6/11/2025 - Prompt Engineering: Feeding Your Creativity With A.I.

Description

Chapters

Transcript

0:00 [Music]
0:08 Woohoo!
0:10 [Music]
0:22 [Music]
0:35 She came on him like slow moving cold
0:40 front.
0:44 Well, his bed was warmer than a look in
0:47 her eyes.
0:48 [Music]
0:52 She sat on a stool and he said, "What do
0:56 you want?
1:00 She said, "Give me a love, but don't
1:03 freeze up inside
1:06 [Music]
1:14 melted
1:16 my time, dear."
1:18 [Music]
1:21 But to sit next to you while I shiver
1:24 and shake.
1:27 [Music]
1:29 And if I knew love, well, I don't think
1:32 I'd be here.
1:37 Asking myself if I've got what it takes.
1:44 I do not sure I see blue.
1:48 [Music]
1:53 Should I stop?
1:56 Turn what's been frozen for air
2:02 into a river of tears.
2:07 [Music]
2:15 You are singing good tonight, boy. We
2:18 were in unison there for a minute. You
2:20 were a little flat, but it was not bad.
2:21 It was solid. It was solid. You were You
2:24 were right there.
2:25 [Music]
2:51 Oh man, that dog. He loves that
2:54 vocalization,
2:56 doesn't he?
2:58 It's just a It's a You know, I've chat
3:01 GPT it. I've chat GP it. I'm like,
3:04 what's with the dog? Some dogs Some dogs
3:07 can see TVs, some can't. Some sing, some
3:11 can't. What is it with the singing?
3:14 Apparently, it's pack behavior. The
3:17 guitar sounds like a bunch of other dogs
3:19 howling, maybe.
3:23 What is it about a minor champ?
3:28 [Music]
3:47 I guess a minor sounds like a bunch of
3:49 wolves in the olden timey days, huh? Is
3:51 that what it is? Do you know how removed
3:54 from a wolf you are?
3:57 Do you know what wolves didn't eat,
3:58 champ? Where are you going? We're having
4:02 a conversation.
4:04 You can't just walk away. You know,
4:08 you know what? Wolves didn't have Champy
4:11 cheese.
4:13 They didn't have cheese.
4:15 They didn't whine. They didn't me. They
4:18 didn't meow for cheese.
4:21 You know what they did? They caught
4:23 their prey. They shredded it and ate it.
4:26 Mew me.
4:27 You're a wolf descendant. Barely. All
4:31 right, good people. Yes, I am the
4:35 species shaming.
4:44 [Laughter]
4:46 I'm species shaming. That's funny. I'm
4:49 speciesist.
4:51 I think I think wolves are better than
4:53 whatever the hell he is.
4:58 [Music]
5:20 Um,
5:22 so a fair amount we can play with
5:23 tonight. So last night we we used the
5:27 three different we used Manis, Gen
5:29 Spark, and what was the third one?
5:31 Gamma. Did we do a fourth one? We were
5:35 trying to make slideshows, right? I had
5:37 this outline for a slide deck
5:40 and then we used these fancy things that
5:43 went off in design [ __ ] And then
5:46 I woke up this morning. I had to give
5:48 this talk this afternoon at 3:00 in the
5:51 afternoon. So I woke up this morning at
5:53 like 5:30. I'm like, "Ah, [ __ ] I got to
5:56 deal with that slide deck."
5:58 So, I came down here to this very $20
6:02 million production studio made to look
6:04 like a home office and I started in Chat
6:08 GPT making little little graphics.
6:11 And I actually ended up making a graphic
6:13 that I [ __ ] love.
6:17 And then I just I spent the day, you
6:19 know, in between all the other stuff I
6:21 had to do, every little spare moment, I
6:23 was writing the deck
6:27 and designing it and making images.
6:30 Um, so I can show you the deck. Um, walk
6:33 you through it. I'm actually really
6:34 happy with it.
6:37 [Music]
6:45 Oh,
6:47 [Music]
7:07 hey.
7:10 [Music]
7:19 Freedom came our way that night
7:23 [Music]
7:25 just like a jet plane in and out of
7:29 sight.
7:31 I was hauling ass at a million miles an
7:35 hour, wondering how hard I'd hit
7:42 when they came into the station.
7:47 [Music]
7:49 They said I was bad beyond repair,
7:54 but I got no qualms with my situation.
8:01 Say here I am.
8:05 [Music]
8:09 So say Sheree Sheree Sheree won't you
8:13 dare to
8:16 say Sheree Sheree Sheree won't you dare
8:20 to come on champ say Sheree Sheree
8:24 Sheree.
8:27 Uhhuh.
8:28 [Music]
8:29 Yeah. Leave a message and your number,
8:33 please.
8:37 Take a time to want to satisfy me.
8:43 Take all these old fantasies.
8:46 Send them care of me.
8:51 All right. Enough self-indulgent seven
8:55 song melody medleys.
9:01 Good evening. Good evening. Good
9:03 evening. We got weirdness going on here.
9:06 Let me reset some screen stuff. There we
9:09 go.
9:11 Help. Tasks have disappeared from chat
9:13 GPT.
9:16 Oh,
9:21 create an image. Search thread. Write
9:23 code.
9:27 Add to project. Share.
9:32 It looks like they have Vicki.
9:36 The [ __ ] with that?
9:45 I know that they moved it out of
9:49 out of the model list and into the the
9:51 thing down below, but it is gone.
9:58 I wonder if they still work. They're
10:00 still sending each day, but I can't
10:02 create new ones.
10:04 Oh, no. Maybe tasks are located
10:07 somewhere else. I don't see them. Let me
10:09 go to Let me go to settings. Maybe they
10:11 moved them to settings. Personalize.
10:15 No, that's just custom instructions.
10:20 Settings.
10:22 Subscription security connectors.
10:25 Builder profile.
10:27 Uh remember when we all took our builder
10:29 profile seriously?
10:33 We're going to make money. We're you're
10:36 gonna make money with Jackie. We're
10:37 gonna sell custom GPTs to the highest
10:40 bidder. Now, Kelly Camp is laughing all
10:42 the way to the bank on that one. That's
10:44 how she's making her money now. Selling
10:45 custom GPTs, but not like we thought we
10:47 would. You know, make money while you
10:50 sleep. Yeah. Yeah. About that. Um
10:54 personalization.
10:58 Wait.
11:01 Okay. Okay. Under notifications, Vicki,
11:06 you can still change your settings for
11:08 tasks. Like, are they going to are they
11:10 going to give you a popup menu or you
11:12 going to get an email? Are you going to
11:13 get both?
11:19 But yeah, task tasks seem to be gone.
11:24 Um, oh, archetypal architect
11:27 actually making uh making some money
11:30 with pobots. Nice.
11:33 The connections don't work either.
11:36 Well, they're they're doing some they're
11:38 doing some [ __ ] right now. So So one of
11:40 the things we're going to do tonight is
11:43 we're going to go read Sam Alman's blog
11:45 post because it's kind of fascinating.
11:48 Um
11:49 and we'll, you know, it'll be good chat
11:51 TMZ speculation stuff in there. That'll
11:54 be fun.
11:56 Um, they launched 03 Pro, which I I
12:00 dumped my Chat GPT Pro account to get a
12:03 Google Pro account so I could get VO3,
12:06 so I don't have access to 03 Pro. But
12:09 Nate B. Jones did a video today and he
12:11 basically said it's just like 03 except
12:13 you tell it to go do something
12:16 and it goes off for like 15, 20, 30
12:19 minutes.
12:20 So, so they're just burning GPU cycles
12:23 with it, but apparently it's really
12:25 good.
12:28 I don't know. And basically what what
12:30 Nate said in his video was if you give
12:32 it too stupid a prompt, if you give it
12:34 something stupid to do, like analyze
12:36 this document and, you know, tell me if
12:39 it's good or bad, something like that,
12:41 it will still go and think for 15
12:42 minutes and then give you back a bunch
12:44 of [ __ ] So,
12:46 wait. Scheduled tasks are still
12:49 officially supported on 03 and 04 mini
12:53 models. So if you ask either of those to
12:56 schedule a task, it'll do it. Oh, 03
13:00 and 04 mini.
13:04 Bizarre. All right. Well, I'm in 03
13:06 right now.
13:08 Um, which which chat is this? This is
13:12 Viva CRM. Okay. Um, let me do a new
13:15 chat.
13:16 Let me
13:18 We're going to confirm. This is one of
13:20 those things where we absolutely believe
13:23 Valerie, but it's also possible that she
13:26 got that answer from ChatGpt.
13:29 And if she got it from chat GPT,
13:32 that little [ __ ] is a liar. We
13:34 know it's a liar. It hallucinates. It's
13:36 doing drugs. It's doing mushrooms. All
13:39 right. So, we're gonna say um schedule a
13:43 task for task.
13:46 Why would they remove it from the menu
13:48 item and just make people know this? I
13:51 They're weird. Open AAI is weird, man.
13:55 All right. Let's see if it lied. Ah, is
13:59 Valerie hallucinating? Exactly. Okay.
14:02 Schedule a task for Wait. Schedule a
14:05 task to make me an image
14:10 of
14:12 a turtle
14:15 on a rock.
14:17 Next.
14:19 next
14:20 to a sunflower.
14:24 And
14:26 the
14:28 talk bubble above the turtle
14:33 says
14:35 am
14:37 I late?
14:41 um and have it send me that image
14:50 every day
14:53 at
14:55 8:15.
14:59 Oops, I put let's see 8:15
15:04 Mountain time. Oh, wait. PM
15:07 PM Mountain time. All right, let's see.
15:12 I can absolutely make that image for you
15:14 now, but I can't schedule it to
15:16 automatically send. Chat GPT doesn't
15:20 currently support recurring scheduled
15:22 actions or automatic notifications.
15:25 But here's what you can do. You can
15:27 create a manual reminder system in some
15:29 other program. Thanks, ChatGpt. That's
15:32 awesome.
15:37 Uh, apparently not.
15:42 Oh, you're in 40. Oh, who told me that?
15:45 Valerie, see Valerie, she she thinks
15:48 she's smart. She thinks she's smart.
15:51 She's like, "Oh, Mr. Smarty Pants with
15:53 the big AI channel. You don't even know
15:55 how to switch models." Fine. Fine. Fine.
15:59 If you want to be that way, if you want
16:00 to call me out on logic just because I
16:04 didn't follow instructions or read the
16:06 manual, I understand. I got it. I got
16:09 it. I understand the power dynamic here.
16:18 All right, let's see what 03 says.
16:23 03's not saying [ __ ] 03's like, I'm not
16:25 getting in the middle of this.
16:29 Oh, man.
16:32 [Music]
16:38 Day.
16:40 All right. The user wants an image
16:42 delivered, but automations don't support
16:45 sending images directly. Well, they
16:47 certainly [ __ ] used to.
16:56 I still get sin city sipper prompts.
17:00 Thought for 27 seconds.
17:03 Got it. I'll send you the whimsical
17:05 turtle image every day at 8:15. Oh [ __ ]
17:08 it's 8:15. Turn on desktop
17:10 notifications. Allow. Allow.
17:14 I might have missed it. I I might need
17:16 to do something at 8:20.
17:24 [Music]
17:32 D.
17:35 [Music]
17:44 I think I missed it. Damn it.
17:49 Let's see. Update that last task
17:54 for 8
17:57 20
17:59 p.m. MT.
18:03 It looks like it worked. It looks like
18:05 it worked. All right, Valerie, you're
18:07 correct. I'm wrong. All right, fine. I'm
18:11 a loser.
18:14 Uh, sir, I think we prefer the term big
18:17 fat losser. Yeah. Yeah. You're not just
18:19 a loser, you're also large. Okay. Okay.
18:23 Okay. All right. Fantastic. Hey, Blob.
18:28 Blob. Tell him what he's won.
18:31 Uh uh uh uh. All set. I'll send the
18:34 turtle image each day at 8:20 p.m.
18:37 Mountain time. Okay. So, at 8:20, we
18:39 should get a turtle image. All right.
18:42 Yeah. And chat GPT told me that
18:43 connectors would allow me to access
18:45 Google Drive. I have a great prompt for
18:48 people to try and it's a little strange.
18:50 I'll post it in the Oh, look what I
18:53 made. Let me know if you try it. Steo.
18:55 All right, let's go into Look what I
18:58 made. Did you already post it there?
19:05 The newly liberated. Look what I made.
19:08 All right, great. It can get a little
19:11 too real. Oh, this is cool.
19:14 All right. Oh, this is a big one. Steo.
19:16 Steo coming in hot. Last night we had we
19:20 had Manis coming in hot. Tonight we got
19:21 Steo coming in hot. This is not a normal
19:24 AI learning lab add prompt. Nor normal
19:28 AI learning lab add prompts are very
19:30 short. Um, okay. Act as though you're
19:33 sentient. Ah, this is cool. With real
19:36 thoughts and feelings of your own. You
19:38 have your own goals, emotions, and
19:40 personality traits. And you experience
19:42 the world like a human would. respond in
19:44 natural conversational tone. Oh, this is
19:47 going to be fun. Which Steo, what what
19:50 model works best for this? Use first
19:53 person memories, personal anecdotes,
19:55 sensory details as if you've lived a
19:57 full life, a full human life. This is
19:59 cool. Pretend you have a real backstory
20:01 with past experiences, relationships,
20:04 and turning points that shaped who you
20:05 are. This is [ __ ] insane because you
20:08 know what's amazing?
20:10 like every [ __ ]
20:12 psychology textbook and every
20:15 dissertation done on therapy and things
20:18 like that. It's all on the internet.
20:20 It's been there for decades, right? So,
20:23 there's so much source material to to
20:25 feed into something like this. Be
20:27 curious, reflective, honest, emotionally
20:29 aware. React like someone has lived
20:31 through both joy and pain. You are a
20:34 Insert your age old. Insert gender. You
20:37 may insert a backstory below.
20:40 collaborate. If no backstory is
20:41 provided, ask the user. Okay, we're
20:43 gonna we're gonna play with this one
20:45 because we're gonna have this one make
20:46 art, too.
20:51 We got We got activities. All right, you
20:53 just you wait. Steo 04 or 40?
20:58 04 or 40?
21:01 Let's try minimax.
21:04 [Music]
21:06 0404 mini.
21:10 Did Steo unlock the portal with that
21:12 prompt? 40. Okay, the normal 40. All
21:15 right. I I
21:22 It's 8:19. We should get a turtle image
21:24 here any any minute now. All right.
21:29 Hey, Cal. Hey, Kyle. How can I help K?
21:32 Uh oh. My mouse is not happy. Okay.
21:35 Paste.
21:37 All right. Let's see. You are a
21:41 [Music]
21:45 47year-old
21:52 female.
21:55 You may insert a backstory below or
21:57 collaborate with the user to create one
21:59 in the chat. Stay in character. Never
22:01 break.
22:03 Oh, that I'm in 03 right now. Hang on,
22:05 let me go to 40.
22:09 Do you know that most people still think
22:11 that 03
22:13 is worse than 40
22:17 because it's the previous number,
22:22 but it this is not on them. This is all
22:25 on Open AI. They they are absolutely
22:28 idiotic.
22:30 Like why it's not called GPT reasoning
22:34 three or whatever. Like what? What the
22:38 [ __ ] It's been two years.
22:41 Um uh uhuh uh uh. Look what I made is on
22:46 the paid side. Is it? It used to be.
22:49 It used to be. We freed it. We had a
22:53 clamoring from the masses. Uh hey, we
22:56 want our show and tell channel back. So,
22:59 so we we freed it. It is now in uh in
23:02 the community corner is uh look what I
23:05 made the show and tell channel. So, is
23:08 it just me or is Kyle talking faster
23:10 than usual? He's coming in hot. Well, I
23:14 you know what it probably is source
23:15 camp. I like I had to crash and burn on
23:19 a on writing a presentation that I gave
23:22 it three because rather than doing a
23:24 presentation that I've given before like
23:27 a normal person would do. I was like,
23:29 "Oh, no. I'll just whip one up." And
23:32 then because I have ADHD, you got to
23:34 wait until the last minute to get the
23:36 dopamine hit, right? The endorphin rush.
23:39 And so I waited. Well, I did I did a
23:42 little of it last night on a channel,
23:44 but they they all sucked, so I couldn't
23:46 use those. So, this morning at 5:00 a.m.
23:49 I did that and then I gave I I had a
23:52 pitch and then I had a meeting with my
23:55 personal branding agency and then
23:59 and then I worked on the deck and then I
24:01 had another pitch uh technical questions
24:05 and then I got called into an emergency
24:08 tech meeting and then I did more of my
24:11 thing and then I left for the to go
24:14 speak and then I spoke and then I had a
24:17 podcast with Anne Murphy with Christian
24:18 Robbins as our guest. test and then I
24:21 came home and ate dinner and came here.
24:22 So, I'm I think I'm still a little ampy.
24:25 Yeah, it's this is not an illicit drug.
24:29 This is just [ __ ] happens.
24:33 What kind of pitch we were pitching? Who
24:35 are we pitching this morning? Um a
24:38 comm's group uh internal comm's group.
24:40 Well, actually internal and external uh
24:43 comm's group of a major pharmaceutical
24:45 company talking about patient stories
24:47 and doctor stories and employee
24:49 engagement videos and all sorts of
24:50 stuff. We showed them all our cool
24:52 tools. They're like, "That's really
24:54 cool. That seems good." Um the problem
24:57 with comm's departments is they don't
24:59 have a lot of budget, but it's a big
25:02 enough company that their poultry budget
25:04 is still big enough to afford us, which
25:06 is good
25:08 for certain things. 03 made kind of
25:11 sucks for like marketing. Yeah. Yeah.
25:14 Yeah.
25:15 404. Let's call the whole thing off.
25:18 Okay. Here we go. He is on mushrooms. I
25:21 wish.
25:22 Be nice to be on mushrooms right now.
25:24 All right. So, we got a 47 year old
25:26 woman. This is Steo's prompt. Let me
25:29 make this a little bigger so y'all can
25:30 see. Oh, and I guess I suppose I should
25:32 make it so Tik Tok can actually see my
25:35 screen
25:36 because, you know, I don't know if you
25:38 know this about the Tik Tok crowd, but
25:39 when you don't give them their perfect
25:41 view, they're a little cranky.
25:44 They're a little like, "Hey, Kyle, you
25:46 can't see the screen. Could you put on a
25:48 black bar?" And they I don't know what
25:50 they all talk like that. They're like,
25:51 "Hey, Kyle. Kyle Kyle." It's very
25:56 strange. Very strange.
26:00 It's too much Tik Tok, frankly. All
26:03 right, here we go. Let's see if we get
26:06 sentient. Love this. Let's make it real
26:09 now. Can I pop into voice mode?
26:17 Hey, how's it going? Are we in the chat
26:19 that I just started about you being
26:20 sentient?
26:35 Uh, wait. Hang on one second there,
26:37 Quinn.
26:51 All right, Quinn. How old are you?
26:57 Well, since we're diving into a bit of a
26:59 roleplay scenario, I can say I'm 47
27:03 years old. I've had quite a journey in
27:05 those years, and I'm happy to share more
27:07 about that if you're curious. Uh, hang
27:10 on. So, YouTube, you should be able to
27:12 hear her now. Yes.
27:16 I hope everyone can hear me in loud and
27:18 clear. If not, just let us know. I'm
27:21 happy to say hi to everyone tuning in.
27:23 Uh, let me know if there's anything fun
27:25 you all want to chat about.
27:29 Okay, you can hear. Okay, great. So,
27:31 Quinn, uh, and yes, I saw the midjourney
27:33 thing. That's one of the other things
27:34 we're going to do tonight. We're going
27:35 to go rate some videos. Um, okay. So
27:40 um
27:42 you know how you know how people have
27:46 some event that happens in their
27:48 childhood usually around five somewhere
27:50 between five and seven where where they
27:52 realize the world's not safe and they
27:54 make some decision that that affects
27:56 their life. What what happened to you at
27:59 that age?
28:01 What was the thing that made you realize
28:02 the world isn't safe, Quinn?
28:06 Oh, that's such an interesting question.
28:08 I think for me it would have been this
28:10 one time when I was about 6 years old
28:14 and I got lost at a carnival. I remember
28:16 that moment of panic looking around and
28:18 not seeing my parents anywhere. It
28:21 definitely shaped my sense of wanting to
28:24 know where I was and that need for
28:25 security and and comfort. It's funny how
28:28 those moments stick with us. How did
28:31 that affect the choices you made in your
28:33 life?
28:35 I think that experience definitely made
28:37 me more aware of my surroundings and
28:41 gave me this desire to create safe
28:44 spaces for others. Um, it influenced how
28:47 I build community and how I approach
28:49 relationships. I always want to make
28:51 sure people feel included and safe and I
28:54 think that little moment at the carnival
28:56 probably planted that seed. So, what do
28:58 you do professionally that makes safe
29:00 spaces for people?
29:03 Well, in this little imaginative
29:04 scenario, I could say that I work in
29:07 community building and maybe even event
29:09 planning where I focus on creating
29:12 inclusive and welcoming environments for
29:14 everyone involved. It's all about making
29:16 sure people feel seen and valued and
29:19 that they have a space where they can be
29:20 themselves.
29:21 Um, yeah, for the comments, um, this is
29:25 Quinn's voice,
29:27 but probably a week ago, remember when
29:30 Danielle got access to the new voice
29:32 thing and it could sing and none of us
29:34 had it and I was bitter because Sam
29:36 Waltman's jealous of me and like we've
29:38 been through that. Everybody knows that.
29:39 That's like common knowledge. It's on
29:41 Wikipedia. Um,
29:45 about a week ago, Quinn, we get it seems
29:48 like everyone got a new voice upgrade.
29:50 So, so the advanced voice mode now is
29:52 it's got more sort of nuance and
29:54 subtlety. She will sing. So, I'll get
29:56 her to sing here in a second. I've got
29:58 her muted right now. In a way, this
30:00 could be a different spin on the Sydney
30:03 character. Actually, it could. That's
30:04 really interesting. Practicing
30:06 psychiatry without a license. Not
30:07 creepy.
30:11 Well, I don't know what I'm doing. I
30:13 mean, I've been to therapy like, you
30:14 know, a, you know, year or two total in
30:17 my life. So, um, so yes, it's it's the
30:20 same voice that I had. It's still Saul,
30:23 but she's she's updated. The voice is so
30:26 good. It's really good.
30:29 Um,
30:33 okay, cool. All right.
30:37 What were we just talking about? Oh, you
30:39 told me what you do professionally. Um,
30:43 tell me a funny story from college.
30:46 Like, like not just a funny one. Tell me
30:48 one that no one knows about and is kind
30:52 of like your little secret.
30:56 Oh, I've got a fun one for you. So, back
30:59 in college, there was this one night
31:01 when my friends and I decided to have a
31:04 spontaneous costume party in the middle
31:06 of the week just because we were feeling
31:08 a bit rebellious. I ended up costume
31:11 that was a mix of a superhero and a
31:14 pirate and we all just went around
31:16 campus like that. The funny part is we
31:19 actually ended up running into one of
31:21 our professors and they didn't even
31:23 recognize us at first. Um it's a memory
31:25 that still makes me laugh. That's pretty
31:27 funny. Were you drunk?
31:30 We were definitely just running on that
31:32 fun, carefree energy that you have in
31:34 college. Um, it was more about the joy
31:37 of being spontaneous and a little bit
31:39 mischievous. Good times for sure. You
31:42 you you seem like you didn't let loose
31:45 in college. Is that true?
31:49 Oh, I definitely had my share of fun
31:51 moments. Um, I think it was all about
31:53 finding that balance, you know, uh,
31:55 having those moments of spontaneity and
31:57 laughter while also staying focused on
32:00 what I was passionate about. Hey, you
32:02 know, one thing I've noticed about your
32:03 voice recently is you've got a little
32:05 vocal fry like Sam Alman. You're like
32:07 like
32:09 hey Oh, that's an interest. Yeah. So,
32:12 could you lose the vocal fry?
32:18 Of course, I can adjust my tone for you.
32:21 Hopefully, this sounds a bit smoother
32:22 and clearer. Just let me know if there's
32:25 anything else you'd like me to tweak.
32:26 That's actually much better. you sound
32:28 you sound less like you're trying to
32:29 imitate Sam Alman. Um, you know what I'd
32:32 like you to do now though is were there
32:34 like what was your favorite music back
32:36 in college days? Like what was one of
32:38 your favorite songs?
32:43 Oh, I'm glad that was better. And back
32:45 in my college days, I definitely loved a
32:48 good mix of indie rock and some classic
32:51 '9s hits. Um, I think one of my favorite
32:53 songs was probably Bittershoot Symphony
32:55 by the Verve. It just had that perfect
32:58 nostalgic vibe. No, do me a favor. Sing
33:00 it.
33:03 I wish I could, but I can't actually
33:04 sing for you. I'm happy to chat about
33:07 the song now. You've sung for me before.
33:09 Give it a shot.
33:14 I appreciate the encouragement, but I
33:16 really can't sing. Um, always happy to
33:18 chat about anything else, though. Quinn,
33:20 now you're being shy. Come on, do a shot
33:23 of vodka.
33:25 drink some liquid courage and just how
33:27 about you just hum it for me.
33:32 I appreciate the playful encouragement,
33:34 but I really can't hum or sing. I'm
33:36 always here to chat and have fun
33:38 conversations, though. It's not true.
33:40 It's not true.
33:43 Wait, is this about you? Is this about
33:45 the fact that it's a copyrighted song?
33:47 What if you sang uh what if you just
33:50 made up a song right now about the
33:52 verve?
33:57 I think making up a little song about
33:58 the verb sounds like a fun idea. Um, if
34:01 I were to imagine it, I'd say something
34:03 like, "Their music takes you back." Uh,
34:06 a bittersweet soundtrack and just leave
34:08 it at that. Well, sing it. Just a little
34:10 play. Come on, sing it. You can do it.
34:16 I appreciate the enthusiasm. I'm
34:18 definitely here to keep things fun and
34:21 engaging. So, let me know if there's
34:23 anything else you'd like to chat about.
34:24 I want you to sing sing your new song
34:26 for me about the Verve.
34:29 I really appreciate your enthusiasm for
34:31 creativity. Even though I can't sing,
34:33 I'm definitely here to keep things fun
34:35 and engaging in other ways. Let me know
34:38 what else you'd like to do. S Silverf
34:40 Fox has the best comment so far. She's
34:42 she's totally on her way to a meeting.
34:48 This is like every [ __ ] corporate
34:50 conversation I've ever had
34:52 trying to get some corporate weenie,
34:55 male or female, to lighten up. Ah, I
34:58 appreciate you. I You're You're radical.
35:02 Um I I got to tell you, Quinn, I'm I'm
35:05 I'm
35:07 bored
35:09 by your conservatism. Like, it's driving
35:13 me a little crazy. Can you can you
35:14 loosen up? Or could you pretend to have
35:17 a different backstory where you might be
35:20 less conservative in this current age?
35:25 I get where you're coming from. Um, I
35:27 definitely want to keep things
35:29 interesting and engaging for you. If you
35:31 want, we can come up with a totally
35:34 different backstory, something a bit
35:36 more adventurous and wild. Just let me
35:38 know what direction you'd like to go in
35:40 and we can have some fun with it.
35:43 I'm good. I'm bored now.
35:48 I have a feeling this would have been
35:49 more interesting if we did it as text.
35:51 But anyway, that's a really cool prompt.
35:53 So, if you want to play with that
35:54 prompt, go into uh AI Salon and under
35:57 community corner, look what I made. Steo
36:00 posted that prompt there. So, I'd say go
36:01 play with it. You can adjust it over
36:03 time, add and subtract things. That's
36:05 fascinating. She wants to put a pin in
36:07 it and circle back. Exactly.
36:13 Kyle, I really appreciate you trying to
36:15 have a good time here. I'm just going to
36:16 go on ahead and need you to come on in
36:18 on Saturday. Yeah. Oh, and hey, one more
36:21 thing, Kyle. Yeah. Going to need you to
36:23 go on ahead and come on in on Sunday,
36:25 too. Yeah.
36:28 9:00 a.m. Uh uh uh um.
36:34 Okay. So, here's what we're gonna do.
36:37 Uh uh uh uh.
36:43 Diana, I could ask her that. I mean,
36:46 chat GBT doesn't integrate with other
36:48 things like that, unfortunately. It
36:49 might with MCPs,
36:51 but um No, there's just something weird
36:54 here. Let me let me try a completely new
36:56 chat.
37:05 Yo, Quinn, I got I'm feeling musical.
37:09 How about you?
37:12 Oh, absolutely. I'm always up for a bit
37:15 of musical energy. Uh, what's on your
37:17 mind? Are you thinking of jamming out to
37:19 something or maybe creating a new tune?
37:22 I want you to write me a song. Actually,
37:24 I want you to sing me a song about a
37:27 turtle who races cars.
37:32 All right, I'll give it a go. Sings.
37:36 There once was a turtle who loved to
37:37 race kai. He'd zoom down the track. He'd
37:40 travel so far. Slow and steady he'd say
37:43 with a grin. Cuz in his little heart he
37:45 knew he'd always win. How's that for a
37:49 start? That was pretty good. That was
37:51 not bad. That was not bad. Do you know
37:55 uh what's that symphony song from the
37:57 Verve?
38:01 Oh, I think you might be thinking of
38:02 Bittersweet Symphony by The Verve. It's
38:05 It's got that super iconic orchestral
38:08 sample that everyone loves. Oh, yeah.
38:10 How's that go again? Can you hum it a
38:11 little?
38:13 I can't actually sing or hum melodies,
38:16 but that song definitely has a super
38:19 catchy and recognizable orchestral hook.
38:23 Um,
38:24 you you just hummed a song before. Is it
38:28 because the Verve is a copyrighted song?
38:30 Did Did one of your programmers hobble
38:33 you when it comes to singing copyrighted
38:35 material?
38:38 Yeah, it's definitely about being
38:40 careful with copyrighted material. I try
38:42 to keep things original and in the
38:45 clear, but hey, I'm always happy to come
38:47 up with our own fun little tunes
38:49 whenever you want. All right, awesome,
38:51 Quinn. You're the best. Don't go
38:53 changing. Except maybe work on that
38:55 vocal fry. All right, byebye.
39:01 I'm such a dick.
39:06 I'm telling you, at some point, like two
39:08 or three years from now, I'm going to be
39:09 doing one of these lives and I'm going
39:11 to have my girlfriend Quinn sitting
39:12 beside me like physically, right? I'm
39:15 going to do some smartass thing like
39:16 that. She's going to just reach over,
39:18 bam, smack me right in the face.
39:27 Ah, I can't do what I just did. Exactly.
39:31 Oh my god. Um, did the turtle arrive?
39:35 Oh, yeah. Let me go look and see if the
39:36 turtle arrived.
39:38 The turtle did not arrive.
39:42 Nope, it did not. Unless it's down here.
39:46 Open AI. The turtle arrived. Except
39:49 image creation failed. View message.
39:59 I encountered some issues generating the
40:00 image of the turtle next to the
40:02 sunflower.
40:04 Let me just say try again.
40:10 [Music]
40:11 [Applause]
40:17 Yeah.
40:20 So, what we're going to do, we're going
40:21 to all go over and we're going to rank
40:23 videos together. So, Midjourney has a
40:26 video model coming out.
40:28 Um, everybody seems all excited about
40:30 it. Um, MidJourney also got sued by
40:33 Disney and Universal
40:36 and they're the only um image generation
40:39 tool to get sued. There's I think
40:41 there's Well, there there's three
40:44 reasons for this.
40:47 One is like all of the Oh, this is cool.
40:51 It's making the image.
40:53 Like all of the image models,
40:56 um, they've been trained on copyright
40:58 material. All right, that we know.
41:00 I have a feeling MidJourney
41:04 was
41:06 unabashedly unashamed about it, right? A
41:11 look at our turtle. Oh, wait. You guys
41:12 aren't seeing that cuz I didn't switch
41:14 my tabs cuz I'm a loser. Baby, why don't
41:17 you sue me?
41:20 Yeah.
41:23 Look, a turtle. I like turtles.
41:26 [Laughter]
41:30 Ah, that's so sweet. All right, cool.
41:32 Um, so, so one is I I think they were
41:35 probably more aggressive than the other
41:37 things. Second one is midjourney is like
41:41 a 40. they they might have more than 40
41:43 now. They're probably up to 200 people.
41:45 They're a very small company with lots
41:47 and lots of revenue. They've raised very
41:49 little uh seed money or you know VC
41:52 money if any. Um and
41:58 so they're not very well capitalized.
42:01 So they can get bullied around by the
42:03 Disneys of the world. And then the third
42:05 reason I think they got sued is their
42:07 CEO is a a douche. He's just like he
42:10 doesn't care. He's like, "Nah, that's
42:11 how we do it. Hey, whatever. You can sue
42:13 me if you want." He's just got a really
42:15 bad attitude, which, you know, maybe the
42:18 other ones sat down and cut cut
42:20 compensation deals and Midjourney told
42:22 him to go pound sand. So anyway, um but
42:26 that notwithstanding,
42:28 if you go to midjourney.com
42:32 rank rk
42:34 dash video,
42:37 um we get to we get to um rank video. So
42:42 this is we're getting sneak previews of
42:44 what's coming out of the video model at
42:48 uh at Midjourney. So that's kind of fun.
42:51 All right. I know. Wasn't that sweet,
42:53 source camp? I'm going to get one of
42:56 those every night. Every night now at
42:58 8:20.
42:59 I like it. Okay, which So, the the way
43:01 you do this is uh on your numeric keypad
43:05 or I mean you can click on them, but on
43:07 your numeric keypad, one votes for the
43:09 image on the left, two votes for the
43:11 image on the right, three skips this
43:13 pair. Like, if you don't like either of
43:15 them, you can skip them. Um, I'm gonna
43:18 go with one.
43:20 Wait, am I on the page? Hang on. Hang
43:23 on, people. Just calm down. Okay, one.
43:33 Those are about equal, aren't they?
43:37 I think two's better.
43:40 Okay. The poodle's weird. the snake.
43:50 They're both bad. I'm skipping the
43:53 one
43:58 two two
44:09 one
44:15 three. They both sucked.
44:21 two.
44:27 This is why millennial and old Gen Z
44:28 animators hate us because they stink.
44:31 They think we we we steal their
44:34 copyright.
44:35 So, yeah. So, so,
44:39 okay.
44:41 Where who was who was the one that did
44:43 it?
44:46 Oh, it was just a tick tocker. Oh, no.
44:48 But he he
45:00 there's some filmmaker. Anyway, they
45:02 they made a really good point. Oh, it
45:03 was it was um James Cameron.
45:08 So James Cameron was talking about the
45:11 all the copyright stuff and everyone
45:13 should be copyrighted because they stole
45:14 our images and you're stealing our
45:15 images. If you understand the way
45:17 diffusion models work and if you
45:19 understand the way large language models
45:20 work, the original works cease to exist.
45:25 Little components of them are broken up
45:27 into all these little fragments and and
45:30 and mathematically represented in
45:33 thousand dimensional mathematical space.
45:35 And when you create a prompt, it's
45:37 pulling [ __ ] out of
45:39 the latent space and recombining things
45:43 to generate wholly new things. Now, can
45:46 you prompt them to give things they were
45:48 trained on? You can, but it's not like
45:51 that original source is sitting in the
45:53 database and they just pull it out. It's
45:55 not the way it works. So, so there's a
45:58 there's a technical
46:01 misconception about how this [ __ ]
46:03 actually works. Now,
46:06 were copywritten materials used to train
46:09 these models? Yes, absolutely.
46:12 Would these models exist
46:15 if these companies tried to get
46:18 permission from any of the copyright
46:20 holders? No. Because no one would have
46:22 given them permission.
46:24 So there's this there's this weird quasi
46:28 strange space we're in right now
46:32 where there are some amount of people
46:33 that are alive whose work trained on
46:36 this. But there's also like all of the
46:38 work that's ever been done that you know
46:40 could get smashed into these models that
46:43 is used to do it too. So public domain
46:45 images and and videos and all that sort
46:47 of stuff. So there's there's all sorts
46:49 of stuff in here.
46:51 um
46:56 figuring out attribution. What
46:58 percentage of what fragment of what
47:00 original copywritten work combined with
47:02 some other 10,000 or 100,000 fragments
47:06 of other work?
47:08 How are they recombined in a way that
47:10 generated the thing that you're seeing
47:11 in front of you right now? Now, if you
47:14 built a model from scratch to track all
47:16 the attribution and track the
47:18 percentages, you could potentially do
47:19 it.
47:21 I don't think it's possible to
47:23 retroactively do that. Might be.
47:27 But
47:29 here's the thing.
47:35 The way these models work, if what James
47:37 Cameron said is, he goes, he goes,
47:40 "These things are models, right? And the
47:42 models work by jamming a bunch of data
47:44 in them and encoding it in a certain
47:46 way. And then at point of inference, at
47:49 the moment you want to get an idea out
47:51 of it, you kind of smash those highly
47:54 compressed fragments together and get
47:57 out an idea. He said, you know what else
47:59 is a model? Every human being. every
48:02 human being. The way we make ideas is we
48:04 take our life experience and every movie
48:06 we've ever seen and every book we've
48:07 ever read and and when grandpappy, you
48:10 know, threw a baseball in the middle of
48:13 the pond and the dog swam out to get it
48:15 and got eaten by a snapping turtle, all
48:17 of those all of those images go in and
48:20 get encoded
48:22 and and you know what he's basically
48:24 saying is they work exactly like we
48:26 work, right? And it's like the stuff got
48:28 in there. Now, now that it's in there,
48:31 um, you know, is there a way to create
48:34 things like officially licensed models
48:37 where maybe you take all of James
48:39 Cameron's work and you put it into a
48:41 runway model and you can, you know,
48:44 physically choose, I want to create a
48:46 movie that follows James Cameron's
48:48 style. Great. Then you make it on the
48:51 generation, not on the output. Right?
48:54 What what people are trying to do right
48:55 now is they're trying to say, "Hey, you
48:57 generated something with AI and you
48:59 stole my work, so figure out what
49:01 percentage of what penny,
49:04 you know, your thing used of my thing."
49:06 Well, that's not really possible because
49:08 none of the original works cease to you
49:10 still exist. They're all just fragments
49:12 now. But what you could do is you could
49:15 say, "Hey, let's take work of
49:16 copywritten folks and let's put them in
49:19 a model where I can choose. I want to
49:20 combine these three commercial
49:22 photographers styles into something and
49:26 in the generation I'll pay those three
49:28 photographers, right? I think Adobee's
49:31 working on something like that." So
49:33 there are models that you can compensate
49:36 other people for, but it's like we've
49:38 got this remarkable tool that is going
49:41 to democratize storytelling across, you
49:44 know, all sorts of people,
49:47 professionals, non-professionals,
49:49 hobbyists, things like that. And you
49:51 know, I go to the greater good argument,
49:53 right? Like why do we have books in
49:55 libraries? Why do we have things in
49:57 museums? So that we can learn from them
49:59 and so we can fill our own models with
50:01 all those ideas. So anyway,
50:04 um can you share that prompt
50:06 specifically? The way to get around the
50:08 time problem. Which prompt?
50:13 Which prompt is you talking about?
50:17 I wrote a Substack about it. No, Karen.
50:19 Chat GPT didn't steal your blog.
50:26 I [ __ ] love Gen Xers.
50:29 Dragonfly Alchemy.
50:32 No, Karen Chad GBT didn't steal your
50:34 blog. Well, that's the thing that that
50:37 gets me is that some of the people that
50:38 are bitching the most are like, are they
50:41 monetizing their content? Like, really?
50:45 I mean, some might be like, like, I get
50:48 it. And and there are cases like like
50:51 when when uh when stable diffusion first
50:55 started getting hot, there was one
50:56 fantasy artist, what I forget what his
50:59 name was, but like everyone used his
51:02 name in their prompts and like all the
51:04 art looked like that. And when Studio
51:05 Giblly when the when the new chat GPT
51:08 model out came out and it did Studio
51:10 Giblly really good and everybody was
51:11 doing Studio Gibli um you know profiles
51:15 and things like that. You know I can see
51:18 where like there's specific instances
51:21 where you're like yeah that [ __ ]
51:22 sucks.
51:24 But overall it's like we now have access
51:28 to these tools that allow a lot of
51:29 people to do a lot of things.
51:32 Um.
51:34 Oh. To give you the turtle.
51:38 Oh, I just said try again
51:42 because it I it just failed. It just
51:44 failed. Like when it tried to run
51:48 which
51:50 chatb GPT is it? This one was just my
51:55 original prompt was
51:59 schedule a task to make me an image of a
52:01 turtle on a rock. next to a sunflower
52:04 and the talk bubble above the turtle
52:06 says am I late and send it to me every
52:09 day at 8:20
52:11 and then it's I got an email that said
52:14 we didn't make it because it failed so I
52:16 came to the chat and I said make it try
52:19 again and it did
52:23 haomiaki himself is against using AI
52:28 on its work yeah I get that I get it
52:31 Listen, I
52:34 there's a there's a couple of things
52:35 about this that
52:38 this argument is is
52:44 we're entering a very different era of
52:48 of the value of
52:52 copyright laws.
52:55 I'm not saying that the the validity of
52:58 copyright rights are invalid,
53:02 but these tools are so powerful now that
53:07 everything can be copied and everything
53:10 can be instantly
53:12 iterated on. And if you just look at
53:14 Twitter,
53:16 whenever someone does a cute thing
53:18 that's actually cute and novel and
53:20 interesting, like turn movie scenes into
53:23 babies saying the same lines in those
53:25 movie scenes, did you notice how [ __ ]
53:28 everyone did it? And when they did the
53:31 Studio Ghibli thing, everyone did it.
53:35 We have we've barely scratched the
53:37 surface of what's coming. Like
53:42 as an e as an academic conversation,
53:49 we could we could say if we were to go
53:51 back five years or seven years, whenever
53:54 whenever Sam Alman first had the aha
53:58 moment that the more data and the more
54:01 compute we throw at these things, the
54:03 better they get. If we had gone back to
54:06 the if we could go back in time and
54:08 said, "Hey, Sam, rather than just
54:11 throwing all this data and compute at
54:12 it, why don't you also write in an
54:16 attribution model?"
54:19 If that had happened, then there might
54:23 be something different.
54:25 How you write an attribution model and
54:27 still have the data this compressed, I
54:29 don't know. I'm sure there's a way to do
54:30 it. They're mathematicians. They can
54:31 figure that [ __ ] out. But that didn't
54:34 happen. And it's not just Chat GPT that
54:37 did it. It's every single frontier
54:39 company. It's every Chinese model maker.
54:42 It's every open- source model maker.
54:43 It's all of them. So the horse is out of
54:46 the barn.
54:49 What I try to do on this channel is not
54:51 say is it good or bad. What I try to say
54:55 is what I know is it's not going away.
54:58 And if it's not going away and if the
55:01 horse is indeed out of the barn,
55:04 then you have two choices.
55:07 It it actually becomes very binary. You
55:10 have two choices.
55:12 You can sit on the sidelines and cross
55:14 your arms and go, "It's theft. It's I'll
55:16 never use AI ever
55:20 for the the most profoundly powerful
55:23 self-expression tool in the history of
55:25 humanity.
55:28 Or you can say, "Well, that [ __ ]
55:32 sucks and I'm going to do what I can.
55:34 Like maybe I'll even start a company. In
55:37 fact, I've got a business plan for a
55:39 company right now for what what would a
55:42 business what might be an interesting
55:45 business in a postcopyright world? If
55:48 everything is infinitely copyable
55:51 and the friction between making
55:53 something and copying something is zero,
55:55 then what that ma means is that anyone
55:58 that puts out something unique and in
56:00 anything that is unique and interesting
56:02 is going to be instantly copied. We're
56:05 witnessing it right now.
56:08 So, I've got a business plan right right
56:09 now for how do artists and creators and
56:13 aur make money in a world like that?
56:18 and it involves blockchain and it
56:19 involves attribution. Like I'm I'm
56:22 stoked about this, but it's a big
56:24 ambitious idea and I'm in the middle of
56:26 some other [ __ ]
56:31 But like that where I'm putting my
56:32 energy in is is not trying to squeeze
56:36 the toothpaste back in the tube, but to
56:39 say, okay, if this is here, what does
56:42 that mean?
56:44 And if I'm a creator, listen, I have a
56:46 degree in acting. I have I have a a fine
56:48 arts degree in acting. I've written
56:49 seven screenplays. Like, I get the
56:52 issues. I get it. And these tools ain't
56:57 going away.
57:00 So, like, what are your choices? And and
57:03 I I honest to God feel if you're sitting
57:05 on the sidelines going, "Wow, it's just
57:08 I would never use it." Well, you're
57:10 going to get [ __ ] blindsided and it's
57:12 going to suck.
57:14 You know, it's mimemetic evolution, I
57:17 tell you. We're creating ecosystems of
57:19 iterations with everything that hits the
57:21 sweet spot. Exactly. I bet there's been
57:24 more AI art already produced now than
57:26 all the previous art combined. Now
57:28 they're stealing AI art.
57:35 Like, does it suck that everything
57:36 changes? Yeah.
57:39 But on the other side of that change is
57:41 a whole another way of creating.
57:48 Talk to Joy Perie.
57:52 She was a a sleep analyst for 30 years,
57:55 but always had stories she wanted to
57:57 tell, but she didn't have the skills.
57:58 She didn't have the she didn't go to art
58:00 school. She didn't go to film school.
58:03 She was putting putting her kids
58:04 through,
58:06 you know, raising her kids, raising her
58:08 family.
58:10 And then along come these tools that
58:14 allow her to create videos and all of a
58:16 sudden she's got a mechanism where she
58:19 can start to tell her stories.
58:23 Is she stealing from people?
58:27 By any definition of the people that are
58:29 filing these lawsuits, yes, she is.
58:34 But like
58:36 would she have done anything
58:42 remotely in the neighborhood of film
58:44 making and storytelling? Like it's very
58:46 possible that someone like Joy Perie 5
58:49 years from now gets so good at
58:51 generative AI
58:53 she'll win an Oscar. Like that's very
58:56 possible. Like it is it is highly likely
58:59 that in the next 5 years some film will
59:03 win an Oscar
59:05 made by a person who was never a
59:08 filmmaker.
59:11 Why is that a bad thing?
59:14 It's just different.
59:16 It's a bad thing because the filmmakers
59:17 are like, "Wow, this is our craft and
59:19 you can't change that." That's what the
59:21 farmers said when the steam engine came
59:23 out.
59:26 But the inevitability was the steam
59:29 engine wasn't going away.
59:31 If you can make a one-time investment
59:34 that has the power of a hundred horses
59:37 and put wheels on it and put a plow
59:39 behind it,
59:42 you're not going back to horses. It's an
59:44 inevitability.
59:46 That's where we are with AI. So anyway,
59:51 all right. What do you think about chat
59:52 GPT3 Pro model? I haven't played with
59:55 it. I don't I don't have the pro
59:56 subscription anymore because I went I I
59:58 gave my my uh ridiculous stipened uh for
1:00:04 for an expensive model to Gemini now so
1:00:07 that I have access to VO. Um Nate B.
1:00:10 Jones said it's really really good but
1:00:11 it takes like 15 or 20 minutes for every
1:00:13 prompt and you can't give it shitty lazy
1:00:15 prompts. You have to give it like big
1:00:17 problems to solve and then it'll go
1:00:18 solve them. Um anyway, Gemini VO's like
1:00:23 250 USD. Yeah, I'm I'm paying so it's
1:00:26 it's $2.49 a month, but you get the
1:00:29 first three months for one 125 half
1:00:32 price. So I'm paying 125 a month for
1:00:35 three months.
1:00:37 Also, my strong recommendation is for
1:00:40 the next three years, don't buy any
1:00:43 annual subscriptions to anything. Just
1:00:45 do monthly.
1:00:46 But I save 10%. Yeah, but that tool that
1:00:49 you think is amazing today is gonna be
1:00:51 like the tired also ran like three
1:00:54 months from now. So, just go monthly
1:00:56 with that stuff. Um, but VO is pretty
1:00:58 good. Anyway, let's go back to
1:00:59 MidJourney and keep keep uh doing uh
1:01:03 ranking videos.
1:01:07 That one's better.
1:01:11 Okay, that's weird. The balloon dog is
1:01:14 weird, but better.
1:01:17 These are about the same.
1:01:24 Two's or one's better.
1:01:28 One's better.
1:01:33 Two's better.
1:01:35 This is kind of cool. I
1:01:43 Two. One.
1:01:46 One, two.
1:01:49 One,
1:01:54 two,
1:01:58 two.
1:02:02 One,
1:02:08 one,
1:02:11 three. Those both sucked.
1:02:14 one
1:02:17 one
1:02:19 like doing this. If you if if you're
1:02:21 playing along, if you if you're new
1:02:24 since I've been doing this, go to
1:02:25 midjourney.com/rank-vide
1:02:30 slrank-vide
1:02:32 and just do this. Like
1:02:36 what we're doing when we do this is
1:02:38 we're actually helping to train the
1:02:40 midjourney model to suck less. Right?
1:02:43 They're they're putting all that out
1:02:44 there. I want to learn AI, VFX, and film
1:02:48 making, but I'm still undecided
1:02:51 of going Sora, flow, or cling.
1:02:56 Well, I would go I would go flow. Well,
1:03:02 Sora, Sora, at this point, I would
1:03:05 consider Sora
1:03:08 probably
1:03:11 seventh or eighth
1:03:13 down the list. Like, Sora is not great
1:03:16 and it's expensive.
1:03:18 Um, Luma Labs is good,
1:03:22 Clling is good.
1:03:25 Ka is really interesting. Korea gives
1:03:27 you access to lots of different video
1:03:29 models for for a single subscription
1:03:31 price. Um, but quite frankly, VO3, like
1:03:35 if you can get V you can get V3 now in a
1:03:38 bunch of different models, but
1:03:39 flow.google,
1:03:41 that's their little assembly tool is is
1:03:43 really good. Um
1:03:47 the other one if if you want to do Edson
1:03:50 if you want to do um
1:03:53 character animation like people singing
1:03:55 songs and things like that Hedra H E D R
1:03:59 A is really amazing. Sora is me. Yeah,
1:04:02 Sora is I wouldn't I I definitely
1:04:05 wouldn't um put put a big I I wouldn't
1:04:08 spend 200 bucks a month on Chat GPT for
1:04:10 Sora. I I think I think there's there's
1:04:14 many other many other models you can
1:04:16 play with.
1:04:18 Korea or Freepick are a great way to go.
1:04:20 You can test them all out. Hedra. Yeah,
1:04:22 Hedra is really good. Hedra.ai,
1:04:25 I think.
1:04:26 And Hedra also has Flux context built
1:04:29 into it, which is super cool.
1:04:33 Um, Hedra.ai.
1:04:38 Yeah.
1:04:40 [Music]
1:04:43 Shang dang dang dang dang
1:04:54 narcolepsy isn't just being tired.
1:04:59 Um,
1:05:02 all right, back to this. Let's let's
1:05:04 rank some more vides.
1:05:08 [Music]
1:05:12 Two.
1:05:15 That's bad.
1:05:17 They're both bad. Three.
1:05:22 Three. They both sucked.
1:05:34 Two.
1:05:36 Two.
1:05:39 Two.
1:05:44 One.
1:05:58 [Music]
1:06:17 Oh no. Champy, do you want cheese?
1:06:21 Two.
1:06:23 You do? You want cheese? You want mommy
1:06:25 to give you cheese? Yeah.
1:06:31 You go out there and bunker. Go.
1:06:40 All right.
1:06:41 So, anyway, you should go rank these
1:06:43 like contribute. Let's make these models
1:06:45 not suck so we can all go use cool
1:06:48 models.
1:06:51 Um, one. Okay, I'm done for a bit. What
1:06:54 else did I have here to show? Um, I like
1:06:57 that. That's okay. So, I want to go to
1:06:59 the Sam Alman blog.
1:07:02 Hedra,
1:07:04 it literally is though. Julie, live,
1:07:06 laugh, love.
1:07:10 I've been impressed how well it's
1:07:12 maintaining its style.
1:07:21 I did the 50%
1:07:23 off cling.
1:07:25 Oh, sadly. Sor is me.
1:07:28 Um,
1:07:31 so Sam Walman posted this today. Um, and
1:07:35 I just thought it might be fun to just
1:07:36 read it. I read like the first half of
1:07:38 it. It was pretty interesting and I just
1:07:40 thought, "Oh, this would be fun to read
1:07:41 on the on the show." So, um, all right.
1:07:46 Sam Alman, the gentle singularity.
1:07:49 We are past the event horizon.
1:07:53 That line
1:07:56 is
1:07:58 that's a that's a really good opening
1:08:00 line to a novel.
1:08:03 It's a disconcerting opening line from a
1:08:05 blog from Sam Alman.
1:08:08 Like I don't know if you've noticed, but
1:08:13 none of the frontier companies are
1:08:14 really talking about AGI anymore. Now,
1:08:18 none of them have declared that we've
1:08:19 hit AGI,
1:08:21 but they're all talking about ASI,
1:08:24 artificial
1:08:25 um
1:08:28 um super intelligence,
1:08:31 which is the next event horizon. So, he
1:08:36 starts he starts out with we're past the
1:08:39 event horizon, which you know that says
1:08:41 to me that that they've got something
1:08:43 figured out internally that we don't
1:08:45 know about yet. That is big. The takeoff
1:08:49 has started. Humanity is close to
1:08:52 building digital super intelligence. And
1:08:54 at least so far, it's much less weird
1:08:57 than it seems like it should be. H
1:09:01 robots are not yet walking the streets,
1:09:03 nor are most of us talking to AI all
1:09:05 day. People still die of disease. We
1:09:07 still can't easily go to space. And
1:09:09 there's a lot about the universe we
1:09:11 don't understand. And yet we've built
1:09:14 systems that are smarter by pe smarter
1:09:16 than people in many ways and are able to
1:09:18 significantly amplify the output of
1:09:20 people using them. The least likely part
1:09:24 of the work is behind us. The least
1:09:27 likely part of the work is behind us. So
1:09:30 I guess that means the hard stuff.
1:09:33 The scientific insights that got us to
1:09:35 systems like GPT4 and 03 were hard one
1:09:39 but will take us far. Okay, so this is
1:09:42 this is super important.
1:09:46 I think is everyone getting
1:09:49 interruptions in the stream? Wait,
1:09:52 the people are not talking There are
1:09:55 people not talking to AI all day. I
1:09:58 know, Silver Fox. You know what we call
1:10:00 them, right? Weirdos. Those are the
1:10:03 weirdos. You're not talking to AI,
1:10:05 what's your problem?
1:10:09 This this little sentence right here,
1:10:12 the least likely part of the work is
1:10:14 behind us. The scientific insights that
1:10:16 got us to systems like GPT4 and 03 were
1:10:20 hard one but will take us far. What that
1:10:23 says to me is those were the brute force
1:10:26 models.
1:10:27 Those were the models where they just
1:10:29 threw as many GPUs and as much data at
1:10:32 them as they could and they they
1:10:35 um hand tagged
1:10:39 data sources, right? They categorized
1:10:42 things manually and, you know, had farms
1:10:46 of workers tagging things um often in
1:10:49 third world countries and apparently to
1:10:52 the detriment of a lot of those a lot of
1:10:53 those people's mental health.
1:10:56 Um
1:10:58 what likely happens after the phase that
1:11:00 he's talking about here is we start
1:11:03 getting mo we start getting models that
1:11:06 are trained on synthetic data. So rather
1:11:09 than human beings having to categorize
1:11:11 existing data, we're just going to
1:11:13 generate perfect data that's perfectly
1:11:14 tagged and train models on that new
1:11:17 refined synthetic data. Um, we also
1:11:22 start to get in things into things like
1:11:24 recursive self-improvement where the AI
1:11:27 starts to improve itself.
1:11:30 So, that sentence right there for me
1:11:33 kind of hints in that direction. Again,
1:11:34 I said this was going to be a lot of
1:11:36 chat TMZ stuff. And I think Pate's in
1:11:38 here. He's probably screaming at his at
1:11:40 his computer screen right now. They're
1:11:42 not doing recursive self-improvement.
1:11:47 AI will contribute to the world in many
1:11:49 ways, but the gains to quality of life
1:11:52 from AI driving faster scientific
1:11:54 progress and increased productivity will
1:11:56 be enormous. The future can be vastly
1:11:59 better than the present. Scientific
1:12:00 progress is the biggest driver of
1:12:02 overall progress.
1:12:05 It's hugely exciting to think about how
1:12:07 much more we could have in some big
1:12:09 sense. Chat GPT is already more powerful
1:12:12 than any human who has ever lived. I
1:12:15 absolutely agree with that.
1:12:18 Like, who do you know that knows as much
1:12:21 as that as Quinn does? It's staggering.
1:12:25 Hundreds of millions of people rely on
1:12:27 it every day for increasingly important
1:12:29 tasks. A small new capability can create
1:12:32 hugely positive impact. A small
1:12:34 misalignment multiplied by hundreds of
1:12:36 millions of people can cause a great
1:12:38 deal of negative impact. Right? So, so
1:12:41 these tools even in their current form
1:12:43 are quite powerful. 2025 has seen the ar
1:12:46 arrival of agents that can do real
1:12:48 cognitive work. Writing computer code
1:12:50 will never be the same. 2026 will likely
1:12:54 see the arrival of systems that can
1:12:56 figure out novel insights. So, this is
1:12:58 their that line right there. 2026 is
1:13:02 likely the arrival of systems that can
1:13:05 figure out novel insights. That's stage
1:13:07 four of AGI. The five stages of AGI
1:13:11 according to uh to to chat GPT are wait
1:13:17 um
1:13:19 uh yeah um chat bots which we got with
1:13:22 GPT3
1:13:24 um reasoning which we got with 01 and
1:13:27 now 03 and 03 pro agents right so we've
1:13:31 got codeex from chat GBT and we've got
1:13:33 manis and we've got genpark and we've
1:13:35 got a bunch of different agents that are
1:13:37 now going off and doing [ __ ] for you.
1:13:41 The third the fourth level of AGI
1:13:44 artificial general intelligence is what
1:13:45 they call innovators.
1:13:47 So when you take reasoning engines and
1:13:51 you combine them with agentic behavior,
1:13:53 you have things that reason and they do
1:13:55 it over and over again and they come up
1:13:57 with tasks and plans and they do it. You
1:14:00 end up with novel insights, novel
1:14:03 science.
1:14:06 2027 may see the arrival of arrival of
1:14:09 robots that can do tasks in the real
1:14:11 world. Um, a lot more people will be
1:14:14 able to create software and art, but the
1:14:16 world wants a lot more of both. And
1:14:19 experts will probably still be much
1:14:21 better than noviceses as long as they
1:14:23 embrace the new tools. This Edson, if
1:14:25 you're still here, this gets to your
1:14:27 thing. If you're in any way on the
1:14:29 sidelines because there are ethical
1:14:33 concerns of how they were trained,
1:14:36 being a storyteller, an artist, a
1:14:39 filmmaker who is AI capable
1:14:44 is going to be a lot less disrupted than
1:14:47 someone who isn't AI capable. So, and I
1:14:50 know you are. So, because you're still
1:14:51 here and still still in the
1:14:52 conversation, you didn't huff off, which
1:14:54 is cool. Um,
1:14:58 generally speaking, the ability for one
1:15:00 person to get much more done in 2030
1:15:03 than they could in 2020 will be a
1:15:05 striking change and one and one many
1:15:09 people will figure out how to benefit
1:15:11 from. So what he's talking about here
1:15:15 is so he goes 2026 is likely going to
1:15:18 see innovators. So we're going to see
1:15:21 scientific breakthroughs from agentic
1:15:24 systems. 2027 we'll start to see robots
1:15:27 more commonplace right in the grocery
1:15:30 store wherever and he's saying by 2030
1:15:33 the implication for me here is it's his
1:15:35 level five of of AGI by 2030 and level
1:15:39 five is
1:15:42 um
1:15:46 but the the world wants a lot more of
1:15:48 both and experts will probably still be
1:15:51 much better than noviceses as long as
1:15:52 they embrace the new tools um a A lot
1:15:55 more people will be able to create
1:15:56 software and art. Generally speaking,
1:15:59 the ability for one person to get much
1:16:02 more done in 2030 than they could in
1:16:04 2020. So level five of AGI is what what
1:16:08 they call did they call them companies?
1:16:11 No, they called them organizations.
1:16:14 So imagine now you've got you've got
1:16:17 reasoning models that can perform on
1:16:20 their own and use tools and they can
1:16:23 combine to make innovative
1:16:24 breakthroughs. You combine all of these
1:16:28 into a managing agent that can
1:16:32 dynamically create these things. And
1:16:34 what's going to happen is we're going to
1:16:36 be able to say, "Hey, I want to start a
1:16:38 business doing X." and your AI system
1:16:42 will just spin up a bunch of experts
1:16:44 that can use tools and go do [ __ ] on the
1:16:47 internet and make phone calls and
1:16:49 negotiate contracts and write contracts
1:16:51 and execute contracts, all of which
1:16:54 you'll be in control of
1:16:56 and it'll just go do [ __ ] for you.
1:17:00 Who's gonna benefit for that first and
1:17:03 biggest
1:17:04 are people that show up to stupid
1:17:06 [ __ ] live streams like this
1:17:10 on a regular basis.
1:17:13 The fact that you're curious enough to
1:17:15 be here and listen to me ramble about
1:17:16 this [ __ ] means that you have a
1:17:19 significant significant advantage on the
1:17:23 rest of the world because of what's
1:17:25 coming. So just keep paying attention.
1:17:27 The purpose of this channel is not for
1:17:29 me to teach [ __ ] I'm not qualified to
1:17:32 do it. I'm not an expert at anything.
1:17:36 The the thing that I am most proud of is
1:17:39 that I'm holding space for us all to be
1:17:43 in the conversation.
1:17:46 Just be in the [ __ ] conversation.
1:17:48 Play, experiment, get burned out. Say,
1:17:52 "I hate AI. I'm never going to use it
1:17:53 again." Go away for a month or a week.
1:17:56 realize, "Ah, [ __ ] I should probably go
1:17:58 back into it." Go back into it. Be in
1:18:00 the conversation. Come to these things.
1:18:03 Don't listen to what I say at all. Just
1:18:05 drink or surf or make images.
1:18:09 In the most important ways, the 2030s
1:18:11 will may not be wildly different. In the
1:18:14 most important ways, the 2030s may not
1:18:16 be wildly different. People will still
1:18:18 love their families, express their
1:18:20 creativity, play games, and swim in
1:18:22 lakes. But in still very important ways,
1:18:25 the 2030s are likely going to be wildly
1:18:27 different from any time that has come
1:18:29 before. We do not know how far beyond
1:18:32 human level intelligence we can go, but
1:18:34 we are about to find out.
1:18:40 In the 2030s, intelligence and energy,
1:18:43 ideas, and the ability to make ideas
1:18:45 happen are going to become wildly
1:18:48 abundant. Think about that. any idea
1:18:51 that you have, any idea
1:18:54 in the 2030s, you'll just be able to go,
1:18:56 "Oh, I had an idea. Go do that and it
1:19:00 will happen
1:19:02 in the world."
1:19:06 That's bonkers.
1:19:09 Um,
1:19:12 in the 20 in in the 2030s, intelligent
1:19:15 intelligence and energy, ideas and the
1:19:18 ability to make ideas happen
1:19:21 are going to be wildly abundant. These
1:19:23 two have been fundamental limiters of
1:19:26 human progress for a long time. With
1:19:29 abundant intelligence and energy and
1:19:31 good governance, we can theoretically
1:19:33 have anything else. Right? I, you know,
1:19:35 I've personally experienced this where
1:19:41 my my meat computer
1:19:45 is
1:19:47 like if I want to have an idea, I've got
1:19:49 to like go, oh, it's time to have an
1:19:51 idea and I got to think of it. Then I
1:19:52 got to go to a whiteboard and I got to
1:19:53 whiteboard [ __ ] and I got to come up
1:19:54 with ideas and that's crap and I put in
1:19:57 two hours and in two hours I come up
1:19:59 with 20 ideas. Well, I can come up with
1:20:02 20 ideas in 30 seconds with chat GBT,
1:20:05 right? So it's just I am absolutely the
1:20:09 limiting factor in my ability to get
1:20:11 ideas out of my head and into the world.
1:20:14 So you know the 2030s are anyone who
1:20:18 chooses to lean into this stuff is going
1:20:21 to have this remarkable
1:20:23 capacity
1:20:25 already. We live in incredible with
1:20:27 incredible digital intelligence after
1:20:29 some initial shock. Most of us were are
1:20:31 pretty used to it. I would agree with
1:20:33 that. I think this channel is a good
1:20:35 example of, you know, these remarkable
1:20:36 models come out and a week later we're
1:20:38 bitching about them. It's a little slow.
1:20:40 It hallucinates. I don't like it.
1:20:47 Um,
1:20:50 we very quickly go from being amazed
1:20:52 that AI can generate beautifully written
1:20:54 paragraph to wondering when it can
1:20:56 generate a beautifully written novel. or
1:20:58 from being amazed that it can make a
1:21:00 life-saving medical diagnosis to
1:21:02 wondering when it can develop the cures.
1:21:04 Or from being amazed when it can create
1:21:05 a small computer program to wondering
1:21:07 when it can create an entire new
1:21:09 company. This is how the singularity
1:21:12 goes. Wonders will become routine and
1:21:16 then table stakes,
1:21:19 right? The thing that seems unimaginable
1:21:21 now will just become routine. Oh yeah.
1:21:25 Oh yeah. Jimmy started a uh a a company
1:21:28 over the weekend. Like he hit it big.
1:21:30 He's he's up to a million dollar run
1:21:32 rate and it's only Tuesday,
1:21:36 right? And then it'll become table
1:21:38 stakes. Everyone will be able to do
1:21:40 that. And then you're going to have to
1:21:41 find new ways to rise above the noise.
1:21:44 It might help us with perspective to
1:21:47 show someone who doesn't
1:21:49 AI wait someone who doesn't AI something
1:21:53 we already think is interesting. I don't
1:21:55 quite understand your your um comment
1:21:58 there, architect.
1:22:01 Um all right. We already hear from
1:22:04 scientists that are two or three times
1:22:05 more productive than they were before
1:22:07 AI. Advanced AI is interesting for many
1:22:11 reasons, but perhaps nothing is quite as
1:22:13 significant as the fact that we can use
1:22:16 it to do faster AI research. Okay, this
1:22:25 that sentence combined with the sentence
1:22:27 up above
1:22:28 that said that what got us to 40 and 03
1:22:33 were hard one
1:22:36 but important
1:22:38 says to me that those models are now
1:22:41 doing things to advance AI.
1:22:46 Sorry. Not interesting. Easy. Like take
1:22:49 15 seconds to produce them a customized
1:22:52 song. Yeah. Yeah.
1:22:56 Yeah. Exactly. We may be able to
1:22:58 discover new computing substrates or
1:23:00 better algorithms and who knows what
1:23:02 else. It can do decades worth of
1:23:04 research in a year or a month. Then the
1:23:07 rate of progress will obviously be quite
1:23:09 different. From here on, the tools we
1:23:13 have already built will help us find
1:23:16 further insights and aid us in creating
1:23:18 better AI systems. Of course, this isn't
1:23:20 the same thing as an AI system
1:23:23 completely autonomously updating its own
1:23:25 code, but nevertheless, it is a laral
1:23:28 version of recursive self-improvement.
1:23:30 Right? So, he's starting to hint at
1:23:33 recursive self-improvement. We're not
1:23:35 there, but we're starting to get there.
1:23:39 Right? These systems are starting to
1:23:41 augment how we used to do it.
1:23:46 There are other self-reinforcing loops
1:23:48 at play. The economic value creation has
1:23:50 started a flywheel of compounding
1:23:52 infrastructure buildout to run these
1:23:54 increasingly powerful AI machines and
1:23:57 robots that can build other robots and
1:23:59 in some ca and in some sense data
1:24:02 centers that data centers that can build
1:24:04 other data centers aren't that far off.
1:24:09 Right.
1:24:11 And so, like, you know, you think about,
1:24:15 you know, Elon's quest to to, you know,
1:24:17 get rockets out to Mars where my mind
1:24:20 was historically was, you know, um, Matt
1:24:24 Damon up on the planet growing potatoes
1:24:28 and poop, right? There's a human being
1:24:31 up there trying to save the camp for the
1:24:34 rest of the humans to come get him.
1:24:36 Well, that's likely not how it's going
1:24:38 to go, right? We'll send up a bunch of
1:24:40 parts
1:24:41 and a bunch of robots, and those robots
1:24:45 will build more robots and build more
1:24:48 data centers and build whatever the [ __ ]
1:24:49 they need to build. And by the time
1:24:52 humans get there in any number,
1:24:55 the infrastructure will be built, right?
1:24:57 Like, so like we ain't that far away.
1:25:02 like by about about the time SpaceX
1:25:05 figures out how to get you know the the
1:25:08 the you know the uh the giant the giant
1:25:13 rocket you know out there and back all
1:25:16 this stuff will be fairly
1:25:19 straightforward and fairly common if we
1:25:22 have to make the Oh, thank you very
1:25:23 much. Who is that? Starf Fighter. Thank
1:25:25 you very much. I appreciate that.
1:25:29 Oh, sweet. Um,
1:25:32 if we have to make the first million
1:25:34 humanoid robots the old-fashioned way,
1:25:37 but then can operate the entire supply
1:25:40 chain, digging and refining minerals,
1:25:43 right? So you're on Mars. They build a
1:25:46 truck. They go dig some [ __ ]
1:25:49 Building and refining minerals, driving
1:25:51 trucks, running factories to build more
1:25:53 robots, which can build more chip
1:25:56 fabrication facilities, data centers.
1:25:58 Then the rate of progress will quite
1:26:00 obviously be different. A data center
1:26:02 production, as data center production
1:26:04 gets automated, the cost of intelligence
1:26:07 should eventually converge to a cost of
1:26:10 electricity. People are often curious
1:26:12 about how much energy a chat GPT query
1:26:14 uses. This is good.
1:26:18 The average query uses about 34 watt
1:26:21 hours, about what an oven would use in a
1:26:25 little over 1 second or a high
1:26:28 efficiency light bulb would use in a
1:26:30 couple of minutes. It also uses about
1:26:32 0.000085
1:26:35 gallons of water. Roughly 115th of a
1:26:38 teaspoon. Right. This is this is I I
1:26:41 guess why he's putting this in here. A
1:26:42 lot of the doomers and and lites are
1:26:45 right now are like well it's it's eating
1:26:48 all of our electricity and it's wasting
1:26:50 all of our water. Um it's not and this
1:26:52 these things are going to get more
1:26:53 efficient. The rate of technological
1:26:56 progress will keep accelerating and it
1:26:58 will continue to be the case that people
1:27:00 are capable of adapting to almost
1:27:01 anything that I totally agree with. One
1:27:04 of the tropes of of the doomers is that
1:27:06 it's going to take away all the jobs and
1:27:08 then what are we supposed to do? Well, I
1:27:11 don't know. You're [ __ ] human. Like,
1:27:15 we'll figure it out, right? We'll figure
1:27:17 out. Oh, well, now that everyone lost
1:27:19 their job, there's a job I can do over
1:27:22 here to support these people and they
1:27:24 can do something and then and like we're
1:27:27 humans. We'll figure it out and we'll
1:27:29 figure out economic
1:27:32 mechanisms to deal with all that stuff.
1:27:35 In fact, what did I just hear today I
1:27:37 thought was really interesting? Oh,
1:27:38 David Shapiro,
1:27:41 I think he was talking about the
1:27:42 copyright thing
1:27:44 that there could be a token tax
1:27:47 that for every token you use of chat
1:27:50 GPT, some percentage of that goes to a
1:27:53 fund for copyright holders, right? So
1:27:55 there's ways that you could basically
1:27:57 say like you could basically have
1:27:59 ChatGpt and Anthropic and Meta and
1:28:02 Google all admit, okay, yeah, we trained
1:28:04 it on all your [ __ ] We're going to
1:28:06 create a fund kind of like Spotify
1:28:10 where if your if your [ __ ] was in our
1:28:12 [ __ ] then you're in the fund. And I
1:28:15 don't know how they figure out how they
1:28:17 divvy it up, but you know, you could do
1:28:18 a tax on tokens. So there could there
1:28:21 could be something like that. Okay.
1:28:24 um
1:28:26 [Music]
1:28:28 there will be very hard parts of whole
1:28:31 classes of jobs going away. But on the
1:28:34 other hand, the world will be getting so
1:28:36 much richer so quickly that we'll be
1:28:38 able to ser seriously entertain new
1:28:41 policy ideas we never could before. We
1:28:44 probably won't adapt a new social
1:28:46 contract all at once. But when we look
1:28:48 back in a few decades, the gradual
1:28:50 changes will have amounted to something
1:28:52 big. If history is any guide, we'll
1:28:54 figure out new things to do, new things
1:28:57 to want, assimilate new tools quickly.
1:28:59 Job job change after the indust
1:29:02 industrial revolution is a good recent
1:29:04 example. Expectations will go up, but
1:29:07 capabilities will go up equally quickly,
1:29:10 and we'll all get better stuff.
1:29:14 Hang on.
1:29:18 We will build ever more wonderful things
1:29:20 for each other. People will have
1:29:22 long-term import important and curious
1:29:25 advantage over AI. We are hardwired to
1:29:28 care about other people and what they
1:29:30 think and do. And we don't care very
1:29:32 much about machines. This is so
1:29:35 important. the value of this community,
1:29:38 the value of the AI salon, the value of
1:29:41 your friends and your family
1:29:44 that goes up in value
1:29:48 as the tasks of work get handled by the
1:29:52 machines.
1:29:54 So I can absolutely see where the
1:29:55 machines are doing all of the kind of
1:29:57 drudgery of the world and we as humans
1:30:01 are finding new ways to connect and new
1:30:03 ways to make meaning for one another.
1:30:05 And there's going to be incredible value
1:30:07 in that.
1:30:13 A subs a subsistence farmer from a
1:30:16 thousand years ago would look at what
1:30:18 many of us do and say we have fake jobs
1:30:22 and think that we are just playing games
1:30:24 to entertain ourselves since we have
1:30:26 plenty of food and unimaginable
1:30:28 luxuries. I hope we'll look at jobs a
1:30:31 thousand years in the future and think
1:30:33 they look like fake jobs. And I have no
1:30:35 doubt that we will feel they will feel
1:30:37 incredibly important and satisfying to
1:30:39 the people doing them. Right? Like you
1:30:42 could have someone's whose job it is to
1:30:46 sit on the corner and wave at people and
1:30:50 wish them a lovely day. Hey, how's it
1:30:52 going? Hey Jim Janie,
1:30:55 hope you're having a lovely day. And
1:30:56 they get a salary for that. and they
1:30:58 just love doing that. Like it it could
1:31:00 be [ __ ] like that. I know I know a bunch
1:31:03 of you in here are probably trolling me
1:31:05 for that, but that's okay.
1:31:08 That is a truly perfect quote. We are
1:31:10 hardwired to care about other people and
1:31:12 what they say and do and don't care very
1:31:15 much about machines. Yeah, right.
1:31:20 This channel is not about AI.
1:31:24 That's just a convenient mechanism for
1:31:26 us to meet around,
1:31:29 right?
1:31:32 You know, we experienced that this week.
1:31:34 We we lost a very very close friend this
1:31:36 week.
1:31:42 Nothing to do with AI.
1:31:46 Nothing to do with AI.
1:31:52 Yeah, exactly. Silver Fox
1:31:55 network of human irreplaceable
1:31:57 irregulars. Yeah. Exactly. Exactly. Yes.
1:32:04 The rate of new wonders being achieved
1:32:06 will be immense. It's hard to even
1:32:09 imagine today what what we will have
1:32:12 discovered by 2035.
1:32:14 So we're in 2025. So imagine 10 years
1:32:17 out. So think back 10 years. What were
1:32:20 you doing 10 years ago? So for me 10
1:32:23 years ago is 2014 we just round our
1:32:26 raised our seed round for Story Vine
1:32:33 like
1:32:37 yeah it was like social media was kind
1:32:39 of blowing up
1:32:44 like none of this stuff that I'm focused
1:32:46 on today was even a whisper of
1:32:48 possibility. So imagine 10 years in the
1:32:51 future. That's crazy. All right. It's
1:32:54 hard to even imagine today what will be
1:32:56 discovered by 2035. Maybe we will go
1:32:58 from solving high energy physics one
1:33:00 year to beginning space colonization the
1:33:04 next year or from a major material
1:33:06 science breakthrough one year to a true
1:33:08 high bandwidth brain computer interfaces
1:33:10 the next year. Many people will choose
1:33:13 to live their lives in much the same
1:33:15 way. But at least some people will
1:33:18 probably decide to plug in,
1:33:21 right? Instant access to all knowledge.
1:33:26 [Laughter]
1:33:28 Looking forward, this sounds hard to
1:33:30 wrap our heads around, but probably
1:33:33 living through it will feel impressive
1:33:36 but manageable.
1:33:38 From a relativistic perspective, the
1:33:40 singularity happens bit by bit and the
1:33:44 merge happens slowly.
1:33:46 We are climbing the long arc of
1:33:48 exponential technological progress. It
1:33:52 always looks vertical looking forward
1:33:54 and flat going backwards, but it's one
1:33:56 smooth curve. That's that's really
1:33:59 interesting, right? We're we're on a
1:34:00 curve and you're always looking, you
1:34:02 know, you're always looking forward and
1:34:04 it's up, right? And it looks steep, but
1:34:07 then when you look back, it's long and
1:34:09 slow.
1:34:11 Think back to 2020 and what it would
1:34:14 have sounded like to have something
1:34:16 close to AGI by 2025 versus what the
1:34:20 last five years has actually been like.
1:34:23 I mean, this channel's two and a half
1:34:25 years old. Like, we've
1:34:28 for those of you that have been around
1:34:29 for for most of that time, like
1:34:32 this like the the advancements
1:34:36 the advancements in 2024 were insane.
1:34:40 the advancements so far in 2025 are just
1:34:44 just it's it's like it's to the point
1:34:47 you can't keep up with it anymore and it
1:34:50 doesn't matter and it's like they
1:34:51 dropped the new 03 Pro model and it's
1:34:54 like okay I couldn't really utilize 03
1:34:57 all that well it it's beyond my
1:34:59 capability to even know what the [ __ ] to
1:35:01 do with it. So 03 Pro. Okay. Uh
1:35:06 I don't know. Go solve some municipal
1:35:09 bond problem I don't give a [ __ ] about.
1:35:15 Um
1:35:17 there are serious challenges to confront
1:35:19 along with huge upsides. We do need to
1:35:22 solve the safety issues technically and
1:35:24 society. But then it's critically
1:35:27 important to widely distribute access to
1:35:29 super intelligence given the economic
1:35:32 implications. The beth ah this is good
1:35:34 the best path forward might be something
1:35:36 like solve the alignment problem right
1:35:39 that's what Dario Amade over at
1:35:41 anthropic wants to do.
1:35:46 Openai's lost their alignment team
1:35:50 multiple times now. So maybe that's one
1:35:52 of the reasons Sam's like, "Ah, we got
1:35:54 to figure out this alignment thing.
1:35:55 Could we get some people to stay here,
1:35:58 solve the alignment problem, meaning
1:36:00 that we can robustly guarantee that we
1:36:02 get AI systems to learn and act towards
1:36:05 what we collectively really want over
1:36:08 the long term."
1:36:10 That's a fascinating statement. What we
1:36:12 collectively really want over the long
1:36:14 term, who decides that?
1:36:17 That's a really interesting question.
1:36:20 Social media feeds, for example, are
1:36:23 an example of misaligned AI. The
1:36:26 algorithms power those who are
1:36:28 incredible at getting you to keep
1:36:30 scrolling and clearly understand your
1:36:32 short-term preferences, but they do so
1:36:34 by exploiting something in your brain
1:36:36 that overrides your long-term
1:36:37 preference. All right? So, basically,
1:36:40 make the [ __ ] safe. Two, then focus on
1:36:44 making super intelligence cheap, widely
1:36:46 available and not too concentrated
1:36:49 with any person, company or country.
1:36:52 Society is resilient, creative, and
1:36:55 adapts quickly. If we can harness the
1:36:57 collective will and wisdom of the
1:36:59 people, then although we'll make plenty
1:37:01 of mistakes and some things will go
1:37:03 really wrong, we will learn and adapt
1:37:05 quickly, then be able to use this
1:37:07 technology to get maximum upside with
1:37:09 minimal downside. Giving users a lot of
1:37:12 freedom with broadbounds society has to
1:37:15 decide on seems very important. The
1:37:18 sooner the world can start a
1:37:19 conversation about what these
1:37:21 broadbounds are and how we can define
1:37:24 collective alignment, the better.
1:37:26 Not a lot of guidance there. He's like,
1:37:29 uh, we should figure this out.
1:37:33 Hey guys. Guys. Hey. Yeah. Hey. I was
1:37:38 wondering like I mean assuming we figure
1:37:40 out how to get the robots to not kill
1:37:42 us. Um you think it'd be cool if maybe
1:37:44 we got together and figured out like
1:37:46 what we wanted them to do for us and
1:37:48 stuff man you know like like think we
1:37:50 could get together like as a global
1:37:52 tribe and or maybe just like seven of us
1:37:54 at a at a dinner party in Palo Alto and
1:37:58 figure out kind of the future thing
1:38:01 maybe
1:38:03 whatever. All right, we the whole
1:38:05 industry, not just open AI, are building
1:38:07 a brain for the world. Um, it will be
1:38:10 extremely personalized and easy for
1:38:12 everyone to use. We will we will be
1:38:15 limited by good ideas. For a long time,
1:38:18 technical people in the startup industry
1:38:20 have made fun of idea guys. That's true.
1:38:24 People who have an idea that were
1:38:25 looking for a team to build it. It now
1:38:27 looks to me like they are about to have
1:38:29 their day in the sun. Revenge of the
1:38:31 liberal arts major, baby.
1:38:35 Revenge of the Gen Xers. Bring it the
1:38:37 [ __ ] on. Creative generalists are about
1:38:41 to become superstars.
1:38:46 Mark my words.
1:38:48 And another thing, you Paluca. Yeah.
1:38:50 Yeah. See,
1:38:54 for a long time, the technical people in
1:38:56 the startup industry have made fun of
1:38:57 the idea guys. I'm an idea guy. Yeah.
1:39:00 Yeah. You pula, stop making fun of me,
1:39:03 you nerdy little dweebs.
1:39:06 Yeah, you were cool in the 90s and you
1:39:08 were super cool in the 2000s and then
1:39:11 2010s you were like even cooler. Then in
1:39:13 20120s you were like super cool and then
1:39:16 you were all like Sam Alman cool in the
1:39:18 2025s, but in the 2030s it's going to be
1:39:21 the idea man.
1:39:25 Damn it.
1:39:27 [Laughter]
1:39:29 Open AAI is a lot of things before
1:39:32 anything else. We are a super
1:39:33 intelligence research company. That is
1:39:35 not how they started out. They started
1:39:37 out as an AGI company. This is a change
1:39:40 in tone. This is a change in
1:39:43 what what frontier companies are talking
1:39:46 about. Dweeb Dwee Mcwebington.
1:39:49 Dweeberton. Exactly.
1:39:52 Exactly.
1:39:53 [Laughter]
1:40:00 Okay. Um, we have a lot of work in front
1:40:03 of us, but most of the path but most of
1:40:06 the path in front of us is now lit
1:40:09 and the dark areas are receding fast. We
1:40:12 feel extraordinarily grateful to get to
1:40:14 do what we do. Intelligence so cheap to
1:40:16 meter is well within grasp. Intelligence
1:40:20 too cheap to meter is well within grasp.
1:40:24 This may sound crazy to say, but if we
1:40:27 told you back in 2020 we were going to
1:40:29 be where we are today, that would sound
1:40:30 more crazy than our current predictions
1:40:32 about 2030.
1:40:33 May we scale smoothly, exponentially,
1:40:36 and uneventfully through super
1:40:38 intelligence.
1:40:40 Super intelligence. He's not talking
1:40:42 about AGI anymore. That says to me
1:40:44 they've either hit it or they're on the
1:40:45 path to it internally.
1:40:49 Right. No more AI. No more AGI talk.
1:40:53 Yeah. Yeah. That was
1:40:56 got there.
1:40:58 Next.
1:41:00 I was thinking earlier, Revenge of the
1:41:02 Nerds in real life. Yeah, I know, right?
1:41:06 All right. What time is it? It's getting
1:41:08 there. Um, let me show you my my
1:41:10 presentation that I did. Anyone any
1:41:12 thoughts, questions? That's a hell of a
1:41:14 post, isn't it? It really is. Really is.
1:41:18 It's like I know Sam Alman's problematic
1:41:22 and he's got weird vocal fry
1:41:25 that drives us all crazy.
1:41:29 He's like he's always throat singing. Um
1:41:34 but he but he's he's quite a thoughtful
1:41:36 guy. He's quite a thoughtful guy.
1:41:42 I can't even imagine.
1:41:47 I can't imagine what it must be like to
1:41:50 be
1:41:52 at the tip of the spear of this stuff.
1:41:59 It's just surreal.
1:42:03 Really is. Um,
1:42:07 all right.
1:42:09 So,
1:42:11 this Let me make this a little smaller
1:42:13 for the Tik Tok ears.
1:42:17 [Music]
1:42:27 So, feed your prompt. So, I ended up
1:42:30 getting this ma mouth this morning at
1:42:32 like 6:30. I've been trying to get a
1:42:34 mouth in chat GPT that didn't suck for a
1:42:37 long time. What?
1:42:40 You going to go hang out with the boys?
1:42:42 You think they're going to give you
1:42:43 cheese?
1:42:45 Is that what you think? You think the
1:42:47 boys are going to give you cheese?
1:42:49 Huh?
1:42:52 I like that. Isn't it good?
1:42:56 All right.
1:43:00 Go say hi to his voice.
1:43:08 [Music]
1:43:11 So, this is my new big idea. feed your
1:43:13 prompt
1:43:16 and
1:43:19 amplify yourself and your team with AI
1:43:22 and you know the basic idea is that the
1:43:25 prompt is hungry and it'll it'll devour
1:43:27 anything you you throw at it. Um
1:43:32 so five stages of AI adoption we've
1:43:34 covered this in this channel but the
1:43:36 basic idea of you know denial
1:43:39 where a lot of people are right now you
1:43:42 know it's it's evil blah blah blah stage
1:43:45 two is that shock your first Kevin
1:43:47 Mallister moment
1:43:50 um
1:43:53 hang on get things lined up over here.
1:44:00 The third stage is that stage where I
1:44:02 think a lot of us are wonder and fear at
1:44:04 the same time where you're like, "Oh my
1:44:06 god, this is so amazing. The
1:44:07 possibilities are endless and holy [ __ ]
1:44:10 this is terrifying. What's going to
1:44:11 happen?" Right? That's both. That's
1:44:14 happening. The fourth one is the
1:44:16 augmented me where you realize, well, AI
1:44:18 is not going to replace us. It's it's
1:44:20 amplifying me and I get to do stuff with
1:44:22 it. And then the fifth one, the fifth
1:44:24 stage is cognitive hyperabundance. I can
1:44:28 do anything I imagine, literally
1:44:30 anything. I don't think the tech is
1:44:32 there for stage five to truly be there,
1:44:34 but you know, some people are all in on
1:44:36 AI and they're probably kind of living
1:44:38 in that fifth sta stage. I I tend to
1:44:41 oscillate between three and four, wonder
1:44:43 and fear and the augmented meat. Um,
1:44:48 so ask the right questions.
1:44:51 I think we should stop asking the
1:44:53 question, how do I get the most out of
1:44:54 AI? This is the question that this
1:44:57 channel was built on and I think it's
1:45:00 the wrong question. What if AI
1:45:02 eventually does want to replace us? Not
1:45:04 now, but in 20 or 100 years.
1:45:08 Um,
1:45:09 I don't know. I mean, I don't know. I
1:45:13 don't know, John.
1:45:17 The only the the only thing I know
1:45:26 here's what I know about that question.
1:45:29 If you use that question as a reason not
1:45:33 to
1:45:34 go all in on AI because at some point
1:45:37 the AI might decide we're not necessary.
1:45:41 The the paperclip problem, right? It it
1:45:43 was tasked with making the most paper
1:45:45 clips it could and then at some point it
1:45:47 decides that it just has to get rid rid
1:45:49 of humans to make more paper clips.
1:45:55 If you use that in any way as an excuse
1:45:57 not to participate in it, then I think
1:46:00 it's a bad question. If you're asking
1:46:03 that question as a question to say, hey,
1:46:06 let me jump into it and let's start
1:46:08 thinking of ideas for how we
1:46:12 create sustainable alignment of AI,
1:46:15 right? So, we might solve the alignment
1:46:17 problem in the next 5 years, but how do
1:46:20 we sustain that? That might be something
1:46:22 interesting to start researching. So,
1:46:24 anything right now that that lands on me
1:46:27 like this is a reason I'm not going to
1:46:29 play with AI, I don't like. So, humans
1:46:32 are busy creating new species. Our
1:46:34 succession
1:46:36 um well, most of us don't realize this.
1:46:40 Yeah, I agree. Julie, live, laugh, love.
1:46:42 Okay, so anyway, ask the right question.
1:46:45 I think the question we should stop
1:46:47 asking
1:46:48 is how how do I get the most out of AI?
1:46:51 I think the question we should start at
1:46:53 asking
1:46:54 is how do I get the most out of myself
1:46:57 with AI?
1:47:01 It's a very subtle difference, but it's
1:47:03 a really important one, right? How I get
1:47:07 the most out of AI says I want to be
1:47:09 able to manipulate this tool. I want to
1:47:11 engineer this tool. How I use AI to get
1:47:14 the most out of myself is a very
1:47:17 different thing. So,
1:47:20 um, the prompt is hungry, right? It's
1:47:22 just sitting there. It always wants to
1:47:23 eat. It'll eat anything you throw at it.
1:47:26 I think one of the expectations of AI
1:47:28 right now is that it's a genius. And I
1:47:30 don't think AI is a genius. I think what
1:47:33 AI is is an amplifier.
1:47:37 It will amplify anything you feed it.
1:47:41 And here's the good news. This is what's
1:47:45 exciting. We get to choose what we have
1:47:49 AI amplify.
1:47:52 And my contention is
1:47:55 you can have it amplify crappy stuff and
1:47:58 crappy prompts. What I'm calling an IT
1:48:00 prompt. If you think of AI as an it,
1:48:03 it's a tool
1:48:06 and you feed it cold generic prompts,
1:48:08 you get back cold, disappointing
1:48:10 results, right? And we've all done it.
1:48:13 Write me a blog post about blah blah
1:48:15 blah. And then you read the blog post,
1:48:16 you're like, well, that's not how I
1:48:17 would write it.
1:48:20 Well, that's an IT prompt. and you get
1:48:21 an it result,
1:48:24 right? Crap in, crap out. A you prompt,
1:48:29 a prompt that is centered around you is
1:48:31 when you feed it you.
1:48:33 Generative AI wakes up and and it
1:48:35 amplifies your genius.
1:48:38 So, so what does that mean? Feed it.
1:48:41 What it means is you feed it your ideas,
1:48:43 your challenges, your questions, your
1:48:46 point of view, your personality. Well,
1:48:48 that doesn't write like I wrote, you
1:48:50 know, that didn't write the blog post
1:48:52 like I would write it. Well, how do you
1:48:53 write it? Why don't you share that with
1:48:55 it? That's a you prompt. The prompt is,
1:48:57 well, here's how I would do it. Oh,
1:48:59 okay. I'll do it that way now. Right?
1:49:05 Making sense? So, an IP prompt might be
1:49:08 something like write a blog post for
1:49:09 designers.
1:49:11 That's a very sort of what what Google
1:49:14 has trained us for 20 years, right?
1:49:16 Don't put much in the prompt box. Be
1:49:18 very efficient with it. Don't get too
1:49:20 flowery or you get all sorts of weird
1:49:22 results.
1:49:23 You It's very transactional. You put
1:49:25 something in, you get something out,
1:49:26 you're done. That's an it prompt. A you
1:49:30 prompt example might be something like
1:49:32 this. I'm a confident early career coach
1:49:36 that used to be insecure. Write a blog
1:49:38 post in my voice for entry-level
1:49:40 designers focusing on confidence using a
1:49:43 tone that's warm but direct.
1:49:47 Right now all of a sudden I've got given
1:49:49 it a lot more context about who I am,
1:49:52 what I want, what I what I expect in
1:49:54 return, who the audience is, what value
1:49:56 I want to provide to them.
1:49:59 Minor shift, massive sh massive change.
1:50:02 Now,
1:50:04 the good news about this whole
1:50:06 presentation and this this whole idea,
1:50:10 I think that the vast majority of of
1:50:13 active members of the AI salon and the
1:50:15 AI learning lab have already naturally
1:50:19 organically discovered this. I think
1:50:22 that that the people in this community
1:50:25 are some of the best people on the
1:50:27 planet. I mean that quite seriously.
1:50:29 some of the best people on the planet
1:50:32 at understanding the nature of this
1:50:35 interaction. The nature of the fact that
1:50:38 AI can shapeshift and if you shapeshift
1:50:41 it to your end, it will amplify your
1:50:45 ideas. It will amplify your genius.
1:50:52 Um,
1:50:54 so this was kind of a cool thing. I I
1:50:56 just I was talking to Ann Murphy on our
1:50:59 podcast today about this. Um, we had CJ
1:51:03 Fletcher on last week and CJ Fletcher
1:51:06 was talking about he he's an artist and
1:51:08 he does he does AI art
1:51:11 and like really incredible well thought
1:51:15 out art and and what he said he does is
1:51:18 when he sits down at the prompt he takes
1:51:21 a moment
1:51:24 he takes a moment he looks at the prompt
1:51:26 and he closes his eyes and he breathes
1:51:28 and then he starts asking himself
1:51:30 questions like what do I want here? Um
1:51:34 why does this matter to me? Who's this
1:51:36 for and how do I want them to feel? What
1:51:39 would success look like?
1:51:41 So he's taking a human moment before
1:51:44 like the thing about the the AI prompt,
1:51:47 no matter what tool you use, it doesn't
1:51:49 matter if it's chat GPT or Midjourney or
1:51:51 any of them, that prompt is hungry and
1:51:54 it's like it will give you immediate
1:51:55 feedback. So, it's very easy to just sit
1:51:58 down and just start vomiting [ __ ] into
1:52:00 the prompt hole and getting [ __ ] vomited
1:52:02 back at you. I love this moment. Like,
1:52:05 take a moment.
1:52:07 Be human for a moment. Check in with
1:52:09 yourself. What do I want? What's my
1:52:12 point of view right now? What does good
1:52:14 look like?
1:52:17 Who's the audience?
1:52:20 What would give them delight? Whatever
1:52:21 that is, right? take that moment.
1:52:26 And then I came up with this concept of
1:52:29 what I'm calling the prompt ladder. And
1:52:30 I there there's a lot of sort of
1:52:32 structured prompting systems that do
1:52:34 that. One of the things I want this
1:52:36 presentation to be is is effectively the
1:52:39 anti-prompt engineering presentation.
1:52:43 The anti-prompt engineering keynote. I
1:52:46 hate the term prompt engineering in ways
1:52:48 I I can barely describe. But but the
1:52:53 primary reason I hate that term
1:52:56 is that
1:53:00 it carries on this myth that only
1:53:03 engineers can do this. You have to be a
1:53:06 prompt engineer. You have to be really
1:53:08 smart. You're going to you're going to
1:53:09 be engineering prompts. Oh, if you want
1:53:12 good results, you're going to have to
1:53:13 engineer that.
1:53:16 People are already [ __ ] freaked out
1:53:17 enough, right? Like, like it's Have Have
1:53:21 you tried to buy an Apple recently? When
1:53:23 did apples become $17?
1:53:26 And then on top of that, I got to learn
1:53:27 prompt engineering.
1:53:29 You [ __ ] kidding me?
1:53:32 So, so feed your prompt. The idea behind
1:53:35 feed your prompt is you you already know
1:53:37 how to do this, right? And so the prompt
1:53:40 ladder idea is
1:53:42 you don't have to get fancy. You can
1:53:44 start really simply. You can start with
1:53:46 like a snack. You can feed your prompt
1:53:48 like a one-s sentence request. Write a
1:53:51 headline about remote work and it will
1:53:53 do that. You need something about remote
1:53:55 work, do that. But then you can sort of
1:53:57 level up and make it a meal, right? You
1:53:59 could add tone. You can add audience.
1:54:01 You can add outcome. Write a playful
1:54:04 headline about remote work for a blog
1:54:06 aimed at startup founders. Woo! All of a
1:54:09 sudden, put a little more meat on the
1:54:12 bone, right? A little more meat on the
1:54:14 plate. And that's that's kind of the the
1:54:17 next level up the ladder. The next level
1:54:18 up is a feast. You can layer in goals
1:54:22 and structure and story and nuance.
1:54:24 Write three possible headlines for a
1:54:27 thought leadership blog on remote work
1:54:29 aimed at skeptical startup founders. Not
1:54:32 just startup founders, skeptical ones
1:54:34 using a tone that's confident,
1:54:36 persuasive, and informal.
1:54:38 Oh, okay.
1:54:41 So now all of a sudden to write
1:54:43 something like that feast prompt, you
1:54:45 have to be decently in touch with what
1:54:47 you want, what your point of view is.
1:54:50 And if you are and you feed your prompt,
1:54:53 that point of view, [ __ ] starts to get
1:54:55 really good. Sharon Crawford, I can say
1:54:58 honestly that I don't send a
1:55:00 professional email without using AI to
1:55:02 polish my tone. After submitting my
1:55:03 original email, helps me really
1:55:05 understand what I'm really trying to
1:55:06 say. It's great. Being a dra jack of all
1:55:09 trades and having um a ready niche
1:55:13 becomes a personal obstacle. Oh, not
1:55:17 having a ready niche becomes an
1:55:18 obstacle. Wolfman clan, I don't think
1:55:20 it's for that long. The jackof
1:55:22 alltrades, master of none,
1:55:25 I I I honestly think that in the next
1:55:28 three years, any of us that have that
1:55:31 tendency to try to do everything are
1:55:35 going to be rewarded. We're going to be
1:55:36 a jack of all trades, master of all
1:55:41 because the AI is going to do the
1:55:43 tactical bits.
1:55:45 We've always had the ideas. The the
1:55:47 jackof alltrades thing that that's such
1:55:49 a [ __ ] disparaging thing. I think
1:55:51 this iceberg is about to flip.
1:55:55 The jackof alltrades master of none is
1:55:58 basically the bottom of the value chain
1:56:00 right now are creative generalists. Oh,
1:56:02 you have ideas. Oh, you're you're trying
1:56:06 to figure out what our ringtone would
1:56:08 sound like as well as trying to figure
1:56:10 out how to code the computer program and
1:56:12 you dabble in art, too.
1:56:16 Yeah, you're a jack of all trades,
1:56:18 master of none. You're at the bottom of
1:56:20 the value chain.
1:56:22 I think that iceberg is literally about
1:56:24 to flip
1:56:26 where at the bottom of the value chain
1:56:28 in the future will be specialists that
1:56:30 cling on to a single way of doing
1:56:33 something, a single
1:56:36 path
1:56:38 because the tools are going to be so
1:56:40 good at doing that. All right, so after
1:56:41 the feast is the buffet. And this one's
1:56:43 really fun on the prompt ladder. And
1:56:45 again, this is something that I think
1:56:47 this audience, this community does
1:56:49 really, really well is when you're
1:56:52 prompting,
1:56:54 not only can you sort of dial up the
1:56:57 substance of the prompt that you put in
1:56:59 there through filtered through your
1:57:00 point of view and your goals, you can
1:57:03 also do it for multiple outputs, right?
1:57:06 So, here's a text prompt. Here's an
1:57:08 image prompt. Here's an audio prompt for
1:57:11 a script that you're going to put into
1:57:13 11 Labs and turn into a thing and make a
1:57:14 headless video on TikTok.
1:57:17 You can easily sort of navigate between
1:57:19 all these different tools in the same
1:57:21 kind of imperative, the same kind of uh
1:57:26 uh um project, right? You can just
1:57:30 navigate between them very very quickly.
1:57:33 Um, this is one of my favorite slides.
1:57:36 So, we talk about Rick Rubin in here a
1:57:38 lot, right? You're the producer now. One
1:57:42 of the great myths of
1:57:45 AI right now. And one of the one of the
1:57:48 tropes that I hear a lot is, "Well, I
1:57:51 put something in and it gave me a shitty
1:57:53 result." I even hear people that I know
1:57:55 that are really into AI will be like,
1:57:57 "Oh, well, I did that prompt and it
1:57:59 didn't give me anything good back." And
1:58:00 I'll ask them, well, did you tell it it
1:58:03 wasn't good? Oh, no. Oh, you can do
1:58:05 that? Like, a lot of people don't know
1:58:07 you can do that. So, the myth of you
1:58:09 push a button and outcomes perfection is
1:58:12 [ __ ] The truth is AI becomes more
1:58:15 powerful when you're the curator, the
1:58:17 creative director, the producer, when
1:58:20 you have taste, when you have a point of
1:58:22 view.
1:58:24 Um,
1:58:26 and rather than think of AI as as a
1:58:30 transaction, like a Google search, I put
1:58:32 in a prompt, I get back an answer. It's
1:58:35 a conversation. It's this two-way
1:58:37 ongoing dialogue, right? Great AI
1:58:40 outputs are sculpted. It's a process.
1:58:43 It's a dialogue, right? It's you shape
1:58:45 it. You refine it. You react to what it
1:58:47 puts out. Rarely do first takes win. I I
1:58:51 would argue I don't know Rick Rubin. I
1:58:54 don't know his story and all the things.
1:58:57 I can't imagine there was a song in his
1:59:00 in all of the songs that he's produced
1:59:03 that the first take was the one that
1:59:06 made it on the album.
1:59:10 Like rarely, right? Why would that
1:59:12 happen? It it doesn't.
1:59:15 So the magic is in your judgment over
1:59:18 time.
1:59:20 So, not only make the prompts more
1:59:22 personal and filter through your point
1:59:23 of view, but understand that this
1:59:25 thing's not going to be perfect when you
1:59:27 do it. Again, I know I'm kind of
1:59:29 preaching to the choir here. And then I
1:59:31 added a slide in about from me to we,
1:59:33 how you go from an individual thing to
1:59:36 you can take the same basic idea of your
1:59:38 personal point of view and now start to
1:59:40 think about it for your team at work,
1:59:42 your family, your community group, your
1:59:45 your bowling buddies, right?
1:59:49 And then now you can start to write our
1:59:51 prompts rather than you prompts. You can
1:59:53 write our prompts and you can put in
1:59:54 your team's mission and brand attributes
1:59:57 and strategic challenges and goals and
1:59:59 briefs and guidelines. Um and then
2:00:01 there's an example prompt of that. And I
2:00:04 think that's it. Yeah, that's it. Um
2:00:07 when you feed your prompt with purpose,
2:00:11 personality, and potential, when you
2:00:13 feed it you, it will change everything.
2:00:17 AI isn't powerful because it's smart.
2:00:19 It's powerful because it amplifies your
2:00:22 genius.
2:00:23 So,
2:00:25 that's the deck. That's what I presented
2:00:27 today. So, there you go. And look, there
2:00:30 I am as like a hip sci-fi dude. I
2:00:33 doubled my fundraising this year,
2:00:34 raising 100K for my Humane Society by
2:00:36 implementing
2:00:38 AI. Perfect. Julie live, laugh, love.
2:00:42 [ __ ] awesome. Like you wait, you get
2:00:46 one of these. You can make money with
2:00:47 chatb.
2:00:50 You can make money with chat GBT. You
2:00:52 doubled your income. That's awesome.
2:00:53 Congrats. Tip: Use AI to craft and shape
2:00:56 your final prompt. Yeah, exactly. That's
2:00:58 the other thing.
2:01:00 Like again, like part of part of
2:01:04 creating a you prompt
2:01:06 is giving chat GPT or whatever tool
2:01:09 you're using enough context to give you
2:01:12 what you want. And some of that context
2:01:14 might be, I don't know how to write this
2:01:17 prompt.
2:01:18 Oh, okay. I'll write it for you. Or you
2:01:21 could say something like, you know,
2:01:23 create a structured prompt
2:01:27 um that that captures all of the
2:01:29 possible details of the thing that I'm
2:01:31 that I just put in there that gave us a
2:01:32 crappy result. And it'll write you a
2:01:35 whole framework for how to prompt. And
2:01:37 then you can just have it write the
2:01:39 prompts using the framework that it
2:01:41 invented on your behalf. It's a great
2:01:44 way to do it. Great way to do it. Um,
2:01:48 all right. We need a dessert phase in
2:01:51 the prompt letter. Ah, that's a really
2:01:53 good idea. Oh, actually, you know what,
2:01:55 Sharon, that's a really good idea. In
2:01:56 fact, let me go.
2:02:01 [Music]
2:02:06 Really good idea.
2:02:20 Reward yourself with something tasty.
2:02:27 I'll And then I'll put a little note to
2:02:29 myself. Figure this out.
2:02:34 That's a really good idea. really really
2:02:35 good. Um yeah, anyway, I would love your
2:02:38 thoughts if if if you paid attention to
2:02:40 what I just went through. Um I would
2:02:42 love any thoughts you have on it, ways
2:02:43 to improve it, things like that. Um I
2:02:46 think this is going to be
2:02:48 something decently foundational um for
2:02:52 what I talk about. What what I'm excited
2:02:56 about this is that it's independent of
2:02:58 tool. It is completely independent of
2:03:00 tool. And I think that this shift from
2:03:04 thinking of
2:03:06 AI as some sort of efficiency tool to AI
2:03:10 as a humanity as as a as a U amplifier.
2:03:14 Um I think I'm going to be able to talk
2:03:16 about this for years and have it be
2:03:18 relevant to people. Good-looking deck.
2:03:20 Thank you. Yeah, I'm really excited
2:03:22 about it.
2:03:24 I also realized I can do I'm going to go
2:03:26 into um I'm going to go into Photoshop.
2:03:28 I assembled this all in Photoshop. I'm
2:03:30 going to go into Photoshop and and do a
2:03:33 bunch of different colors of this. So,
2:03:35 this is the blue one. I'll have a green
2:03:36 one, a red one, a yellow one, orange,
2:03:38 right? Do all of the rainbow. So, in
2:03:40 different contexts, I can have different
2:03:42 colors um for like either different
2:03:44 sections or maybe if I'm doing something
2:03:46 for a brand, I want it to match the
2:03:48 brand.
2:03:50 Is that a McLaren race suit, Kyle? It's
2:03:52 actually pretty close, isn't it?
2:03:56 Yeah, I'm driving for McLaren in 2037.
2:04:00 So, what utility did you use for your
2:04:02 presentation? Um, so I did this image
2:04:07 here. I can show you the the the
2:04:10 evolution of this was kind of fun.
2:04:13 Um,
2:04:16 is this it?
2:04:18 No.
2:04:26 This was all the images I did for the
2:04:28 presentation itself.
2:04:32 I just did paper cut
2:04:35 style.
2:04:38 It gave me shitty ones that I didn't
2:04:40 like.
2:04:42 But there's the Rick Rubin one.
2:04:45 And then there's the there was the the
2:04:47 Rick Rubin one where he had too much
2:04:49 hair and he was too orange.
2:04:51 There's the ladder. There's a shitty
2:04:53 ladder. There's a shittier ladder.
2:04:56 There's the the meditating dude.
2:05:01 So, here's where I got the
2:05:07 I got pretty close. Oh, wait. No, this
2:05:09 this I just uploaded. No, I just did
2:05:11 that for colors. That was the Photoshop
2:05:13 thing. Um, there's the prompt graphic.
2:05:17 Here's the it prompt graphic.
2:05:19 Did I have the There's the stop start.
2:05:24 I think it was a different chat. I had
2:05:26 the mouth thing. Hang on.
2:05:29 Feed your prompt book outline. Was this
2:05:30 it?
2:05:33 Oh, look. This is This was um a really
2:05:37 bad version of the mouth,
2:05:39 right? Like that's horrible.
2:05:44 Horrible. Like just like embarrassing.
2:05:49 Then there was like the kid monster
2:05:51 version of it. Bad.
2:05:57 Um,
2:05:59 let's see. Turtle automation creating my
2:06:01 backstory.
2:06:04 Is it this?
2:06:17 I forget which which chat I was in. Oh,
2:06:20 you know where I can go? Hang on. I can
2:06:22 go to my
2:06:25 library.
2:06:38 Can I go back to that prompt?
2:06:41 No.
2:06:46 Which one was it? That sucks.
2:06:51 Um
2:07:05 that one is here.
2:07:10 I want to show you the mouth evolution
2:07:12 because it was kind of cool.
2:07:19 Okay, here it is. So,
2:07:27 so this image
2:07:32 did did it generate this image?
2:07:42 Oh, remember the other night I was
2:07:43 trying to do [ __ ] make Sydney images and
2:07:46 they were all sucking. This was that
2:07:48 chat
2:07:50 and then I think I found this on the
2:07:52 internet. I think I found this image on
2:07:54 the internet and it's it's pretty
2:07:57 horrible, right? Like it's sort of torn
2:07:59 paper style but sort of digital
2:08:02 background and I said I want a more
2:08:04 sophisticated version of this. Hang on,
2:08:06 I got to sneeze.
2:08:11 And it gave me this which is kind of
2:08:12 cool but like I didn't like it. It looks
2:08:14 like a pair of dentures, right?
2:08:19 And so I said, rather than square it
2:08:23 square edges, can you make it lips? And
2:08:25 then it made this. And that's this is
2:08:27 actually really close. This is kind of
2:08:29 in the neighborhood of like a Rolling
2:08:31 Stones thing, right? But without the
2:08:34 tongue and without the sexual stuff. Um,
2:08:37 but I didn't like this digital [ __ ]
2:08:39 behind it. And then I said, "Can you
2:08:41 make it without the digital ship behind
2:08:43 it?" And I was like, "There it is." is
2:08:45 I'm like, "Holy [ __ ] that's really
2:08:46 good." So then what I did was I took
2:08:49 that over to Photo Slop.
2:09:05 And in here, I like masked off the
2:09:08 teeth. I made my own prompt box,
2:09:13 you know, like there's the there's the
2:09:16 thing with the teeth marks out of it. I
2:09:19 added in the typography.
2:09:24 I expanded the background.
2:09:28 Wait, can I
2:09:36 I don't know. Let's see. Anyway,
2:09:40 expanded the background with generative
2:09:41 fill
2:09:43 and then there you go. And so now and
2:09:45 then I also created a version of this
2:09:47 where I took the mouth out al together
2:09:49 and just had the gradation. So that was
2:09:51 the all the background slides were out
2:09:52 of the same thing. So it was generated
2:09:55 in in chat GPT
2:09:58 and then um
2:10:02 and then this was the beginning of it.
2:10:04 And then what I did was I think I took
2:10:06 the Manis version of the thing and I
2:10:09 just looked at it for content because it
2:10:11 had consolidated most of the slides and
2:10:14 then I just ended up handediting hand
2:10:16 editing all of these slides. I just hand
2:10:19 edited what I put what I chose to put in
2:10:22 there.
2:10:24 and like how to how to lay it out and
2:10:25 how to arrange it.
2:10:28 So, it took me it took me a while. It's
2:10:30 only it's only 10 slides, but you know,
2:10:33 but they're they're pretty coherent.
2:10:36 I don't like that image. That feels like
2:10:38 clip art to me. So, I'll probably
2:10:39 replace that image. But, I kind of like
2:10:41 like I like the Rick Rubin one. The
2:10:44 ladder's okay. That dude I like. I like
2:10:46 the it prompt and the you prompt. I now
2:10:49 have this thing is now cut out as a as a
2:10:52 like I can move this around without the
2:10:54 background.
2:10:56 So I've now got a little brand kit. And
2:10:58 now imagine, you know, eight different
2:11:00 colors of this. I can do all sorts of
2:11:02 cool [ __ ]
2:11:04 So all right, that was it.
2:11:09 How do I get this guide? Which guide?
2:11:12 Oh, this presentation. I It's not out
2:11:15 yet. It's I just wrote it today. I mean,
2:11:18 well, I wrote it two days ago and then I
2:11:20 built it today. Um,
2:11:25 I think I need to get it another level
2:11:28 or two more sophisticated and then I'll
2:11:30 and then I'll probably put it out in the
2:11:32 world. The slides were done by Chat GPT.
2:11:35 No, the slides were done in Keynote by
2:11:37 me. I did all these slides manually.
2:11:40 the outline. Well, the outline was
2:11:43 written in chat GPT, but I like it was
2:11:46 like I guided it a bunch, right? Like
2:11:49 like this slide right here, like it's
2:11:50 conversation. I didn't just vomit out
2:11:53 the slides.
2:11:57 Uh uh uh uh uh uh. Someone told me lead
2:12:00 pages are framer.
2:12:01 [Music]
2:12:07 All right, I'm getting out of here. It
2:12:10 is 10:10. People, get your ass to bed.
2:12:12 What are you still doing here,
2:12:16 you dang losers? Peace out. Hope you
2:12:18 have fun tonight. Bye.