they're probably going to want you
54:13 to not only, you know, learn coding, but
54:15 be really [ __ ] good at it and be able
54:18 to use all these AI amplification tools
54:22 like cursor co-pilot and, you know,
54:24 cursor and copilot and, you know, all
54:26 all of the replet and things like that.
54:28 Um, if you want to be building models
54:31 and fine-tuning models, if you want to
54:33 be digging deep into um the open-source
54:36 world, if you want to develop your own
54:38 applications, even if you're going to
54:39 vibe code them, really understanding
54:42 databases and security and O and all of
54:46 the stuff that you need to do to build a
54:47 worldclass application. Right now, these
54:51 vibe coding tools suck at all that
54:53 stuff.
54:55 But
54:57 here's here's the butt. Here's where you
54:59 don't have to necessarily learn coding.
55:01 It depends what you want to do.
55:04 One of the most important roles in the
55:07 next five years is is going to be what I
55:09 call a creative generalist. And that and
55:11 that is someone who can look at the
55:13 entirety of a project, the entirety of a
55:16 of an idea. So you're working for a
55:19 company and you need to revamp your I
55:21 don't know your
55:23 sales outreach.
55:26 One way you could do that is you could
55:27 go into application development and you
55:29 could get with a UX designer and you
55:30 could
55:33 you could run down that deep rabbit hole
55:35 and that's going to take you a long time
55:36 to build. The other way is to be a
55:39 creative generalist that sits above all
55:41 of the tactical execution but can
55:43 actually understand all of the different
55:45 things that are required to do that.
55:47 Okay, there's going to be some change
55:48 management required. There's going to be
55:50 some systems development required. We're
55:52 going to need to really understand our
55:53 customers. We're going to really need to
55:55 understand how our current sales um
55:58 approach isn't working. Where's it
56:00 breaking down? We're going to need to do
56:02 some data analysis.
56:05 So, imagine being so fil with AI tools
56:10 that you could just sit there like the
56:13 the Wizard of Oz or like the the
56:16 conductor of a symphony
56:18 and just go, "Okay, I'm going to pull in
56:20 this tool and do the data analysis. Then
56:22 I'm going to take the answer of that.
56:23 I'm going to put that into a
56:24 presentation. And this afternoon, I'm
56:26 going to give that presentation to the
56:27 boss. I'm going to get the boss's
56:29 blessing. And then I'm going to go do
56:30 all this. And then and then I'm going to
56:32 write scripts. And then I'm going to
56:33 create training materials.
56:36 Like that kind of person is going to be
56:39 they're going to be like [ __ ]
56:40 magicians. It's going to be it's
56:42 literally going to be like the Wizard of
56:43 Oz. They're going to have that kind of
56:45 power
56:47 that doesn't require coding,
56:50 right? That requires broad thinking.
56:54 That requires understanding lots and
56:57 lots of elements. I think the day of the
57:00 hypers specialist is numbered.
57:03 There will still pe people that hypers
57:05 specialize, but they're going to have to
57:06 get better at going horizontal.
57:09 Um, that's me, the creative generalist.
57:11 Uh, there was a Tik Tok pin, but I
57:13 missed it. Uh, uh, uh, I agree with you.
57:16 to understand the basic code even
57:20 blender but limited all obsolete. Yeah,
57:22 exactly. Okay. Communication skills will
57:25 be really important. Yes, communication
57:27 skills are going to be critical,
57:30 right? Empathy,
57:33 um critical thinking, creative thinking,
57:36 curiosity,
57:37 adaptability. Holy [ __ ] You want to
57:40 know you want to know what to study
57:43 rather than coding? Learn enough code so
57:46 you understand the components of coding.
57:50 But
57:52 you want to really master something, get
57:54 really masterful at something, become
57:56 the most adaptable person on on the
57:59 block.
58:01 Become someone that the minute something
58:03 new comes out, you've got curiosity
58:04 enough to go, "Oh, I'm going to go learn
58:06 that." And then you can immediately go,
58:09 "Is that better or worse than what we
58:11 have today?" If it's the same or worse,
58:14 I don't give a [ __ ] about it. If it's
58:16 dramatically better, I adapt all of the
58:18 other stuff that I do and how I think
58:20 about stuff to this new reality.
58:23 And that's just going to that's going to
58:25 keep rolling for the next five or 10
58:27 years.
58:29 Those people that have curiosity and
58:32 hunger and ambition and adaptability,
58:37 holy [ __ ]
58:39 you're going to be valuable. like deeply
58:43 valuable like you know
58:47 won't ever be looking for work valuable
58:51 and there going to be a lot of people
58:52 looking for work
58:56 um
58:58 new pin I don't see a pin okay
59:01 hit you on LinkedIn great this is a
59:04 refreshing take from sea levels
59:06 expecting raw 40% increase in output
59:10 yeah so listen I'm glad you hit me up on
59:12 LinkedIn. I'm happy to chat. Um and and
59:14 excited to I'm I'm I'm excited by your
59:17 things. Um the MIT study that just came
59:20 out
59:22 that that said 95% of AI projects have
59:26 failed
59:27 0% return 0% value. They said 95% of of
59:33 the things you've got to look at the
59:35 methodologies of that study
59:38 like how they're defining success is
59:41 increased revenue.
59:44 So like there could be no other value
59:46 from an AI implementation than increased
59:49 revenue, right? What no they also they
59:54 didn't define they like it's not based
59:56 on quantitative data. It's based on 54
59:59 qualitative interviews where they said,
1:00:02 "Did did you make more money or not?"
1:00:05 No. Oh, okay. So, that's a zero.
1:00:08 Like, that study is [ __ ] ridiculous.
1:00:11 And so, CEOs that expect a 40% bump in
1:00:16 revenue because we gave people access to
1:00:18 J Chat GPT is [ __ ] [ __ ]
1:00:22 Nobody knows how to use it.
1:00:25 That's what this channel is about is
1:00:28 nobody, nobody,
1:00:32 Hang on, blowing my vocal cords out
1:00:34 here.
1:00:38 Nobody knows how to use this [ __ ]
1:00:42 One of the things I'm doing in
1:00:43 consulting right now is anyone in the
1:00:45 sea suite. If I'm going to work with
1:00:47 anyone in the seauite and they don't
1:00:49 [ __ ] use generative AI, I'm not going
1:00:53 to work with them
1:00:55 because this is not the kind of thing
1:00:57 that you can go I'll just delegate it to
1:00:59 the IT group. I'll just delegate it to
1:01:01 the people that handle that. I I'll just
1:01:03 delegate it to the innovation group. No.
1:01:06 You've got to understand
1:01:10 as someone who's trying to drive change
1:01:11 in an organization, you've got to
1:01:13 understand
1:01:15 that what generative AI is doing to
1:01:17 computers is making them behave in ways
1:01:19 that they've never behaved before.
1:01:23 This is not an efficiency play.
1:01:27 Can you use AI to make more efficient
1:01:30 ways of doing what we've already done?
1:01:32 Yes.
1:01:37 It's not all that interesting.
1:01:40 We've had Zapier for years. We've had
1:01:43 automations for years. We've had
1:01:45 specialized AI for years. That's not
1:01:48 what we're talking about here. We're
1:01:50 talking about generalized intelligence
1:01:54 that does stuff. It generates.
1:02:00 No one in an organization knows exactly
1:02:03 what that means for their department,
1:02:04 for their job. No one. So, the only way
1:02:08 we're going to get through this is to
1:02:10 collaborate and to mix ideas together
1:02:12 and be in conversations like this. And
1:02:14 Murphy in the house was happening
1:02:18 just in time for my favorite topic. Um,
1:02:21 what's we're we're going deep here. We
1:02:23 are going deep here. Um, we paid 240
1:02:26 bucks for the tool. source camp picked
1:02:28 up three new clients in 24 hours
1:02:32 because I can adapt to my clients pain
1:02:35 points. Yeah, this is
1:02:39 this is the [ __ ] thing about AI right
1:02:42 now is
1:02:45 h it's it's it's so [ __ ] exhausting.
1:02:48 It and it's only exhausting because
1:02:50 we've been through this before, right?
1:02:52 Every time there's a new technology, we
1:02:54 go through this. In the mid 90s,
1:02:58 the number of conversations I had with
1:03:00 seauite executives that were saying, "We
1:03:02 need a website and I would ask,"Wh do
1:03:05 you need a website?" And they go, "Cuz I
1:03:08 was golfing with my buddy and he's got
1:03:09 one and I got a I'm" I'm like, "What do
1:03:11 you want it to do?" I don't know. Well,
1:03:14 then how do you know you need a website?
1:03:16 You might need a TV ad.
1:03:19 You might need a better sales force. You
1:03:21 might not need a website. And it's the
1:03:24 same [ __ ] with AI right now. We need AI.
1:03:26 Why? What do you want it to do? Do you
1:03:29 know? Do you even know what it makes
1:03:30 possible?
1:03:32 Well, I just need it. Okay, I got chat
1:03:36 GPT for the staff. They're going to go
1:03:38 figure it out, are they?
1:03:43 It's like, [ __ ]
1:03:46 What source camp's doing is going in
1:03:48 going, got a problem? Oh, do you want me
1:03:50 to show you how we can solve that
1:03:53 problem? You know the problem that you
1:03:55 haven't been able to solve for two years
1:03:58 and you think it's not solvable? Do you
1:04:00 want to watch me solve it in an hour?
1:04:03 That's what she does.
1:04:06 It's remarkable.
1:04:08 And then she and then she's she like
1:04:11 solves their problem in real time in
1:04:13 front of them. And they're like, I how's
1:04:16 that possible? She's like, "Well, keep
1:04:18 paying me and and we'll knock a bunch of
1:04:22 other things down like that."
1:04:26 That's because she's living in this
1:04:29 creative generalist role where she can
1:04:32 look at any problem in any business and
1:04:35 go, "Okay, um, if I use that tool and
1:04:37 that and I put those together, right?"
1:04:40 She's like the conductor.
1:04:43 So those people that that get fil with
1:04:46 lots of tools and even tools that are
1:04:49 way outside their expertise like coding
1:04:52 or images or music
1:04:57 they're going to be the ones that going
1:04:58 to be able to walk into any situation
1:04:59 and go what's your problem? Oh we can we
1:05:01 can crack that nut.
1:05:04 All right seuite is uniquely unqualified
1:05:08 to strategies for AI. Yes, I have to
1:05:10 educate them under the cloak of
1:05:12 darkness. Actually, you know what, Anne
1:05:14 Murphy, that's a really good point. I'll
1:05:16 tell I'll tell you something fun.
1:05:19 When when when we first started
1:05:21 agency.com, the the original name of the
1:05:24 company was the agency. And the reason
1:05:27 we called it the agency was we said we
1:05:30 understand what websites are. We really
1:05:32 didn't like, you know, we understood as
1:05:34 much as anyone because it was just being
1:05:35 born, right?
1:05:38 But but what our hypothesis was was
1:05:40 there's going to be a lot of seuite
1:05:42 executives that are going to want a
1:05:44 website, but they're not going to want
1:05:45 to admit that they don't know what a
1:05:47 website is. And so we talked about the
1:05:51 agency being like this, you know, cloak
1:05:53 and dagger agency that we would sneak
1:05:55 in, you know, the the service entrance
1:05:58 of the big building, you know, go up to
1:06:00 the sweet sea suite, teach them what
1:06:02 websites were, and then leave. You know,
1:06:04 we'd come build your website, then go
1:06:05 out the back door. Um, I think there's
1:06:08 something there. I think with generative
1:06:10 AI, what what you've got, especially in
1:06:13 big companies, is
1:06:16 you've got executives who their
1:06:19 organization has had AI for decades.
1:06:24 They're like, "Oh, yeah, we got AI. We
1:06:26 got people for that." They don't
1:06:28 actually understand at all that what
1:06:31 generative AI is is different and and
1:06:34 and that the implications of it are are
1:06:37 profound.
1:06:38 Um Boopsy, good to see you. I haven't
1:06:41 seen you in a while. I use chat GPT to
1:06:43 prepare for an appeal with the country.
1:06:47 Oh, with the county. I was able to make
1:06:49 them admit
1:06:51 fault. Awesome.
1:06:54 Yes. Oh, you want to know something cool
1:06:56 today?
1:06:57 So, um, so my wife's an artist and she
1:07:00 got, um, selected to be interviewed for
1:07:03 an article
1:07:05 and she had a brain injury at some
1:07:06 point. So, she gets overwhelmed in
1:07:08 conversations and things like that. And
1:07:09 so, when this reporter was interviewing
1:07:12 her, she was getting overwhelmed and she
1:07:14 gets, "Can you just send me the
1:07:16 questions?" And so, the woman sent the
1:07:18 questions and there was a lot of
1:07:19 questions and a lot of them were
1:07:21 repetitive and things like that. So what
1:07:23 we did was I said, "Oh, this is a
1:07:26 perfect use for chat GBT,
1:07:29 not to write not to write the questions,
1:07:31 but just to make sense of let her sort
1:07:33 of ramble, right, in in all sorts of
1:07:36 order and then let chat GPT make sense
1:07:39 of it." And so I would read the question
1:07:42 and she'd think about it for a bit and
1:07:44 she, you know, we'd go back and forth a
1:07:45 little bit and then I'd say I put on
1:07:47 chat GPT and I go, "Okay, Quinn, you
1:07:49 know, here's the question." and then
1:07:51 Gabby's gonna answer and then I want you
1:07:52 to make sense of it. And so I would ask
1:07:54 the question, then Gabby would answer
1:07:56 and she would say this idea and that
1:07:58 idea and and you know she'd go all over
1:08:00 the place and then Quinn would go,
1:08:02 "Well, here's how you might say that."
1:08:03 She would write a perfect like retelling
1:08:06 of what Gabby just said. Well written,
1:08:10 wellcrafted. I would literally copy and
1:08:12 paste that, remove m dashes because
1:08:15 there were too many of them. And it was
1:08:17 good to go. It was amazing. It was
1:08:19 absolutely amazing. Um, yeah, that that
1:08:23 ability to to take the thing you want to
1:08:26 do, right? You want to get the the
1:08:29 county to admit that they [ __ ]
1:08:30 something up
1:08:32 and and they've got all sorts of ways of
1:08:35 avoiding responsibility or admitting
1:08:37 that they [ __ ] something up. Well, all
1:08:39 of that's been documented somewhere. So,
1:08:41 that's all in the learning model, right?
1:08:43 So, you can go into it and go, "Okay,
1:08:45 here's my situation. Here's what the
1:08:47 countyy's saying. Here's my evidence.
1:08:50 Give me an outline. Give me a speech.
1:08:53 Beautiful. Brilliant. Love it.
1:08:56 Tik Tok pin. Um, no pin there right now.
1:09:02 Repin.
1:09:08 Cavuno. My VP is meeting with our CHRO
1:09:13 and I prepared her for their AI
1:09:16 conversation. So, so Kuno works for a
1:09:20 big company and and uh there's a lot of
1:09:24 people. So, so if you're new here, it
1:09:26 seems like we've got some new folks
1:09:27 here. The people that show up to this
1:09:29 channel night after night. I'm here five
1:09:31 nights a week. I do this for one and a
1:09:33 half, two hours, five nights a week,
1:09:35 starting at 8:00 p.m. Mountain time.
1:09:36 It's a little late, but [ __ ] it. What
1:09:38 else are you going to do? Um the people
1:09:41 that show up here night after night,
1:09:42 they're called the irregulars because
1:09:44 they're just not normal.
1:09:46 Because why would you watch this, right?
1:09:50 Why would you watch an old man rambling
1:09:52 about [ __ ] AI? Because what they
1:09:55 understand is if they're in the
1:09:56 conversation
1:09:58 and they can be just a little bit
1:10:00 smarter about AI than the people around
1:10:02 them and if they can talk intelligently
1:10:06 about it, then good things are going to
1:10:08 happen.
1:10:10 And and a number of people on this
1:10:12 channel started out as what Ethan Mullik
1:10:16 described as the secret cyborgs. They
1:10:18 were interested in AI, but they were in
1:10:20 the shadows. And one of the things we
1:10:22 talk about in the AI salon is play
1:10:24 first, mindfully create, and generously
1:10:27 lead. And the generously lead part looks
1:10:30 like talk about AI. Talk about what
1:10:33 you're learning. Talk about what it's
1:10:35 good at. Talk about what you love. Talk
1:10:37 about what you hate. Just talk about it.
1:10:39 Talk about it at work. Talk about it at
1:10:41 the dinner party. Talk about it on
1:10:43 [ __ ] Facebook. You want to get some
1:10:45 hate mail? Talk about AI on Facebook. I
1:10:49 don't do it. It's bad enough when I get
1:10:52 my ass handed to me on LinkedIn.
1:10:55 But just talk about it because what it
1:10:58 forces you to do is it forces you to
1:10:59 think about, well, wait, what is my
1:11:01 opinion here? And then you put it out
1:11:03 there and you establish yourself as
1:11:05 someone who is thoughtfully
1:11:07 engaging with AI. You establish yourself
1:11:10 as some kind of expert.
1:11:13 None of us are experts right now. Zero
1:11:16 people, zero percentage of us
1:11:22 because the stuff's changing too fast.
1:11:23 But you can be ready for AI. Ann Murphy
1:11:27 who's here, she and I put together with
1:11:29 Vicky Baptiste a thing called the AI
1:11:31 readiness training program. It's a
1:11:34 fivemodule comprehensive training
1:11:37 program.
1:11:40 Yeah. If you go to are you
1:11:41 readyforai.com,
1:11:42 it doesn't teach you any tools. It's not
1:11:44 about tools. It's about shifting your
1:11:47 mindset.
1:11:48 How do you think about being creative
1:11:50 with AI? How do you think about business
1:11:51 with AI? what you know, how do you think
1:11:53 about the ethics and security and
1:11:55 privacy? How do you apply this to
1:11:58 business?
1:12:00 And so there's all this remarkable video
1:12:02 content in there from from last year's
1:12:03 AI Festivus. And then we pulled the
1:12:06 lessons out of, you know, all of those
1:12:08 videos that were kind of consistent
1:12:10 across them. And that's what that
1:12:12 training is about.
1:12:15 And it's amazing.
1:12:18 And so
1:12:19 if you do that, what happens? Like one
1:12:22 of the things that happened with Kuno
1:12:23 is,
1:12:25 you know, she got asked to do some talks
1:12:28 and then she got asked to do more talks
1:12:30 and then she someone at one of the talks
1:12:32 said, "Oh, you should talk to the VP and
1:12:34 then she gave a talk to the VP and then
1:12:35 she gave and then she was on the AI
1:12:37 steering committee." I think I think is
1:12:39 that right? And now there's all sorts of
1:12:42 things where she's training up people
1:12:43 high in her organization because I
1:12:45 guarantee you this, she is probably one
1:12:48 of the smartest people in her
1:12:49 organization
1:12:51 about AI,
1:12:53 about generative AI,
1:13:00 you know. I don't I don't even know your
1:13:02 background career. I don't I don't know
1:13:04 what you did before. Was it was in
1:13:06 instructional design? I don't know.
1:13:10 But why she's expert right now is is
1:13:13 because she's thinking broadly and
1:13:15 thoughtfully and critically about what
1:13:18 AI is good at, what it's not good at,
1:13:19 and how you apply that to a business in
1:13:21 a responsible way. Very few people are
1:13:25 doing that right now. So if you do that
1:13:26 in a loud and visible way, you will get
1:13:29 noticed. Okay.
1:13:32 Um
1:13:47 Okay. Um, the other thing, um, since Ann
1:13:51 Murphy's here, so I mentioned AI
1:13:53 Festivus.
1:13:55 If you were not here last year for AI
1:13:58 Festivus, shame on you. And if you were
1:14:00 not here the year before when we did
1:14:02 GPTs for good,
1:14:05 shame on you. So, every Christmas
1:14:09 holiday, Ann Murphy and I seem to lose
1:14:12 our mind
1:14:14 and we say, "Hey, wait. We've got two
1:14:16 weeks off. Why don't we fill that with a
1:14:18 high stress live event that covers 24
1:14:22 hours?"
1:14:26 And so, so two years ago, we decided to
1:14:30 do 24 hours of GPT building for
1:14:32 nonprofits. And we created more than 500
1:14:35 custom GPTs for nonprofits. And last
1:14:38 year we did this thing called AI
1:14:39 Festivus. And it was 34 speakers from
1:14:42 four different communities that just
1:14:45 came and it's the Friday and Saturday
1:14:48 between Christmas and New Year's or
1:14:50 Hanukkah and New Year's or whatever you
1:14:52 celebrate there and New Year's.
1:14:56 It's like like you got you got to
1:14:58 include them all now. Um so that Friday
1:15:02 and Saturday. So this year
1:15:05 it's December 26th and December 27th. We
1:15:08 start at 9:00 in the morning, we go till
1:15:09 9 at night, 12 hours on Friday, 12 hours
1:15:13 on Saturday. And you're like, "What are
1:15:15 you [ __ ] insane?" Yes.
1:15:18 And we're going to do it again this
1:15:20 year, and it's going to be bigger and
1:15:21 better than last year. Last year we got
1:15:25 some boosts from some some people in
1:15:27 other communities
1:15:29 and we thought we might be doing this
1:15:32 for about 400 people if we were really
1:15:34 lucky and got our [ __ ] together. We
1:15:36 ended up having 3,500 people um sign up
1:15:41 for festivists. We had 200 people show
1:15:43 up and we averaged 540 concurrent
1:15:47 sessions every hour for 24 hours over a
1:15:51 36-hour period. It was absolute insanity
1:15:55 and it was amazing. And people said it
1:15:57 changed their lives. And like we we uh
1:15:59 we on our podcast on the AI readiness
1:16:02 project podcast um this week we
1:16:04 interviewed someone that said, "We have
1:16:07 a question. How did you get into AI?" He
1:16:08 said, "I got into AI because of AI
1:16:10 Festivus. changed his life.
1:16:14 So, we're doing that again this year.
1:16:16 So, save those dates. If you know
1:16:19 someone that should speak at it, let us
1:16:20 know. Um, and if you know someone who
1:16:23 wants to sponsor it. The the thing that
1:16:25 we're committed to doing, which is what
1:16:27 we did last year, is that it's free for
1:16:29 everyone.
1:16:31 And so, we're committed to doing that.
1:16:34 It's not free for us because because
1:16:36 what we found out, we did this on Zoom
1:16:38 last year and we found out if you want
1:16:40 to have more than like 20 people on your
1:16:42 Zoom call, you got to you got to [ __ ]
1:16:44 step up. You got to give up some coin.
1:16:48 And luckily luckily we had some people
1:16:49 come in and sponsor at the last minute.
1:16:52 But um but yeah, that's that's coming
1:16:55 up. All right. It's so cool.
1:16:59 What time is it? It's 9:18. I I do want
1:17:02 to go do some Gemini um some Gemini
1:17:05 banana time.
1:17:08 And here's why. Um
1:17:13 you could absolutely argue
1:17:16 Gemini,
1:17:18 go to gemini.google.com.
1:17:21 Okay, you can flip the screen up there.
1:17:22 There you go, Brandon. Um,
1:17:25 you could absolutely argue
1:17:28 that image generation inside a large
1:17:31 language model, you can do it inside
1:17:33 chat GPT, you could do it with image gen
1:17:35 inside Gemini before and now we can do
1:17:38 it with this new model called nano
1:17:40 banana.
1:17:42 Why take any time on that? Why why spend
1:17:45 I've now spent this is now my third
1:17:47 night where I'm spending time working on
1:17:48 this. Why am I doing that? Because it's
1:17:52 different. There's something different
1:17:53 about this one and I don't know how to
1:17:55 use it yet. And that drives me [ __ ]
1:17:58 crazy. And like that's that's an
1:18:00 attribute if you can if you can develop
1:18:03 that thing of like I got to I got to
1:18:05 know. I got to know how does why does
1:18:07 this how do you make it do the good
1:18:09 stuff? Why why is it not doing what I
1:18:11 want it to do? Why is it not wanting to
1:18:13 Right? If you can develop that hunger
1:18:16 for understanding these things, you're
1:18:18 going to be in really good shape. Like
1:18:19 that's the core skill for the next 3 to
1:18:21 5 years. Just that curiosity and that
1:18:24 drive to be like, "How does this [ __ ]
1:18:26 thing work?"
1:18:28 Um, I'm clueless on how this thing
1:18:30 works. I can do a little bit with it. I
1:18:33 understand what it's doing. I just don't
1:18:34 understand how to make it sing yet. And
1:18:36 so that's why I want to play with it.
1:18:38 All right.
1:18:40 Um, why can't OpenAI come up with fun
1:18:42 names like Nano Banana? Well, but you
1:18:44 know what's funny, Source Camp?
1:18:46 The name of this is not Nano Banana. The
1:18:49 name of this is 2.5 flash image
1:18:52 generation.
1:18:54 Like nowhere does it say nano banana.
1:18:58 That was the code name of the project
1:19:00 that everyone on the internet said this
1:19:02 nano banana model is [ __ ] amazing.
1:19:06 And because it had a fun name, people
1:19:07 are sticking with that. And now Google's
1:19:09 leaning into it a bit and they're making
1:19:11 banana images and things like that.
1:19:13 That's not the name of it.
1:19:16 The name of it is yes. Um it uh the we
1:19:20 talked to the the branding professionals
1:19:22 here here at Google. It's that's at
1:19:24 google.com
1:19:25 and uh w within within the the branding
1:19:28 framework of of the Gemini artificial
1:19:31 intelligence suite. uh the uh the the
1:19:35 image generation well it's natively
1:19:37 within uh the Gemini 2.5 flash which is
1:19:40 the model of course and it's a
1:19:41 multimodal model and within that is the
1:19:43 image generation and and that's what was
1:19:46 the image portion of that is is is what
1:19:48 was codenamed nano banana for testing
1:19:51 purposes of course that's not the
1:19:52 official the official name is is just
1:19:54 it's Gemini 2.5 flash
1:19:59 no
1:20:01 call it [ __ ] nano banana Hannah
1:20:04 idiots. Okay.
1:20:07 [Laughter]
1:20:12 Oh my god. All right.
1:20:16 Um, okay. Let's see. So, if you're at
1:20:20 gemini.google.com,
1:20:21 you have to be in 2.5 Flash. This is
1:20:24 another confusing thing. If you're in
1:20:25 2.5 Pro, this does not have the newest
1:20:29 image generation model in it. So the
1:20:31 less capable large language model has
1:20:33 the more capable image generation model.
1:20:36 Remember that. Put that in your pipe and
1:20:38 smoke it. Okay. So what we're going to
1:20:40 do here is we're going to say I want you
1:20:44 to create
1:20:46 a J jso n. If you don't know what jso n
1:20:50 is, it's JavaScript
1:20:52 object notation.
1:20:54 JSON or some people call it JSON, but
1:20:57 that just sounds like a dude. So I call
1:20:59 it JSON. So I can sound impressive. All
1:21:03 right. I want you to create a JSON
1:21:05 framework
1:21:08 to describe
1:21:11 all of the elements
1:21:16 that might be in an image
1:21:21 that
1:21:28 that might be in an image. So let's just
1:21:29 see what it comes up with.
1:21:36 This JSON framework includes a
1:21:38 structured way. So we got a time stamp,
1:21:41 overall scene description, camera
1:21:43 parameters, elements, person label,
1:21:46 pedestrian,
1:21:49 segmentation, mask, URL. Good l um
1:21:53 gender value female. All right. Good
1:21:57 lord, that's a lot.
1:21:59 Oh, and here's a description. Now, now
1:22:01 we have documentation of our of our JSON
1:22:04 framework.
1:22:06 Okay. So, then
1:22:09 what? How do I get out of this? Okay.
1:22:12 Then I'm going to say um
1:22:15 I want you to write
1:22:18 a JSON prompt
1:22:22 for a cinematic
1:22:27 shot.
1:22:28 from a movie in
1:22:31 16 by9 aspect
1:22:37 ratio
1:22:40 with
1:22:42 a
1:22:44 an unforgettable
1:22:47 character
1:22:52 in an unforgettable
1:22:56 place
1:22:58 with an unforgettable
1:23:01 object
1:23:06 and
1:23:09 some unforgettable
1:23:12 action.
1:23:14 Okay, let's see. We'll see what it comes
1:23:16 up with. So what I'm doing here, the the
1:23:18 the reason to play with JSON is you
1:23:22 don't have to do this at all, but it
1:23:24 does seem that that this image model
1:23:28 does better the more context you give
1:23:30 it. So um there's a there's a music
1:23:34 generation tool right now called
1:23:36 producer.ai
1:23:39 that you create songs in the context of
1:23:42 a conversation. And it's really amazing
1:23:44 because you can just go, "No, make it
1:23:46 more boppy and it'll make the song more
1:23:48 boppy." Like it just it understands
1:23:51 imprecise
1:23:53 creative direction.
1:23:55 This nano banana thing doesn't. Like
1:23:58 you'll get a really good shot and then
1:23:59 you'll give it an imprecise creative
1:24:01 direction and it'll just make a
1:24:03 completely different image. So what I'm
1:24:06 trying tonight is is trying something a
1:24:09 little different. All right. Shot ID.
1:24:11 unforgettable climax shot 01 aspect
1:24:15 ratio camera angle low tracking shot. So
1:24:18 here's the other thing. If you don't
1:24:19 know anything about cinematography
1:24:22 and if you don't know anything about
1:24:24 movie making and and visualization
1:24:29 um
1:24:34 oh last night I was oh thank thank god I
1:24:36 said that that nano banana is not the
1:24:39 actual name of the product. Yeah. No,
1:24:40 you can't find Nano Banana. That Yeah.
1:24:43 No, what you can find is Gemini 2.5
1:24:46 Flash. And if you don't know that it's
1:24:47 Nano Banana in there, then nanny nanny
1:24:50 boooo on you. So, I'm glad you found it,
1:24:53 Chef Kelly. Okay. Um, but
1:24:58 large language models know all of the
1:25:00 cinematography terms. They know all the
1:25:02 photography terms. They know colors.
1:25:04 They know lighting. They know positions.
1:25:05 They know blocking. All the [ __ ] that's
1:25:08 ever been written about visual
1:25:10 communications is in the large language
1:25:12 models. So if you just say come up with
1:25:15 a JSON framework, it understands that
1:25:17 what what JSON what a JSON framework is
1:25:19 is it's just a structured list of
1:25:23 attributes of this image. And what's
1:25:26 cool is once you have it, you can just
1:25:28 tell it to fill in this framework
1:25:31 for any image you want.
1:25:33 Okay.
1:25:37 Low angle, dramatic backlighting,
1:25:39 desaturated blues and grays with with
1:25:42 spot red accent.
1:25:44 Frame composition rules rule of thirds
1:25:47 leading lines toward the character.
1:25:49 Great name. Elias the whisperer. Okay.
1:25:52 Unforgettable place. The obsidian spires
1:25:55 of Athlgard. Great. A vast desolate
1:25:59 plateau where colossal jagged spires of
1:26:02 black crystallin rock pure. Fine.
1:26:05 Beautiful. Okay. So, we're going to take
1:26:08 this whole thing,
1:26:10 this prompt.
1:26:15 And then we're going to do this. So,
1:26:17 what was the character's name?
1:26:24 Elias. Okay. So, I have I have an idea.
1:26:29 So, we're going to do we're going to do
1:26:31 new new chat.
1:26:35 I'm gonna h maybe I shouldn't do a new
1:26:37 chat. Maybe I'll go back. Maybe I'll do
1:26:38 it in the context of this other chat.
1:26:41 Yeah, let's do that. Okay.
1:26:44 So, I'm going to go
1:26:48 here's an image prompt
1:26:53 that I that
1:26:57 I want you to make.
1:27:00 The uploaded
1:27:06 picture
1:27:08 is who I want. who I want Elias to look
1:27:14 like. All right. And then I'm going to
1:27:19 paste in that JSON prompt, but I'm going
1:27:21 to upload an image.
1:27:32 Kyle Shannon Dreams.
1:27:39 Those are bad.
1:27:42 Let's see.
1:27:56 Um,
1:28:00 let's see.
1:28:02 Hold, please. I got to go find
1:28:03 something. Go to Kyle Shannon Dreams.
1:28:06 Ah, here we go. This is the guy I want
1:28:08 right here.
1:28:11 So, we're going to export this. Uh, mind
1:28:14 your tabs. Am I not sharing the right
1:28:17 tab? I am not, but I'll be back. That's
1:28:20 okay. Thank you.
1:28:25 Guitar desktop. Go.
1:28:29 We're going to come here. We're going to
1:28:31 go desktop.
1:28:34 All right. That's the guy. You can't see
1:28:36 it.
1:28:38 All right. That's the dude. He's got
1:28:41 this sort of crazy split beard.
1:28:45 So, we're gonna we're gonna give it this
1:28:47 intense, highly structured prompt. And
1:28:51 we're going to give it a dude for him to
1:28:52 look like. And then we're going to make
1:28:54 this. All right.
1:28:56 Here we go. Is this gonna be any good? I
1:28:59 don't know. But mind your black bar.
1:29:01 Shut up. I know. Producer Brandon.
1:29:05 All right.
1:29:06 I don't know what this banana talk is
1:29:09 about, but it sounds cool. Oh, so uh uh
1:29:13 nano banana was the code name of the new
1:29:16 image generation tool.
1:29:19 Wait, what the [ __ ] just happened?
1:29:29 What did it do?
1:29:41 It just wrote us an application.
1:29:46 It just coded us a website.
1:29:54 Holy [ __ ]
1:29:57 Okay. Do do you remember when I said
1:30:00 before I don't know how to use this
1:30:02 thing?
1:30:07 Okay.
1:30:12 Um
1:30:15 when companies talk about models being
1:30:18 multimodal
1:30:22 and and uh OpenAI talked about this
1:30:24 today with their speechtoech synthesis
1:30:27 model.
1:30:32 When all this stuff first came out, all
1:30:34 these models were distinct models. You'd
1:30:36 have a language model, you'd have an
1:30:37 image model, you'd have a sound model,
1:30:39 you'd have a code model.
1:30:42 code and language went together pretty
1:30:44 quick, but you had you had specialized
1:30:45 code models.
1:30:49 The Gemini series starting I think
1:30:51 starting with 2.0 and then it got really
1:30:54 good in 2.5.
1:30:56 Um, these things are multimodal, meaning
1:31:00 if you upload a video into this thing,
1:31:02 it can understand the video, all of it,
1:31:04 the images, the audio, everything about
1:31:08 it. Same with audio.
1:31:11 What this did, which I don't understand
1:31:15 why it did it, but because I gave it
1:31:17 code and I said, I want you to generate
1:31:19 this image. It generated the image, but
1:31:22 it it it it put the the image prompt
1:31:28 inside an application and it created the
1:31:32 unforgettable cinematic moment
1:31:34 generator.
1:31:36 You got to be kidding me.
1:31:38 This is crazy.
1:31:52 Generate a new image.
1:32:00 All right. Same image. Okay. So, what
1:32:03 I'm going to say is
1:32:06 um
1:32:11 right now the prompt is hardcoded.
1:32:21 Vicki pin on Tik Tok. I told you about
1:32:24 this yesterday. That's what it did to
1:32:26 me. I didn't I didn't take it in, Vicki,
1:32:29 that it did that. Wait, I I'm I'm mind
1:32:33 blown
1:32:36 because this is not um
1:32:41 this is kind of an amazing behavior. We
1:32:43 asked it for an image and it gave us a
1:32:45 website that generates images. It
1:32:48 generates the same image.
1:32:52 That's bonkers. Oh, you posted on
1:32:55 irregulars.
1:32:57 I didn't see it. That's amazing.
1:33:05 That's crazy. This crazy talk. Right
1:33:08 now, the prompt is hardcoded.
1:33:12 Um,
1:33:17 let's make it so you can upload
1:33:23 an image
1:33:25 and give it a short
1:33:29 an image of the character
1:33:35 and give it a short prompt.
1:33:41 and it
1:33:45 will use AI to fill in the
1:33:54 JSON
1:33:56 framework
1:33:58 we created.
1:34:01 This is amazing. Let's see if Well, I'm
1:34:05 not going to spend a lot of time vibe
1:34:06 coding here. I just want to see what it
1:34:08 does.
1:34:11 Yeah, I got that Brandon, I think.
1:34:13 Unless there was another one that I
1:34:14 missed.
1:34:27 I'll proceed with updating the canvas.
1:34:30 So, is it rewriting the code now?
1:34:39 Yeah, it looks like it looks like it
1:34:40 rewrote the code. Oh, here we go.
1:34:43 Dynamic shot. No file up. Okay, choose a
1:34:46 file. So, let's put the dude in there.
1:34:49 Oh, that's cool.
1:34:51 Short cinematic idea. Um, Elias
1:34:56 stands
1:35:01 in the middle of Time Square
1:35:07 in the rain
1:35:13 after
1:35:16 all the people left.
1:35:24 generate a cinematic shot.
1:35:29 It's generating a cinematic shot.
1:35:38 Oh, and I hit it tw Oh, wait. Gener. Oh,
1:35:40 look. Generated JSON prompt. Holy [ __ ]
1:35:43 Oh my god.
1:35:47 This is amazing.
1:35:51 Yeah, this is a Kevin Mallister moment
1:35:53 for me.
1:36:07 And here's the JSON prompt that it
1:36:09 created this big ass JSON prompt.
1:36:14 And can I can I do another one?
1:36:17 Choose file. Yeah. Let's go get a
1:36:20 different.
1:36:22 All right. There we go.
1:36:26 You don't see that, but okay. So,
1:36:28 there's a woman. There's a black and
1:36:29 white photo of a woman. So, we're going
1:36:31 to say um
1:36:35 we'll call it Gabby. Gabby
1:36:41 is riding
1:36:44 on a horse.
1:36:49 in a
1:36:54 cowboy outfit
1:36:58 in the dusty
1:37:03 desert
1:37:08 west.
1:37:11 Uh uh
1:37:18 at sunset
1:37:20 with
1:37:26 a cloud
1:37:30 of dust behind her
1:37:34 lit
1:37:36 by the setting sun.
1:37:42 I I want to actually I'm going to read
1:37:43 the I want to read the JSON prompt and
1:37:45 see what it does with that. See if it
1:37:47 generates cinematic shot.
1:37:51 Now I can share this. Copy the contents.
1:37:55 Can I publish it?
1:37:59 Show console.
1:38:07 This is crazy.
1:38:13 So what now? We put the we put the image
1:38:17 in full description
1:38:20 of the scene into cling AI. No, what I
1:38:23 would do is well so so we can add to
1:38:26 this. We can say uh well I won't I won't
1:38:29 add a database to it, but I'll say um I
1:38:32 want to add a button
1:38:40 to download
1:38:43 the
1:38:46 file
1:38:49 and be able to name it.
1:38:54 I also want it to
1:39:00 download
1:39:02 a TXT version
1:39:06 of the
1:39:09 Oh [ __ ]
1:39:11 Uh,
1:39:16 I want to create a txt version of the
1:39:19 JSON prompt.
1:39:26 Let's try and let it fix the error that
1:39:28 it it just had.
1:39:33 Okay. You know what I said before that I
1:39:35 felt like an idiot and I felt like a
1:39:37 dumb dumb cuz I really didn't know how
1:39:38 to use
1:39:41 um this thing.
1:39:43 I feel
1:39:48 I feel
1:39:50 well I'm excited
1:39:53 but like I don't know what this means at
1:39:55 all. Like is this expected behavior? If
1:39:59 you give it a JSON prompt it it's going
1:40:01 to it's going to assume you want code. I
1:40:04 don't I don't know. This isn't
1:40:07 necessarily a bad thing. It's a cool
1:40:09 thing. All right. Let's see.
1:40:13 All right. Um,
1:40:16 Gabby wears
1:40:19 a cowboy
1:40:22 outfit
1:40:25 in 18
1:40:28 94
1:40:32 in the Mesa Wild West.
1:40:38 Riding at speed at sunset.
1:40:42 Okay. Create a cinematic shot.
1:40:51 This is crazy people. It's crazy people.
1:40:55 It's crazy. Crazy crazy.
1:41:02 It's very cool.
1:41:04 All right. It's I think it broke the app
1:41:09 quite frankly.
1:41:12 because we we don't have the JSON prompt
1:41:14 anymore.
1:41:16 Add Gemini features. What's that? Oh,
1:41:19 what what the hell did I just do?
1:41:24 I enhanced the cinematic shot with two
1:41:26 new LLM powered features using the
1:41:29 Gemini API to give you more control and
1:41:31 inspiration.
1:41:33 Here's the updated code.
1:41:43 Huh?
1:41:51 Oh, look. It added it added uh it added
1:41:54 AI stuff.
1:41:57 Expand the idea. Generate the shot. Wow.
1:42:02 Generate an image. Okay. Go get the
1:42:03 image. Got the image.
1:42:07 character name Gabby.
1:42:11 Um,
1:42:13 Gabby. Gabby on a horse
1:42:18 in the 1800s.
1:42:22 And then I'm going to say expand idea
1:42:37 Gabby not just on a horse but one with
1:42:40 her horse. A powerful slick Mustang
1:42:44 whose main Jesus generate a cinematic
1:42:47 shot out of that. Now
1:42:51 I think it I think it broke our thing. I
1:42:54 think it's broken, but we'll see.
1:42:57 This is crazy.
1:43:03 If you're new here, if you haven't been
1:43:06 paying attention, I'm playing around
1:43:08 with image generation within Gemini 2.5
1:43:12 Flash.
1:43:14 I had Gemini generate
1:43:16 a JSON prompting framework for an image.
1:43:20 I then had it generate that image and
1:43:22 rather than just generating the image,
1:43:24 it generated a website to generate
1:43:26 images.
1:43:28 So, so I'm trying to get my head my my
1:43:32 jaw back up off the floor to understand
1:43:34 how do I even like make sense of this?
1:43:40 Watch her be a centaur. Yeah, that's
1:43:42 true. Can it make a custom ratio like
1:43:45 2.1? If we add that, I'm going to just
1:43:48 tell it it broke it.
1:43:51 Um, it looks like
1:43:54 the app is no longer
1:43:59 generating the JSON
1:44:05 prompt and there for
1:44:12 not generating the image. It hangs
1:44:19 on generation.
1:44:23 Fix that.
1:44:25 And I want to be able to add a button to
1:44:28 download it. End the prompt. I also want
1:44:32 an aspect
1:44:35 ratio
1:44:37 chooser
1:44:40 that includes
1:44:44 vertical,
1:44:46 square, and
1:44:50 cinematic
1:44:52 wide ratios.
1:44:57 All right, let's see what we get here.
1:44:59 Upper left says canvas. What?
1:45:01 Unfortunately,
1:45:03 something went wrong. Fix error.
1:45:22 HTTP error.
1:45:28 All right, it's fixing it. All right,
1:45:30 people. I'm getting I'm getting crunchy.
1:45:33 We didn't We didn't do much image
1:45:35 generation, did we?
1:45:38 Okay, tomorrow night. So, it's Thursday
1:45:40 night. I made this announcement last
1:45:41 night and I was wrong. But this night,
1:45:44 it's correct.
1:45:46 Tomorrow night is Friday night date
1:45:48 night. If you haven't been to a Friday
1:45:49 night date night, do yourself a favor
1:45:52 and show up. And you might ask, well,
1:45:55 what's different on Friday night date
1:45:57 night? I don't know. Well, if you have a
1:45:58 date, that'd be different. If I'm your
1:46:01 date, well, welcome. Uh, we do this.
1:46:05 We'll we'll be I'm going to definitely
1:46:06 be playing with Nano Banana tomorrow.
1:46:09 Um,
1:46:13 so do that. And then tomorrow at 11:00
1:46:16 a.m. Mountain time is AI office hours.
1:46:19 So, I have AI office hours. Uh, if you
1:46:22 go to my LinkedIn channel,
1:46:25 Kyle Shannon on LinkedIn, go to my
1:46:28 events, you'll see AI office hours mixed
1:46:30 in with these recordings and just click
1:46:33 on one of those and the URL there is uh
1:46:39 is the one for tomorrow at 11. Okay.
1:46:46 Enhanced error reporting,
1:46:49 flexible as aspect ratio, but there's no
1:46:54 Oh, yeah, we now have aspect ratios.
1:47:00 All right. Well, they're shitty, but
1:47:01 okay. What we can, but now we have them.
1:47:04 So, okay. So, we'll do we'll do this as
1:47:07 a square image just to test it. We're
1:47:11 going to upload an image,
1:47:14 Gabby.
1:47:16 We're going to say Gabby.
1:47:18 We're going to say rides a horse
1:47:23 in the 1800s Wild
1:47:27 West.
1:47:29 And I'm not going to I'm not going to
1:47:31 improve it,
1:47:34 but I am going to say generate the shot.
1:47:49 Hey, Pate, did you know did you know
1:47:52 that Gemini will just randomly build you
1:47:54 a website if you ask it to make an image
1:47:57 using JSON prompting?
1:48:00 Because that's what happened here.
1:48:03 This is crazy.
1:48:05 Uh, it still doesn't look like it did
1:48:08 that. So, um,
1:48:12 still breaking
1:48:18 before it generates the JSON
1:48:26 prompt.
1:48:33 I've been using a little JSON and chat
1:48:36 GPT and now it asks me if is
1:48:39 If I want JSON appendix added
1:48:42 to deep research. Oh, that's
1:48:44 interesting. I haven't had that problem
1:48:47 specifically. I don't know that it's a
1:48:49 problem paid. I think it's actually
1:48:50 really interesting. Like it it seems to
1:48:52 me like it it sensed that I wanted code
1:48:55 or that I, you know, that I had code
1:48:58 and that maybe I wanted something more.
1:49:01 But I don't know. It's kind of crazy.
1:49:04 But it's written us a pretty
1:49:05 sophisticated app here.
1:49:10 and it generates images.
1:49:14 Check out QC coder.
1:49:17 Yeah, this is crazy.
1:49:20 All right, dynamic shot generator. Your
1:49:23 cinematic shot will appear here. All
1:49:24 right, let's try it again.
1:49:27 Gabby.
1:49:30 Gabby.
1:49:33 Um, cinematic. Uh, Gabby rides
1:49:38 a Mustang horse
1:49:41 in
1:49:43 1800's
1:49:45 Wild West.
1:49:49 Yeah, that's it. That's a good prompt.
1:50:11 Oh, QCoder writes everything. You can
1:50:14 chat and it it does everything in Visual
1:50:16 Studio. It's like this. Yeah, like a lot
1:50:18 of these vibe coding platforms are
1:50:20 they're pretty pretty [ __ ] powerful.
1:50:25 Um I think this has failed again.
1:50:27 Anyway,
1:50:29 um I guess I can come back to this chat
1:50:32 if I wanted to work on this. I just want
1:50:35 it to work and it's not. So, it's
1:50:36 pissing me off now. All right, [ __ ] it.
1:50:39 Enough. Done for the night. Um
1:50:42 All right. I'm crispy. That was a peppy
1:50:45 night. Um thanks for the good
1:50:47 interactions tonight and the really good
1:50:48 questions. Um it makes these things much
1:50:53 uh it makes these evenings
1:50:57 I I think these kind of evenings are
1:50:59 much more important where we get to
1:51:01 really talk about what's going on with
1:51:04 AI and what they do and how they're
1:51:06 going to affect us. Those are really
1:51:08 good talks. So, I really appreciated
1:51:09 that tonight. All right, let me jump
1:51:11 over to the irregularize
1:51:14 and reload.
1:51:18 It looks like we've got some activity.
1:51:21 [Laughter]
1:51:23 Kevin Mallister moment. Yep. I had one
1:51:25 of those tonight.
1:51:29 That was kind of cool.
1:51:31 There's me pushing the button.
1:51:34 Thank you, Lord. Digital gods.
1:51:38 Very nice. All right, everybody. Um,
1:51:42 Givi. So, tomorrow, tomorrow is just a
1:51:46 regular Friday.
1:51:49 So, Friday night date night, 8:00 p.m.
1:51:52 Mountain time here. be there. Be square.
1:51:55 I'll give you homework for the weekend.
1:51:57 And then 11:00 a.m. tomorrow is AI
1:52:00 office hours and you should come to
1:52:02 that. Okay.
1:52:06 You start making them input to each
1:52:08 other becomes interesting what you can
1:52:11 make. Yeah. Yeah. There's all sorts of
1:52:13 cool ideas.
1:52:15 And the fact that you could just add
1:52:17 like I just hit the button and it said
1:52:19 add Gemini features to this app that I
1:52:22 just built you and it just did it. That
1:52:25 was pretty [ __ ] slick. All right,
1:52:28 we're in new territory, people. When
1:52:30 when your image generating tool starts
1:52:32 writing you an application to generate
1:52:34 images, you know you're in different ter
1:52:36 we ain't in Kansas anymore.
1:52:41 That was absolutely trippy. So anyway,
1:52:44 hope you had fun tonight and uh I'll see
1:52:46 you tomorrow. Peace out.