7/23/2025 - Taming Hallucinations in Large Language Models

Description

Chapters

Transcript

0:01 You ready, Chippy?
0:07 [Music]
0:18 [Music]
0:29 10,000 words swimming around my head. 10
0:32 million more in books written beneath my
0:36 bed.
0:41 And I wrote or read them all when
0:43 searching in the swamps. Still can't
0:46 find how to hold my hands.
0:53 And I know you need me in the next room
0:56 over. I'm stuck in here all paral.
1:06 I kept myself in ruts. Too much time
1:09 spending mirrors framed in yellow walls.
1:17 Ain't it like most people? I'm no
1:20 different. Like to talk on things we
1:22 don't know about.
1:24 [Music]
1:28 Well, like most people, I'm no
1:31 different. like to duck on things we
1:33 don't know about.
1:37 [Music]
1:44 [Music]
1:49 Do
1:52 [Music]
2:01 no.
2:16 No.
2:17 [Music]
2:32 Good day. Good people. It's a little
2:34 moist in here. It's a little moist in
2:36 here. Do you feel it? Can you feel it?
2:38 I'm a little
2:40 [Laughter]
2:40 [Music]
2:49 We're gonna talk about artificial
2:51 intelligence tonight. I don't know if
2:53 you've heard this term.
2:55 There's intelligence. We've got that.
2:58 And then they figured out a way to wire
3:00 some computers together in such a way
3:04 that uh it's like that but not with
3:09 machines.
3:11 So they they call it machine learning.
3:15 I I don't think machines learn.
3:18 You program machines. We're the humans
3:21 here. We've got the intelligence. You
3:23 tell the machine what you want it to do.
3:27 The machine doesn't tell us what it
3:29 wants to do.
3:32 So we're going to we're going to explore
3:33 that tonight. We're going to figure it
3:35 out.
3:37 We're going to figure out the
3:38 mathematics behind it.
3:40 We're going to optimize some tensor
3:42 processors.
3:44 We've got we've got engineers and
3:45 mathematicians from Google that hang out
3:48 here.
3:52 We're going to we're going to probably
3:54 order a couple of H100 GPUs from
3:58 Invidia.
4:00 Is it inv video or is it N vid?
4:07 What? Why do they spell it like that?
4:12 And do we want an A100 or an H100?
4:14 Because I think an H100 is kind of the
4:16 cheap shitty thing. And then we'll get
4:18 an A100. Anyway, we're going to use all
4:21 of the different libraries
4:24 from Nvidia. Thank you very much there,
4:27 Forever Hooked. Very kind of you.
4:33 And then we're going to we're going to
4:35 learn programming. We're going to learn.
4:37 We're gonna learn uh there's this thing
4:39 there's a paper out called uh attention
4:42 is all you need
4:44 and it's about insecure theatricals.
4:48 kids that grew up just wanting
4:50 attention, wanting attention, wanting
4:52 attention. And then they went to theater
4:54 school and then out of theater school
4:57 they moved to New York City and they
4:58 started and and ran a theater company
5:02 and after four and a half years they got
5:04 burned out
5:06 and so they they decided to write
5:08 themselves an acting career. So they
5:09 wrote seven screenplays in two year. No,
5:13 that's
5:15 that's not what that paper was about.
5:16 that that was that's my life anyway. But
5:20 attention is all you need is this paper
5:23 that introduces this thing called the
5:24 transformer. So we're going to we're
5:26 going to learn what a transformer is and
5:28 then we're going to go collect all of
5:30 the information
5:32 that humanity has ever put on the
5:34 internet and we're just going to gather
5:36 it. We're going to we're going to stick
5:38 it in a box and and we're going to put
5:40 it in a transformer and it's going to
5:42 embed it. And I from what I understand,
5:44 we're going to we're going to take all
5:46 of those documents and and they're going
5:48 to they're going to be shattered into
5:51 trillions and trillions of fragments.
5:53 Yes, Brandon.
5:55 >> Hey, before we all sounds amazing, but
5:58 before we do all of that, could we uh
6:00 answer Ginger Jay's question on Tik Tok?
6:03 >> Oh, yes.
6:06 Hi, Ginger J. What's happening? Did you
6:09 have an actual question about actual AI?
6:12 Oh, wait. I lost the pin. Someone put
6:14 the pin back up there, please.
6:18 Okay. Can you comment? I use chat GPT.
6:21 How do I get it to stop lying and make
6:24 up answers that sound good? Um, okay.
6:29 So,
6:31 good question. and
6:34 all that horse hockey I was just talking
6:36 about before that that was in the
6:38 neighborhood of what what actually
6:39 happened with creating this stuff. Okay,
6:42 so there's a couple of things. Um,
6:49 one of the things that we all have to
6:52 grapple with right now is that the way
6:58 large language models chat GPT cla
7:03 um,
7:07 are we good? Yeah, I think we're good.
7:08 Oh, wait. I got to move this up here.
7:10 Hang on a second.
7:13 The way these things work. Hang on. I
7:14 got to turn on my air, too. It's a
7:15 little It's a little hot. It's a little
7:17 hot.
7:21 The way these things work is not the way
7:24 computers have ever worked in history.
7:28 Um, they used to be called, I think,
7:31 probabilistic is the is the or
7:34 deterministic. Computers used to be
7:36 deterministic. You would program them,
7:38 you would put logic in them, and they
7:39 would always behave the same way every
7:41 single time.
7:42 large language models don't do that.
7:45 And the nature of of how they were built
7:50 um
7:54 means that
7:57 for one thing they can they can be
7:59 creative which one of the things that
8:01 you'll hear about that that people will
8:04 say about AI is well it's just a it's
8:06 just a probability calculator and it can
8:09 never be creative and it can never come
8:10 up with an original thought. I actually
8:12 don't agree with that. I think it I
8:13 think it can I think the way they work
8:15 is very similar to the way our brains
8:17 work. So there are some cases when
8:19 hallucinations is what they're is what
8:21 what you're referring to is called
8:22 hallucinations. There are some cases
8:24 where hallucinations can actually be
8:26 beneficial, right? And that would be if
8:28 you're doing creative development. You
8:30 want some story ideas and it will be
8:32 creative and it won't just give you the
8:33 same answer every time. It'll give you
8:35 something different every time. So there
8:37 are some cases where that's a good
8:39 thing. There are other cases if say
8:41 you're trying to research something and
8:44 you want facts
8:46 and and the large language model
8:50 what I call it is mansplaining as a
8:52 service it mansplains to it gives you an
8:54 answer confidently even if it doesn't
8:56 know what it's talking about right like
8:58 every good man in your life um
9:02 and that could be really frustrating now
9:04 there are things you can do to mitigate
9:06 it there are things that you can do that
9:07 that that make it better. So, let me let
9:11 me talk through a couple of them. Um,
9:15 and and I'll talk this came from Andre
9:17 Karpathy, who was the one of the
9:19 co-founders and the chief scientist at
9:21 OpenAI. Um, and he was describing
9:24 someone had asked him the question about
9:26 about why large language models
9:28 hallucinate. And and he used a
9:30 mathematical example and I'm not going
9:33 to give you the math. I have an acting
9:34 degree. I don't I don't know about math.
9:37 But he was talking about um if you ask
9:41 about a particular mathematical
9:43 conjecture, right, a particular
9:45 mathematical challenge
9:48 and you just ask chat GPT about that
9:51 challenge, it will basically go look at
9:54 everything it's been trained on, which
9:55 is essentially everything that's ever
9:57 been on the internet, and it will sort
9:59 of cobble together an answer for you,
10:01 and it will answer you. And very often
10:03 it'll be wrong. and he was he was
10:05 describing why that might be. So let's
10:07 say you've got this mathematical formula
10:10 and you just ask about it and he said
10:12 well
10:13 in chat GPT are not only the the
10:17 professor that came up with that
10:19 conjecture that formula and the answer
10:21 to it but every student that he ever
10:24 taught and every student that ever tried
10:26 to attempt it every student that ever
10:28 read about that conjecture and and so
10:31 not only is the correct answer in the
10:33 training set so is every other wrong
10:36 answer in the training set. And so the
10:39 way I like to think of what they call
10:41 the latent space, the latent spa, think
10:43 of the latent space is you take all of
10:45 the knowledge of humanity, you fracture
10:47 it into like tiny little shards of glass
10:50 and they all get sort of clustered into
10:53 thousand dimensional mathematical space
10:55 and it's just this massive ball of
10:57 fragments of knowledge and when you put
10:59 in a prompt, it extracts out of that
11:02 these little fragments and sort of
11:03 stitches them back together for you.
11:05 It's actually quite remarkable the way
11:06 it works.
11:08 If you think about it like the Grand
11:10 Canyon, if you take all these shards of
11:12 glass and you dump them into the Grand
11:14 Canyon, if you just ask it about that
11:16 that mathematical formula, it'll look at
11:18 all the shards of glass in all of the
11:20 Grand Canyon. But if you narrow the
11:24 latent space by providing an additional
11:26 context, I want you to act like a a a a
11:31 mathematics a polymath, you know,
11:34 professor that specializes in this
11:36 branch of mathematics. And I'm
11:38 particularly interested in this
11:39 particular formula. And I really only
11:42 want to hear from, you know, the most
11:44 experts in the world about this formula.
11:46 What it's what you're doing is every
11:48 time you add context like that, you're
11:50 going into an increasingly small subc
11:53 canyon of the Grand Canyon and you're
11:55 cutting out all that other noise. So,
11:58 one of the best ways to make chat GPT
12:02 less bullshitty
12:04 is to provide it more context. So, if
12:06 you just say, "Write me an article about
12:09 X," it's going to write you a shitty
12:10 article. If you say, "Hey, I'm an expert
12:15 in
12:16 zucchini gardening, and I live in
12:19 Denver, Colorado, and I've got a raised
12:21 bed, and we're in our third year of
12:23 gardening, and I put in a zucchini
12:25 plant, and one of them I fertilized, and
12:28 one of them I didn't, and for whatever
12:30 reason, the plant that I didn't
12:32 fertilize is doing better. And I want
12:34 you to write me a LinkedIn article about
12:38 zucchini gardening in raised beds.
12:42 all of a sudden you've narrowed sort of
12:44 the world that it's going to draw from
12:46 into this into this much smaller place.
12:49 Does that make sense? Um
12:53 so that's a thing. Now
12:56 you can also use different models. Let
12:58 me share I'm sharing my screen. I'm
13:00 going to share it in a different way
13:02 here.
13:03 Follow a second.
13:08 [Music]
13:15 Okay,
13:17 there's another way you can do this. So
13:19 if you haven't played with switching
13:20 models
13:22 in chat GPT, you have these different
13:24 things called models and they behave
13:26 very very differently and they are very
13:29 very very very very very
13:33 confusing
13:34 because a bunch of engineers named them
13:37 and they did it really poorly. So you
13:39 have this thing called 40
13:42 then you have a thing called 03
13:44 Tik Tok pin.
13:47 Here's your article. Overfertilized.
13:56 That's funny. I didn't I couldn't see
13:57 the bottom of it. There's Tik Tok has
14:00 this weird interface where whenever
14:01 someone pins a comment, I can read the
14:03 first two lines and if I want to expand
14:04 it, I always it always does other [ __ ]
14:07 I don't actually expand it about once
14:09 every five times. Anyway, you got this
14:12 thing called 40. This is the base model
14:14 that most people use. That's the
14:15 default. I would argue that most people
14:19 that use chat GPT never never touch
14:21 these models.
14:23 Um
14:26 03 is not lower than 40.
14:32 It's actually a completely different
14:34 kind of model
14:38 and it's called a reasoning model. See
14:39 it says uses advanced reasoning and then
14:41 and then you've also got 04 mini and 04
14:43 mini high. So, so 03 is the is the kind
14:47 of MacDaddy reasoning model and then 04
14:51 mini and 04 mini high are smaller
14:53 versions of the next
14:55 evolution of 03, but they're they're
14:59 kind of the smaller previews of of
15:01 what's coming with 04 Pro or whatever
15:04 the the big 04 is,
15:07 but they're not the same at all. And and
15:09 there's no way you would know that.
15:11 There's no way you would understand
15:13 that. But
15:15 what 03 does when it talks about
15:17 reasoning, what happens with 40 is you
15:20 ask it a question, it gives you an
15:22 answer. So even if you do all of the
15:24 context providing that I gave you, it
15:26 can still get it wrong.
15:29 What 03 does is it actually thinks about
15:32 it and talks to itself before it answers
15:35 you. So if you ask it a question about
15:38 raised bed zucchini gardening, it will
15:41 say, "Oh, the user wants to, you know,
15:43 talk about zucchini gardening, maybe I
15:45 should go out and I should do some
15:46 research. Let me go out and and research
15:49 some some websites about gardening and
15:52 that particular thing and understanding
15:54 about different species of zucchini and
15:57 and it will just talk to itself and go
16:00 do research on the web and think about
16:02 it and analyze stuff. Sometimes it'll
16:04 even use what are called tools where it
16:06 will write Python applications and
16:09 generate spreadsheets and do math and
16:11 all just all sorts of weird [ __ ] So, so
16:15 that's that's a next way to reduce
16:17 hallucinations is use a reasoning model.
16:19 They can still hallucinate. Don't be
16:22 clear of that. And then the final thing
16:24 that you can do is you can um there's an
16:29 option in chat GPT called deep research.
16:31 So if you combine 03 which is this
16:34 reasoning engine with deep research it
16:36 will really dig deep and research deep
16:39 and check its work and things like that
16:40 and it gets better and better and
16:42 better. Even that can still hallucinate.
16:45 So you're not going to completely
16:46 eliminate hallucinations until these
16:49 models probably hit another generation
16:52 or two or three. Um and even then maybe
16:55 not. Um just by the nature of how they
16:58 are. Okay. If I've trained in 40, does
17:02 any of that transfer to the other mo
17:04 models? Um, what do you mean if you've
17:06 trained like if you've put in custom
17:08 instructions and and you're doing
17:11 projects and things like that? Yes. The
17:13 memory. So, couple of things are
17:15 happening in chat GPT. If you go to your
17:18 settings,
17:20 go to go to a section called
17:22 personalization and you'll see a memory
17:25 section. Um,
17:29 and you'll also see custom instructions.
17:31 So, if you go into custom instructions,
17:34 um, custom instructions are you're
17:37 telling chat GPT what you want it to do.
17:39 One of the things you could say in
17:40 custom instructions, I don't know if it
17:42 would work, but it might, is say, I want
17:44 you to do everything you can to to
17:46 reduce hallucinations. So, before you
17:48 answer me, check your work. Now, I don't
17:50 think that'll work with 40, but that
17:52 might work with 03. Um, so custom
17:56 instructions can be good. Um, but you've
17:59 also got reference saved memories and
18:02 reference chat history. So as of about
18:05 three or four months ago, chat GPT has
18:08 access to your past, I think it's year
18:10 of chats.
18:12 And so that applies to all models. So
18:15 you can be in a 40 conversation and you
18:19 can flip over to the 03 reasoning engine
18:22 and just keep asking in the same
18:24 conversation. There are some things that
18:26 that mess it up like if you generate
18:28 images within it that used to mess it
18:30 up. Things like that. Memories light
18:34 the corn memories light the corners of
18:36 my mind. Everybody know memories
18:40 like the corners of my mind.
18:46 Oh, and by the way, speaking of which,
18:50 let me share one other thing here,
18:52 people. Let me share another thing. Look
18:55 at that. You see that right there?
18:59 Next week, starting on Monday, um I'm
19:03 doing a fiveday
19:06 back to basics crash course where um
19:10 kind of all the silly talk I did at the
19:12 beginning about how large language
19:13 models work. Um, on Monday we're going
19:17 to talk about AI, why it's a big deal,
19:19 how it compares to the early days of the
19:21 worldwide web, how large language models
19:24 work, how diffusion models, the image
19:27 generation models work. Um, and so
19:29 that's Monday. Tuesday is going to be a
19:31 deep dive into chat GPT. So all the
19:33 stuff I'm talking about here, we'll do
19:35 hands-on like a hands-on hour with chat
19:37 GPT and unpack a bunch of stuff. Um,
19:41 Wednesday is going to be a creative day
19:42 where we're going to play with image
19:44 generation models, video, music, audio,
19:47 things like that. Um, Thursday is where
19:50 we're going to bring it all together.
19:51 We'll do some some business use cases,
19:54 figuring out how do you determine which
19:56 model to use when. Um, and then Friday
19:59 will be an open-ended Q&A. All right. So
20:01 that's next
20:03 next week
20:06 here on Tik Tok
20:08 from 8 to 8:30 Mountain time will be a
20:11 pre-show where we'll just hang out. And
20:14 then starting at 8 8:30 we're going to
20:16 turn off Tik Tok and just go completely
20:18 to YouTube.
20:20 And let me see. I've got
20:24 I think I have these slides, Brandon.
20:26 Let me grab them.
20:30 Oops.
20:47 [Music]
20:52 So, here is the YouTube channel.
21:00 Wait. Oh, add the stage. There we go.
21:03 Oh, that's producer Brandon. Wait,
21:05 why did I not?
21:09 Hang on, people.
21:12 I guess I just didn't share. There we
21:13 go.
21:15 So, there's the uh there's the YouTube
21:19 channel. If you're if you're just on
21:20 TikTok right now, learning lab-ai is the
21:23 YouTube channel. So 8:30 next week, 8:30
21:27 Monday, Tuesday, Wednesday, Thursday,
21:28 Friday, um is where we'll be streaming
21:31 live. Okay? So you can zap that QR code
21:33 if you want to do that or just type
21:35 inlab-ai
21:37 on YouTube and subscribe to the channel.
21:40 Okay.
21:43 Okay.
21:45 Beautiful. Which which other slide? This
21:48 other slide.
21:53 Ah, and then the other thing that you
21:56 should do if you haven't yet is you
21:58 should join the AI salon. So if you go
22:01 to community.thesalon.ai,
22:05 that's going to take you to our
22:06 community site. Um, we have a special
22:09 channel in there just for the AI
22:10 learning lab called the Irregulars.
22:13 Um, but there's all sorts of cool stuff
22:15 in there. um and all sorts of amazing
22:20 people and we talk about uh this journey
22:22 of AI literacy we call it where you play
22:25 first mindfully create and generously
22:28 lead. Um, and if you're just getting
22:31 started with AI and trying to figure
22:33 this out and you're asking questions
22:34 like, "Hey, why the hell does this does
22:37 this thing lie so much? Um, everybody's
22:40 dealing with it. Everybody's dealing
22:42 with the jank, I call it, right? Embrace
22:45 the jank. These tools are remarkable and
22:48 weird. They they computers just have not
22:52 ever behaved as unpredictably as these
22:55 things behave. Um, they've also never
22:58 been able to do things quite as
22:59 remarkable as what these things can do.
23:02 They they they are, you know, they are
23:05 just on the verge of magic. They are
23:08 they are all math, but it's it's
23:10 mathematical as we call it here. All
23:13 right.
23:15 [Music]
23:24 So, go join the salon and and you'll see
23:26 when you get there there's kind of a
23:27 sevenstage welcome process. The second
23:29 one is introduce yourself. So, go
23:31 introduce yourself. You should go do
23:32 that.
23:36 All right, let me change my sharing one
23:39 more time
23:41 and then we're going to do a little
23:43 movie night.
23:45 We are gonna do a little movie night and
23:48 I think it's going to be a lot of fun. I
23:50 think it's going to be a lot of fun
23:52 unless it's boring.
23:53 then we'll do something else. All right.
23:56 So, I'm going over to the AI sell
24:00 and I am jumping down to the irregulars
24:02 channel. So, down the left hand side
24:03 here, right under clubs and hubs, the
24:06 first one is the irregulars
24:09 and we've got a
24:14 little image from Steo. It's a little
24:16 moist in here tonight. Steve capturing
24:19 the essence of the uh of the of the
24:22 early part of the uh the broadcast this
24:25 evening.
24:29 Thank you, Stea. Very nice. Very nice.
24:31 Very nice.
24:33 Um
24:36 what else we got in here?
24:40 What' Claire have?
24:43 Oh, a version of Black Betty for Kyle to
24:45 practice. Very nice.
24:49 It's funny. Our uh our sales guy for
24:51 Story Vine has has the Tom Celich
24:53 mustache now. It's pretty funny. These
24:56 are all cute. All right. Groovy. Groovy.
25:00 Groovy. Groovy. Groovy. Groovy.
25:04 Groovy. So, in one of those Well, I
25:08 didn't have that on my bingo card for
25:11 2025.
25:14 Um, I don't know if you all know the
25:15 comedian Theo Vaughn.
25:19 Um,
25:20 very funny dude, but very unassuming, a
25:24 very weird kind of comedy. Um, I could
25:28 see a lot of people not getting his
25:30 comedy, but he's he's like a super
25:32 genuine dude. um
25:37 he got and and so he's been
25:40 he's been kind of elevating his stature
25:42 on the you know comedians they all sort
25:45 of go on each other's podcasts. He's
25:48 been doing a lot of podcasts and getting
25:49 a lot of interesting interviews and it
25:50 looks like he's got this new higher
25:52 production value podcast called This
25:55 Past Weekend and he just did an
25:58 interview with Wen Phoenix and it was
26:00 really trippy and weird because he's
26:03 weird and we Phoenix is weird and they
26:05 just kind of weirded out with each
26:07 other. um
26:10 he got an interview with Sam Alman
26:13 [Laughter]
26:18 and so I figure we'll watch some of it
26:21 tonight because it's [ __ ] odd.
26:25 It's just odd. Now, if anyone's got any
26:28 specific questions on AI, I'm happy to
26:30 answer questions. I'm happy to demo demo
26:32 [ __ ] I'm happy to go deal with the dog.
26:36 Hold, please.
26:39 Stop.
26:49 Champ just saved us from the evildoers.
26:53 The evildoers out on the street. We're
26:55 safe now.
26:59 All right. Sharon Crawford. Huh? No, he
27:02 did not. This is a 90minute interview
27:05 with Sam Alman.
27:07 I I just and I saw a couple of the clips
27:10 at at one point at one point Theo Vaughn
27:13 goes he goes, you know, essentially you
27:18 tech people are weird and he points out
27:20 Peter Teal as this really weird he is he
27:23 like an evil tech overlord and Sam
27:25 Alman's like we're actually really good
27:28 friends and Theo Vaughn was like oh we
27:30 don't have to talk about him.
27:35 Oh my god. So anyway, so I figure we'll
27:37 play we'll play a little. All right.
27:43 Okay. Fantastic.
27:46 Fantastic.
27:52 Get off my lawn.
27:55 All right, let's go. Let's just watch
27:58 it. We'll watch it for a bit.
28:12 Let's let's just stop it right there.
28:14 This this totally feels like the uh
28:18 Bryant Gumble Katie Turk interview in
28:21 1994
28:23 or five where they're like, "What's that
28:26 little the the A with the circle around
28:29 it?" I think it's about it's what is
28:33 this internet thing anyway? And he's
28:36 like he he's a big player in AI. Oh, no
28:40 audio. Oh, that's cuz I'm sharing it
28:41 wrong. Cuz I'm a loser, baby. Why don't
28:45 you sue me? Yeah.
28:50 Fantastic.
28:51 Okay. What we're going to do here is
28:53 we're going to need you to go on ahead
28:55 and start that over so the good people
28:58 on the YouTube there can see the
29:00 beginning of this thing. You see what
29:01 I'm saying? All right. It's not the tab.
29:08 What?
29:14 I think this is right.
29:17 Yes, it's right. I was saying it. We
29:19 need to turn this into the tab learning
29:21 lab.
29:22 >> Oh,
29:24 that was comedy. I thought I was
29:27 actually getting a an actionable note.
29:29 No, I was just being made fun of. Okay,
29:31 here we go.
29:35 Today's guest is uh well, dude's a
29:37 straightup tech lord, let's be honest.
29:39 He's uh he's one of the leaders, the
29:41 world leaders in the development of AI.
29:45 Um he started open AI which is known for
29:49 uh having chat GPT.
29:52 Uh we had a fascinating chat about the
29:54 pros and cons um the fears and hopes
29:57 everything I could learn about uh about
30:00 artificial intelligence and where we're
30:01 headed. TBD baby today's guest. I mean
30:08 this oddly feels like a big deal to me.
30:13 It it it does like Theo Vaughn's got an
30:16 audience that probably ain't deep into
30:19 AI like just the way he talked about it
30:21 just the way he pronounced it. Chat J
30:24 pay
30:25 [Laughter]
30:27 >> is Mr. Sam Alman and I'm very thankful
30:30 for his time.
30:33 [Music]
30:40 [Applause]
30:42 [Music]
30:42 [Applause]
30:44 [Music]
30:47 You know, we had a residential architect
30:49 do this office. We wanted it to feel
30:50 like someone's like really comfortable
30:52 like country house or something like
30:53 that.
30:54 >> Yeah.
30:54 >> Not like the big corporate like sci-fi
30:56 castle.
30:57 >> Yeah. That's what I was I was like a
30:58 little bit like, oh, is it going to be,
31:00 you know, will there be a drawbridge?
31:02 Will we be uploaded into a suite? Like
31:04 what will happen to us? You know,
31:05 >> we don't want that for like residential.
31:07 >> Yeah. I was like, how do we even get
31:09 through the firewall? How many like hit
31:11 points will we need to get through? You
31:12 know, it got very Dungeons and Dragons
31:14 uh in some of my like um imagination
31:18 sometimes.
31:18 >> Yeah, we want people to feel like super
31:19 comfortable and tried to get pretty far
31:21 in that direction.
31:22 >> It feels like it. Your staff's very
31:24 sweet, nice people. Um you have Thanks
31:27 for hanging out, man.
31:28 >> Absolutely. Really appreciate it. Uh
31:31 >> yeah, I haven't seen you since I fell
31:32 out of my
31:32 >> f chair at the inauguration. That was
31:34 really like quite a way to meet you.
31:36 Yeah, I felt so embarrassed and you were
31:38 one of the faces that I looked up and
31:39 saw and I was like, "God and that was my
31:41 first moment like AI build us a better
31:43 chair to be honest with you." You
31:44 >> and you did nothing, right? You were
31:46 just sitting there and it just
31:47 collapsed. Nothing.
31:48 >> I remember that.
31:49 >> And it was just so embarrassing. I was
31:51 like, "Oh, of all people me and here I
31:53 am in this place and uh
31:54 >> I think it was perfect because
31:55 everybody's got to have some sto when
31:57 people are like, "Oh, it was the
31:57 inauguration." Like, everybody's got to
31:59 have some story to tell.
32:00 >> Yeah.
32:00 >> And that was an incredible story for us
32:02 all to tell.
32:02 >> That's a good point. I do remember
32:04 looking at people for help though. And
32:06 oddly your eyes I I was like, "Oh my
32:08 god, he
32:10 >> Yeah, Danielle. Yeah, this is the Today
32:12 Show moment of 2025." It totally is. It
32:15 totally is. Yeah. And Brandon, they've
32:17 had Wait, they they've had this set at
32:19 OpenAI the whole time. They They could
32:21 be doing non-awwkward product launches
32:23 on this set. Yeah, they could.
32:27 >> Could help. You did look like a beacon
32:28 of help in the distance.
32:29 >> I tried to help.
32:30 >> Um,
32:31 you have a baby. You have a new
32:33 >> Yeah.
32:34 >> child.
32:35 >> It is.
32:38 >> There have been like a lot of
32:39 experiences in life where everyone tells
32:41 you something's going to be great and
32:42 then it's like, okay, the people are
32:44 right, the consensus is right. It's like
32:45 even better than I thought it was going
32:47 to be. But this has been the strongest
32:48 example of that ever.
32:50 >> Like I knew it was going to be great and
32:51 it's like way way better. It's
32:53 impossible to describe. There's nothing
32:54 I can say that's not like very cliche
32:56 and it's totally amazing.
32:58 >> What is like one of your And it's a you
33:00 have a young boy. Yeah. And what's
33:02 something like that you think is like
33:04 neat or like what's one thing that kind
33:06 of like is bringing you joy with it?
33:09 >> Watching
33:11 the speed with which he like learns new
33:14 things or gains new capabilities. Yeah.
33:16 Is just unbelievable. It's like every
33:18 day it's like oh man
33:19 >> just couldn't do that before and now
33:20 he's like grabbing stuff and passing
33:22 between his hands and uh getting to like
33:26 watch it dayto day is just an amazing
33:28 rate of change. You know what's amazing
33:30 is I feel like
33:33 his insight as to what it's like
33:36 watching his baby develop and gain new
33:39 knowledge while he's spent the past
33:41 seven or so years getting machines to do
33:44 that and watching them progress. That's
33:46 a fascinating that's a fascinating
33:48 insight.
33:50 >> And then I don't like again I realize
33:51 it's like
33:54 >> how how is Sam Alman odder than Theo
33:56 Vaughn?
34:02 I love you all. Oh my god. You can make
34:05 money off chat GBT.
34:07 You know, I realize that like everything
34:09 about babies are very finely tuned over
34:12 >> and we're not going to necessarily watch
34:14 this whole thing. It's 90 minutes, but I
34:16 just I figure we'll watch it and get
34:18 some interesting insights here. I just I
34:21 just know that he's going to ask Sam
34:23 questions that no one else will ever ask
34:25 him.
34:27 It's just such an odd pairing
34:30 for
34:30 >> a long period of evolution to make us
34:32 like love them and be fascinated by
34:33 them. And it's like a neurochemical
34:35 hack. But I love it. It's great. It's so
34:37 strong. It's so intense.
34:38 >> So, it's really like almost like a
34:40 coffee for your heart or something kind
34:41 of.
34:43 >> I don't even know how to find I've tried
34:45 to like come up with an analogy to tell
34:46 because now I'm like telling everybody
34:47 you got to have a lot of kids. It's
34:48 really important.
34:49 >> Yeah. Um,
34:50 >> and I've been looking for an analogy of
34:51 what to explain. And then I always just
34:53 say like I I don't know how to explain
34:54 this. It's just it is the best thing
34:56 I've ever done by far. I feel like
34:59 >> a completely changed person. And I was I
35:02 was like thinking the other day like
35:05 there used to be all these other like at
35:07 this point all I do is work and hang out
35:08 with my family. I like I don't I don't
35:10 like really get to do a lot of hobbies
35:12 anymore.
35:12 >> Silver Fox, how old is the baby? 4.01
35:15 months.
35:21 right? Busy time at work. I don't get to
35:23 hang out with my friends that much. Uh
35:26 and I and I don't you know there were
35:27 like all these things where people tell
35:28 you like oh you got a baby coming. You
35:30 got to go you know take that spontaneous
35:31 international trip cuz you're not going
35:32 to be doing that again for a long time.
35:34 And I was like oh that is kind of sad.
35:37 In practice you don't do it that often.
35:39 I at least didn't do it that often. And
35:40 I don't miss it at all. I like remember
35:42 that that used to be a possibility. Now
35:44 I can see that's not going to be a
35:45 possibility for a long time and I'm
35:46 thrilled with the trade.
35:47 >> You're moved on.
35:48 >> I'm so happy.
35:49 >> Um, how old is your child?
35:51 >> Four months.
35:51 >> Oh, that's a funny like at five or six
35:53 months they start to get like fun and
35:55 you can like they're still like they
35:57 can't go anywhere, you know, but they're
35:59 like intrigued and stuff. They start to
36:01 like smile or process more. I don't know
36:03 how you guys say it. But um
36:04 >> yeah, he's totally like turned on
36:06 though, like really aware, understands
36:07 things. It's super cool. I have a
36:09 thought sometimes that this will be one
36:10 of the last like maybe 40 years that we
36:14 conceive children in the body. Did you
36:16 have any thoughts about that?
36:19 >> I've definitely heard a lot of people
36:20 say that. Um,
36:22 >> okay. That's the best [ __ ] question
36:25 ever.
36:29 This is the last 40 years or so that
36:33 children are going to be se conceived
36:35 inside the body. And Sam's like, "H,
36:40 [Laughter]
36:46 I haven't thought about it hard myself,
36:47 but yeah, I guess it does make sense."
36:49 Like,
36:51 I guess that does make sense.
36:52 >> Like, God, you were in your mom's. It's
36:54 crazy, you know? You pervert or
36:56 whatever. Like, like I think in the
36:57 future people will be it'll be kind of
36:59 done like in a
37:00 >> in a vet or something.
37:01 >> Yes. In like a nice vet. You can go see
37:03 it on the weekends or whatever. And like
37:04 >> doesn't that just feel like off to you?
37:07 Like I can totally intellectually like
37:09 understand that that may be the better
37:10 way to do it.
37:11 >> Oh yeah, it feels way off to me. I was
37:13 trying to
37:14 >> look at look at Sam Alman trying to
37:16 trying to go like dude like don't go
37:18 this dark.
37:20 Sam Alman's defending humanity here.
37:25 >> Thought you would like it. You know, I
37:27 thought
37:29 >> I thought you would like it.
37:37 Theo Theo trying to suck up to what he
37:40 thinks Sam Holman is.
37:44 Oh my god, this is great. I mean
37:47 >> like or I thought you that would be like
37:49 a thought like I guess I like that for
37:50 me that's one of like my futuristic
37:52 thoughts you know
37:54 >> like I can totally accept that that will
37:56 be what everybody does and that it's you
37:59 know easier and we can like make it
38:00 healthier for the child and mother you
38:02 know the mother doesn't take the health
38:03 risk but
38:04 >> but man
38:06 >> so intellectually I can say that and
38:08 then like emotionally it feels like h
38:10 something is off of that yeah
38:13 >> oh yeah yeah cuz then the family like on
38:15 the weekends the parents would come and
38:17 like tinker on the glass or whatever or
38:19 the dad would put like a um you know
38:21 like a go falcon sticker on the thing or
38:23 you know what I'm saying people would
38:24 like decorate it all up or write little
38:26 messages on there. Um,
38:28 >> you know, I think there's another like
38:30 another take I have on all of this is
38:31 that there in this world that we're
38:34 heading to of like crazy sci-fi
38:35 technology becoming reality, the the
38:37 sort of like the deeply human things
38:39 will become the most precious, sacred,
38:41 valued things
38:43 >> and that we'll really care about like
38:45 the
38:46 >> See, I I I deeply agree with that. I I
38:49 do think that AI is going to push us to
38:51 this weird place where
38:56 like human interactions become elevated.
39:00 They become more important.
39:03 It's interesting to hear Sam do that. I
39:06 What a what a fascinating take from from
39:09 Theo Vaughn. Like I thought this is who
39:12 you were. I thought you'd like this
39:15 dystopian future. And Sam's like, "Uh,
39:17 we just talked about my baby, dude.
39:21 >> The human experience more than ever."
39:23 And maybe it won't go that way. I don't
39:24 know.
39:24 >> Yeah,
39:25 >> man.
39:26 >> Do you uh No, and that's some of the
39:27 stuff we want to talk about. And thanks,
39:29 thanks so much, man, for sitting down.
39:30 Um, do you think your child will go to
39:33 college? Do you think like what do you
39:35 kind of think that looks like?
39:36 >> Probably not. Um, if I had to guess,
39:38 like I I think Well, I only went to half
39:40 of college.
39:41 >> You You did you drop out?
39:42 >> Yeah,
39:43 >> dude. You guys all I freaking dropped
39:45 out. I didn't get [ __ ] You dropped out.
39:47 Wang dropped out. Zuckerberg dropped
39:50 out.
39:51 >> Um,
39:51 >> probably a lot of other people.
39:52 >> And you?
39:53 >> Yeah. Yeah. Okay. Well, hey, we're both
39:55 here, so Oh, it worked out fine.
39:59 >> And you
40:04 Oh my god.
40:08 Oh my god.
40:09 >> Guys, you're right. You know, you're
40:10 right. Never mind. I'm sorry. I'm being
40:11 self-defeating.
40:13 >> Um, yeah. What does that look like when
40:15 you think about that like
40:17 >> yeah with AI with so much new
40:18 information coming online right and so
40:20 much like data being collected and like
40:22 um information being uh carpulled and
40:26 and maybe which is a term
40:28 >> so you you and I never grew up in a
40:30 world that didn't have computers like
40:32 and our parents were like oh this there
40:34 weren't computers and then there were
40:35 and it was this big crazy adjustment it
40:37 took them a long time to figure it out
40:38 but to us like computers just always
40:40 existed they were just I mean maybe they
40:42 were kind of new but they were always
40:44 around
40:45 >> and and then like you know a kid that is
40:47 like
40:49 there was there was this video on
40:50 YouTube I saw like maybe 12 years ago
40:53 something like that that 14 years ago
40:55 that just really stuck with me. It was
40:56 like a little baby in a dentist waiting
40:58 room or something picking up one of
41:00 those old glossy magazines and going
41:02 like this.
41:02 >> Oh, I remember that.
41:04 >> And to that kid it was just like a
41:05 broken iPad because that kid had just
41:07 like grown up in a world where like
41:09 there were touch screens everywhere. And
41:12 my kid will never grow up will never
41:16 ever be smarter than an AI.
41:17 >> Yep.
41:18 >> That will never happen.
41:20 >> That's fascinating. My kid will never
41:22 ever be smarter than an AI. That will
41:25 never happen.
41:27 Right. I've talked about this on this
41:29 channel that we are we are in this very
41:32 rare
41:33 slice in history, incredibly rare slice
41:36 in history
41:39 where any of us who if you're conscious
41:41 enough to watch the AI learning lab and
41:44 demented enough to watch the AI learning
41:45 lab,
41:49 we get the privilege and I think this is
41:51 an absolute [ __ ] privilege of knowing
41:55 what the world was like before AI
41:59 and we're going to know what the world
42:01 is like after AI.
42:03 We're the cusp generation
42:07 of this transformation
42:09 and it is such an incredibly rare moment
42:12 in history.
42:15 And his kid is part of the generation
42:18 that will never
42:20 know what it's like to not have an
42:22 entity, an AI entity that's smarter than
42:25 humans. It's amazing.
42:29 >> Unless it's implanted, right? Yeah.
42:31 Exactly.
42:32 >> You know, kid born a few years ago, they
42:33 had a brief period of time. My kid never
42:35 will be smarter, but also
42:38 they'll never they'll never know a world
42:40 where like products and services aren't
42:42 way smarter than them and and super
42:45 capable. They can just do whatever you
42:46 need. And in that world, I think
42:48 education's going to feel very
42:50 different. I already think college is
42:51 like maybe not working great for most
42:53 people, but yeah, I think fast forward
42:56 18 years, it's going to look like a
42:57 very, very different thing.
42:59 >> Yeah. Yeah. Do you think there Oh,
43:01 here's that video right here. This kid.
43:03 >> Yeah. Yeah. Yeah. All right. I was wrong
43:04 about the dentist. It was Or maybe
43:06 there's a few of these.
43:06 >> He's like, "Somebody charge this
43:08 magazine." He's yelling. How would you
43:09 recommend to a parent right now to
43:11 prepare their children for like an AI?
43:13 >> Yeah, that's good. Um, source camp. Some
43:15 of us have gone from pre pre- microwave
43:18 to post AI. Exactly. Exactly.
43:23 Or pre video recorder. I remember
43:29 we had a this was 19
43:35 like 78 or nine.
43:38 We had a Magnavox. I It must have been
43:41 like a 27 or 32 inch television. It was
43:44 this giant white white box with a TV in
43:49 it and on top of it was a VHS player
43:53 that was it was like the size of a trunk
43:55 of a car
43:57 and the thing would come out of it.
44:01 Some of us are about to go from pre
44:03 prejetsons to postJetsons. That's really
44:05 good.
44:07 God damn this this interview like
44:14 I think Theo Vaughn is being smacked in
44:18 the face here with
44:30 the the future is not the cliche he
44:32 thought it was. He doesn't know what it
44:35 is yet,
44:37 but like he's coming in with these
44:39 preconceived notions about what the
44:41 future's going to be like and who who
44:43 the overlords of that future are.
44:46 Fascinating. Just fascinating.
44:48 >> My future kind of like are there certain
44:50 curtails that you would start to put in
44:51 now? Are there certain like um you know
44:54 adjustments where you like get them in a
44:56 certain training or have them start to
44:57 watch certain models of things online?
44:59 Like what is that?
45:01 >> You know, I I actually think the kids
45:02 will be fine. I'm worried about the
45:03 parents.
45:04 >> If you look at the history of the world
45:06 here, when there's new technology, like
45:08 people that grow up with it, they're
45:09 always fluent. They always figure out
45:10 what to do, they always learn the new
45:12 kind of jobs. But if you're like a
45:14 50-year-old
45:16 and you have to like kind of learn to do
45:18 things in a very different way, that
45:19 doesn't always work. Yeah.
45:20 >> So, I think the kids are going to be
45:22 fine. I mean, I do have
45:24 >> That's all of you. He's talking about
45:26 you, you old fat people.
45:30 Definitely not me because I'm not old or
45:33 fat.
45:34 >> Shut up.
45:35 >> Worry. Like I do have worries about kids
45:38 in technology. Like I think this
45:40 scrolling the kind of like you know
45:43 >> silver fox down boy
45:45 >> short video feed dopamine hit. It feels
45:48 like it's probably messing with
45:49 >> Sharon Crawford. Not me.
45:51 >> Kids brain development in a super deep
45:53 way. So it's not that I have no worries.
45:54 I have like extremely deep worries about
45:56 what technology is doing to kids. But in
45:58 terms of kids ability to like be
45:59 prepared for the future and use a new
46:01 technology, they seem really good at
46:03 that.
46:03 >> Yeah.
46:03 >> Always through history.
46:04 >> That's a good point actually. Yeah. It's
46:06 like if you just grow up with it, it's
46:07 just like having it's just totally
46:09 normal. It's like having kneecaps or
46:10 whatever. You're just kind of used to
46:11 it.
46:12 >> You you can't imagine the world where
46:16 >> it's like having kneecaps.
46:18 [Laughter]
46:23 Oh my god.
46:27 This This is just the bro most broken
46:29 interview. It's so good. It's It's so
46:33 good.
46:34 >> It doesn't exist. You just Right. Yeah.
46:36 Yeah. That's a good point.
46:38 >> I remember when I was
46:39 >> You know what it is about Theo Vaughn is
46:41 he's so sincere. Like he just like just
46:44 the way his brain works like the dots
46:47 that he connects. It's just it's [ __ ]
46:51 off. It's like a dude that huffed paint
46:54 as a teenager trying to navigate the
46:57 world
47:02 >> uh in school in like junior high and
47:05 Google first came out
47:07 >> and all the teachers like freaked out
47:09 and they're like this is the end of
47:10 education you know I know if you why do
47:12 you have to memorize history facts in
47:14 history class if you can just look them
47:15 up instantly on the internet you don't
47:17 even have to learn to go to the library
47:18 and the answer is like yeah maybe
47:20 memorization is less important but with
47:22 these new tools, you can think better,
47:23 come up with new ideas, do new stuff.
47:25 I'm sure the same thing happened with a
47:26 calculator before.
47:27 >> Yeah.
47:28 >> And you know, now this is like this is
47:29 just a new tool that exists in the tool
47:31 chain.
47:32 >> And what about like say if there is
47:34 somebody though that's like learning
47:35 history right now, like they just
47:36 started their second year of college.
47:38 >> Oh, that Celsius. Yeah, that thing will
47:39 definitely you won't be able to blink
47:41 for a month, homie. That thing will
47:42 Yeah, you'll sneeze and release. [ __ ]
47:46 >> You won't be able to blink for a month.
47:50 He gave him an energy drink. Oh my god.
48:00 Oh, something I wonder about is because
48:02 I'm deep in AI, I don't come up with new
48:05 fresh perspectives. Keep trying new
48:07 things. That's interesting.
48:10 That's interesting. I mean, I think
48:12 there's a couple of ways that you could
48:13 look at that, Gareth. I mean, the other
48:14 one is that maybe you're maybe you're
48:18 considering wider perspectives than you
48:20 would have otherwise because you can
48:22 travel so far knowledgewise, you know,
48:25 laterally, but who knows?
48:27 >> 5.0, dude. You'll freaking Are you guys
48:29 at 4.5 already?
48:30 >> We're 4.5 already. 5.0 is uh I think
48:32 it's going to be great.
48:33 >> Oh, it'll come out fast if you had that
48:34 Celsius. I'm saying you
48:35 >> maybe the researchers need it, not me.
48:37 But uh you know, we'll get them some.
48:38 >> Yeah, that'll get you there, man. Um so,
48:41 say there's somebody just for example,
48:42 like that's learning history right now.
48:43 They're in their second year of college.
48:44 They're they're taking history. Is that
48:46 are there some subjects in like like
48:49 they they're going to be a historian? Is
48:51 that still a viable space of work? Uh as
48:54 AI moves forward, do you think?
48:56 Honestly,
48:58 >> I I assume there will be some version of
49:00 it that is uh I I think it's very hard
49:05 to predict exactly how something
49:07 evolves. Um I or predict exactly the
49:10 jobs of the future going to be like the
49:14 you know not that long ago it would have
49:18 been very hard to predict either of our
49:20 jobs if you go back 100 years the idea
49:22 of like this CEO of an AI company or a
49:26 podcaster like you know probably would
49:29 have been things that didn't seem to be
49:31 the most obvious evolutions of the
49:33 things people were doing at the time.
49:35 >> Yeah. You just seemed almost probably
49:36 crazy even in trying to explain those to
49:39 someone.
49:39 >> You would. And now, in fact, two of the
49:41 job I I heard that the job that young
49:43 people most want is some version of your
49:46 job.
49:47 >> The job that young people most want is
49:48 to be uh, you know, podcast influencer,
49:51 uh, YouTube, they want a YouTube
49:52 channel, like whatever it is, they they
49:54 like six, seven year olds. I don't know
49:55 how to describe it, but that's what they
49:56 want.
49:57 >> And a lot of people also want my job.
49:59 They want to do like a startup or they
50:00 want to work on AI. And these just
50:02 didn't exist.
50:03 >> Yeah. So like the rate with which the
50:04 new things come along is is fast and
50:08 also trying to predict what they are. I
50:10 don't know. The thing I say all the time
50:11 is no one knows what happens next. It's
50:13 like we're going to figure this out.
50:14 It's this weird emergent thing. Does the
50:16 current job of a historian exist in the
50:19 same way? I'll bet not quite. But
50:22 another thing I believe is that humans
50:24 are obsessed with other people. Like we
50:26 are so deeply wired to care about other
50:27 people to care about stories and history
50:29 our own history is extremely interesting
50:31 to us. So I would say somehow or other
50:34 we're still going to care about that.
50:36 There's going to be some kind of job
50:37 doing that.
50:39 >> Man, that's cool. G I guess I I if
50:42 when I take that avenue of thought like
50:44 okay there will still be this historian
50:46 or some it'll be some evolution of that.
50:49 Right. That does seem kind of cool to me
50:51 because there's a level of creativity in
50:53 there. There's a level of like faith and
50:55 spontaneity in there that I think is
50:57 kind of exciting. So yeah, I guess I
50:58 hadn't really thought about that.
50:59 Sometimes I get stuck in this doomsday
51:01 thing like I just see like you know like
51:03 the history book closes and they're like
51:05 we have enough we have all the history
51:06 over here you know
51:08 >> it you know people used to say like oh
51:10 there's no need for more music we've
51:11 made perfect music like why does anyone
51:13 need anyone to create anymore and that's
51:15 obviously ridiculous. Yeah.
51:16 >> Or they would say there's that famous
51:17 patent office quote, everything that
51:19 humans ever possibly need has been
51:21 invented. There's nothing left to do.
51:22 >> I have heard that. But here we are.
51:24 >> Here we are. And and like
51:26 >> Yep. And I know Shadow, I saw your
51:28 comment there about, yeah, let's pollute
51:29 the environment even more. Yeah. I mean,
51:31 listen,
51:34 so so these tools right now take a lot
51:36 of energy. These tools are also going to
51:39 help us solve energy problems and
51:41 they're going to help um come up with
51:45 new kinds of energy or new kinds of
51:47 energy production. They're also going to
51:49 get a lot more efficient over time. Um
51:53 but they're not going to go away. Like
51:55 part of the part of this channel is is
51:58 not to judge it good or bad, but just to
52:00 say if it's not going away, how do we
52:02 deal with it? So, um so that's that's
52:04 what we're doing here. Someone asked me
52:06 the other day like, you know, how long
52:07 is it until you can make like a AI AI
52:10 CEO for OpenAI? And I was like, probably
52:13 not that long. And they were like, well,
52:14 aren't you really sad about that? And I
52:16 was like, no, I think it's awesome. I'm
52:18 for sure going to figure out something
52:19 else to do. I'm excited to do that.
52:20 Like,
52:21 >> I think that's great.
52:23 >> Right. So, you could create something
52:24 that would have your job, but then you
52:28 could do something else.
52:29 >> Totally.
52:30 >> But then how do you know that you'll
52:31 still get paid for your job? I guess
52:33 like
52:34 >> well
52:34 >> it's kind of a big question. I I kind of
52:36 think that
52:37 >> but yeah I guess the framing of that
52:39 question might be better like say there
52:40 are jobs that get curtailed by
52:41 >> there will be some
52:42 >> okay
52:42 >> I think it's important to be honest
52:44 >> there will be some there's going to be a
52:45 lot there's going to be a lot but what
52:48 no one talks about is jobs going to be
52:50 created so we'll see we'll see where Sam
52:51 goes with this this is interesting
52:53 >> it's about that there will be some jobs
52:54 that totally go away
52:56 >> but mostly I think we we we will rely on
52:59 the fact that people's desire for more
53:02 stuff for better experiences for, you
53:06 know, a higher social status or
53:07 whatever, that seems basically
53:09 limitless. Human creativity seems
53:11 basically limitless. And human desire to
53:13 like be useful to each other and to
53:15 connect with each other and do stuff for
53:16 each other and focus on other people
53:18 seems pretty limitless, too. So, I think
53:21 throughout all of history, there have
53:22 been these predictions like ah, you
53:24 know, we're gonna
53:27 like all be on the beach and work an
53:29 hour a day or hour a week or whatever
53:31 and we're going to have unlimited wealth
53:32 and and
53:34 >> I've never heard that one. I would I
53:35 love that.
53:36 >> I mean, they used to say this. They used
53:37 to like the industrial revolution,
53:39 people were like, "Oh, you know, we've
53:41 just figured out how to automate like
53:42 man's lot in life. There's nothing left
53:43 to do. We're going to have these
53:44 machines do all the work."
53:45 >> Makes sense, probably. and you watch
53:47 these machines doing all this stuff that
53:49 only people used to physically do and
53:52 everybody panicked and said there's
53:53 going to be no more jobs and we figured
53:55 out
53:57 >> new stuff to want. Now here's an
53:58 interesting thing. If you could go back
53:59 to that industrial revolution time and
54:01 people before that were, you know,
54:03 really,
54:04 >> which by the way, I think the industrial
54:06 revolution to me feels like the thing
54:08 that is the closest the closest parallel
54:11 to what's going on with AI. Where the
54:14 industrial revolution augmented our
54:16 physical strength, AI is augmenting our
54:19 intellectual strength, our intelligence.
54:21 Um,
54:24 and if you look at the absolute
54:26 transformation that happened, you know,
54:28 from the the late 19th century through
54:31 the 20th century, um,
54:35 like
54:37 that's about to happen with with our
54:41 intellect, right? So, just like these
54:43 machines got way stronger than any human
54:46 and we figured out other things to do,
54:48 that's about to happen with our
54:49 intellect. Um, I think it's going to
54:51 happen a lot faster than the transition
54:54 uh in the industrial revolution just
54:57 because there's not so much physical
54:59 stuff to do. Um, and I think because of
55:02 that the transition is going to be
55:05 really profound and and terrifying
55:09 >> on the grind.
55:10 >> Yeah. Exactly. P M the industrial
55:13 revolution didn't happen this fast.
55:15 Exactly. Yeah. You had time from from
55:18 the day they invented the steam engine,
55:20 this thing that went kadunk,
55:22 you know, and could make a belt, you
55:24 know, go around like this till when they
55:26 put that steam engine into a tractor and
55:29 then, you know, made that tractor pull a
55:31 plow, right? There there were like
55:33 years, decades in between th those major
55:36 chunks of things. But you did have, you
55:40 know, spans of 20 or 30 years where you
55:43 had, you know, massive amounts of new
55:47 inventions that changed everything.
55:51 So what will happen with massive info
55:54 overload and organizing
55:56 everyone sharing and doing research
55:58 exponent exponentially? Fascinating to
56:01 think about and position for. Josh,
56:04 that's a great question. I like implicit
56:07 in your question is that we humans are
56:10 going to have to deal with the
56:12 information overload. We may not, right?
56:15 It might be the kind of thing where the
56:17 AI is
56:21 at any point if any human needs
56:23 anything,
56:25 the AI either makes it or goes and finds
56:27 it. And so, so I think I think
56:32 information overload is is what Google
56:34 tried to solve 25 years ago, right? They
56:38 said, "Hey, if we could index all of
56:41 this information
56:44 and we could make a little search box
56:47 and we could do that, well, then someone
56:49 rather than having to navigate
56:52 by clicking on all these hyperlinks,
56:54 they could just ask our thing and we'll
56:57 tell them where it is in this giant
56:59 information space."
57:03 AI represents all of that information is
57:06 now compressed into this magic box that
57:10 we just get to tap into and knowledge
57:12 emerges out of it. And so,
57:16 so while massive amounts of information
57:18 are being created by AI, I don't think
57:21 we're going to have to process all that
57:23 much. I mean,
57:26 you know, your friends and co-workers
57:28 that use AI to write those 20page
57:31 proposals for you that you then put into
57:34 chat GPT and distill it down into five
57:36 bullet points.
57:39 There'll be a lot of that.
57:44 Super hard trying to like kind of have
57:46 enough food to survive. Go back to those
57:49 people. Look at our jobs today. Would
57:52 those people say we have real jobs or
57:54 would they say you have unbelievable
57:55 abundance, unbelievable wealth, so much
57:57 food to eat, incredible luxury, and you
58:00 guys are just like playing a game to
58:01 entertain yourselves. Is that a real job
58:03 or not? And they would probably say
58:04 where they sit, what you guys do is not
58:06 a real job. You guys are, you know,
58:09 you're too rich. You're wasting your
58:10 time. This is this is the argument that
58:12 a lot of creatives are making right now
58:14 that if you just push a button and it
58:15 generates something, that's not real
58:17 creative, right? Is is what he's talking
58:20 about now. that you know someone from
58:21 the industrial revolution looking at
58:23 what we do now that's not real work
58:25 >> you're trying to
58:26 >> right real work is men lifting things
58:28 >> like
58:29 >> yeah you guys are a couple of dang zest
58:31 lords out there freaking playing uno in
58:32 the park or whatever they would they
58:34 would not I don't think my grandfather
58:35 would be like you have a job he would
58:36 still be like you need to get a job
58:38 >> yeah totally and when we look forward
58:40 another hundred years of what people are
58:41 doing they'll probably think they're
58:43 working very hard it'll feel very
58:45 satisfying very intense to them they're
58:46 really like they'll feel engaged they'll
58:48 be making my people happy they'll be
58:49 creating value for each other. But if we
58:51 could look forward that hundred years at
58:52 those guys, do you think we would say
58:54 they're working or like man you have
58:56 like AI doing everything for you? You're
58:57 just trying to entertain.
58:58 >> Yeah. Like I could see so so it's like
59:01 if if if AI ends up handling most of the
59:05 tasks that we do now
59:08 what might become valuable is like
59:11 taking a walk with a human being. And so
59:13 there might be professional hikers that
59:16 just walk with other people,
59:19 right? I just want to like walk and hang
59:21 out with someone and I'm willing to pay
59:23 for that and that's a job. And there
59:25 might be someone else that's just like,
59:26 you know, teaching juggling on a street
59:28 corner might become a respectable job.
59:37 We'll see.
59:38 >> Yeah. Like, oh, you guys have it so
59:40 easy, right? But I think that's
59:42 beautiful. I think it's great that those
59:44 people in the past think we have it so
59:45 easy. I think it's great that we think
59:46 those people in the future have it so
59:47 easy.
59:48 >> Yeah. Valerie Cox, I hope life work
59:50 balance swings toward more life
59:52 enjoyment.
59:54 And that's that's kind of what I'm
59:55 saying that that
59:58 a lot of the tasks that we call work
1:00:01 today.
1:00:03 We take pride in doing our work, but a
1:00:06 lot of the tasks that we do are
1:00:08 mind-numbing,
1:00:10 shitty tasks. Imagine if those
1:00:14 mind-numbing shitty tasks were done,
1:00:17 handled.
1:00:18 Assume we navigate through the money
1:00:20 piece of this, right? Everyone's got to
1:00:22 figure out how to get paid or whatever
1:00:24 that looks like in the future. Assume
1:00:26 we're going to navigate through that
1:00:27 because we're humans and we adapt. And
1:00:29 then what's on the other side of that?
1:00:32 Well, we would probably start to
1:00:34 prioritize things like life enjoyment.
1:00:36 Well, what gives me more joy in life?
1:00:38 Well, I like camping. Well, I like
1:00:40 riding bikes. Well, I like doing this. I
1:00:41 like doing that. Maybe the things that
1:00:44 seem like leisure, the thing we call
1:00:47 leisure right now,
1:00:50 which we [ __ ] all over, right? You only
1:00:52 do leisure on the weekends. We don't do
1:00:54 leisure for a job. We work for a job.
1:00:58 Well, maybe that shifts in the future.
1:01:01 >> Like that is the beautiful story of
1:01:04 >> Sher Sharon Crawford. There's a chance.
1:01:07 >> Oh [ __ ] I I lost I just moved
1:01:09 something.
1:01:10 >> Celsius. Yeah, that thing will
1:01:11 definitely you won't be able to blink.
1:01:12 >> A historian exists in the same way. I
1:01:14 would bet not quite. But another thing I
1:01:17 believe is that humans are obsessed with
1:01:19 other people. Like we are so deeply
1:01:21 wired to care about other people, to
1:01:22 care about stories in history, our own
1:01:25 history is extremely interesting to us.
1:01:27 So I would say somehow or other we're
1:01:30 still going to care about that. There's
1:01:31 going to be some kind of job doing that.
1:01:33 >> Man, that's cool. Yeah, I guess I I if
1:01:37 when I take that avenue of thought like,
1:01:39 okay, there will still be this
1:01:41 >> live live for I like that, Rose. Live
1:01:43 for the hobbies. The hobification of
1:01:44 humans. Yeah. Like like I I I think the
1:01:48 point that that they're both saying here
1:01:51 or that Sam's saying to him is like
1:01:54 Theo, look at your job right now. Your
1:01:57 job is to like sit and talk to people
1:02:00 and let other people watch that.
1:02:03 Like
1:02:04 someone from a hundred years ago, that's
1:02:06 not work. What are you doing? Go do
1:02:10 something with your life, you [ __ ]
1:02:12 slacker, and get a haircut.
1:02:19 Shave off that mullet. Historian or
1:02:21 somebody, it'll be some evolution of
1:02:24 that, right? that does seem kind of cool
1:02:26 to me because there's a level of
1:02:27 creativity in there. There's a level of
1:02:29 like faith and spontaneity in there that
1:02:31 I think is kind of exciting. So, yeah, I
1:02:34 guess I hadn't really thought about
1:02:34 that. Sometimes I get stuck in this
1:02:36 doomsday thing. Like I just see like,
1:02:38 you know, like the history book closes
1:02:40 and they're like, "We have enough. We
1:02:41 have all the history over here." You
1:02:43 know, it,
1:02:43 >> you know, people used to say like, "Oh,
1:02:45 there's no need for more music. We've
1:02:47 made perfect music. Like, why does
1:02:48 anyone need anyone to create anymore?"
1:02:50 And that's obviously ridiculous. Or they
1:02:52 would say there's that famous patent
1:02:53 office quote everything that humans ever
1:02:55 possibly need abundance unbelievable
1:02:57 wealth so much food to eat
1:02:58 >> that space right like the like the
1:03:00 movement that happens with AI and with
1:03:02 just technology which will advance
1:03:04 >> and when we look forward another hundred
1:03:05 years of what people are doing they'll
1:03:07 probably
1:03:08 >> think they're working very hard it'll
1:03:09 feel very satisfying very intense to
1:03:11 them they're really like they'll feel
1:03:12 engaged they'll be making people happy
1:03:14 they'll be creating value for each other
1:03:15 but if we could look forward that
1:03:17 hundred years at those guys do you think
1:03:19 we would say they're working or like man
1:03:20 you have like AI doing everything for
1:03:21 you. You're just trying to entertain
1:03:22 yourselves.
1:03:23 >> Yeah. Like, oh, you guys have it so
1:03:25 easy. Right.
1:03:26 >> But I think that's beautiful. I think
1:03:27 it's great that
1:03:29 >> Jeff Flanigan, George Jetson, I'm so
1:03:31 tired of pushing this button.
1:03:35 Like on Star Trek. Yeah, exactly. Shane
1:03:38 Saw. I blindly trust AI will fix a lot
1:03:41 of things. I don't have faith that
1:03:42 humans will allow being fixed.
1:03:46 Well, yeah. Humans are always going to
1:03:48 be messy. Humans are always going to be
1:03:51 messy, but a lot of the a lot of the
1:03:53 [ __ ] angst of the world right now
1:03:58 is the like the nature of the work we
1:04:02 do. Like people are stressed out and
1:04:07 like a lot of our economy is based on
1:04:10 complexity
1:04:12 and a lot of people's jobs are dealing
1:04:15 with complexity, right? So, like if you
1:04:18 think about, I don't know, an insurance
1:04:21 adjuster, their job is to make it as
1:04:25 complicated as possible
1:04:28 to try to get someone to process their
1:04:32 claim so that they don't have to pay out
1:04:34 the money so they can make more profit.
1:04:36 Like, by design, it is overly
1:04:39 complicated. The US health care system
1:04:41 is overly complicated by design.
1:04:45 So as many people as possible get paid
1:04:48 in the process of
1:04:52 making you well. And then on the other
1:04:55 side of that equation is entities trying
1:04:58 not to pay for any of that, right? And
1:05:01 making it as complicated as possible so
1:05:04 as many people as possible can take a
1:05:07 toll out of the transaction of getting
1:05:10 you better. Right?
1:05:14 That's most of our society today. The
1:05:16 global economy is based on
1:05:20 either
1:05:22 evolved complexity or designed
1:05:24 complexity.
1:05:26 AI is going to simplify a lot of that
1:05:28 and that that starts to shift things and
1:05:31 maybe we become less shitty to each
1:05:32 other.
1:05:33 >> Those people in the past think we have
1:05:34 it so easy. I think it's great that we
1:05:36 think those people in the future have it
1:05:37 so easy. Like that is the beautiful
1:05:39 story of us all contributing to human
1:05:41 progress and everybody's lives getting
1:05:43 better and better.
1:05:44 say we're able to get to that space,
1:05:46 right? Like the move like the movement
1:05:48 that happens with AI and with just
1:05:49 technology which will advance quicker I
1:05:51 think which is one thing that AI feels
1:05:53 like to me it's a fast forward button on
1:05:55 technology and on uh possibility because
1:05:59 things can be information can be
1:06:00 quantified so quick and a lot of like uh
1:06:02 more menial tasks even though they're
1:06:04 not really menial in people's lives um
1:06:06 but menial hypothetically uh can be done
1:06:09 quicker to get a lot of the framework
1:06:11 for things done fast. But how will
1:06:13 people survive? Like how do we adjust
1:06:16 our structure of finan of like if some
1:06:20 people own the companies that have the
1:06:22 AI and then a lot of people um are just
1:06:26 using the AIs and the agents created by
1:06:28 AIS to do things for them. How will
1:06:31 society like societal members still be
1:06:33 able to financially survive? Will there
1:06:35 still be money? What is that?
1:06:37 >> This is a great question. This is like
1:06:39 th this is this is the question no one
1:06:42 wants to ask because they don't want to
1:06:44 hear the answer, right? What the [ __ ]
1:06:47 happens? We we get to that future where
1:06:49 AI does everything. How are people
1:06:51 getting paid?
1:06:53 It'll be interesting to hear Sam's
1:06:55 answer. There were some interesting uh
1:06:56 posts on YouTube. Um Josh Groves, that's
1:06:59 why I made right node
1:07:03 uh an app. It shouldn't be this way.
1:07:06 Stop the madness. Make it easier. Bill
1:07:07 Gates said he hires lazy people because
1:07:09 they find the easiest route to um they
1:07:14 they find the easiest route. AI can help
1:07:16 us with this art form. This is why
1:07:19 master synthesizers
1:07:21 um and system thinkers will win. Like
1:07:23 you always say, Kyle, fighting for the
1:07:24 future. Yep. The hardest problem in
1:07:27 therapy was always convincing people to
1:07:29 loosen their grip on the problems they
1:07:32 have to grab different ones.
1:07:33 Interesting.
1:07:35 If AI takes away the financial
1:07:36 requirements, there's still the human
1:07:38 need for control. Interesting.
1:07:40 Fascinating. Fascinating.
1:07:43 This is cool. I'm I'm digging this. I
1:07:45 listen, this is a non-standard. It's
1:07:48 movie night at the at the learning lab.
1:07:51 Um I'm fascinated by this interview
1:07:54 because again, I think this is the most
1:07:56 unlikely pairing I could have imagined
1:07:58 in 2025.
1:08:00 And uh I love that Theo Vaughn, you can
1:08:05 tell he he's not played with AI. I love
1:08:08 that he's coming in with these
1:08:09 assumptions and he's asking the
1:08:11 questions everyone's probably wondering.
1:08:13 It's really really good.
1:08:15 >> Does that make any sense? That question.
1:08:16 >> It totally makes sense. Uh sorry. I
1:08:18 don't know. Neither does anybody else.
1:08:19 But I'll tell you my current best guess.
1:08:21 >> Okay.
1:08:22 >> Well, I'll say two guesses. one I think
1:08:23 it is possible that we put you know GPT7
1:08:28 or whatever in everybody's chat GBT
1:08:30 everybody gets it for free and everybody
1:08:33 has access to just this like crazy thing
1:08:35 such that everybody can be more
1:08:37 productive make way more money doesn't
1:08:39 actually matter that you don't like own
1:08:40 the cluster itself but everybody gets to
1:08:42 use it and it turns out even getting to
1:08:45 use it is enough that people are like
1:08:47 getting richer faster and more
1:08:48 distributed than ever before that could
1:08:50 happen I think that really is possible
1:08:52 there's Another version of this where
1:08:57 the most important things that are
1:08:58 happening are these systems are
1:09:00 discovering you know new cures for
1:09:02 diseases new kinds of energy new ways to
1:09:05 make spaceships whatever and most of
1:09:07 that value is acrewing to the like
1:09:09 cluster owners us just so that I'm not
1:09:11 dodging the question here and then I
1:09:13 think
1:09:13 >> an AI cluster owner is an individual or
1:09:17 entity responsible for the design
1:09:19 deployment management and ongoing
1:09:21 maintenance of an AI I compute cluster.
1:09:25 So, so Altman's saying it's possible
1:09:28 that
1:09:29 because the AI will produce all this
1:09:31 remarkable stuff that the people with
1:09:34 those clusters will basically hoard the
1:09:36 compute and they'll do all the fancy
1:09:39 good [ __ ] and keep all the money. So,
1:09:40 let's see how he answers it.
1:09:43 >> Will very quickly say, "Okay, we got to
1:09:45 have some new some new economic model
1:09:47 where we share that and distribute that
1:09:49 to people." To be honest, I've never
1:09:51 heard of Theo Von. Yeah, if you're not
1:09:53 if you're not kind of deep in the comedy
1:09:55 world, he's he's um he's an acquired
1:09:58 taste. I think he's funny as [ __ ] but
1:10:01 like a lot of times I don't know that he
1:10:04 knows why he's as funny as he is. Um he
1:10:08 just sees the world in this very
1:10:10 different way and that he's interviewing
1:10:12 Sam Alman like I don't I want to know
1:10:14 how this [ __ ] thing came about.
1:10:16 Although at right at the beginning um
1:10:19 apparently they were at both at the
1:10:20 inauguration and Theo Vaughn's chair
1:10:23 fell apart right in front of Sam Alman.
1:10:25 So I think that's how they met um was
1:10:28 Theo Vaughn crashing crashing to the
1:10:30 ground at the uh the Trump inauguration.
1:10:36 >> Uh I used to be really excited.
1:10:38 >> What about people's competitive nature?
1:10:40 I'm not sure how you can completely
1:10:43 level the playing field. So you know
1:10:45 what, Rose? I don't I don't think you
1:10:47 have to I don't think I don't think
1:10:50 human nature goes away, right? I think
1:10:54 people will find new ways to compete if
1:10:57 if the AI
1:11:01 like like one of the things that Sam was
1:11:03 just talking about here before where he
1:11:05 said if everyone's got access to chat
1:11:07 GPT7 sa Sam Alman about a year and a
1:11:10 half or two years ago talked about this
1:11:12 concept of universal basic compute and
1:11:16 and you've probably heard of universal
1:11:18 basic income where you know basically
1:11:20 you take the profit from all these
1:11:22 machines means you tax these frontier
1:11:25 model companies, you take, you know,
1:11:27 some percentage of the profits, you
1:11:29 distribute that back out to the people
1:11:30 that lost their jobs, right? That's one
1:11:32 model, universal basic income. Sam was
1:11:35 talking about the concept of universal
1:11:37 basic compute. And what he was basically
1:11:39 saying is you give every person some
1:11:42 amount of compute per month that maybe
1:11:46 that amount of compute is the equivalent
1:11:49 of having 10 employees. And these are
1:11:52 like 10 of the best employees you could
1:11:54 ever have or 100 employees that every
1:11:56 human being could have a hundred
1:11:57 employees and those employees would do
1:11:59 anything that that human wanted them to
1:12:01 do. Well, some humans are going to be
1:12:04 ambitious and they're going to want
1:12:05 they're going to say like I want more
1:12:07 than a 100red employees. I want to I
1:12:08 want a thousand employees or I want
1:12:10 10,000 employees. And then other people
1:12:13 are going to be like I don't really have
1:12:15 use for 100 employees. I don't care. So
1:12:18 why don't you just pay me
1:12:21 what it's worth to have these hundred
1:12:22 employees go work for you, right? And so
1:12:26 so that's the model he's talking about
1:12:27 where if everyone's got access to these
1:12:29 super powerful things, I think human
1:12:31 dynamics still play out in that. Some
1:12:33 people are going to be more ambitious,
1:12:34 some are going to be less. People are
1:12:36 still going to compete, but we're just
1:12:38 you we're just going to have different
1:12:40 tools of the game to be able to do to to
1:12:42 to be be able to compete. And then if we
1:12:45 do move to the hobbyification of humans
1:12:48 where you know hobbies become as
1:12:52 important as jobs, well then maybe we
1:12:54 get into competitive needle point,
1:12:56 right? There's the national needle point
1:13:00 um speed needle point competition. Like
1:13:02 I I think human nature is still human
1:13:04 nature. It's just where we put our
1:13:06 energy is going to shift
1:13:07 >> about things like UBI. I still am kind
1:13:09 of excited like universal basic income
1:13:10 where we just give everybody money.
1:13:12 >> Yeah, you hear that term a lot. Yeah,
1:13:13 universal basic income. Yeah, I heard
1:13:15 you and Rogan talk about that too a
1:13:16 while back.
1:13:16 >> I still am kind of excited about that.
1:13:18 But I think people really need agency.
1:13:21 Like they really need to feel like they
1:13:22 have a voice in governing the future and
1:13:26 deciding where things go. And I think if
1:13:27 you just like say, okay,
1:13:30 AI is going to do everything and then
1:13:32 everybody gets like a, you know,
1:13:34 dividend from that, it's not going to
1:13:36 feel good. And and I don't think it
1:13:39 actually would be good for people. So, I
1:13:41 think we need to find a way where we're
1:13:43 not just like if we're in this world
1:13:45 where we're not just distributing money
1:13:47 or wealth like actually
1:13:50 I I don't just want like a check every
1:13:52 month. What I would want is like a
1:13:53 ownership share in whatever the AI
1:13:55 creates so that I feel like I'm
1:13:56 participating in this thing that's going
1:13:57 to compound and get more valuable over
1:13:59 time. So, I sort of like universal basic
1:14:01 wealth. By the way, if you're curious
1:14:04 about this whole topic of universal
1:14:06 basic income, universal basic compute,
1:14:09 you should follow a guy on YouTube named
1:14:11 David Shapiro. I talk about him a lot.
1:14:14 Um, and he's got a his whole new thing
1:14:16 recently is post labor economics. He's
1:14:19 got a whole model for this where you
1:14:21 basically bring things down to the
1:14:23 municipal level and you basically use AI
1:14:26 to exploit the resources in your local
1:14:29 region and everybody participates in
1:14:31 that. It's a fascinating model, but
1:14:33 there's there's
1:14:35 we're going to obviously need to figure
1:14:37 something out. There's no good answer
1:14:38 for this, but if you want someone who's
1:14:40 thinking deeply about it, David Shapiro,
1:14:42 >> better than universal basic income. And
1:14:44 I think I don't like basic either. I
1:14:46 want like universal extreme wealth for
1:14:47 everybody. Um, but but even then, like I
1:14:50 think what people really want is the
1:14:52 agency to kind of co-create the future
1:14:54 together. And
1:14:58 and in a world where it's like the AI is
1:15:01 mostly coming up with the new scientific
1:15:03 inventions, at least we've got to still
1:15:05 have humans like invent the new culture
1:15:07 and have that be a very distributed
1:15:09 thing.
1:15:10 >> Okay. I guess yeah, I I I see what
1:15:12 you're saying, but would that be like an
1:15:15 American thing, do you think? Like since
1:15:16 it were invented here, or do you think
1:15:19 I'm just wondering what does that look
1:15:21 like? You know,
1:15:22 >> the economic model of it all, or the
1:15:23 whole thing?
1:15:24 >> Yeah. or like is there a dividend of the
1:15:26 company that then is divided up between
1:15:30 the masses sort of
1:15:31 >> I mean a crazy idea but in the spirit of
1:15:34 crazy ideas is that if the world there's
1:15:36 like eight roughly 8 billion people in
1:15:37 the world
1:15:38 >> if the world can generate like
1:15:42 eight quintilion tokens per year if
1:15:44 that's the world actually let's say the
1:15:46 world can generate 20 trillion quintil
1:15:48 20 quintillion tokens per year
1:15:50 >> tokens of
1:15:51 >> like each word generated by an AI
1:15:53 >> okay
1:15:53 >> just making up a huge number here. We'll
1:15:55 say, okay, 12 of those go to, you know,
1:15:57 the normal capitalistic system, but
1:15:59 eight of those eight quintilion tokens
1:16:01 are going to get divided up equally
1:16:03 among 8 billion people. So, everybody
1:16:05 gets one trillion tokens.
1:16:07 >> And that's your kind of
1:16:08 >> he's talking this is his universal basic
1:16:11 compute idea right here.
1:16:13 >> Universal basic wealth globally. And
1:16:16 people can sell those tokens. Like if I
1:16:17 don't need mine, I can sell them to you.
1:16:19 We could pull ours together for some
1:16:21 like new art project we want to do. But
1:16:23 but instead of just like getting a
1:16:24 check, you're get everybody on Earth is
1:16:26 getting like a slice of the world's AI
1:16:28 capacity and then we're letting the like
1:16:31 massively distributed human ingenuity
1:16:34 and creativity and economic engine do
1:16:35 its thing.
1:16:36 >> I mean that's like a crazy idea. Maybe
1:16:38 it's a bad one, but that's the kind of
1:16:40 thing that I think sounds like someone
1:16:41 should think about it more.
1:16:43 >> One of the big fears is like purpose,
1:16:44 right? Like human purpose. Like work
1:16:46 gives us purpose. And also I think the
1:16:48 idea that we are the ones advancing
1:16:52 humanity gives us purpose like we are
1:16:55 the
1:16:57 like yeah like we have some
1:17:01 control over our own destiny maybe gives
1:17:04 us this sense of purpose and it feels
1:17:07 like that we would lose a sense of
1:17:08 purpose or that purpose would be
1:17:10 adjusted like if AI is to really you
1:17:13 know continue to advance so quickly. It
1:17:16 feels like our sense of purpose would
1:17:18 start to really disappear.
1:17:20 Do have you had thoughts about that?
1:17:22 >> I worry about this a lot. It's so I
1:17:24 think people have worried about this
1:17:25 with every big technological revolution,
1:17:27 but I agree that this time it feels
1:17:29 different. Like
1:17:30 >> Okay. Yeah. Because if say you had an
1:17:31 axe and somebody came out with a saw,
1:17:33 you're like you're like, "Yeah, that's
1:17:35 it."
1:17:35 >> Or even if they come out with like a a
1:17:37 robot that cuts the tree down, it still
1:17:38 feels fine. But like creativity
1:17:41 intelligence I think cuts so deeply at
1:17:43 the core of whatever we are and how we
1:17:46 how we value ourselves. Um
1:17:50 one example we can look at this right
1:17:52 now I think one area where AI is having
1:17:53 a big impact is on how people write
1:17:55 software for a living and AI is really
1:17:57 good at that. It's really changed what
1:17:59 it means to be a software developer.
1:18:02 I haven't heard any of those software
1:18:04 developers say that they even though
1:18:05 their job is different that they don't
1:18:06 have meaning they still enjoy it.
1:18:08 they're operating at a higher level. Um,
1:18:11 and I'm hopeful, at least for a long
1:18:13 time, you know, 100 years from now, who
1:18:15 knows? But I'm hopeful that that's what
1:18:17 it'll feel like with AI is even if we're
1:18:19 asking it to solve huge problems for us,
1:18:21 even if we ask it to say like, you know,
1:18:23 go discover a cure for cancer,
1:18:27 there will still be a lot of things to
1:18:29 do in that process
1:18:31 that feel valuable to a person.
1:18:35 >> You'll still asking it the questions.
1:18:36 You're still like helping guide it.
1:18:37 you're still framing it or whatever it
1:18:39 is, you're still like talking to the
1:18:40 world about it. And
1:18:46 and I think all of human history
1:18:48 suggests that we find a way to put
1:18:51 ourselves at the center of the story and
1:18:54 feel really good about it. Like you know
1:18:55 if you kind of think like
1:18:59 we used to think that the earth was the
1:19:01 center of the solar system and then
1:19:02 we're like very humanentric view and
1:19:04 then we're like okay fine the sun is the
1:19:06 center of the solar system but the solar
1:19:07 system is at least the center of the
1:19:09 galaxy and now oh man there's a lot of
1:19:11 galaxies and oh man now we're this like
1:19:12 tiny speck in this like very huge
1:19:14 universe
1:19:16 and
1:19:18 and yet we still manage to feel all like
1:19:20 a lot of main character energy. And so I
1:19:23 somehow think even in a world
1:19:25 >> Did you already said that was great? We
1:19:27 we still feel like we we still feel like
1:19:30 we have a lot of main character energy
1:19:33 that that even though the more
1:19:35 insignificant we realize we are, we
1:19:38 still find a way to put ourselves at the
1:19:40 center of the story. I think that's
1:19:41 great. Yeah. That that's the thing. The
1:19:44 thing that baffles me about doomers that
1:19:48 that say that AI is going to erase our
1:19:52 meaning is they talk about humans as if
1:19:55 they're static. Like like our current
1:19:57 point of view is always going to be our
1:19:59 current point of view. [ __ ] changes.
1:20:02 You adapt to it. You shift your point of
1:20:05 view, right? Like we're going to adapt.
1:20:08 Like humans are going to adapt. We're
1:20:10 not going to not adapt.
1:20:12 That's a a thing I've never gotten about
1:20:14 that particular argument.
1:20:16 >> AI is doing all of this stuff that
1:20:17 humans used to do. We are going to find
1:20:19 a way in our own telling of the story to
1:20:21 feel like the main characters. And I
1:20:23 think in an important sense we will be
1:20:25 and that's really good. I also like you
1:20:29 know probably already today
1:20:32 there could be a very compelling version
1:20:35 of two AIs talking like this.
1:20:37 >> Yeah.
1:20:38 >> And I don't think I would want to watch
1:20:39 that.
1:20:39 >> Right. Like I think I I really do feel
1:20:42 deeply wired to like th
1:20:43 >> This is an important point. So So when
1:20:47 when Big Blue beat Gary Kasparov Gary
1:20:50 Kasparov at chess,
1:20:53 the predictions were no one will ever
1:20:55 play chess again. And we'll just have
1:20:57 machines playing chess against each
1:20:58 other. And guess what? Nobody gives a
1:21:02 [ __ ] about two machines playing chess
1:21:03 against each other. You know what? What
1:21:06 all of that did? It made chess more
1:21:10 popular and it made the appreciation of
1:21:13 humans that are really good at it more
1:21:16 appreciated,
1:21:19 right?
1:21:22 Care about the real person behind it. I
1:21:24 think that's like deep in the biology,
1:21:26 right?
1:21:27 Yeah. That's the part that I think a lot
1:21:29 of times it's like even though you can
1:21:30 get into like these wormholes of like
1:21:33 possibility and these fear holes of
1:21:35 possibility or um kind of this dystopian
1:21:39 ideas that in the end I'm like I'd
1:21:42 rather probably watch something that's
1:21:43 real, you know? It's like because I'm
1:21:45 real. You know what I'm saying? Like I
1:21:47 don't want to talk really to a robot.
1:21:49 I'd ra you know. Yeah. I think in the
1:21:51 end there's going to be a part of you
1:21:52 that wants to continue to just talk to
1:21:54 um talk to humans. Do you uh
1:21:59 what's like one of your fears? Like
1:22:01 what's a fear you have of AI? Like if
1:22:04 you have like a fearful space that it
1:22:06 could go like I know you mentioned it a
1:22:08 little bit.
1:22:09 >> This morning I I was testing our new
1:22:11 model and I got a question. I got
1:22:13 emailed a question that I didn't quite
1:22:15 understand. Uh and I put it in the
1:22:18 model, this GPT5, and it answered it
1:22:21 perfectly.
1:22:23 And I really kind of sat back in my
1:22:25 chair and I was just like a oh man here
1:22:28 it is moment
1:22:30 and I got over it quickly. I got busy
1:22:32 onto the next thing. But it was like a I
1:22:34 mean this what kind we're talking about.
1:22:35 I felt like useless relative to the AI
1:22:37 in this thing that I felt like I should
1:22:40 have been able to do and I couldn't and
1:22:41 it was really hard but the AI just did
1:22:42 it like that.
1:22:43 >> Yeah.
1:22:44 >> It was it was a weird feeling.
1:22:47 >> Yeah. Yeah, I think that's I think that
1:22:48 feeling right there that's the feeling a
1:22:50 lot of people kind of have like what's
1:22:53 going you know when does it happen?
1:22:55 What's going to happen? Um but I think
1:22:58 some of it is it's like yeah you it's
1:23:00 hard to conceptualize until you're
1:23:02 further along.
1:23:04 >> I I'm all to totally I don't think we
1:23:06 know quite how that's gonna
1:23:07 >> but Sam actually answered it.
1:23:11 He said he said he had this weird moment
1:23:13 where he felt inadequate as a human and
1:23:16 then he moved on, right? Like like it's
1:23:19 like that's the thing is like yes, we're
1:23:22 going to confront these things that
1:23:23 we've never had to confront before and
1:23:26 then we're going to go, "Oh, okay. Well,
1:23:28 I guess I don't have to do that
1:23:29 anymore."
1:23:31 And then we get to choose what to do
1:23:33 next, right? Like I don't know feel you
1:23:35 just have to like approach it step by
1:23:37 step. Another thing I'm afraid of, and
1:23:40 we had a,
1:23:42 you know, a a a real problem with this
1:23:45 earlier, but it can get much worse, is
1:23:47 just what this is going to mean for
1:23:48 users mental health.
1:23:50 >> Um, there's a lot of people that talk to
1:23:52 chatt all day long. There are these sort
1:23:54 of new AI companions that people talk to
1:23:56 like they would a a girlfriend or a
1:23:58 boyfriend. Um,
1:24:01 and we were talking earlier about how
1:24:02 it's probably not been good for kids to
1:24:04 like grow up like on the dopamine hit of
1:24:07 scrolling, you know, or whatever.
1:24:09 >> Yeah. Do you think that that how do you
1:24:10 keep like um AI from having that same
1:24:14 effect like that negative effect that
1:24:15 social media really has had?
1:24:17 >> I I'm I'm scared of that. I don't I
1:24:19 don't have an answer yet. Uh I don't
1:24:21 think we know quite the ways in which
1:24:22 it's going to have those negative
1:24:24 impacts. Uh, but I feel for sure it's
1:24:27 going to have some and we'll have to I
1:24:29 hope we can learn to mitigate it
1:24:30 quickly.
1:24:32 >> Um, can AI can they pull up pornography
1:24:34 and stuff like that too or No.
1:24:36 >> Sure.
1:24:36 >> Oh my god.
1:24:38 >> God, I didn't know that.
1:24:43 >> The von talks a lot about sex addiction,
1:24:46 porn addiction, and stuff like that. So,
1:24:48 he just asked the question. Everyone in
1:24:51 America is hoping to know, can they pull
1:24:53 up porn?
1:24:59 Oh my god, that's funny.
1:25:04 I wonder if brain implants will become
1:25:07 similar to getting braces. Probably. So
1:25:09 the the way that uh Neurolink the robots
1:25:13 that they've made to implant those
1:25:14 things are are uh
1:25:18 you know um fairly automated. Um
1:25:22 I have a strong suspicion that that uh
1:25:26 human uh brain computer interfaces are
1:25:29 not going to require implantation for
1:25:31 super long.
1:25:32 >> No, it's fine. I Yeah, but I just Yeah,
1:25:35 I don't even need to know that. I'm
1:25:36 gonna have that stricken from my own
1:25:38 record.
1:25:39 [Laughter]
1:25:42 >> All right, we'll stop it right there.
1:25:46 I'm gonna have that stricken from my own
1:25:47 record.
1:25:53 Oh man. Well, so that's the that's about
1:25:56 the first half hour of of a 90minute
1:25:59 interview. So, I'm obviously going to go
1:26:00 watch the rest of that. You should, too.
1:26:02 Um,
1:26:04 I just like like it feels important to
1:26:07 me. It it feels like a um
1:26:12 the the Today Show moment for AI.
1:26:16 It It really does.
1:26:19 That that moment where Theo Vine where
1:26:22 where he goes, "Don't don't you think
1:26:24 we're just going to be growing babies in
1:26:26 vats?"
1:26:28 And then Sam Holman's like, "That's kind
1:26:30 of weird, dude." and he's like, "Oh, I
1:26:32 thought you would like it."
1:26:38 Oh my god, that's hilarious.
1:26:41 Wow. Wow. Wow. Wow. Wow. All right,
1:26:44 there you have it, man. That is that is
1:26:47 uh
1:26:50 that's a good interview. That's a good
1:26:52 interview. The Theo interview with Waqen
1:26:54 Phoenix is also similarly good because
1:26:58 they're just both uncomfortable human
1:27:00 beings trying to communicate.
1:27:04 Um,
1:27:06 and they they keep sort of blowing each
1:27:08 other away in terms of their humanity
1:27:12 and they like respect each other's
1:27:14 humanity. And this one is similarly
1:27:16 awkward where they're just in these two
1:27:19 guys are in such different planets.
1:27:23 They're like not on the they're not on
1:27:25 the same planet
1:27:28 and yet they're trying to work it out.
1:27:29 Yeah. It's fascinating. I saw a source
1:27:31 camp said I she has a strong suspicion
1:27:33 they're going to become friends. Yeah,
1:27:35 they probably will. And you know, I I I
1:27:38 promise you this. I that that Theo vaugh
1:27:42 leaves the Open AI offices having played
1:27:44 with AI probably more than he ever has
1:27:48 and and maybe understanding it in a
1:27:50 different way. F absolutely fascinating.
1:27:56 Chat GPT, you've talked all day. I have
1:27:58 a headache. You need to go outside.
1:28:00 Yeah, exactly.
1:28:02 Oh my god. Sam is much more human than
1:28:06 Zuck. Yeah. Glad he's in the lead. Yeah.
1:28:09 Every time I see one of these things
1:28:11 with Zuck, I'm just like, "Dude, [ __ ]
1:28:14 get a spokesperson."
1:28:16 Like Sam
1:28:19 Sam is smart enough to know like like
1:28:23 he's at least smart enough to know that
1:28:25 he's got to acknowledge the scary [ __ ]
1:28:28 Zuck just doesn't. Zuck's just like,
1:28:30 "We're going to make a bigger computer.
1:28:33 It's going to be good. we're going to
1:28:34 put it in the metaverse. It's going to
1:28:35 be awesome.
1:28:37 And it just he always seems disconnected
1:28:39 from reality. Sam at least has the
1:28:41 skill. He's at least smart enough to
1:28:43 understand he's got to acknowledge
1:28:45 reality. Now, whether or not
1:28:48 his company takes actions consistent
1:28:50 with that,
1:28:52 that remains to be seen. Um, absolutely
1:28:55 fascinating. Other thoughts?
1:29:00 Should
1:29:03 [Music]
1:29:08 be a very interesting autumn. Yeah, it's
1:29:10 going to be interesting.
1:29:13 I got to go watch the rest tomorrow.
1:29:15 Pretty funny. Yeah, I know. Me, too.
1:29:19 Oh, careful with the corn word. Oh, you
1:29:21 can't you can't say P O R N on Tik Tok.
1:29:24 Really? You can't even say the word.
1:29:28 Unbelievable.
1:29:30 Well, we won't have to deal with it
1:29:32 after September. So, there you have it.
1:29:36 Um,
1:29:40 what do I like on Ronin?
1:29:43 What you like on Ronin? I don't know
1:29:45 what that means.
1:29:47 Um, now I need to go watch the rest of
1:29:50 it. One more plug for B2B.
1:29:56 Oh, yeah. Yeah, yeah. Let's do
1:30:00 so next week people. Next week
1:30:05 [Music]
1:30:14 [Music]
1:30:16 um
1:30:19 do me a favor. So So next week I'm doing
1:30:22 a fiveday back to basics AI crash
1:30:24 course. Doesn't cost a thing. It's still
1:30:27 it's basically just like show up and
1:30:28 I'll keep doing this [ __ ] except instead
1:30:31 of movie night,
1:30:33 we're gonna do five different nights.
1:30:35 First night's going to be what is AI?
1:30:37 How does it work? Why is it a big deal?
1:30:40 Why is it historically like something
1:30:43 I've seen before? Um,
1:30:47 really get your head around what it is.
1:30:49 Tuesday night is going to be chat GPT
1:30:53 101. Like what is chat GPT? What are the
1:30:56 different models? What can you do with
1:30:57 it? Like like like I don't know if you
1:31:00 knew this. Did you know that chat GPT
1:31:02 can see you can upload a picture of a
1:31:05 whiteboard and it'll understand
1:31:06 everything on the whiteboard including
1:31:08 pictures and words
1:31:10 and context.
1:31:14 Did you know that chatbt can talk? You
1:31:16 can just have a full-on conversation
1:31:18 with it. Um, did you know that it can
1:31:21 use tools that it can that it can do
1:31:23 multi-step reasoning?
1:31:26 Reasoning people will tell you that it's
1:31:28 not actually reasoning, but it can do
1:31:29 multi-step internal prompting, let's
1:31:31 call it. Um, and it can use tools and
1:31:34 write Python and and and do all sorts of
1:31:36 crazy [ __ ] like that. So, that's Tuesday
1:31:38 night. Wednesday night is going to be a
1:31:41 creative night. We're going to be
1:31:43 looking at creative tools and creativity
1:31:45 and making images and making video and
1:31:48 making music and making sound effects
1:31:50 and voice synthesis and all that sort of
1:31:53 [ __ ] Thursday night's going to be we
1:31:55 bring it all together
1:31:57 and solve like complicated use cases
1:32:01 that require multiple tools coming
1:32:03 together. And then Friday night's going
1:32:05 to be just an open-ended Q&A to talk
1:32:08 about all the stuff that we've learned
1:32:10 during the week. So, here's my request.
1:32:12 My request is go to
1:32:17 YouTube
1:32:19 right now and subscribe to the AI
1:32:22 Learning Lab. So, it's learning lab-ai.
1:32:26 I didn't I wasn't quick enough to get AI
1:32:28 learning lab. So,
1:32:31 oh, and by the way, the the crash course
1:32:33 next week is free. It's free to
1:32:35 subscribe to the AI Learning Lab. The
1:32:38 course next week is free. I think that
1:32:41 the course we will probably make
1:32:43 available on demand in the future, but
1:32:45 if you just come next week live, you get
1:32:47 it for free. Okay? Um, but go subscribe
1:32:51 to YouTube now and then next week the
1:32:53 sessions will all start at 8:30 p.m.
1:32:56 Mountain time Monday through Wednesday.
1:32:59 Um,
1:33:01 we'll have to tell my seniors to tune
1:33:04 in. Yes. Yes. Bring your seniors. So
1:33:07 that's my other request. So the first
1:33:08 request is go subscribe to the AI
1:33:11 learning lab uh on YouTube. The second
1:33:15 request is tell people tell people in
1:33:19 your lives like I know that there were
1:33:20 some people in here earlier who are you
1:33:22 know AI haters.
1:33:25 Come as an AI hater like at least here's
1:33:28 my request. Here's the thing about this
1:33:31 channel.
1:33:33 What you will not get what what you will
1:33:35 get on this channel is a lot of optimism
1:33:38 about the possibility of AI.
1:33:42 What you're not going to get a lot of
1:33:43 from me is is the technology itself good
1:33:46 or bad? Is it ethical or not? Was it
1:33:51 trained illegally or not? Those are all
1:33:54 really important debates,
1:33:57 but that's not what this channel is
1:33:58 about. What this channel is about is
1:34:01 this is a profoundly powerful technology
1:34:04 that's not going away
1:34:08 and it's getting better dramatically
1:34:11 fast.
1:34:13 And if it's not going away and if it's
1:34:15 evolving incredibly quickly, then you
1:34:18 only have two choices.
1:34:20 You can learn about it or not. And
1:34:23 there's a lot of people right now
1:34:25 choosing to not learn about it. It takes
1:34:28 too much energy. Eh, it hallucinates.
1:34:30 It's stealing from artists. Eh, and
1:34:33 they're sitting on the sidelines with
1:34:34 their arms crossed.
1:34:39 My deep deep request is let's get as
1:34:42 many people as possible who are sitting
1:34:44 on the sidelines to at least understand
1:34:47 what it is and what it makes possible.
1:34:50 Because what it is when used well is
1:34:55 it's an amplifier of you.
1:35:00 It's really good at taking your ideas
1:35:03 and getting them out of your head and
1:35:05 into the world quickly and more
1:35:07 powerfully and then expanding those
1:35:10 ideas and then amplifying those ideas
1:35:12 and then spreading those ideas to
1:35:14 different outputs.
1:35:16 It is incredibly empowering. It is
1:35:19 incredibly inspiring.
1:35:22 But you literally cannot
1:35:26 understand that
1:35:28 without using it.
1:35:32 So here's the good thing. You could come
1:35:35 next week, start the week hating AI,
1:35:38 learn all about it, and then at the end
1:35:40 of the week still hate AI and never use
1:35:43 it again. Cool.
1:35:47 But if you're just on the sidelines
1:35:48 assuming you know what it is, assuming
1:35:51 you understand its motivations, it
1:35:54 doesn't have motivations. It's just a
1:35:56 reflector. It's just reflecting you.
1:36:00 Whatever you put in it, it reflects back
1:36:02 at you in this amplified, magnified way.
1:36:08 So my request is tell people about this.
1:36:11 Let's get some people here. Okay?
1:36:16 Um,
1:36:20 so that's my second request. Here's my
1:36:22 third request. My third request
1:36:25 is join the AI salon. So the AI salon is
1:36:29 a place where we can continue the
1:36:30 conversation after these nightly things.
1:36:33 We can share things. We can talk to one
1:36:35 another. But you can also meet people.
1:36:39 Connecting with other people in the face
1:36:42 of all this change is probably the
1:36:44 single most important thing you can do.
1:36:47 Yeah, you got to learn AI tools. Yeah,
1:36:49 you got to get fluent. Yeah, you got to
1:36:51 become ready for AI. All that sort of
1:36:53 [ __ ]
1:36:56 But it's connecting with other people
1:36:58 who are thinking about this stuff
1:36:59 thoughtfully, creatively, ethically
1:37:06 that will open a whole new world for
1:37:07 you. So, if you go to
1:37:09 community.thesalon.ai
1:37:12 or or do the little uh QR code there,
1:37:17 go join. All right.
1:37:20 Beautiful. Beautiful people.
1:37:24 Um, shout out to Gwen uh Shafitz for um
1:37:29 for co-hosting the AI readiness project
1:37:32 with me me today. Great job, Gwen. And I
1:37:34 don't know if Mr. Kay is here, but Mr. K
1:37:36 was our our special guest today and he
1:37:38 did great. It was a it was a really good
1:37:40 a good episode. So if you want to see
1:37:42 that you can go to um airelinesspro.com
1:37:48 and you can see the um
1:37:52 the recording of it. It was really good.
1:37:54 So anyway, so thank you Gwyn. All right,
1:37:58 I'm going to get on out of here. Uh I
1:37:59 hope you had fun tonight watching Theo
1:38:01 Vaughn.
1:38:04 I think this is the last 40 years where
1:38:06 we'll have babies in human bodies. Sam
1:38:09 Alman, huh?
1:38:15 Uh Danielle wants to know if I played
1:38:17 with Pika Social yet. I played with it a
1:38:19 little bit last night. Mostly just
1:38:21 flicking through other people's stuff. I
1:38:23 I haven't uh made my own things yet. So,
1:38:26 I have it. It's installed. Code code
1:38:29 worked. Thank you. Uh, and then uh we
1:38:32 will uh I I'll play with it and then
1:38:35 I'll I'll form some opinion on it
1:38:37 sometime this week. So anyway,
1:38:41 all right everybody, uh hope you had a
1:38:43 good time and I will talk to you uh
1:38:46 tomorrow. Tomorrow is Thursday. Yeah,
1:38:48 today's Wednesday. All right, peace out
1:38:49 everyone.
1:38:51 Bye-bye.