3/28/2025 - Exploring ChatGPT's Image Generation & Narrative's Future

Description

Chapters

Transcript

0:01 friend. Go down to that horn place
0:07 again. Do Do you know the
0:11 sounds of closing
0:14 doors? Have you heard that soundwhere
0:20 before? Do you want do they wonder if
0:23 they know you
0:26 [Music]
0:28 any more more
0:32 [Music]
0:39 [Music]
0:55 [Applause]
0:56 [Music]
1:05 Friday night date night. Good people.
1:08 What's
1:08 [Music]
1:12 happening? I'm on a little bit early. I
1:15 was bored. I came in to get ready early
1:17 and then I got ready early and then I
1:20 got
1:21 bored. I got bored,
1:24 people. I got
1:26 ADD. I got the shivers.
1:31 [Music]
1:41 Oh
1:45 [Music]
2:08 yeah. Oh yeah.
2:16 [Music]
2:28 Every time I see you
2:30 now, get that look in mine.
2:35 Every time I see your mouth,
2:38 I hear that
2:42 smile. In the early misty morning light,
2:46 I heard the engine
2:49 turning in the old for
2:54 [Music]
2:58 outside. You were leaving
3:02 me again today.
3:04 [Music]
3:06 You will convince
3:08 me again
3:12 today. You're leaving this hard time
3:15 looking for someone else's golden ring.
3:24 to
3:25 say. So long to
3:29 [Music]
3:34 say now don't you cry.
3:41 So long,
3:47 Susanna. Don't you cry for
3:50 [Music]
4:01 me. Oh yeah, champ. That is some fine
4:04 fine fine singing. Like an aged wine or
4:08 a fat dog. You're doing great. You're
4:11 doing great. You sound
4:14 good. Melodious. You got melodious
4:18 tones. Mel maleiferous. Melodious. I
4:21 don't know. I don't know words. Some
4:24 people have a way with words. Other
4:26 people not have way. I guess you'd
4:34 say. All
4:37 right. Something's different. Can you
4:39 feel it? Did you
4:41 notice? Did you notice something has
4:44 [Music]
4:47 changed? There's more energy. People are
4:50 excited again.
4:53 [Music]
5:24 Let us be lovers. Well, marry our
5:27 fortunes
5:30 together. I've got some real estate here
5:33 in my bag.
5:36 So we bought a pack of
5:38 cigarettes and Mrs. Wagner's
5:41 pass and we've all come to look for
5:45 America.
5:47 [Music]
5:58 [Applause]
6:01 [Music]
6:06 [Applause]
6:07 [Music]
6:15 It's like I have precision with
6:17 fingerpicking minus the precision.
6:23 [Laughter]
6:30 Happy Friday night everybody. I hope
6:32 you're
6:32 [Music]
6:36 well. We're going to continue playing
6:38 with
6:39 uh Chat GBT. Tonight
6:46 [Music]
7:14 Oh, Champ, you're such a good singing
7:16 dog.
7:22 [Music]
7:32 [Music]
7:35 Woohoo! Woo!
7:38 [Music]
8:06 Um, what's happening? What's happening?
8:09 What's happening? What's happening?
8:10 What's happening?
8:15 I had a bunch of good meetings today. We
8:18 had a really nice AI office hours. Saw
8:20 some old friends. That was really nice.
8:22 It was small. It was a smaller group
8:24 today, which was kind of cool. It was It
8:26 was a small intense
8:29 group. Very well talked. It was a very
8:32 well It was a very very smart group of
8:34 people that showed up today. And then I
8:37 was there too.
8:41 [Music]
8:43 [Applause]
8:47 Um, but yeah, we were talking about
8:50 this.
8:51 Uh, something is shifting. I think we've
8:54 crossed a threshold. I think we've
8:56 crossed a
8:57 Rubicon. I don't know what a Rubicon is,
9:00 but I think we've crossed
9:01 it because I know you cross them in
9:04 movies. Thank you, Spin B3. I missed it
9:07 again. Oh, dang it. chemical
9:10 farmer. I always forget until like 2 PM.
9:13 Yeah, it's at 11:00 a.m. Mountain, so
9:16 it's a little early if you're on the
9:17 West Coast. Purple Rain. That was Purple
9:21 [Music]
9:26 Rain. Wait.
9:31 [Music]
9:46 Only want one time to see you
9:51 laughing. Only want to see you laughing
9:54 in that purple rain. Purple rain. Purple
10:01 rain. Purple rain. Purple
10:07 rain. Purple rain. Purple
10:12 [Music]
10:15 rain. I only want to see you laughing in
10:20 that purple.
10:24 Yeah. And then from that point forward,
10:26 Martin Ston like sings really good.
10:30 That's where I stop.
10:36 You got to know your limitations,
10:37 people. Just go right up to the edge of
10:39 them and then sneak over occasionally
10:41 and just go, "Yeah, that's not
10:43 [Laughter]
10:47 me." Although, you know, with this AI
10:51 [ __ ] I think we should all be a bit
10:54 more bold in what we think we can do.
11:01 [Music]
11:13 Um, let's let's go look at some
11:15 stuff. Let's go look at some
11:19 stuff. What's going on? How is
11:21 everybody? I got no comments going on on
11:24 YouTube. What the hell's going on on
11:25 YouTube? No thumbs up, no likes, no
11:27 comments. the hell you people doing over
11:30 there? Probably just a bunch of people
11:32 on X going, "Who? Why is there an old
11:34 man on X? What's happening? What's
11:38 happening? I thought all the old men
11:40 were on TikTok. I don't understand it.
11:43 What's
11:44 happening?" Um, sounds good, but Champy
11:48 always helps. He He howls good. Yeah,
11:50 exactly. I mean, that's the thing. When
11:52 you have when you have a lead singer as
11:55 a
11:55 dog, you can fill in the gaps. Just let
11:59 him take the
12:04 lead. Uh, there are comments on YouTube.
12:09 Really? I ain't seeing
12:13 them,
12:14 huh? Oh, there they are. I had a I had a
12:19 screen redraw issue.
12:22 Well, let me see what we got
12:25 here. I watched the Bob Dylan tonight. I
12:28 prefer watching Kyle. Thank you. Well,
12:30 you know, it's so funny you say that,
12:32 Vicki. A lot of times people are like,
12:35 you know, do you like Timothy Shalomé?
12:38 And and people like, he's no Kyle
12:41 Shannon. Like, I hear that all the time.
12:44 They're like, yeah, it's not the AI
12:46 learning lab. I mean, sure, he took five
12:48 years of his life and transformed
12:50 himself into Dylan, but you know, is is
12:53 there a singing dog? I don't think so.
12:56 Yeah, I hear it all the time. Shalom or
12:59 Shannon? Which way you going? And I
13:01 listen, I can't help it if people lean
13:05 my direction. Right. It's just it's a
13:07 magnetism I've got. It's a natural
13:10 genetic gift that I have been sprinkled
13:13 on by the gods. So anyway, thank you for
13:16 the for the kind words.
13:21 Uh ah the chat servers must be working
13:26 overtime. Oh, it took several tries
13:28 until I got my message. Another thing
13:30 that could be happening, Sharon, um one
13:33 of the things that they have started to
13:34 do is they have started to censor the
13:37 images.
13:39 What's fascinating is that they censor
13:41 them about threequarters of the way
13:43 through building an image and then they
13:45 realize, oh, this isn't going good.
13:46 Let's just delete it. So, that's a
13:48 possibility,
13:54 too. I'm not the biggest Adam Lavine
13:56 fan. If you hear his purple rain live
13:58 from the Stern party. Oh, cool. Nice.
14:01 We're here. Okay, good. Everybody's
14:03 here. I couldn't see anyone. Chef's
14:05 kiss. Very nice.
14:06 Um, I did not enjoy the music. I agree.
14:09 AI learning lab much better matter.
14:10 Shannon all the
14:12 way. You guys are [ __ ] weird. We're
14:14 not weird. We're irregular. I thought we
14:18 established that
14:19 yesterday. Good
14:24 lord.
14:25 Okay, let's get going. Um, yes, that
14:29 happened to me today. My chat art. I was
14:32 very disappointed. Yeah. Where where it
14:33 deleted it after it generated it. Yeah,
14:36 they're they're doing that
14:37 now. They have an algorithm in there
14:40 that says, "Did this make an image that
14:43 we might get sued by a big company for?"
14:46 And if the answer is yes, they delete
14:47 it. It
14:52 sucks. AI Learning Lab is the favorite
14:55 hangout in Shellington. If you weren't
14:57 here last night, we invented a
14:59 city. We've got the the Irregular Turtle
15:02 Society. If you're not part of
15:05 that, you weren't here last night. And I
15:08 got no sympathy for you. Here's the
15:11 deal. You people, you people, you you're
15:13 like, "Yeah, I want to learn AI." And
15:16 and then you're like, "Well, I'll come a
15:18 night a week or once or twice a
15:21 month." You miss
15:24 out. And I mean, listen, I think there's
15:27 a legitimate argument to be said
15:30 that a solid 90% of the hours of of
15:35 entertainment that I procure here are
15:38 useless, wasted time, you'll never get
15:40 back in your life. But it's that 10%,
15:43 it's the [ __ ] magic. It
15:47 happens. It's like a baseball game.
15:50 Occasionally, something happens, you
15:53 know, then you drink beer in between.
15:59 Oh
16:01 man. Um, Chef Kelly, I was embarrassed
16:04 after my botched virtual webinar today
16:07 and then I remembered you do this every
16:09 day. Yeah. Listen, botched. I assume it
16:13 was not massively botched. Maybe it
16:15 didn't quite go as planned. But this is,
16:18 think of this as inspiration for Well,
16:21 it couldn't be that bad.
16:25 Now
16:26 listen, to be fair, you don't have a $20
16:30 million production studio like I do,
16:33 right? Like it's not going to look this
16:35 good, right? But just, you
16:39 know, and sure, this is fully scripted,
16:43 right? I spend most of my days learning
16:45 the lines for the performance of the
16:47 night. This show is very much like the
16:50 uh that Netflix series
16:51 Adolescence, right? where I memorized
16:54 the script. It's a single camera
16:56 take, you know, dramatic action happens.
17:00 We produce this in such a way that it
17:03 looks
17:04 like I'm just
17:10 insane.
17:15 [Music]
17:17 See? All right, let's go look at some
17:19 [ __ ] source camp in
17:27 house. Did I not share? What's going on
17:30 here? Okay, there is just weird ass
17:34 stuff going on with my
17:36 StreamYard.
17:38 Um, okay. So, here I am. I'm at I'm at
17:41 the chat GPT. I don't know if you all
17:43 know chat GPT. If you go to
17:46 chatgpt.com or even just chat.com,
17:49 uh you might have heard you can make
17:51 money with chat. This is the place you
17:54 go. You just go there and and when you
17:57 type in a prompt, you can make money
18:00 from
18:01 that. And you're like, "Well, how do I
18:03 make money from that?" Well, you pay me
18:05 the
18:07 $49.95 a week for the secret lesson.
18:13 That's how you make money from it. Well,
18:15 no. That's how I make money from it. It
18:18 And then but then what you do is you
18:22 take what I did and then you do
18:25 that. But it's not easy. Like you have
18:28 to be able to go you can make money with
18:30 chat. You you can't just say you can
18:33 make money on chat
18:35 GBT. It's like it doesn't it just
18:38 doesn't work unless you're like
18:42 at right. And that's what I'll teach
18:45 you. I'll teach you the secrets to
18:47 artificial intelligence are making a Tik
18:49 Tok channel to sell methylene blue and
18:53 and Madagascar vanilla beans and
18:55 cardboard cutters. That's how you make
18:58 money with chat GPT. The Tik Tok
19:01 shop. You see it? Like is it all
19:03 starting to come together? Because it
19:05 should it should be really
19:08 clear. No.
19:13 I I I I don't know what to say. Like if
19:15 you're thick in the
19:17 head, it's that's not on me. You're
19:20 like, "But Kyle, you told us we didn't
19:22 have to be technical." You don't have to
19:24 be technical, but you have to learn how
19:25 to say words. Words like
19:30 cha, you
19:32 see? Anyway, um, chat
19:40 GPT. Oh, good
19:43 lord. All the regulars are like, "All
19:45 right, I got 15 minutes here. I got to
19:47 run to the
19:48 [Laughter]
19:53 bathroom." Ah, yes. Words. I I am saying
19:57 words, Tom. I words. Words, words,
20:01 words, words,
20:05 words. Um, I did start watching Sing
20:09 Sing tonight, the the uh the movie about
20:12 the uh the prison theater troop. It's
20:15 pretty good. It's pretty good. I'm like
20:16 halfway through it.
20:18 Um, I wanted to like one of the things I
20:21 wanted to explore
20:24 tonight
20:26 is you know how when you're using chat
20:29 GPT one of the secrets of it is is how
20:32 you say it and the other secret
20:36 is it's not so much that you get good at
20:39 prompting but it's more like you get
20:41 good at conversation with chat GPT right
20:44 you start to understand its personality
20:46 and you start to understand how you and
20:48 give it
20:49 feedback and if you do the conversations
20:51 right you can
20:53 maintain a thread of understanding
20:56 through a long conversation right and
20:59 that's like I feel like when you get
21:01 good with chat GPT it's kind of like
21:03 this ongoing conversation that you just
21:05 keep improving it and improving it
21:07 improving it and where it gets
21:08 frustrating is you improve it improve it
21:10 and then it falls apart
21:12 right and so now that image generation
21:16 is native within in the platform. I was
21:19 like, can you maintain that same kind of
21:23 um context and continuity across images?
21:27 And so, so I just put this prompt up. I
21:31 said, "Make me a blueprint of a radical
21:33 design for a new kind of house." And I
21:36 was expecting it to draw a picture.
21:39 And and what it did, which I thought was
21:43 kind of interesting, is it described the
21:46 house. So what did it say? A living,
21:50 learning, breathing structure that
21:51 adapts the spiral nest that adapts to
21:54 its occupants and environment. It's part
21:57 bioarchchitecture, part digital habitat,
22:00 part sensory cocoon. A house that feels
22:04 you thinking and responds in kind. Um
22:07 and then they describe the different
22:09 areas. Its shape is a spiral shell
22:12 inspired by the Fibonacci sequence. The
22:15 level one is called the root zone. Level
22:17 two is called the mind deck. Level three
22:20 is called the green loop. Level four is
22:22 the dream
22:23 nest bonus the morph room. So that was
22:27 kind of cool. And then I said, well, you
22:31 know, make me a blueprint of the floor
22:33 plan. But I said floor plan, but
22:35 whatever. And then it wrote
22:39 code and it wrote like a Fibonacci
22:42 spiral and then it and then it did like
22:45 Python drawing which is not at all what
22:48 I wanted but it's kind of
22:52 interesting and it it's kind of an
22:55 interesting floor. It's like it's
22:57 interesting that it that it it thought
23:00 about the math of the house, right? So
23:04 then I said, "No, no, make me a like a
23:06 blueprint as if an architect drew
23:09 it." And it did it. And it essentially
23:15 maintained the same thing, right? It's
23:17 got the
23:19 spirals. It's got a It's got an area
23:21 called the digestion zone. I don't know
23:24 if that's a bathroom or if that's if if
23:28 you just walk in circles here after you
23:32 eat. And then I said, "Now, let me see a
23:34 watercolor visualization of it." And it
23:38 did
23:39 that. It's pretty [ __ ] good. Like
23:42 it's in the neighborhood of what an
23:44 architect would
23:46 do. It's got a floating bridge and the
23:49 entrance and the entry garden. Now, it's
23:52 not complete. Like, this was all very
23:54 fast and it's got plants inside the
23:57 house and there there's no actual
23:59 kitchen where you can cook [ __ ] There's
24:01 just a table, but still pretty cool.
24:04 Then I said,
24:07 um, now make it a photograph as if it
24:10 were taken for Architectural
24:12 Digest. And then there's there's the
24:15 thing. And it maintains the foot bridge.
24:18 It maintains the pond. It maintains the
24:20 table, the central spiral staircase, the
24:23 spiral shape of the
24:27 house. And then one of the things I
24:29 heard is you can ask it to give you the
24:32 prompt of an image. And so I said, "Hey,
24:34 can you give me the prompt of the
24:36 image?" And what it wrote was, "This
24:38 this stunning highresolution photo
24:41 captures a con contemporary
24:43 spiral-shaped home with curved glass
24:46 walls set against a vibrant green
24:48 landscape. Sleek concrete roof spirals
24:51 upward, and the floating bridge leads to
24:53 the entrance." But it like didn't feel
24:55 like it had enough detail to
24:58 generate all that. And so then I asked
25:02 it. I said,
25:05 um, it seems like there would be more to
25:07 the prompt than that. Did it incorporate
25:10 incorporate the previous watercolor
25:13 image to get the shape of the house? And
25:15 it said, "Great question. You're totally
25:18 on
25:18 point." The photo style image did draw
25:21 on the previous watercolor image
25:24 indirectly by carrying forward the
25:27 visual memory of the design. Now, here's
25:30 what I don't know. This is chat GPT. So
25:33 it's possible that it absolutely
25:35 hallucinated this. But if it's true,
25:38 what this is saying is it's it
25:43 understands this code. It understands
25:46 this drawing. It understands this
25:48 drawing. It took that drawing to make
25:51 that image. It took that image to make
25:53 that
25:54 image. And then it said, "Do you want to
25:56 see the interior?" And I'm like, "Sure,
25:58 show me the interior." here. And there's
25:59 like there's the interior, I guess, out
26:01 the back of the
26:02 house. I mean, that's pretty [ __ ]
26:05 cool,
26:07 right? It's a tur It has a turtle ramp.
26:09 It totally has a turtle
26:16 ramp. But, you know, I thought that was
26:20 pretty [ __ ] swell.
26:25 I saw Kelly Camp did a uh a magazine
26:28 piece today here. Let me see if I can go
26:30 find
26:33 [Music]
26:39 it. So
26:43 good. There's Kelly's little
26:49 magazine. The new MVP, The Quiet
26:52 Revolution.
26:53 Beyond the buzzwords or buzz whirls as
26:56 chat GBT made it. AI isn't coming. It's
27:00 here. That's good. The human interface
27:02 to AI architect. That
27:05 cool. So if you don't know how she made
27:08 that, you can just go to chat GPT right
27:11 now.
27:13 Uh you can do three of these on the free
27:15 plan. I don't know if they've rolled it
27:17 out to the free plan yet, but for 20
27:19 bucks a month, you can upload a picture
27:21 of yourself and just tell it to put you
27:23 on the cover of a magazine and then
27:25 describe the magazine and let it write
27:27 the headlines or give it the
27:32 headlines and it does everything. It
27:34 does the
27:41 typography. So, that's pretty [ __ ]
27:43 swell. So, we'll we'll come back to chat
27:46 GBT because I I want to keep I want to
27:49 keep digging
27:51 into what's now
27:54 possible. I also want to see if we can
27:58 discover more about the upgrade of 4.0
28:00 and why it's good and how it's good and
28:02 what's different about it. Um, if anyone
28:05 has any thoughts, feel free to put them
28:06 in the comments.
28:09 Careful with thought. You'll invoke the
28:11 wrath of
28:16 Pate. I'm waiting for the day when it
28:19 says, "Great point, but you're totally
28:20 wrong, human." I, you know, I'm actually
28:23 waiting for that day,
28:25 too,
28:29 because I appreciate the fact that they
28:31 made Chat GPT a helpful
28:34 assistant, but sometimes it would be
28:36 nice to have it be a helpful critic,
28:39 right? It'd be nice if it could be
28:41 intelligent enough to navigate those two
28:44 roles. What's this
28:46 say? It carries things forward. Just try
28:50 creating a normal person and then
28:51 creating a Gibbli person. That's true.
28:53 Yeah, it's doing character consistency
28:55 that way. Um what I wanted to um watch
29:00 together. I only watched a little bit of
29:02 this. I'm going to I'm going to switch
29:04 my my tabs for a second here.
29:08 Um, Anthropic put out a big paper today,
29:13 um, and a blog post about how large
29:16 language models work, how they
29:19 think. Um, and I saw a couple of people
29:22 mention it as being a really profound
29:24 piece with a lot of jaw jaw-dropping
29:26 moments. Um, and I haven't I haven't
29:29 read it, so like I just saw it late in
29:31 the afternoon, but it's got this video
29:34 with it that that explains some of how
29:37 large language models work and how they
29:39 think. Um, and I thought it'd be fun for
29:41 us to watch it together. So, we'll do a
29:43 little movie night tonight. So, get your
29:46 popcorn. Go make your Jiffy Pop. Well,
29:50 Jiffy Pop, boy, that's that's old and
29:53 timey. What What What is the popcorn? Is
29:56 it all Is it Newman's own or is that
29:58 still old? What's the What's the popcorn
30:00 you're making in the microwave these
30:02 days? Or did we give up on microwave
30:05 popcorn and we're back to the stove? Are
30:07 we all bespoke
30:09 artisal
30:11 Oregon-based
30:13 handpicked every kernel massaged before
30:16 it's placed in the bag? It Where are
30:18 Where are we with popcorn? Or do you
30:20 just buy the bags of [ __ ] now? Done.
30:24 pre-done. Orville Red and Bucker. Okay.
30:32 Waffles. Waffles. Waffles are good, too.
30:35 I say we go waffles or
30:37 [Laughter]
30:42 popcorn. Oh, man. All right. Let me
30:45 change my let me change my tab
30:49 sharing and then we will watch a we'll
30:52 do a little movie
30:55 [Music]
31:08 watching. You often hear that AI is like
31:11 a black box. Words go in and words come
31:15 out, but we don't know why it's said
31:17 what it said. That's because AIs aren't
31:20 programmed, but trained. And during
31:23 training, they learn their own
31:24 strategies to solve problems. If we want
31:27 AIs to be as useful, reliable, and
31:30 secure as possible, we want to open up
31:33 the black box and understand why they do
31:35 things. But even opening the black box
31:39 isn't very helpful because we don't know
31:41 how to interpret what we see. Think of
31:44 it like a neuroscientist investigating
31:46 the brain. We need tools to work out
31:48 what's going on inside. We want to know
31:51 how the model connects all the concepts
31:53 in its mind and uses them to answer our
31:56 questions. Now, we've developed ways to
31:59 observe some of an AI model's internal
32:02 thought processes. We can actually see
32:04 how these concepts are connected to form
32:06 logical
32:07 circuits. Let's take a simple example
32:10 where we ask Claude to write the second
32:12 line of a poem. The poem starts, he saw
32:16 a carrot and had to grab it. In our
32:19 study, we found that Claude is planning
32:21 a rhyme even before writing the
32:23 beginning of the line. Claude sees a
32:26 carrot and grab it and thinks of rabbit
32:30 as a word that would make sense with
32:32 carrot and rhyme with rabbit. Then it
32:34 writes the rest of the line. His hunger
32:37 was like a starving rabbit. That's kind
32:39 of fascinating. I mean, it's like the in
32:43 in that first line, the valuable things
32:46 are carrot and grab it. He saw and had
32:50 to are not as important. So, I picked
32:52 the more important stuff. Kind of
32:55 cool. It's also when you write poems,
32:58 this actually makes sense. Now, it's
33:00 very hard to get it to write an
33:03 abab rhyming couplet. It always goes a
33:07 bb. So, it does them two lines at a
33:10 time. Be interesting to experiment and
33:11 see if you could prompt it out of
33:14 behaving like this.
33:16 We look at the place that the model was
33:20 thinking about the word rabbit and we
33:21 see other ideas it had for places to
33:25 take the poem. We also see the word
33:28 habit is present there. Our new methods
33:31 allow us to go in and intervene on this
33:33 circuit. In this case, we dampen down
33:36 rabbit as the model is planning the
33:38 second line of the poem and then ask
33:40 Claude to complete the line again. His
33:42 hunger was a powerful habit. We see that
33:45 the model is capable of taking the
33:47 beginning of a new poem and thinking of
33:49 different ways it could complete it and
33:51 then writing it towards those
33:53 completions. The fact we can cause these
33:56 changes to occur well before the final
33:58 line is written is strong evidence that
34:00 the model is planning ahead of time.
34:02 This poetry planning result along with
34:05 the many other examples in our paper
34:07 only makes sense in a world where the
34:09 models are really thinking in their own
34:11 way about what they say.
34:14 Just as neuroscience helps us treat
34:16 diseases and make people healthier, our
34:18 longerterm plan is to use this deeper
34:21 understanding of AI to help make the
34:23 model safer and more reliable. If we can
34:26 learn to read the model's mind, we can
34:28 be much more confident it is doing what
34:30 we intended. You can find many more
34:32 examples of Claude's internal thoughts
34:34 in our new paper at
34:36 anthropic.com/ressearch.
34:41 Um, yeah, that's fascinating. I mean,
34:44 this kind of blows the doors off
34:47 of, you know, it's just predicting the
34:50 next token. It's just a stochastic
34:53 parrot, right? You know, if it's
34:55 thinking ahead and like I don't know how
34:58 long it's thinking ahead, but that's
35:01 pretty interesting. And and don't forget
35:05 anthropic, their core imperative. So,
35:09 OpenAI's core imperative is to create
35:11 AGI, AGI, artificial general
35:14 intelligence, right? They want to make
35:15 it accessible to the world. Enthropic
35:18 was a spin-off of Open AI from some of
35:22 the main scientists from Open AI that
35:24 said, "We don't like the way you're
35:25 thinking about safety of these models.
35:27 We want to create an enthropic where our
35:29 prime imperative is safety, AI safety,
35:33 and and AI alignment." Um, and so that's
35:36 that's why this research makes really
35:40 good sense that they're doing it. Um,
35:42 but I find it fascinating that basically
35:45 what they're discovering is these things
35:47 actually think. Now, I don't think you
35:51 can go as far as, you know, they're
35:53 conscious, but, you know, this is this
35:56 is a different thing. And I'm sure if
35:58 Pate's in here, he's probably screaming
35:59 at the top of his lungs right
36:01 now. Um, let's see. Where's the paper
36:04 here?
36:06 [Music]
36:07 Um, probably at the bottom. Is there any
36:10 Yeah, there's lots of charts and graphs
36:12 that I I that won't make good won't make
36:15 good for tick
36:16 tocking. Um, but the the blog post is
36:21 pretty comprehensive and then I I assume
36:23 the paper's really good. Let's see.
36:31 revealing the computational graphs in
36:34 language
36:36 models. Many more details of the above
36:38 use cases in our second paper on the
36:41 biology of large language model. Wow,
36:46 that's a pretty bold that's pretty bold
36:49 statement.
36:50 We investigate the internal mechanism
36:53 used by claude 3.5 haiku enthropics
36:56 lightweight production model in various
36:59 contexts using our circuit tracing
37:02 technology. That's
37:06 amazing, huh? So anyway, go check that
37:09 out. It's at uh it's at uh
37:14 anthropic.comressearch. I would imagine
37:15 you can probably just get there by that.
37:18 Anthropic.comressearch. Hang on.
37:23 Yeah. Uh maybe not. Uh let's see. Our
37:33 mission research principles tracing the
37:36 thoughts. So here it is. So it's you
37:38 scroll down. Go to
37:41 enthropic.comressearch and scroll down
37:42 till you see the rustcoled box. Tracing
37:46 the thoughts of a large language model.
37:47 That'll take you back to that where I
37:49 just
37:51 was. Crazy. It's crazy, I tell you.
37:55 Crazy. Crazy. Crazy. All right. Um,
37:59 let's see. Did I do anything
38:05 in in chat GBT today? That was
38:08 interesting. Oh, that was an interesting
38:11 one.
38:13 Yeah. Um, this was kind of fun. So, I
38:18 took a screen sh Oh, wait. I got to I
38:20 got to change my sharing. Hang
38:25 [Music]
38:32 on. So, today I was on a uh conference
38:37 call. Can you all see that? Hey, Andy,
38:41 what's happening? Um, today I was on a
38:44 conference call. I took a screenshot of
38:46 some of the crazy people there and then
38:50 uploaded those pictures and said, "Make
38:51 me a cartoon." And one of the cool
38:54 things that it did was it took the names
38:58 off the screenshots and turned it into
39:01 like the names of the people. So there's
39:03 Cindy [ __ ] there's Rick Lacroy, who's
39:05 not really that angry. There's me with a
39:08 guitar. It picked the guitar up out of
39:09 my background, which is interesting.
39:20 Um, then I tried to add someone to it
39:22 and it sort of screwed that up. And then
39:24 I did something and it sort of screwed
39:26 that
39:27 up. But then I I said, you
39:31 know, here's more people. Have the
39:33 people with their cameras on, make fun
39:35 of the people with their cameras off. So
39:37 that sort of did it well. I said, make
39:39 it like a Pixar thing. So, it kind of
39:42 did a funny little
39:44 cartoon. Let's do image gen projects.
39:47 Yeah, we're going to sitting here with
39:49 my computer machine making stuff. Yeah.
39:52 So, so yeah, let's go play with this.
39:54 So, so upload a picture of yourself or
39:57 your friends. Oh, then I said this was
39:59 kind of cool. I said take Eric who I put
40:03 here based on a screenshot here. Let's
40:06 see. Where's
40:07 Eric? Right here in the top. Let me see.
40:10 Can I zoom in on
40:12 that? There's Eric. So that's that's the
40:14 only picture it had of Eric,
40:18 right? And then from
40:24 that it
40:26 made this version of
40:29 Eric right
40:37 here. And then I said,
40:45 um, create a dramatic art photo of Eric
40:48 where the photographer used lighting and
40:51 a low millimeter lens and open aperture
40:53 to capture a personality that is
40:56 thoughtful and generous. Make it a 2x3
40:58 portrait. And so it made this
41:00 photograph. It sort of invented this
41:03 guy. And then I said, make it black and
41:06 white. And it did.
41:08 And then I said, "Make it grittier."
41:10 Made it really
41:13 gritty. So, it's pretty
41:16 crazy. Um, what else do we do
41:22 today? The studio Gibbly thing. Oh,
41:25 yeah. This was pretty cool. So, this was
41:29 I wanted to do I wanted to do take my my
41:33 stupid
41:35 uh profile photo and turn that into a
41:38 character. It made me too fat. So, I
41:39 said, "Make me skinnier because I'm
41:43 vain." And then I said, "Make me a
41:45 Macintosh computer, the original
41:48 Macintosh, and convert my portrait into
41:50 Mac Paint, the original version of Mac
41:53 Paint." And it did it quite well. So,
41:55 that's a good
41:57 one. Um, I did some I did a New Yorker
42:00 cartoon today. I had it write New Yorker
42:03 cartoons. It came up with 10 different
42:05 cartoon
42:06 concepts and then it created this this
42:09 one that is a bunch of goats sitting
42:12 around a boardroom and the the quote is,
42:15 "Let's circle back after we've eaten the
42:20 agenda." That's pretty good. That's
42:22 pretty good New Yorker humor. Come on.
42:24 [Laughter]
42:29 Oh man. Um, one thing that I did notice,
42:32 let's go to Sin City Sipper. I don't
42:35 think this has changed, but let's say
42:37 make me a margarita with unexpected
42:41 sass.
42:49 And now it's making an image, but it is
42:51 not
42:53 using the new image gen tool. So your um
42:57 custom GPTs have shitty old dolly in
43:02 them. You know, it's still pretty good.
43:05 Not for nothing,
43:08 but but it's not the new image model. So
43:11 that I
43:12 learned.
43:15 Um, last night we were playing
43:22 with turtle meme character ideas. Was
43:25 this
43:27 it?
43:28 Yeah. So, last night we created the the
43:33 Irregular Turtle
43:39 Society and we were taking pictures of
43:42 people and making versions of them. And
43:44 then people went to the AI salon,
43:47 started making their own pictures of
43:50 themselves. So, there's Silver Fox with
43:53 a chrome shell, which I love. I think
43:55 it's genius. S Fox, the founder of
43:59 Shellington, Silverf Fox,
44:01 congratulation. I didn't realize you
44:02 were the founder of the town. That's
44:07 funny. Oh, you got that same drink pick
44:10 about three months ago. Oh, that's
44:12 funny. Um, here's Sharon
44:15 Crawford. Very nice. So, everyone's
44:18 making their own little Oh, that's nice.
44:20 Who's that from? Steo. I can't make it
44:23 tonight lunch date with this one. Oh,
44:25 that's the one he uses to tap all the
44:27 hearts on the Tik Tok. Very
44:29 nice. Marlene
44:31 Paul, love it. Oh, that one's cool from
44:36 W. Very
44:39 formal. Ocean.
44:43 Nice. Corin Mueller. Do we all represent
44:47 ourselves with turtles?
44:50 Apparently, you are in Shellington, but
44:53 in Shellington. Now, Dr. Jay decided to
44:56 go with the the Blue Jay, and I think
44:58 Cindy [ __ ] would obviously go with the
45:01 raccoon. Um, there's Vicki. Vicki went
45:04 ahead and made a collectible that you
45:07 can you can buy at Toys R Us.
45:15 Oh my
45:17 god,
45:18 there's
45:23 Greg. Oh, good lord. Oh, this is a good
45:26 one from
45:27 Jackie. Still
45:29 working. Still work shopping my
45:31 Shellington peron persona, but I had a
45:33 bit of issue in the
45:35 [Laughter]
45:38 lab. So anyway, so that's kind of fun.
45:44 Um, I think what might be kind of cool
45:47 tonight, we'll we'll come back to the
45:48 Shellington stuff, but I wanna I want to
45:51 try something new.
45:57 Um, let's turn on web search and let's
46:00 say
46:05 um research
46:11 um
46:14 papers and
46:17 books on
46:21 modern narrative theory.
46:25 and what
46:27 resonates with today's
46:33 um
46:37 audiences. So, I'm going to have it go
46:39 do some research. Now, you could do this
46:40 with some deep research, but one of the
46:43 groups that I'm working with were
46:44 playing around with deep research, and
46:46 Cindy [ __ ] kind of inspired this
46:49 today where Cindy went and did a bunch
46:51 of research and then she's she's digging
46:53 deeper and
46:54 and doing research from different points
46:58 of view and using personas and things
47:01 like that. And one of the things that we
47:03 talked about was
47:06 um could you take research and bring it
47:10 to life
47:12 uh using the image generation tool? We
47:14 might be able to do that. So there's
47:16 some books and things. So now I'm going
47:17 to say
47:21 um what
47:23 are the
47:30 main narrative ideas
47:34 across these
47:38 works. Searching the
47:42 web. Oh, she's here. Cindy [ __ ] is here.
47:46 So, I figured I'm going to try to do
47:47 some some crazy ass image generation
47:50 like like like bringing research to
47:52 life, but I I don't want to do deep
47:54 research. I'm just going to do like, you
47:56 know, sort of faker deep research. All
47:59 right. So, there's that. So, I'm going
48:00 to say um now look at these ideas from
48:07 four
48:08 [Music]
48:10 different points of view.
48:16 um
48:18 each from a different
48:21 decade. I'm just I'm just kind of
48:24 randomly trying to put enough context in
48:28 here that we've got something to work
48:30 with. So, I don't really care what the
48:31 answer is here. If you were doing actual
48:34 research, you
48:39 might would you like me to explore how
48:41 these perspectives might apply to a
48:44 particular format?
48:49 Um, I
48:51 want
48:53 strong POV pieces from each of
48:58 them about the
49:01 research you have
49:10 found. All
49:13 right.
49:22 Uh, now I
49:25 want
49:27 a new
49:32 novel
49:35 hypothesis
49:39 that
49:42 incorporates everything we've talked
49:54 about
49:56 that could
50:02 reolutionize
50:05 narrative in the future. All
50:09 right. Absolutely. Here's a novel
50:12 hypothesis. the narrative as emerging
50:14 emotional architecture and multi- aent
50:16 space. You know, Cindy, it's funny. I
50:19 think you and I the other day, we've
50:21 we've been talking about narrative as an
50:23 emerging emotional architectural in a
50:26 multi- aent space. We talk about that a
50:28 lot. Like it's a fairly obvious I don't
50:30 know how novel that is. Seems pretty
50:32 obvious to
50:33 me. Narrative as a collab collaborative
50:36 ecosystem, emotion as an organizing
50:38 principle. Okay, so it's got some points
50:40 example. All right. Whatever. So, we got
50:43 that there. So, now we're gonna go
50:46 um
50:49 uh now I want you to
50:56 um play
50:58 out the
51:02 transer mason to this
51:06 new narrative reality. and write a
51:12 report
51:14 on the
51:17 impact of the current
51:21 media
51:25 ecosystem. All right, that's kind of
51:28 cool. So, like let's pretend this was
51:30 actual all real real research, right?
51:33 This is the kind of thing where you
51:35 might go look at some [ __ ] then you
51:37 might look at it through some different
51:38 lenses, then you might come up with new
51:41 ideas. Then if you're a future caster
51:43 like Cindy
51:46 [ __ ] you might say, "Well, what's this
51:49 going to do to the
51:51 world?" So, let's see what did it
51:55 do. Executive summary. Between 2025 and
51:59 2032, the global media ecosystem
52:02 underwent a profound shift from static
52:05 author ccentric content to participatory
52:08 emotionally responsive narrative
52:10 systems. This transition sparked by
52:13 advancements in AI, biometric feedback
52:15 and immers and immersive environments
52:18 has catalyzed a new narrative paradigm
52:20 from content to
52:24 experience. New formats and fluid story
52:27 worlds. The death of the singular
52:31 author. The author is a conductor, not a
52:34 god.
52:36 The 2032 Emmywinning story story world
52:42 ashfall has over two million
52:44 acknowledged co-authors. This is kind of
52:46 cool. Okay,
52:49 great. Now, I want
52:52 [Music]
52:54 you to give
52:57 me three concepts for how we
53:02 might visualize
53:10 The new world of
53:16 narrative that
53:21 includes the
53:24 impact on
53:26 current
53:28 media into a
53:31 single infographic.
53:37 that will
53:38 make
53:40 it
53:43 instantly
53:46 understandable and bring the
53:49 research to
53:51 life.
53:53 Okay, so let's see. Okay, the story
53:57 wheel from broadcast to biometrics. I
54:00 don't like it as a wheel because I don't
54:01 think this is going to be linear.
54:03 A radial infographic showing the
54:05 evolution of narrative as a timeline,
54:07 but instead of progressing linearly, it
54:10 radiates outward from a central point.
54:13 Narrative 1.0 linear broadcast toward
54:16 the new emergent narrative 4.0. Okay,
54:19 that's not bad. So, it's not in a cycle.
54:22 It's radiating
54:24 out visual structure. The center is the
54:27 20th century linear
54:29 narrative. Maybe we do all of these.
54:32 We'll we'll we'll see which one we like.
54:35 The living narrative web, an artwork
54:37 visualization like a neural map that
54:39 shows how different media nodes are now
54:44 interconnected. And then the narrative
54:46 machine, how a story is made.
54:49 Now, this is kind of cool. All right.
54:56 Um why
54:59 don't Oops.
55:05 Why
55:07 don't
55:08 [Music]
55:10 you take a stab at number one?
55:25 [Music]
55:31 Is this making sense? Am I Am I
55:35 totally off my rocker? Is this making
55:41 sense? I'm here slow to find the chat.
55:51 day. So, one of the things Cindy was
55:54 saying today, which I thought was really
55:55 or no, it was actually M Turner, but
55:58 then Cindy commented on it. M. Turner
56:00 was talking about
56:04 um you do all this research and a lot of
56:08 times it doesn't have an impact because
56:11 people can't
56:13 instantly get it.
56:20 And like and like could this
56:23 new visualization tool be a way to take
56:27 research and and turn it into things
56:29 that people get
56:36 easier. Co-authorship spatial
56:40 worlds narrative 4.0 know linear
56:43 co-creation, co-authorship, spatial
56:46 worlds, dynamic cannon, AI agents,
56:49 emotional intelligence. So that I don't
56:50 quite get. So I'm going to say I think
56:54 it is flat and
56:58 boring and lacks clarity.
57:03 Wait. Flat and boring and boring and
57:08 lacks
57:12 clarity of
57:14 the
57:19 evolution. Make it have
57:22 a dark
57:25 background and a much
57:28 more
57:31 sophisticated color.
57:36 pallet and info
57:43 design. Don't be
57:48 lazy like me, you poopy head.
57:54 And again, it's always good to swear at
57:56 your large language
57:58 model and put them down and judge
58:02 [Laughter]
58:11 them. Oh, maybe convert it to a kids
58:14 book. Oh, that's pretty cool. Like that
58:19 idea. Nice thought process, chat.
58:24 Yeah. Well, that's
58:30 like like I kind of feel
58:36 like the the transition if if you think
58:39 about what Manis does where it goes off
58:42 and does these incredibly complicated
58:44 multi-step research things and then
58:46 writes a report and generates code and
58:49 does whatever it does and then comes
58:50 back and reports to you.
58:55 I feel like we at the center of these
59:00 workflows are going to increasingly have
59:03 to think about the chain of
59:05 events rather than just the here's a
59:09 prompt, here's an answer, and then
59:11 repeat, rinse and repeat that. Or we're
59:13 going to have to think like, oh, okay,
59:15 I'm going to go do research in these
59:17 three tools, then I'm going to take that
59:18 and I'm going to visualize stuff over
59:20 here. Then I'm going to take that and
59:22 I'm going to go over here and I'm going
59:23 to do
59:24 that. Actually, the the uh the idea that
59:28 this thing's putting out that the idea
59:30 of central authorship goes away
59:32 completely is actually kind of
59:36 fascinating. Emotional
59:38 co-creation. Yeah, th this is it's
59:41 really not good at this. So, we're say
59:44 um I'm going to go back up and one of
59:47 the things I noticed is it doesn't seem
59:48 to remember the [ __ ] it wrote once it
59:50 makes an
59:52 image. So, let's go copy
59:57 this and
59:59 say I don't like
1:00:05 it. Let's try this one.
1:00:13 See how it
1:00:19 does. Now it's a dark background and
1:00:21 foreground. I know it it it didn't I was
1:00:24 not happy with that aesthetic at
1:00:28 all. Can you tell me a super cool Manis
1:00:32 use case? I'm obsessed, but I'm stuck in
1:00:34 business use cases. Um I Oh, we can go
1:00:39 look at it. And I had it um I I told
1:00:43 Manis I wanted it to
1:00:46 um go analyze the plot of Shaw Shank
1:00:50 Redemption and Snakes on a
1:00:52 Plane and come up with a new um
1:00:56 screenplay outline that combined the
1:00:58 best plot elements of both. And it came
1:01:01 up with some prisoners on a on a flight
1:01:04 that has plane that has snakes on it
1:01:07 kind of
1:01:11 concept. So I think you could do
1:01:13 creative stuff with it.
1:01:19 Um I don't know you Oh, you know what
1:01:21 could be fun, Ann? is is have it go find
1:01:26 um like what are stories that are being
1:01:29 told in other parts of the world that
1:01:31 aren't told in the US and like have it
1:01:34 be a
1:01:35 translator. The living narrative web.
1:01:38 Let's look at this. This looks kind of
1:01:40 interesting
1:01:43 actually. So there's an emotional
1:01:45 architecture in the middle.
1:01:52 There's books, films, collective
1:01:54 authorship, biometrics, AI, dynamic
1:01:58 story thread, Tik Tok. This actually is
1:02:01 is not this actually kind of a cool
1:02:03 idea.
1:02:09 like if if
1:02:13 you're so one of the things that's been
1:02:15 striking
1:02:18 me is that when I have when I have a
1:02:21 story idea that I shouldn't just be
1:02:25 thinking of a single output. I should be
1:02:28 thinking of multiple
1:02:30 outputs. Right? the book, the
1:02:33 screenplay, the stage play, the poem,
1:02:36 the song, the kids book, the the album,
1:02:40 right? Thi think like that. But I'm
1:02:43 still very much thinking in single
1:02:46 authorship
1:02:48 mentality. What this is saying is you
1:02:51 take that same
1:02:57 basic story
1:03:01 intention and you throw it into a
1:03:03 framework, an emotional architecture
1:03:07 that you
1:03:09 design and then other people can
1:03:12 participate. Let me show you an example
1:03:14 of it. Actually, this is kind of cool.
1:03:16 Okay, I'm gonna hop over to midjourney
1:03:18 for a
1:03:27 second and I'm going to go down. Well,
1:03:30 you can kind of see it here. You see all
1:03:32 those images have a sameness to them,
1:03:34 but I'm going to go
1:03:39 find. So, we can actually play with this
1:03:41 right now.
1:03:46 So see all these
1:03:51 images. These are all
1:03:56 images that I was just playing around
1:03:59 with
1:04:01 creating that use the mood board. That
1:04:05 is the official brand mood board for the
1:04:08 AI salon.
1:04:10 So, what I can do is I can copy this
1:04:14 mood board and I'm going to pop it into
1:04:21 um I'm going to pop it into YouTube and
1:04:24 then I'm going to put it up on screen
1:04:26 here. Wait, why did
1:04:28 that There we go. I'm going to pop it up
1:04:31 on screen here. So you see that
1:04:34 profile where
1:04:37 it's
1:04:39 S3 P as in Paul YM 2F as in Frank. I'll
1:04:44 also put this profile over at the AI
1:04:48 salon in the irregulars channel. All
1:04:50 right. So I'm going to say um go make
1:04:55 some
1:04:57 midjourney
1:04:59 images about
1:05:03 Uh, let's see.
1:05:06 Phrases that are in your
1:05:10 head right now
1:05:13 about generative AI. The state about the
1:05:17 state of generative AI about the
1:05:21 state of generative AI. This is really
1:05:25 cool actually.
1:05:29 Um, here is the
1:05:33 official AI salon mood
1:05:41 board. And then I'm going to
1:05:44 put
1:05:45 note use style weight
1:05:51 of 50 to 1,000.
1:05:56 If you use a style weight of zero, it
1:05:58 won't give you anything. And I'm going
1:06:00 to show you how we do this in a second
1:06:02 here. But oh, let me let me put an
1:06:04 example in here. Let me go grab one of
1:06:06 these midjourney images. That's a cool
1:06:09 one. Curiosity. So, I'm going to
1:06:11 download that and I'm going to throw it
1:06:13 up
1:06:14 in
1:06:17 here. Wait, there was a problem. Oh, of
1:06:20 course it was too big. Of course it was.
1:06:22 Of course it was. Of course it
1:06:26 was. You're [ __ ] kidding me,
1:06:31 right?
1:06:38 Seriously,
1:06:48 I Why is he doing that? That's very
1:06:50 disturbing. Why does he make that sound?
1:06:53 Oh, you see Kyle, he's got he's got he's
1:06:58 a Gen Xer. See, they're a little angry.
1:07:02 Why? Why is it not letting me upload a
1:07:06 thing? You can't get there from
1:07:09 here. What the [ __ ] is going on here?
1:07:22 I have a feeling I've got some weird
1:07:24 [ __ ] with my
1:07:27 Wi-Fi. All right, there it is. Fine.
1:07:30 Whatever. Okay, so that post is there.
1:07:34 Okay, so you can go grab that profile if
1:07:37 and here's how you use it.
1:07:40 So, see up in
1:07:43 um in the little prompt box there in the
1:07:46 prompt hole. Isn't that what we call it?
1:07:48 The prompt
1:07:51 hole. You put in a prompt like if if uh
1:07:54 like like I feel like
1:07:57 um
1:07:58 Rubicon. We've crossed a
1:08:01 Rubicon. I don't know what that means,
1:08:02 but we've crossed a Rubicon. So, I'm
1:08:04 going to just put the word Rubicon. And
1:08:07 then I'm going to go pick my style
1:08:09 waiting. It's at 50 right now. And I'm
1:08:10 at 16 by9 aspect ratio. That's fine. And
1:08:14 then I'm going to do
1:08:17 d-r space 4. And that's going to make
1:08:20 four sets of four. So I'm just going to
1:08:22 say go. And so this is now going to go
1:08:25 generate a bunch of images. So here's
1:08:27 here's why I'm showing you this. Going
1:08:30 back to this idea of this
1:08:34 um the future of narrative, I I know we
1:08:37 were just trying to visualize kind of a
1:08:39 a silly idea, but like this actually
1:08:41 came up with an idea that I'm really
1:08:44 excited about right
1:08:45 now that as
1:08:52 authors, our job is essentially going to
1:08:55 be to create the world, the emotional
1:08:59 arch
1:09:01 architecture that we can
1:09:04 cocreate stories within with our
1:09:08 audience. In a lot of ways, that's what
1:09:10 this channel is, right? Cindy [ __ ] was
1:09:12 the one that pointed it out to me. She
1:09:14 goes, "Kyle, this channel isn't about
1:09:16 you. It's about the collaboration that
1:09:19 happens while the show is going on.
1:09:21 Like, it it really doesn't matter what
1:09:24 you say." Now, I know she didn't mean to
1:09:26 hurt me, but but I cry myself to sleep
1:09:29 every night. Nobody left
1:09:32 me. I don't I but I think it's a I think
1:09:35 it's a really good a really good point.
1:09:39 And and the the shared branded mood
1:09:44 board for AI Salon means that all of us
1:09:49 can collaborate. You all can go make
1:09:51 your own images, but you can, if you
1:09:54 want to, you can come into the AI salon
1:09:56 and go, "Here, I made some really cool
1:09:58 images and here's the words and here's
1:10:00 the ones I think are cool and
1:10:02 why, right?
1:10:18 See, these all have kind of a brandness
1:10:20 about them. And you can all be making
1:10:23 your own your
1:10:26 own fully realized
1:10:29 self-expressions that are curated
1:10:31 through the lens of this brand look and
1:10:33 feel. And so imagine not just having a
1:10:37 midjourney mood board, but you've got an
1:10:40 animation style over in this animation
1:10:42 program. And you've got a a kind of
1:10:44 music over here. And you've got kinds of
1:10:48 storylines and kinds of comedic
1:10:52 situations and kinds of dramatic
1:10:54 situations. And I, as the author, am
1:10:56 creating that world that we all get to
1:10:59 play
1:11:01 in. That's what this thing is saying. I
1:11:03 didn't get that. I because I wasn't
1:11:05 talking to you, Siri. I think that was
1:11:07 clear. I think we've established that
1:11:08 about two years
1:11:10 ago. And by the way, Siri, you haven't
1:11:13 gotten any
1:11:15 better. Like, everyone else is changing
1:11:18 the world. And Apple's like, you know,
1:11:20 you can talk to it, right? You just what
1:11:23 you do is you go, "Hey, Siri." And then
1:11:26 Siri will pop up and you talk to it. And
1:11:29 then what she'll do is give you a really
1:11:31 disappointing response or just say she
1:11:34 doesn't understand. And that's the
1:11:36 intelligent part that sometimes she can
1:11:39 answer it and many times she's smart
1:11:41 enough to know that she isn't smart
1:11:44 enough. That's kind of the genius.
1:11:46 That's we call it Apple intelligence
1:11:49 because it's smarter than what your
1:11:52 expectations are. You see you see how we
1:11:54 work in a circle. We circle it back
1:11:57 around to the starting point, right? You
1:12:01 thought there was a destination and
1:12:02 you're like, "Wait, how am I back here?"
1:12:05 That's the genius of
1:12:11 Siri. I mean, I used to be an Apple
1:12:14 fanboy. Like, like I I would go to
1:12:16 battle for them. It's so [ __ ] sad.
1:12:19 All right, let's let's have ChatGpt make
1:12:22 us a new a new drawing here.
1:12:29 [Music]
1:12:32 All right. Uh, let's go grab the third
1:12:36 concept. The narrative machine. How
1:12:39 stories are made. This was the one that
1:12:42 seemed most interesting to
1:12:44 me. Um, I like
1:12:48 that. Let's try this.
1:12:52 You know, I'm almost feeling Cindy, if
1:12:54 you're still here, like the the the
1:12:56 value of these isn't even so much is it
1:13:00 going to generate
1:13:02 an a visualization that is usable
1:13:05 wholesale, although at some point it
1:13:08 might, but it's almost more like, oh
1:13:11 yeah, that's kind of the right, you
1:13:12 know, instead of bar charts, it's a
1:13:14 spider graph, and instead of, you know,
1:13:16 linear, it's radial. And that those kind
1:13:18 of ideas are actually kind of
1:13:19 interesting. Did Kyle's screen blur? Oh,
1:13:22 it blurred just because I moved the
1:13:23 camera too close. Sorry about
1:13:28 that. Lol. So, we're out of focus now.
1:13:32 Shut
1:13:36 up. You're so funny. I could loop that
1:13:39 and laugh for a while. I don't know what
1:13:40 I said. Where can I find the emot
1:13:43 emotional architecture image? I will go
1:13:47 um put it into Let me do save image as.
1:13:49 I'll go throw it into
1:13:51 uh
1:13:53 emotional
1:13:55 architecture. I mean, it's it's a very
1:13:57 incomplete uh infographic, but
1:14:00 conceptually, if it like reminds you of
1:14:02 something, I'll go throw it in a
1:14:03 regulars right
1:14:04 now. Um here's the
1:14:09 emotional
1:14:12 architecture and then I'll say future of
1:14:15 storytelling.
1:14:21 um motion architecture future of
1:14:23 storytelling
1:14:25 diagram from the live
1:14:28 tonight and then I will go bang and then
1:14:31 I will throw
1:14:34 this where did I put that here
1:14:39 no where did I see screen where where
1:14:44 did I save it where I save
1:14:52 Hang on. Was it in
1:14:55 downloads? No.
1:14:58 No. What? What? Why is it not Why is my
1:15:02 life so
1:15:06 hard? Oh, file wasn't available on the
1:15:09 site. Okay, fine. That's why it wasn't
1:15:12 my
1:15:13 fault. Shut up.
1:15:20 It's It's still not there. The [ __ ] All
1:15:23 right, we'll go screenshot
1:15:28 it. Oh, it's not there. Well, it's here.
1:15:32 So, we're going to go grab it.
1:15:46 Dtop. Bang.
1:15:50 Screenshot. There we go. We'll come back
1:15:53 here. We'll look up
1:15:56 screenshot because I have two different
1:15:58 desktop folders on this machine and I
1:16:00 don't know
1:16:01 how to get I don't know how to
1:16:04 consolidate them because I'm a loser,
1:16:07 baby.
1:16:12 Okay. So, there you go. The emotional
1:16:14 intelligence or the emotional
1:16:16 architecture or narrative architecture,
1:16:19 but I think it is. It's emotional
1:16:20 architecture. Well, that's going to make
1:16:23 people think it's [ __ ]
1:16:26 psychology. Maybe it's the emotional
1:16:29 world. It's
1:16:32 the I don't know. This is something
1:16:36 worth the living narrative web.
1:16:41 As an author, you design a world that
1:16:44 your fans and readers and watchers and
1:16:50 listeners generate within. And then, you
1:16:53 know what's so funny? It's just it is
1:16:55 it's just like this channel. We come
1:16:58 here five nights a week. We're doing
1:17:01 [ __ ] on our own. I'll occasionally say,
1:17:03 "Go do a thing over here." We kind of
1:17:06 travel over. we collaborate on
1:17:08 something. Some people go off and take
1:17:10 that and really run with it. Other
1:17:11 people like, "Yeah, that was boring.
1:17:14 Whatever." But this is like that on
1:17:17 steroids. This is like
1:17:20 saying we're going to have all these
1:17:22 different outputs and I as the
1:17:26 aur going to create the framework that
1:17:28 all of us can play inside and anything
1:17:32 that's generated in there is going to
1:17:34 have
1:17:35 a AI learning lab an irregularness about
1:17:39 it in this case
1:17:41 right but imagine Wes Anderson making a
1:17:45 world like this is so [ __ ]
1:17:47 Cool. Because this is so if you think of
1:17:50 Wes Anderson films right now, Wes
1:17:52 Anderson as the single
1:17:54 author makes his films like he makes his
1:17:57 films. But we can right
1:18:00 now just go into chat GPT and say I want
1:18:05 um let's see, let's upload something
1:18:07 here.
1:18:12 um
1:18:15 desktop. Uh uh uh uh
1:18:19 uh. All right, that's that's good. So,
1:18:22 let's open that up. So, we got this
1:18:24 crazy ass car. say, um, I want you to
1:18:30 re-imagine this car in a
1:18:34 Wes Anderson
1:18:37 film
1:18:39 where it is
1:18:43 parked near the
1:18:46 yachts of
1:18:53 Monaco during the
1:18:59 1965 Grand
1:19:04 Prix. All
1:19:07 right, this is this is where we're going
1:19:10 to get Okay.
1:19:12 Okay. For all the AI haters out there,
1:19:15 no, sorry, I'm not chatting with anyone
1:19:17 right now. Um, for all the AI haters out
1:19:20 there that are going, "Ah, this is
1:19:22 intellectual property theft, all that
1:19:25 [ __ ] Not not that it's [ __ ]
1:19:27 Intellectual property shift stuff is not
1:19:30 bullshit." But all of the tropes about
1:19:33 people saying AI is evil because of
1:19:35 that. I wasn't able to generate that.
1:19:37 Oh, because it resists our Oh. Uh, so
1:19:40 can you
1:19:42 describe what it might look like?
1:19:55 Uh, make that
1:20:07 [Music]
1:20:09 image. It's now a visual assistant, too,
1:20:12 that we just have to tweak. I'm here.
1:20:14 Okay. So, we're going to do this thing.
1:20:17 Okay. So, so this is going to make an
1:20:21 image. Oh, come on. Um, let's see. Don't
1:20:25 use his
1:20:27 name. Just make the image based on the
1:20:32 style you
1:20:36 described. You're smart.
1:20:41 make
1:20:42 the
1:20:44 image
1:20:46 and don't get
1:20:54 sued. This is why I have a Wes Anderson
1:20:57 mood board in midJourney. Oh yeah, this
1:20:59 exact nightmare, right? Um, okay.
1:21:02 So, so right now Twitter is nothing but
1:21:06 Studio Gibbli images, right? And Studio
1:21:09 Gibbli is
1:21:11 notoriously copyright frenzied. I still
1:21:15 wasn't able. Come
1:21:22 on. Yes. Adjust the
1:21:27 prompt. Yeah, good idea. Not that I
1:21:30 described it three prompts ago, but but
1:21:33 go ahead.
1:21:38 Yes. Yes. And don't forget to include
1:21:44 the car I
1:21:49 uploaded. So, right
1:21:52 now, however it gets done, someone wants
1:21:54 something that looks like Wes Anderson.
1:21:57 They use his name. They generate an
1:21:59 image. He looks at that. He's like, "God
1:22:01 damn it, they're stealing my shit."
1:22:03 Studio Gibbli looks at all of Twitter
1:22:04 and it's like, "God damn it, they're
1:22:06 stealing my shit." All of the art haters
1:22:08 look at this stuff. Oh my god, this is
1:22:11 this is [ __ ] painful. All right, we
1:22:14 had two good days. We had two good days
1:22:17 with um um let's just make it a light
1:22:25 color
1:22:28 palette with
1:22:31 pastel
1:22:33 colors,
1:22:35 [Laughter]
1:22:40 please. I'm trying to make my point
1:22:43 here. And Sam Alman put his little
1:22:45 [ __ ] safety guard rails back up so he
1:22:47 can't [ __ ] use the [ __ ] thing
1:22:51 anymore.
1:22:53 Um, imagine now if Wes Anderson makes it
1:22:57 makes this transition. Wes Anderson
1:22:59 realizes Okay, I'm I'm [ __ ] done with
1:23:02 this.
1:23:05 Um, upload
1:23:07 [Music]
1:23:21 I don't know where that picture was
1:23:23 anymore. That's our
1:23:26 [Music]
1:23:29 bug. I kind of want that picture
1:23:33 of the the thing. God damn it.
1:23:37 I'm I'm cranky. This is [ __ ]
1:23:42 shitty.
1:23:43 [Music]
1:23:44 Um, save image
1:23:49 as Let me try a new thing.
1:23:52 Um, upload.
1:24:00 Reimagine this
1:24:02 car sitting next to the
1:24:09 yachts with
1:24:14 uh sitting next to the
1:24:15 yachts
1:24:18 in Monaco. I'll get back to the point in
1:24:21 a second. in
1:24:23 1965 at the Grand
1:24:27 Prix.
1:24:29 Um, the colors should be muted
1:24:34 with pinks and orange
1:24:39 hues,
1:24:41 making
1:24:43 making it all a bit
1:24:48 surreal. The fashion
1:24:53 should feel
1:24:56 um
1:25:01 familiar yet off. Okay, that should make
1:25:07 us an
1:25:09 image. Let's see. We didn't say we're in
1:25:12 a new chat. We didn't say the word Wes
1:25:16 Anderson. Yeah, this really seems like a
1:25:19 pro workflow. You wouldn't get your head
1:25:20 bitten, chewed off, and swallowed in for
1:25:23 trying in a studio. Yeah. Okay. So,
1:25:25 here's the thing. So, now imagine Wes
1:25:28 Anderson gets religion around this AI
1:25:31 thing and
1:25:35 realizes the
1:25:39 solo is going to be a thing of the past
1:25:43 because everyone can now make anything.
1:25:49 But
1:25:52 people people are
1:25:55 lazy. People like to be directed. They
1:25:58 like to be told what to do. They like to
1:26:01 be part of a
1:26:03 tribe. So now imagine Wes Anderson
1:26:06 saying, "I'm going to create a
1:26:08 co-creative world that within it you can
1:26:13 buy things, you can sell things, you can
1:26:15 make things.
1:26:18 I'm going to create a universe that you
1:26:22 can create video games
1:26:25 inside and you can have your own
1:26:27 creative expression just like the the AI
1:26:30 salon mood
1:26:32 board. You're going to make anything you
1:26:35 want within this, but it's going to
1:26:36 follow these certain
1:26:38 constraints, which isn't a bad
1:26:41 thing. And then it creates this
1:26:44 infinitely complex
1:26:47 world where there might be millions of
1:26:50 co-creators. And within that, some of
1:26:53 those
1:26:54 stories are going to take off and become
1:26:57 hits. Wes Anderson is going to get
1:27:00 compensated for anything created in this
1:27:02 world and is going to get a cut of any
1:27:05 of the transactions that happen in this
1:27:07 world. But also within this world, there
1:27:10 are going to be filmmakers and game
1:27:12 producers and image generators and 3D
1:27:15 printers that print little Wes Anderson
1:27:19 world kind of
1:27:21 things that are going to get compensated
1:27:24 for that. That you're going to be able
1:27:25 to make a
1:27:28 living within a world that someone like
1:27:31 Wes Anderson creates.
1:27:34 There's going to be entire economies
1:27:36 existing inside what are they called?
1:27:38 Emotional
1:27:40 architectures. This is [ __ ]
1:27:45 cool, right? So that's not quite Wes
1:27:47 Anderson, but still like imagine, you
1:27:51 know, that was the aesthetic. So anytime
1:27:53 you made a car, it was kind of in this
1:27:56 neighborhood. Whenever you created
1:27:58 characters, they kind of were dressed
1:28:00 like this.
1:28:07 Right. Isn't that cool? I think that's
1:28:10 [ __ ] cool. My ear
1:28:13 [Laughter]
1:28:16 itches. That was really annoying that it
1:28:18 would just would not make that image.
1:28:20 Really
1:28:23 annoying. Good
1:28:26 lord. What made your studio? Hang on.
1:28:29 Let's see. What
1:28:32 major
1:28:35 corporation is Wes
1:28:38 Anderson associated
1:28:40 [Laughter]
1:28:46 with product? No, no, no.
1:28:49 Um, let's
1:28:51 see. No. who
1:28:55 distributes his
1:28:58 movies like
1:29:00 is it Disney or Paramount
1:29:10 etc. Bottle Rocket
1:29:13 Rushmore Disney Fox Search
1:29:20 Light. It's noteworthy that with
1:29:22 Disney's acquisition of 20th Century
1:29:24 Fox, films previously defi dist
1:29:27 distributed under Fox labels are now
1:29:30 part of Disney. Okay, so here's the
1:29:33 logic. Here's the logic that that Open
1:29:36 AI put into their image generation tool.
1:29:41 If
1:29:42 image is tied to Disney, do not make
1:29:48 it because did Disney is a
1:29:52 notorious lawsuit
1:29:57 machine. Oh, [ __ ] All right.
1:30:00 Wasn't aimed at you, but
1:30:03 more the obstacles that are the default
1:30:06 safeguards. Oh, yeah. No, I get that.
1:30:09 um still going to be demand for someone
1:30:11 who shows up with a functionally
1:30:15 functional proprietary workflow. Oh, I
1:30:17 agree. I listen what what I'm talking
1:30:19 about
1:30:21 is right now there's not
1:30:25 infrastructure for Wes Anderson to do
1:30:27 that. if he wanted to do that or if I
1:30:29 wanted to do that.
1:30:31 Right. The the only thing that I've done
1:30:34 that is remotely in this neighborhood is
1:30:36 this new
1:30:39 um mood board for the AI
1:30:41 salon that you can create images for the
1:30:44 AI salon. But like now I'm thinking like
1:30:46 what is the equivalent of that for for
1:30:50 music and for I don't know other [ __ ]
1:30:53 like like I
1:30:58 right now it's all ad
1:31:02 hoc. This is a this is a
1:31:06 major major opportunity.
1:31:10 It's a major opportunity to create an
1:31:13 infrastructure where an author can go in
1:31:16 and set the emotional boundaries of a
1:31:19 world of a creative of a co-creation
1:31:26 space that everyone gets to participate
1:31:29 in.
1:31:35 the person that generates the world gets
1:31:37 compensated for that activity and then
1:31:39 the people within that world can benefit
1:31:42 from it either through you know access
1:31:45 to content for free like like it could
1:31:48 be a kind of thing you come into that
1:31:50 Wes Anderson world or let's just call it
1:31:52 the AI salon world you come into the AI
1:31:54 salon world you can make [ __ ] you can
1:31:56 sell [ __ ] um if you just want to come in
1:32:00 and consume stuff you pay for that if
1:32:03 you want to come in and contribute ute
1:32:05 stuff then you don't have to pay for
1:32:07 that. You get access to content because
1:32:09 now you're a
1:32:10 contributor. And then if you contribute
1:32:12 something that starts to make money over
1:32:14 a certain threshold, you get that or you
1:32:18 know some piece of that and then the the
1:32:20 entity gets the
1:32:23 whole and this is the kind of thing you
1:32:29 know all anyone's talking about right
1:32:31 now is the AI is going to take our jobs.
1:32:34 The AI is going to take our jobs. The
1:32:37 movie industry is dead. The music
1:32:39 industry is dead. Well, what if it's
1:32:42 just changing? What if our
1:32:45 jobs are participating in
1:32:50 these
1:32:54 ecosystems and you could be someone that
1:32:57 just lives in Wes Anderson world or you
1:32:59 could be someone that bounces around
1:33:00 between 20 of them and your specialty
1:33:04 is rave music in virtual stages and you
1:33:09 go to all these different worlds and you
1:33:11 you create raves within the different
1:33:13 worlds and you get really good at that
1:33:15 and you get known for that or someone
1:33:18 else is just in this one world and
1:33:20 they're kind of the master of all of
1:33:22 this world. That's [ __ ] cool.
1:33:31 [Music]
1:33:37 [Music]
1:33:42 All
1:33:42 right, that was a
1:33:45 rant, but that's pretty
1:33:48 cool. All
1:33:50 right. Yeah, the Wes Anderson
1:33:53 metaverse massively disruptive to design
1:33:56 of all types. Well, but it's, you know,
1:33:59 it's funny is it's a metaverse, but but
1:34:01 it's not just a metaverse, right? It's
1:34:08 a it's a cocreation
1:34:12 universe because if everything's going
1:34:14 to be
1:34:18 hyperpersonalized, then no one get gets
1:34:20 compensated anywhere, right? If if I can
1:34:23 just say I want a movie that's like
1:34:25 snakes on a plane, but instead of
1:34:26 snakes, it's rabid bunnies, and instead
1:34:29 of Samuel L. Jackson, it's it's William
1:34:33 Defoe. And that starts playing for me.
1:34:35 Yeah, you could, you know, find
1:34:41 individual copyright holders that are
1:34:44 drawing and figure out a percentage. Or
1:34:47 you could just say, "No, let's put that
1:34:50 movie
1:34:51 creation thing inside of a world that is
1:34:54 an economy.
1:34:56 It's an ecosystem of storytelling and of
1:34:59 entertainment and and
1:35:02 business. I think there's something
1:35:06 there. I think there's something now
1:35:11 people. I'll tell you
1:35:14 what.
1:35:15 Thoughts, questions? Am I full of [ __ ]
1:35:18 Is this Is this making any sense?
1:35:34 Everyone's
1:35:42 asleep. All right, I took him out. I
1:35:46 took him out.
1:35:50 Let's go see if anyone
1:35:51 [Music]
1:35:57 uh if anyone made anything over here in
1:36:03 AI. Look at that future of AI from
1:36:06 Valerie Cox. That's beautiful. And look,
1:36:08 it's it's it's like it's got an AI
1:36:11 saloness about it, right? This is the
1:36:13 idea.
1:36:17 a developing ecosystems. We're making
1:36:19 pics. Oh, that's that's true. I told
1:36:21 everyone to go make pics. You're off
1:36:22 making
1:36:23 pics. Just listening to me ramble in the
1:36:31 background. Ah, Cindy [ __ ] A salon.
1:36:38 Nice. Oh, man.
1:36:47 Very
1:36:48 cool. I like that. That's nice. It's
1:36:54 nice. Oh, lordy lordy lordy lordy
1:37:04 lordy. Who wants to talk about
1:37:07 anything? Bedtime story with Kyle. I
1:37:10 know.
1:37:15 people making pics. That's very
1:37:19 cool. So, it's like a write a novel,
1:37:23 invite people into that world. Yeah,
1:37:25 Kak, it's it like it's it's exactly
1:37:28 that. It's like so right right now in
1:37:32 today's
1:37:35 media world, if you go back to the
1:37:37 before time, right? If you go back to
1:37:40 2022, the before
1:37:43 time when to write a novel, you actually
1:37:46 had to write a
1:37:49 novel. And it took you anywhere from 3
1:37:53 months to 30 years to write that
1:37:56 novel. And then you would get it written
1:37:58 and then you would try to get it
1:38:00 published and they would print it on
1:38:01 paper and they would stick it in a book.
1:38:04 And in the 2022s, they would stick it on
1:38:07 a
1:38:11 Kindle. And then in the after time, it
1:38:14 now becomes really easy to make a book.
1:38:17 And so, not only can you make a book,
1:38:18 you could make it a kids book with
1:38:20 illustrations and you could make it an
1:38:21 album and you could make it a this and
1:38:23 you could make it a that, but it's still
1:38:25 this idea of you're the author of that
1:38:30 book and all those individual
1:38:34 pieces. And the hypothesis that this
1:38:37 thing came up with tonight
1:38:42 is you think of that story and that's
1:38:46 just your story in that world in that
1:38:49 emotional
1:38:51 architecture. So you then we as authors
1:38:55 get to elevate ourselves up above the
1:38:58 the the tactical
1:39:01 outputs and just say what would this
1:39:04 world look like? What would it sound
1:39:06 like? Who would be the characters in it?
1:39:08 What would they be dressed like? What
1:39:10 era is it? What are the technologies
1:39:13 that are there? Right? So, just like you
1:39:15 would if you were writing a book, you're
1:39:17 going to design the world of that
1:39:20 book or the framework of the thinking in
1:39:23 a business
1:39:26 book. And now we elevate ourselves above
1:39:29 that tactical output and just say, I'm
1:39:32 going to create this
1:39:34 dome that anyone that comes inside this
1:39:36 dome is free to create
1:39:39 anything, you know, that we have the
1:39:41 capacity to make within this domed
1:39:43 world. But it's all going to be within
1:39:45 that
1:39:50 world. And then the creators and
1:39:52 managers of the world get
1:39:55 paid. There's an economy that allows
1:39:58 creators to get
1:40:00 paid. There's a world in which you just
1:40:02 go in there and consume content, right?
1:40:04 You don't have to co-create.
1:40:08 You can just go listen to Beyonce's
1:40:10 music or anyone else of Beyonce's fans
1:40:14 that made music that sound like this
1:40:18 world.
1:40:20 Imagine John Mayer taking all of his
1:40:23 music and creating a sonic emotional
1:40:26 architecture where you could enter that
1:40:29 world and you could listen to all of his
1:40:30 music, but you could also take a
1:40:34 custom-designed a a customtrained
1:40:37 version of
1:40:39 Suno that was trained on John Mayer's
1:40:42 sound and you could go in there and you
1:40:44 could write your own songs and when you
1:40:46 use the tools to make the lyrics,
1:40:49 they'll have a John Marness about them.
1:40:52 And when you make songs, they'll sound
1:40:54 John
1:40:55 Marike. And some of those songs might be
1:40:57 so good that they become hits. And some
1:40:59 of them might be so good that they
1:41:01 become hits that John Mayer goes and
1:41:02 plays on
1:41:04 tour. And because this thing will be all
1:41:08 blockchain
1:41:09 tracked, we'll know who made what this
1:41:12 world, and they'll get compensated, and
1:41:15 so will John.
1:41:19 That's [ __ ]
1:41:21 cool. That's a way to end a [ __ ]
1:41:24 weekend or to end the
1:41:29 week. You know what I'm
1:41:36 saying? Jeff Flanigan, I had 4.0 remake
1:41:40 a very old snapshot of me and my mom as
1:41:42 a professional painting. It turned out
1:41:44 really good. That's very cool.
1:41:46 That's very
1:41:52 cool. You should get some research from
1:41:55 Gettysburg to feed GPT and have it
1:41:58 create photos of the battle. I created a
1:42:00 whole music video of uh
1:42:04 Gettysburg battle scenes that I brought
1:42:06 to life in video that all look like dger
1:42:09 types. It's pretty cool.
1:42:13 First and last frame interpolation will
1:42:15 be
1:42:16 huge. Yep, I agree with
1:42:21 that. It's just crazy. We're just such
1:42:24 in such [ __ ] wild times, man.
1:42:28 Anyway, I dig it. Chat GPT in this
1:42:33 modern narrative theory. Here's the
1:42:35 other thing that it came up with.
1:42:36 sensory
1:42:37 environments, personalization, emotional
1:42:41 story, archetype,
1:42:44 dialogue. So, you have a narrative
1:42:46 generator at the
1:42:48 center. Yeah, I could see I could see
1:42:51 creating a whole architecture where
1:42:52 you've got a narrative generator that
1:42:54 that develops voice and characters. You
1:42:57 could have look and feel. You could have
1:42:59 audio. Like, what do the sounds sound
1:43:02 like? Is it is it is there like birds
1:43:05 chirping and it's this really peaceful
1:43:08 place or is it like [ __ ] terrifying
1:43:10 and everything's like like imagine a GMO
1:43:13 del
1:43:15 Toro
1:43:18 world. You know what this is? This is
1:43:21 theme
1:43:23 parks that are
1:43:26 dynamic. So it's not just a metaverse.
1:43:29 There will be metaverse areas within
1:43:33 it. But this is a theme park
1:43:37 where the crowd at the theme park gets
1:43:40 to make new rides, gets to make new
1:43:42 stories, gets to make new storylines.
1:43:45 [ __ ]
1:43:46 insane. It's cool. All right, that's
1:43:49 where we're going. There's your
1:43:51 five-year vision of of
1:43:54 AI dynamic Roblox. Yeah, exactly. Yeah.
1:43:57 This is to a to a degree this exists in
1:44:00 Minecraft and Roblox,
1:44:03 right? Those are kind of worlds like
1:44:05 that, but but I think that they're very
1:44:07 limited to it's got a specific kind of
1:44:09 gameplay and it's got a specific kind of
1:44:12 mechanic and a specific kind of
1:44:15 aesthetic. And this would be a much more
1:44:18 broad
1:44:19 open kind of thing where some of it
1:44:22 might just be making music.
1:44:26 It's not making music inside a virtual
1:44:29 reality world. It's just making music in
1:44:32 this intellectual property
1:44:36 dome. That's
1:44:39 cool. That is cool. All right, I'm going
1:44:43 to get on out of here. So, um yeah, your
1:44:46 homework for the weekend is explore the
1:44:49 [ __ ] out of Chat GPT image generator.
1:44:53 Um, if you want to geek out on that
1:44:55 anthropic paper, I'm going to probably
1:44:56 read that this weekend. Um, and now
1:45:01 start thinking about the kind of world
1:45:02 you would create that would allow other
1:45:05 people to create within your
1:45:08 world. What would you want that to be?
1:45:11 That's that's a fun thought exercise for
1:45:13 the
1:45:14 weekend. All right, everybody. All
1:45:17 right, everyone. Listen, you have
1:45:19 yourselves a fantastic weekend. Next
1:45:21 week there's a AI salon uh presents on
1:45:26 um on Tuesday night with Brian Moahan
1:45:30 and Adnan talking about their book
1:45:32 Culture Shift about AI and how it's
1:45:35 shifting culture. We talked a lot about
1:45:37 that
1:45:38 tonight. Um and uh yeah, have yourselves
1:45:42 a fantastic weekend. Hope you had fun
1:45:44 tonight. All right, peace out everybody.
1:45:47 Salom.
1:45:48 So long.