Latest / Elon Musk Thinking / Elon Musk's Brutally Honest Interview!!!

Transcript

Elon Musk's Brutally Honest Interview!!! · Elon Musk Thinking

0:00
Speaker 1: Elon, Welcome to AI startup School. We're just really really
0:04
Speaker 1: blessed to have your presence here today.
0:05
Speaker 2: Thanks for having me So.
0:08
Speaker 1: From SpaceX, Tesla, Neuralink, XAI, and more. Was there ever
0:12
Speaker 1: a moment in your life before all this where you
0:15
Speaker 1: felt I have to build something great? And what flipped
0:19
Speaker 1: that switch for you? Well?
0:20
Speaker 3: I didn't originally think I would build something great. I
0:23
Speaker 3: wanted to try to build something useful, but I didn't
0:26
Speaker 3: think I would build anything particularly great. You've said, probabilistically
0:29
Speaker 3: seemed unlikely, but I wouldn't to at least try.
0:32
Speaker 1: So you're talking to a room full of people who
0:34
Speaker 1: are all technical engineers, often some of the most eminent
0:38
Speaker 1: AI researchers coming up in the game.
0:40
Speaker 2: Okay, I I think we should.
0:44
Speaker 3: I think I'd like the term engineer better than researcher.
0:47
Speaker 3: I mean, I suppose if there's some fundamental algorithmic breakthrough,
0:51
Speaker 3: it's a research. Otherwise it's engineering.
0:54
Speaker 1: Maybe let's start way back, I mean, when you were
0:56
Speaker 1: this is a room full of eighteen to twenty five
0:59
Speaker 1: year old excuse younger, because the founder set is younger
1:03
Speaker 1: and younger. Can you put yourself back into their shoe?
1:07
Speaker 1: When you know, you were eighteen nineteen, you know, learning
1:10
Speaker 1: the code, even coming up with this first idea for
1:13
Speaker 1: zip too. What was that like for you?
1:15
Speaker 2: Yeah?
1:16
Speaker 3: Back in ninety five, I was faced with a choice
1:18
Speaker 3: of either do you know Grass Studies PhD a Stanford
1:22
Speaker 3: in material science actually working on ultra capacities for potential
1:27
Speaker 3: use in electric vehicles, essentially trying to solve the range
1:29
Speaker 3: problem for electric vehicles, or try to do something in
1:33
Speaker 3: this thing that most people have never heard of called
1:34
Speaker 3: the Internet. And I talked to my professor was Bill
1:38
Speaker 3: Nicks in the material science problement, and said, like, can
1:41
Speaker 3: I defer for a quarter because this will probably fail
1:45
Speaker 3: and then I'll need to come back to college. And
1:48
Speaker 3: then he said this is probably the last conversation we'll have,
1:50
Speaker 3: and he was right. So but I thought things would
1:53
Speaker 3: most likely fail, not that they would most likely succeed.
1:56
Speaker 3: And then in ninety five I wrote basically, I think
2:01
Speaker 3: the first or close to the first maps directions, Internet,
2:05
Speaker 3: white pages and yellow pages on the Internet. I just
2:08
Speaker 3: wrote that. I just wrote that personally, and I didn't
2:10
Speaker 3: even use a website. I just read the port directly
2:11
Speaker 3: because I couldn't afford h and I could have afforded
2:15
Speaker 3: t one. Original office was on Sherman Avenue in pale Alto.
2:20
Speaker 3: There was like an I s P on the floor below,
2:23
Speaker 3: so I rolled a hole through the floor and just
2:26
Speaker 3: ran a land cable directly to the I s P.
2:29
Speaker 3: And you know, uh, my brother joined me and another
2:32
Speaker 3: co founder of Great Courier who passed away, and we
2:36
Speaker 3: at the time we couldn't even afford a place to stay,
2:38
Speaker 3: so we just the office was five hundred bucks a months,
2:41
Speaker 3: so we just left in the office and and then
2:42
Speaker 3: shouted the YMC A on page mill and al communo
2:45
Speaker 3: and uh yeah, and we I guess we ended up
2:47
Speaker 3: doing a little bit of a useful company as up
2:51
Speaker 3: to in the beginning, and we did built a lot
2:55
Speaker 3: of really really good software technology, but we were somewhat
3:00
Speaker 3: captured by the legacy media companies, and that night Router
3:04
Speaker 3: New York Times hearst whatnot were investors and customers and
3:09
Speaker 3: and also on the board. So they they kept wanting
3:12
Speaker 3: to use our software in ways that made no sense.
3:14
Speaker 3: So I wanted to go direct to consumers anyway. As
3:17
Speaker 3: long story dwelling too much up too, but I really
3:20
Speaker 3: just wanted to do something useful on the Internet because
3:23
Speaker 3: I had two choices, like do a PhD and watch
3:26
Speaker 3: people build the internet or help build the Internet in
3:29
Speaker 3: some small way. And I was like, well, I guess
3:30
Speaker 3: I can always try and fail and then go back
3:32
Speaker 3: to grad studies. And that ended up being like reasonably successful.
3:37
Speaker 3: Sold for like three hundred million dollars, which is a
3:39
Speaker 3: lot at the time these days. That's like I think
3:41
Speaker 3: minimum impulse, but for an AI startup is like a
3:43
Speaker 3: billion dollars. It's like there's so many freaking unicorns. It's
3:46
Speaker 3: like heard of unicorns, point of unicorns, a billion dollars situation.
3:51
Speaker 1: There's been inflation since, so quite a bit more money.
3:54
Speaker 3: Yeah, I mean, like you could probably buy a burger
3:57
Speaker 3: for a nickel. Well not quite, but I mean, yeah,
4:00
Speaker 3: has been a lot of inflation. But I mean the
4:03
Speaker 3: high level of AI is is pretty intense. As you've seen,
4:07
Speaker 3: you know, you see companies that are I don't know,
4:10
Speaker 3: less than a year old getting sometimes billion dollar, multi
4:13
Speaker 3: billion dollar valuations, which I guess could could pan out
4:17
Speaker 3: and probably will pan out in some cases, but it
4:21
Speaker 3: is I ordering to see some of these valuations.
4:23
Speaker 2: Yeah, what do you think.
4:24
Speaker 1: I mean, Well, I'm pretty bullish. I'm pretty bullish, honestly, so.
4:29
Speaker 1: I think the people in this room are going to
4:31
Speaker 1: create a lot of the value that you know, a
4:34
Speaker 1: billion people in the world should be using this stuff
4:36
Speaker 1: and we're not even we're scratching the surface of it.
4:39
Speaker 1: I love the Internet story in that, even back then,
4:43
Speaker 1: you know, you are a lot like the people in
4:45
Speaker 1: this room back then, in that, you know, the heads
4:48
Speaker 1: of all the CEOs of all the legacy media companies
4:51
Speaker 1: look to you as the person who understood the Internet,
4:54
Speaker 1: and a lot of the world, you know, the corporate world,
4:57
Speaker 1: like the world out large that does not understand what's
5:00
Speaker 1: happening with AI, they're going to look to the people
5:02
Speaker 1: in this room for exactly that. It sounds like, you know,
5:04
Speaker 1: what are some of the tangible lessons. It sounds like
5:06
Speaker 1: one of them is don't give up board control or
5:09
Speaker 1: be careful about have a really good lawyer.
5:12
Speaker 3: I guess with the first my first start up, the
5:14
Speaker 3: big really the mistake was having too much shareholder and
5:18
Speaker 3: board control from legacy media companies who then necessarily see
5:23
Speaker 3: things through the lens of legacy media, and that they'll
5:27
Speaker 3: kind of make you do things that seem sensible to them,
5:29
Speaker 3: but really don't make sense with the new technology. I
5:33
Speaker 3: know I should point out that I that I didn't
5:36
Speaker 3: actually at first intend to start a company. I try
5:39
Speaker 3: to get a job at Netscape SMI resumes Netscape and
5:43
Speaker 3: Mark Hendrieson knows about this, but I don't think you
5:46
Speaker 3: have a swim my resume, and then nobody responded. So
5:49
Speaker 3: and then I tried hanging out in the lobby of
5:51
Speaker 3: Netscape see if I could like bump into someone, but
5:53
Speaker 3: I was like too shy to talk and talk to anyone.
5:55
Speaker 3: So I'm like, man, this is ridiculous. So I'll just
5:57
Speaker 3: write sof for myself and see how it goes. So
5:59
Speaker 3: it wasn't actually from the stand point of like I
6:00
Speaker 3: want to start a company. I just want to be
6:02
Speaker 3: part of building you know, the Internet in some way.
6:05
Speaker 3: And and since I couldn't get a job at an
6:07
Speaker 3: ARAC company, I had to start a interac company anyway.
6:10
Speaker 3: The yeah, yeah, I mean from an AI will so
6:12
Speaker 3: profoundly change the future. It's difficult to fathom how much.
6:16
Speaker 3: But you know, the the economy is, assuming we don't
6:20
Speaker 3: things don't go awry, and like AI doesn't kill us
6:23
Speaker 3: all in itself, then you'll see ultimately an economy that
6:28
Speaker 3: is not not ten times more than the current economy. Ultimately,
6:32
Speaker 3: like if we become say or whatever, our future machine
6:34
Speaker 3: descendants or but mostly machine descendants become like a cotttership
6:39
Speaker 3: scale to civilization or beyond, we're talking about an economy
6:43
Speaker 3: that is thousands of times maybe millions of times bigger
6:47
Speaker 3: than the economy today.
6:48
Speaker 2: So yeah, I mean I.
6:51
Speaker 3: Did sort of feel but like, you know, when I
6:52
Speaker 3: was in DC taking a lot of flag for like
6:54
Speaker 3: getting rid of waste and fraud, which is an interesting
6:56
Speaker 3: side quest as side quest go.
6:59
Speaker 1: But but I've got to get back to the main quest. Yeah,
7:02
Speaker 1: I got to.
7:02
Speaker 2: Get back to the main quest here, So back to
7:05
Speaker 2: the main quest.
7:07
Speaker 3: So but I did feel, you know a little bit
7:09
Speaker 3: like there's you know, it's like fixing the government. It's
7:11
Speaker 3: kind of like there's like, say, the beach is dirty
7:14
Speaker 3: and there's like some needles and vcs and like trash,
7:17
Speaker 3: and you want to clean up the beach. But then
7:18
Speaker 3: there's also this like thousand foot wall of water, which
7:21
Speaker 3: is a tsunami of ai, like, and how much does
7:25
Speaker 3: cleaning the beach really matter if you got a thousand
7:27
Speaker 3: foot tsunami about to hit not that much.
7:30
Speaker 1: Well, we're glad you're back on the main quest. It's
7:32
Speaker 1: very important.
7:33
Speaker 3: Yeah, back to the main quest, building technology, which is
7:36
Speaker 3: what I like doing. It's just so much noise like this,
7:39
Speaker 3: The single noise ratio in politics is terrible.
7:42
Speaker 1: So, I mean, I live in San Francisco, so you
7:45
Speaker 1: don't need to tell me twice.
7:46
Speaker 3: Yeah, DC is like, you know, kind of I guess
7:48
Speaker 3: it's all politics in DC. But the if you're trying
7:52
Speaker 3: to build a rocket or cars, or you're trying to
7:56
Speaker 3: have software that compiles and runs reliably, then you have
7:59
Speaker 3: to be maximally true seeking or your software or your
8:02
Speaker 3: hardware won't work like you can't fool like math and
8:06
Speaker 3: physics are rigorous judges. So I'm used to being in
8:09
Speaker 3: a maximally true seeking environment and that's definitely not politic.
8:14
Speaker 3: So an I'm good glad to be back in technology.
8:17
Speaker 1: I guess I'm kind of curious going back to the
8:18
Speaker 1: ZIP too moment, you had hundreds of millions of dollars
8:21
Speaker 1: or you had an exit worths of millions of dollars.
8:24
Speaker 2: I got twenty million dollars.
8:25
Speaker 1: Right, Okay, so you solved the money problem at least,
8:29
Speaker 1: and you basically took it and you rolled. You kept
8:32
Speaker 1: rolling with X dot Com, which is in PayPal and Confinity.
8:35
Speaker 2: Yes, I kept the chips on the table.
8:38
Speaker 1: Yeah, so not everyone does that. A lot of the
8:40
Speaker 1: people in this room will have to make that decision. Actually,
8:43
Speaker 1: what drove you to jump back into the ring.
8:45
Speaker 3: Well, I think I felt with with Zip too, we'd
8:48
Speaker 3: both like incredible technology but never really got used. You know,
8:51
Speaker 3: I think, at least from my perspective, we had better
8:54
Speaker 3: technology than say Yahoo or anyone else, but it was
8:57
Speaker 3: constrained by our customers. And so I wanted to do
9:00
Speaker 3: something that where okay, we wouldn't be constrained by our customers,
9:03
Speaker 3: go direct to consumer. And that's what ended up being
9:07
Speaker 3: like X dot com PayPal, essentially X dot Com merging
9:10
Speaker 3: with Confinity, which together created PayPal, and then that actually
9:15
Speaker 3: the sort of PayPal diaspora has it might have created
9:18
Speaker 3: more companies then, so more companies than probably any anything
9:22
Speaker 3: in the twenty first century. You know, so many talented
9:25
Speaker 3: people were at the combination of Confinity and X dot Com.
9:29
Speaker 3: So I just wanted to, like, I felt like we
9:32
Speaker 3: kind of got our wings clift somewhat with Zip two
9:35
Speaker 3: and it's like, okay, what if our wings on clipped
9:36
Speaker 3: and we go direct to consumer, And that's that's what.
9:39
Speaker 2: PayPal ended up being.
9:41
Speaker 3: But yeah, with I got that like twenty million dollar
9:45
Speaker 3: check for my share of ZIP two. At the time,
9:48
Speaker 3: I was living with in a house with four housemates
9:52
Speaker 3: and had like a ten grand in the bank, and
9:54
Speaker 3: then the check rights in the mail of all placesn't
9:58
Speaker 3: in the mail. And then now then my bank a
10:00
Speaker 3: bounce went from tenenty to twenty million in ten thousands.
10:04
Speaker 3: You're like, well, okay, sluck to pay taxes on that
10:06
Speaker 3: at all. But then I ended up putting almost all
10:09
Speaker 3: of that into x dot com and as you said,
10:13
Speaker 3: just kind of keeping almost all the chips on the table.
10:16
Speaker 2: And yeah, and then.
10:17
Speaker 3: After Paypala was like, well, I was kind of curious
10:20
Speaker 3: as to why we had not sent anyone to Mars,
10:22
Speaker 3: and I went on then went on the NASA website
10:25
Speaker 3: to find out when we're sending people Tomors, and there
10:27
Speaker 3: was no date. I thought, maybe it's just hard to
10:29
Speaker 3: find on the website, but in fact that there was
10:32
Speaker 3: no real plan to send people to Mars. So then
10:35
Speaker 3: you know, I've come to this is such a long story.
10:37
Speaker 3: So I don't want to take up too much time here.
10:39
Speaker 1: But I think we're all listening with rapt attention.
10:42
Speaker 3: So I was actually I was on the long end
10:44
Speaker 3: of the expressway with my friend day O RESSI were
10:47
Speaker 3: like housemates in college, and day was asking me what
10:50
Speaker 3: I'm what we're going to do? What am I going
10:51
Speaker 3: to do after a PayPal And I was like, like,
10:53
Speaker 3: I don't know. I guess maybe I'd like to do
10:55
Speaker 3: something philanthropic in space, because I didn't think I could
10:57
Speaker 3: actually do anything commotional in space because that seem like
11:00
Speaker 3: the purview of nations. So but you know, I'm kind
11:03
Speaker 3: of curious as to when we're going to send people
11:05
Speaker 3: to Mars. And that's when I was like, oh, it's
11:07
Speaker 3: not on the website, and I started digging on not
11:10
Speaker 3: that there's nothing on the NAST website.
11:11
Speaker 2: So then I started.
11:12
Speaker 3: Digging in and and and I'm definitely summarizing a lot here.
11:16
Speaker 3: But my first idea was to do a philanthropic mission
11:21
Speaker 3: to Mars called Life to Mars, where would send a
11:25
Speaker 3: small greenhouse with see some dehydrated nutrient gael land out
11:29
Speaker 3: on mosque and grow, you know, hydrate the jail and
11:32
Speaker 3: then you have this great sort of money shot of
11:35
Speaker 3: green plants on a red background. But the longest time, I,
11:38
Speaker 3: by the way, I didn't realize money shot, I think is.
11:39
Speaker 2: A porn reference.
11:40
Speaker 3: But anyway, the point is that that would be the
11:44
Speaker 3: great shot of green plants on a red background, and
11:46
Speaker 3: to try to inspire you know, NASA and the public to.
11:50
Speaker 2: Send astronauts to to Mars.
11:52
Speaker 3: As I learned more, it came to realize and along
11:55
Speaker 3: the way, by the way, I went to Russia in
11:58
Speaker 3: like two thousand and one and two thousand and two
12:00
Speaker 3: to buy ICVMS, which is like, that's an adventure, you know,
12:03
Speaker 3: you go and meet with Russian high command and say
12:05
Speaker 3: I'd like to buy some ICVMS.
12:07
Speaker 1: This was to get to space, yeah, as a rocket
12:09
Speaker 1: to not.
12:10
Speaker 2: To nuke anyone, but.
12:13
Speaker 3: They had to as a result of onstroduction talks, they
12:17
Speaker 3: had to actually destroy a bunch of their big nuclear missiles.
12:20
Speaker 2: So I was like, well, how about if we take two.
12:22
Speaker 3: Of those, you know, minus the nuke, added an additional
12:26
Speaker 3: upper stage for Mars. But it was kind of trippy,
12:29
Speaker 3: you know, being in Moscow in one of two thousand
12:32
Speaker 3: and one negotiating with like the Russian military to buy ICVM,
12:35
Speaker 3: like that's crazy, but they kept also raising the price
12:42
Speaker 3: on me, so that so like literally it's kind of
12:44
Speaker 3: like the opposite of what negotiation should should do. So
12:47
Speaker 3: I was like, man, these things are getting really expensive.
12:49
Speaker 3: And then I came to realize that actually the problem
12:51
Speaker 3: was not that there was insufficient world to go to Mars,
12:53
Speaker 3: but there was no way to do so without breaking
12:57
Speaker 3: the budget, you know, or even breaking the NASA budget.
12:59
Speaker 3: So that's where I decided to start SAX SpaceX to
13:03
Speaker 3: advance rocket technology to the point where we could send
13:05
Speaker 3: people to Mars.
13:07
Speaker 2: And that was in two thousand and two.
13:09
Speaker 1: So that wasn't you know. You didn't start out wanting
13:13
Speaker 1: to start a business. You wanted to start just something
13:16
Speaker 1: that was interesting to you that you thought humanity needed.
13:19
Speaker 1: And then as you sort of you know, like a
13:22
Speaker 1: cat pulling on you know, a string, it just sort
13:25
Speaker 1: of the ball sort of unravels. It turns out this
13:27
Speaker 1: is could be a very profitable business.
13:30
Speaker 3: I mean it is now, But there had been no
13:34
Speaker 3: prior example of really a rocket startup seating. There have
13:37
Speaker 3: been various attempts to do commercial rocket companies and that
13:41
Speaker 3: all failed. So again with SpaceX, starting SpaceX was really
13:45
Speaker 3: from the standpoint of, like I think there's like a
13:48
Speaker 3: less than ten percent chance of being successful, maybe one percent,
13:51
Speaker 3: I don't know. But if a startup doesn't do something
13:56
Speaker 3: to advance rocket technology, it's definitely not coming from the
13:59
Speaker 3: big defense contractors because they just impedance smashed the government,
14:03
Speaker 3: and the government just wants to do very conventional things.
14:06
Speaker 3: So there's it's either coming from a startup or it's
14:08
Speaker 3: not happening at all. So so like a small chance
14:11
Speaker 3: of success is better than no chance of success, and
14:14
Speaker 3: so that yeah, so SpaceX started that in mid mid
14:18
Speaker 3: two thousand and two expecting to fail, Like I said,
14:20
Speaker 3: probably ninety percent chance of failing. And even like when
14:23
Speaker 3: recruiting people, I didn't like to try to make out
14:26
Speaker 3: that it would probably I said, we're probably going to die,
14:28
Speaker 3: but four chance we might not die. But this is
14:32
Speaker 3: the only way to get people to Mars and advanced
14:34
Speaker 3: state of the art. And then I ended up being
14:38
Speaker 3: chief engineer of the rocket, not because I wanted to,
14:40
Speaker 3: but because I couldn't hire anyone.
14:42
Speaker 2: Who was good.
14:43
Speaker 3: So like none of the good sort of chief engineers
14:45
Speaker 3: would join because it's like this is too risky you
14:47
Speaker 3: were going to die. And so then I ended up
14:50
Speaker 3: being chief engineer of the rocket. And you know, the
14:52
Speaker 3: first three flights did fail, so it's a bit of
14:54
Speaker 3: a learning exercise there, and fourth one fortunately worked. But
14:58
Speaker 3: if the fourth one hadn't worked, I had no money
15:00
Speaker 3: left and that would have been it would have been curtains.
15:03
Speaker 2: So it was a pretty close thing.
15:04
Speaker 3: If the fourth launch of Falcon Nott work, it would
15:07
Speaker 3: have been just curtains and we would have just been
15:09
Speaker 3: joined the graveyard of prior rocket startup. So it's like
15:13
Speaker 3: my estimate of success was not far off. We'd just
15:16
Speaker 3: be made it by the skin of our teeth. And
15:19
Speaker 3: Tesla was happening sort of simultaneously. Like two thousand and
15:22
Speaker 3: eight was a rough year because at mid two thousand
15:26
Speaker 3: and eight we're called summer two thousand and eight, the
15:29
Speaker 3: third the third launch of SpaceX had failed, a third
15:32
Speaker 3: failure in a row. The Tesla financing round had failed,
15:35
Speaker 3: and so Telsa was going bankrupt fast. It was just like, man,
15:40
Speaker 3: this is a grim this is this is going to
15:42
Speaker 3: be a tale of warning of an exercise in hubers.
15:47
Speaker 1: Probably throughout that period, a lot of people were saying,
15:49
Speaker 1: you know, Elon is a software guy.
15:51
Speaker 2: Why are you.
15:52
Speaker 1: Working on hardware? Why would you? Yeah, why would he
15:55
Speaker 1: choose to work on this?
15:56
Speaker 3: Right, Harp said, So you can look at the like
15:58
Speaker 3: the because it's still the you know, the press of
16:00
Speaker 3: that time is still online. You could just search it.
16:02
Speaker 3: And and they kept calling the Internet guy. So like
16:06
Speaker 3: Internet guy aka fool is attempting to build a rocket company.
16:12
Speaker 3: So you know that we got ridiculed quite a lot.
16:17
Speaker 3: And it does sound pretty absurd, Like internet guy starts
16:20
Speaker 3: rocket company doesn't sound like a recipe for sess frankly,
16:23
Speaker 3: so I didn't hold against them. I was like, yeah,
16:25
Speaker 3: you know, admittedly it does sound improbable, and I agree
16:28
Speaker 3: that it's improbable. But fortunately the fourth launch worked and
16:32
Speaker 3: uh and, and NASA warded us a contract to resupply
16:36
Speaker 3: the space station uh and. I think that was like
16:39
Speaker 3: maybe December twenty second. It was like right before Christmas.
16:43
Speaker 3: Because even the fourth launch working wasn't enough to succeed.
16:47
Speaker 3: NASA also needed we also needed a big contract to
16:49
Speaker 3: keep us a life. So so I got I got
16:51
Speaker 3: that call from like the NASA NASA team and I literally,
16:57
Speaker 3: we're rewarding you one of the contracts to re supply
16:59
Speaker 3: the space station, like literally blurt it out, I love
17:01
Speaker 3: you guys, which is not normally you know what they hear.
17:05
Speaker 3: It's usually pretty, you know, sober. But I was like, man,
17:08
Speaker 3: this is a company saver. And then we closed the
17:12
Speaker 3: Tesla financing around on the last hour of the last
17:15
Speaker 3: day that it was possible, which was six pm December
17:17
Speaker 3: twenty fourth, two thousand and eight. We would have bounced
17:19
Speaker 3: payroll two days after Christmas if that round had had closed.
17:22
Speaker 3: So that was a nerve wracking end of two thousand
17:24
Speaker 3: and eight, that's for sure.
17:25
Speaker 1: I guess from your PayPal and zip two experience jumping
17:28
Speaker 1: into these hardcore hardware startups, it feels like one of
17:32
Speaker 1: the through lines was being able to find and eventually
17:35
Speaker 1: attract the smartest possible people in those popular fields. You know,
17:39
Speaker 1: what would I mean the people in this room, like
17:41
Speaker 1: some of most of the people here I don't think
17:42
Speaker 1: have even managed the single person yet, they're just starting
17:45
Speaker 1: their careers. What would you tell to, you know, the
17:47
Speaker 1: elon who's never had to do that yet.
17:51
Speaker 3: I generally think to try to try to be as
17:52
Speaker 3: useful as possible. So it may sound trit but it's
17:55
Speaker 3: it's so hard to be useful, especially to be useful
17:58
Speaker 3: to a lot of people, Where you say the area
18:00
Speaker 3: under the cove of total utility is like how much
18:03
Speaker 3: how useful have you been to your failer human beings?
18:04
Speaker 2: Times?
18:05
Speaker 3: How many people? Like it's almost like like the physics
18:08
Speaker 3: definition of true work. It's incredibly difficult to do that.
18:11
Speaker 3: I think if you aspire to do true work, your
18:14
Speaker 3: probably success is much higher, Like like, don't inspire to glory,
18:18
Speaker 3: aspire to work.
18:19
Speaker 1: How can you tell that it's true work? Like is
18:21
Speaker 1: it external? Is it like what happens with other people?
18:24
Speaker 1: Or you know what the product does for people? Like
18:26
Speaker 1: what you know? What is that for you when you're
18:28
Speaker 1: looking for people to come work for you? Like what
18:30
Speaker 1: you know? What's the salient thing that you look for?
18:32
Speaker 3: Or if they're that's certain question I guess it's I
18:35
Speaker 3: mean in terms of off your in product, you have
18:37
Speaker 3: to say, like, well, if this thing is successful, how
18:39
Speaker 3: useful will it be to how many people?
18:41
Speaker 2: And that's that's what I mean.
18:43
Speaker 3: And then you do you do whatever you know, whether
18:46
Speaker 3: you're CEO or any role in a startup, you do
18:48
Speaker 3: whatever it takes to succeed, like and just and just
18:51
Speaker 3: always be smash smashing your ego like like internalized responsibility.
18:56
Speaker 3: Like a major failure mode is when ego to ability
18:59
Speaker 3: ratio is double graded than sign one. You know, like
19:03
Speaker 3: if you if your ego to ability ratio is it
19:06
Speaker 3: gets too high, then you're you're you're you're going to
19:08
Speaker 3: basically break the feedback loop to reality. And in in
19:11
Speaker 3: AI terms, your your er, you'll have your you'll break
19:15
Speaker 3: your r L loop.
19:16
Speaker 2: So you want you want to don't want to break
19:17
Speaker 2: your art.
19:17
Speaker 3: You want to have a strong r L loop, which
19:19
Speaker 3: means internalizing responsibility and minimizing the ego, and you do
19:22
Speaker 3: whatever the task is, no matter whether it's you know,
19:24
Speaker 3: grand or humble. So I mean that's kind of like
19:27
Speaker 3: why I actually I prefer the term like engineering as
19:30
Speaker 3: opposed to research. I prefer the term And I don't
19:33
Speaker 3: actually don't want it to call x AI lab. I
19:35
Speaker 3: just want to be a company. Like it's like whatever,
19:39
Speaker 3: the what of the simplest, most straightforward, uh ideally lowest
19:43
Speaker 3: ego terms are that those are generally a good way
19:46
Speaker 3: to go. You want to just close the loop on
19:48
Speaker 3: reality hard That's that's a that's super big deal.
19:51
Speaker 1: I think everyone in this room is really looks up
19:53
Speaker 1: to everything you've done around being sort of a paragon
19:57
Speaker 1: of first principles and you know, thinking about the stuff
20:00
Speaker 1: if you've done, how do you actually determine your reality?
20:04
Speaker 1: Because that seems like a pretty big part of it.
20:06
Speaker 1: Like other people people who have never made anything, non engineers,
20:11
Speaker 1: sometimes journalists at time who've never done anything, like they
20:14
Speaker 1: will criticize you, But then clearly you have another set
20:17
Speaker 1: of people who are builders, who have very high you know,
20:21
Speaker 1: sort of area under the curve, who are in your circle. Like,
20:24
Speaker 1: you know, how should people approach that? Like what has
20:27
Speaker 1: worked for you? And what would you pass on like
20:29
Speaker 1: you know, to X to your children? Like you know,
20:32
Speaker 1: what do you tell them when you're like you need
20:34
Speaker 1: to make your way in this world?
20:35
Speaker 2: Here?
20:36
Speaker 1: You know, here's how to construct a reality that is
20:38
Speaker 1: predictive from first principles.
20:39
Speaker 3: Well, the tools of physics are incredibly helpful to understand
20:45
Speaker 3: and make progress in any field. And first principles means
20:48
Speaker 3: just obviously just means, you know, breakings down to the
20:50
Speaker 3: fundamental axiomatic elements that are most likely to be true
20:53
Speaker 3: and then reason up from there as cosionally as possible,
20:56
Speaker 3: as opposed to reasoning by analysis or metaphor, and then
20:59
Speaker 3: it just simple things like like thinking in the limit,
21:02
Speaker 3: like if you strapolate, you know, minimize this thing or
21:05
Speaker 3: maximize that thing. Thinking in the limit is is very
21:07
Speaker 3: very helpful. I use all the tools of physics. They
21:11
Speaker 3: apply to any field. This is like a superpower.
21:13
Speaker 2: Actually.
21:14
Speaker 3: So you can take say, take for example, like rockets.
21:17
Speaker 3: You could say, well, how much should a rocket rocket cost?
21:20
Speaker 3: A typical approach to how to that people would take
21:23
Speaker 3: how much rocket should cost? There is they would look
21:25
Speaker 3: historically at what the cost of rockets are and assume
21:28
Speaker 3: that any new rocket must be somewhat similar to the
21:30
Speaker 3: prior cost of rockets. A first principal's approach would be
21:32
Speaker 3: you look at the materials that the rocket is comprised of,
21:35
Speaker 3: so if that's aluminum, copper, carbon, fiber, steel, whatever the
21:40
Speaker 3: case may be, and say what what how much does
21:43
Speaker 3: that rocket weigh, and and what are the constituent elements
21:45
Speaker 3: and how much does they weigh? What is the material
21:48
Speaker 3: price per kilogram of those constituent elements, In that sense
21:51
Speaker 3: the actual floor on what a rocket can cost. It
21:54
Speaker 3: can asymptotically approach costs the raw materials and then you realize, oh,
21:58
Speaker 3: actually a rocket. The ROM turials of a rocket are
22:01
Speaker 3: only maybe one or two percent of the historical cost
22:04
Speaker 3: of a rocket, so the manufacturing must necessarily be very inefficient.
22:09
Speaker 3: If the if the rom heerial cost is only one
22:12
Speaker 3: or two percent, that would be a first first principles
22:14
Speaker 3: analysis of the potential for cost optimization of a rocket.
22:18
Speaker 2: And that's before you get to reusability.
22:20
Speaker 3: To give an AI sort of AI example, I guess
22:23
Speaker 3: last year went for XCI when we were trying to
22:26
Speaker 3: build a training supercluster. We went to the various suppliers
22:31
Speaker 3: to ask what they said, this is the beginning of
22:34
Speaker 3: last year, that we needed one hundred thousand, h one
22:36
Speaker 3: hundreds to be able to train coherently. And there are
22:40
Speaker 3: estimates for how long it would take to complete that
22:42
Speaker 3: were eighteen to twenty four months. It's like, well, we
22:44
Speaker 3: need to get that done in six months so then
22:47
Speaker 3: or we won't be competitive. So then if you break
22:50
Speaker 3: that down, what are the things you need? Will you
22:52
Speaker 3: need a building, you need power, you need cooling. We
22:56
Speaker 3: didn't have enough time to build a building from scratch,
22:57
Speaker 3: so we've had to find existing building. So we found
23:00
Speaker 3: at a factory that was no longer in use in Memphis.
23:03
Speaker 2: That used to build electrolytes products.
23:06
Speaker 3: But then the input power was fifteen megawats and we
23:08
Speaker 3: needed one hundred and fifteen megawatts, So we read the
23:12
Speaker 3: generators and had generators on one side of the building.
23:15
Speaker 3: And then we have to have cooling, so we rented
23:16
Speaker 3: about a quarter of the mobile cooling capacity of the
23:18
Speaker 3: US and put the chillers.
23:20
Speaker 2: On the other side of the building.
23:21
Speaker 3: That didn't fully solve the problem because the voltage the
23:23
Speaker 3: power variations during training are very very big, so you
23:28
Speaker 3: can have power can drop by fifty percent in one
23:31
Speaker 3: hundred milliseconds, which the generators can keep up with.
23:33
Speaker 2: So then we.
23:34
Speaker 3: Combined we added Tesla megapacks and modified the software in
23:37
Speaker 3: the megapacks to be able to smooth out the power
23:40
Speaker 3: variation during the training run. And then there were a
23:43
Speaker 3: bunch of networking challenges because the networking cables, if you're
23:47
Speaker 3: trying to make one hundred thousand GPUs training coherently, are
23:51
Speaker 3: very very challenging.
23:52
Speaker 1: Almost it sounds like almost any of those things you mentioned,
23:56
Speaker 1: I could imagine someone telling you very directly, no, you
23:59
Speaker 1: can have that, you can't have that power, you can't
24:02
Speaker 1: have this uh, and it sounds like one of the
24:04
Speaker 1: salient pieces of first principles thinking is actually, let's ask why,
24:08
Speaker 1: let's you know, figure that out, and actually let's challenge
24:11
Speaker 1: the person across the table and if they if I
24:14
Speaker 1: don't get an answer that I feel good about, I'm gonna,
24:16
Speaker 1: you know, not allow that to be I'm not going
24:19
Speaker 1: to let that know to stand. Is that? I mean,
24:21
Speaker 1: that feels like something that you know, everyone if someone
24:23
Speaker 1: were to try to do what you're doing in hardware,
24:26
Speaker 1: hardware seems too uniquely and software we have lots of
24:30
Speaker 1: you know, fluff and things that you know, it's like
24:31
Speaker 1: we can add more CPUs for that, it'll be fine.
24:34
Speaker 1: But in hardware it's it's just not going to work.
24:36
Speaker 3: I think these general principles of first principles thinking apply
24:39
Speaker 3: to software and hardware, apply to anything really. I'm just
24:42
Speaker 3: using kind of a hardware example of how we're told
24:46
Speaker 3: something is impossible, but once we broke it down into
24:48
Speaker 3: the constituent elements of we need a building, we need power,
24:51
Speaker 3: we need cooling, we need uh, we need power smoothing,
24:54
Speaker 3: and then and then we could solve those constituent elements.
24:58
Speaker 3: But and then we and then we rand the networking
25:01
Speaker 3: operation to do all the cabling everything in four shifts
25:05
Speaker 3: twenty four to seven, and I was like sleeping in
25:08
Speaker 3: the data center and also doing cabling myself, and there
25:11
Speaker 3: were a lot of both issues to solve. You know,
25:14
Speaker 3: nobody had done a training run with one hundred thousand,
25:17
Speaker 3: h one hundreds training coherently last year.
25:20
Speaker 2: I mean, maybe it's been done this year. I don't know.
25:22
Speaker 3: But then and then we ended up doubling that to
25:24
Speaker 3: two hundred thousand, and so now we've got one hundred
25:27
Speaker 3: and fifty thousand, h one hundred fifty k H two
25:29
Speaker 3: hundreds and thirty k GB two hundreds in the in
25:33
Speaker 3: the Memphis training center, and we're about to bring one
25:36
Speaker 3: hundred and ten thousand GB two hundreds online at a
25:40
Speaker 3: second data center also in the Memphis area.
25:42
Speaker 1: Is it your view that you know pre training is
25:44
Speaker 1: still working and you know larger the scaling laws still hold,
25:49
Speaker 1: and whoever wins this race will have basically the biggest,
25:52
Speaker 1: smartest possible model that you could steal.
25:55
Speaker 3: Well, there's other various elements that side competitiveness for larger
26:00
Speaker 3: I this this for sure, the talent of the people matter.
26:04
Speaker 3: The scale of the hardware matters, and how well you're
26:07
Speaker 3: able to bring that hardware to bear. So you can't
26:09
Speaker 3: just order a whole bunch of GPUs and if they don't,
26:12
Speaker 3: you can't just plug them in. So you've got to
26:13
Speaker 3: You've got to get a lot of GPUs and have
26:15
Speaker 3: them train training coherently and stavely. Then it's like, what
26:20
Speaker 3: unique access to data.
26:21
Speaker 2: Do you have?
26:21
Speaker 3: I guess distribution matters to some degree as well, like
26:24
Speaker 3: how do people get exposed to your AI?
26:26
Speaker 2: Those?
26:26
Speaker 3: Those are those are critical factors for if it's going
26:28
Speaker 3: to be like a large foundation model as competitive, you know,
26:32
Speaker 3: as as many have said.
26:34
Speaker 2: I think my friend eliasit Skier.
26:36
Speaker 3: Said, uh, you know, we we've kind of run out
26:38
Speaker 3: of pre training data of human generated pre like human
26:41
Speaker 3: generated data, you run out of tokens pretty fast, of
26:44
Speaker 3: certainly of high quality tokens. And and then you then
26:48
Speaker 3: you have to do a lot of you need to
26:51
Speaker 3: essentially create synthetic data and and be able to accurately
26:55
Speaker 3: judge the synthetic data that you're creating to very like
26:57
Speaker 3: is this real synthetic data?
26:59
Speaker 2: Well?
26:59
Speaker 3: Is it in a hallucination that doesn't actually mash reality?
27:03
Speaker 3: So achieving grounding and reality is tricky, but we're at
27:07
Speaker 3: the stage where there's more effort put into synthetic data.
27:11
Speaker 3: And right now we're training Grock three point five, which
27:14
Speaker 3: is a heavy focus on reasoning.
27:16
Speaker 1: Going back to your physics point, what I heard for
27:18
Speaker 1: reasoning is that hard science, particularly physics textbooks, are very
27:23
Speaker 1: useful for reasoning, whereas I think researchers have told me
27:27
Speaker 1: that social science is totally useless for reasoning.
27:31
Speaker 2: Yes, that's probably true.
27:33
Speaker 3: So yeah, there's something that's going to be very important
27:36
Speaker 3: in the future is combining deep AI in the data
27:41
Speaker 3: center of Supercluster with robotics. So that's you know, things
27:46
Speaker 3: like like the Optimius humanoid robot, and yeah, Optimist is awesome.
27:50
Speaker 3: There's going to be so many humanoid robots and robots
27:52
Speaker 3: of all robots of all sizes and shapes. But my
27:55
Speaker 3: prediction is that there will be more humanoid robots by
27:58
Speaker 3: far than all other robots can bind by maybe an
28:01
Speaker 3: order of magnitude, like a big difference.
28:03
Speaker 1: And is it true that you're planning a robot army of.
28:07
Speaker 3: A sort, whether we do it or or you know,
28:09
Speaker 3: whether Tesla does it. You know, Tesla works closely with Xai.
28:13
Speaker 3: Like you've seen how many humanoid robots startups are there.
28:16
Speaker 3: Like it's like I think Jensen Bong was on stage
28:19
Speaker 3: with a lot with a massive number of robots, you know,
28:22
Speaker 3: robots from different companies. I think it was like a
28:24
Speaker 3: dozen different humanoid robots. So I mean, I guess, you know,
28:28
Speaker 3: part of what I've been biting and maybe what has
28:30
Speaker 3: slowed me down somewhat is that I'm I'm a little
28:32
Speaker 3: I don't want I don't want to make Terminator real,
28:35
Speaker 3: you know. So I've been sort of, I guess, at
28:37
Speaker 3: least until recent years, dragging my feet on on AI
28:41
Speaker 3: and and humanoid robotics. And then I sort of come
28:44
Speaker 3: to realize that, realizations it's happening.
28:46
Speaker 2: Whether I do it or not.
28:47
Speaker 3: So you got really two choices participate. You could either
28:50
Speaker 3: be a spectator or a participant. It's like, well, I
28:53
Speaker 3: guess I'd rather be a participant than a spectator. So
28:56
Speaker 3: now it's you know, pedals to the metal on humanoid
28:58
Speaker 3: robots and digital superintelligence.
29:01
Speaker 1: So I guess you know, there's a third thing that
29:03
Speaker 1: everyone has heard you talk a lot about that I'm
29:05
Speaker 1: really a big fan of, you know, becoming a multiplanetary space.
29:08
Speaker 1: Where does this fit? You know? This is all you know,
29:10
Speaker 1: not not just a ten or twenty year thing, maybe
29:12
Speaker 1: one hundred year thing. Like it's a multi you know,
29:15
Speaker 1: many many generations for humanity kind of thing. You know,
29:18
Speaker 1: how do you think about it? There's you know AI obviously,
29:21
Speaker 1: there's embodied robotics, and then there's being a multiplant, multiplanetary species.
29:26
Speaker 1: Does everything sort of feed into that last point? Or
29:29
Speaker 1: you know what? What are you driven by right now
29:31
Speaker 1: for the next ten, twenty and one hundred years?
29:32
Speaker 3: See one hundred years, man, I hope civilizations around in
29:35
Speaker 3: one hundred years. If it is around, it's going to
29:37
Speaker 3: look very different from civilization today. I mean, I'd predict
29:41
Speaker 3: that this is going to be at least five times
29:43
Speaker 3: as many humanoid robots as there are humans, maybe ten times.
29:47
Speaker 3: One way to look at the progress of civilization is
29:50
Speaker 3: percentage completion Kladishev. So if you're in a cludos Ship
29:53
Speaker 3: scale one, you've you've hot as all the energy of
29:56
Speaker 3: a planet, and in my opinion, we've only ha maybe
30:00
Speaker 3: one or two percent of Earth's energy, So we've got
30:04
Speaker 3: a long way to go to the Kardschev scale one.
30:06
Speaker 3: Then Karshev two, you've harnessed all the energy of the Sun,
30:11
Speaker 3: which would be I don't know, a billion times more
30:13
Speaker 3: energy than Earth, maybe closer to a trillion, And then
30:16
Speaker 3: Karshev three would be all the energy of a galaxy,
30:19
Speaker 3: pretty far from that. So we're at the very very
30:22
Speaker 3: early stage of the intelligence big bang. I hope we're
30:26
Speaker 3: in terms of be multiplanetary, like I think. I think
30:28
Speaker 3: we'll have enough mass transferred to Mars within like roughly
30:32
Speaker 3: thirty years to make Mars self sustaining such that Mars
30:35
Speaker 3: can continue to grow and prosper even if the resupply
30:38
Speaker 3: shifts from Earth stuff coming, and that greatly increases the
30:42
Speaker 3: probable lifespan of civilization or consciousness and intelligence, both biological
30:47
Speaker 3: and digital. So that's why I think it's important to
30:50
Speaker 3: become a multiplanet species. And I'm so much troubled by
30:53
Speaker 3: the filmy paradox, like why have we not seen any aliens?
30:56
Speaker 3: And it could be because intelligence is incredibly rare and
31:00
Speaker 3: maybe we're the only ones in this galaxy, in which
31:03
Speaker 3: case the intelligence of consciousness is the tiny candle in
31:06
Speaker 3: a vast dogness, and we should do everything possible to
31:09
Speaker 3: ensure the tiny candle does not go out, and being
31:12
Speaker 3: a multiplanet species, or making consciousness multiplanetary, greatly improves the
31:17
Speaker 3: probable lifespan of civilization, and it's it's the next step
31:21
Speaker 3: before going to other star system on you. Once you
31:24
Speaker 3: at least have two planets, then you've got a forcing
31:26
Speaker 3: function for the improvement of space travel, and that ultimately
31:30
Speaker 3: is what will lead to consciousness expanding to the stars.
31:34
Speaker 1: It could be that the Fermi paradox dictates once you
31:37
Speaker 1: get to some level of technology, you destroy yourself. How
31:40
Speaker 1: do we see ourselves? How do we actually what would
31:43
Speaker 1: you prescribe to I mean a room full of engineers, like,
31:46
Speaker 1: what can we do to prevent that from happening?
31:49
Speaker 2: Yeah?
31:49
Speaker 3: How do we avoid the great filters? One of the
31:51
Speaker 3: great filters would obviously be global termin nuclear war, So
31:54
Speaker 3: we should.
31:55
Speaker 2: Try to avoid that.
31:56
Speaker 3: I guess building benign AI robots that that loves humanity
32:01
Speaker 3: and you know, robots.
32:03
Speaker 2: That are helpful.
32:05
Speaker 3: Something that I think is extremely important in building AI
32:08
Speaker 3: is a very rigorous adherence to truth, even if that
32:12
Speaker 3: truth is politically incorrect. My intuition for what could make
32:17
Speaker 3: AI very dangerous is if if you force AI to
32:20
Speaker 3: believe things that are not true.
32:22
Speaker 1: How do you think about you know, there's sort of
32:23
Speaker 1: this argument for open for safety versus closed for competitive edge.
32:28
Speaker 1: I mean, I think the great thing is you have
32:30
Speaker 1: a competitive model. Many other people also have competitive models,
32:34
Speaker 1: and in that sense, you know, we're sort of off of.
32:36
Speaker 1: Maybe the worst timeline that I'd be worried about is,
32:39
Speaker 1: you know, there's fast takeoff and it's only in one
32:41
Speaker 1: person's hands, so that might you know, sort of collapse
32:44
Speaker 1: a lot of things, whereas now we have choice, right,
32:47
Speaker 1: How do you think about this?
32:48
Speaker 2: Yeah?
32:48
Speaker 3: I do think there will be several deep intelligences, maybe
32:52
Speaker 3: maybe at least five, maybe as much as ten. I'm
32:56
Speaker 3: not sure that there's going to be hundreds, but it's
32:57
Speaker 3: probably close. Like maybe it'll be like ten like that,
33:01
Speaker 3: of which maybe four will be in the US. So
33:04
Speaker 3: I don't think it's going to be anyone AI that
33:06
Speaker 3: that has runaway capability.
33:09
Speaker 2: But but yeah, several deep intelligences.
33:11
Speaker 1: What will these deep deep intelligences actually be doing? Will
33:15
Speaker 1: it be scientific research or trying to hack each other?
33:17
Speaker 2: Probably all of the above.
33:19
Speaker 3: I mean, hopefully they will discover new physics, and I
33:22
Speaker 3: think they're very they're definitely going to invent new technologies.
33:26
Speaker 3: I think I think we're quite close to digital superintelligence.
33:29
Speaker 3: It may happen this year, and if it doesn't happen
33:32
Speaker 3: this year, next year for sure. Digital superintelligence defined as
33:35
Speaker 3: smarter than any human at anything.
33:38
Speaker 1: Well, so how do we direct that to sort of
33:39
Speaker 1: superabundance you know we have we could have robotic labor,
33:43
Speaker 1: we have cheap energy, intelligence on demand.
33:46
Speaker 2: You know?
33:47
Speaker 1: Is that sort of the white pill? Like where do
33:49
Speaker 1: you sit on the spectrum? And are there tangible things
33:52
Speaker 1: that you would encourage everyone here to be working on
33:55
Speaker 1: to make that white pill actually reality?
33:57
Speaker 2: I think I think it most likely will be a
34:00
Speaker 2: good outcome.
34:01
Speaker 3: I guess i'd sort of agree with Jeff Hinton that
34:03
Speaker 3: maybe it's a ten to twenty percent chance of annihilation,
34:06
Speaker 3: but look on the right side, that's eighty to ninety
34:08
Speaker 3: percent probably of a great outcome. So yeah, I can't
34:12
Speaker 3: emphasize this enough. A rigorous adherence to truth is the
34:15
Speaker 3: most important thing for AI safe safety, and obviously empathy
34:19
Speaker 3: for humanity and life as we know it.
34:22
Speaker 1: We haven't talked about neuralink at all yet, but I'm curious.
34:26
Speaker 1: You know you're working on closing the input and output
34:29
Speaker 1: gap between humans and gene How critical is that to
34:33
Speaker 1: agi ASI And you know, once that link is made,
34:37
Speaker 1: can we not only read but also write.
34:39
Speaker 3: The neuralink is not necessary to solve digital superintelligence. That'll
34:43
Speaker 3: happen before neuralink is at scale. But what neuralink can
34:48
Speaker 3: effectively do is solve the input output bandwidth constraints. Especially
34:52
Speaker 3: our output bandwidth is very low out The sustained output
34:57
Speaker 3: of a human over the course of a day is
35:00
Speaker 3: less than one but per second, So there's you know,
35:02
Speaker 3: eighty six four hundred seconds in a day, and it's
35:05
Speaker 3: extremely rare for a human to output more than that
35:07
Speaker 3: number of symbols per day, so certainly for several days
35:11
Speaker 3: in a row. So you really with with within neuralink interface,
35:15
Speaker 3: you can massively increase your output bandwidth and your input
35:18
Speaker 3: band with input.
35:20
Speaker 2: Being right to you, you have to do right operations
35:23
Speaker 2: to the brain.
35:24
Speaker 3: We have now five humans who have received the.
35:29
Speaker 2: Kind of the read input where it's reading.
35:32
Speaker 3: Signals, and you've got people with with als who really
35:36
Speaker 3: have their their tetraplegics, but they can now communicate at
35:40
Speaker 3: with similar band to a human with a fully functioning
35:43
Speaker 3: body and control their computer and phone which is pretty cool.
35:48
Speaker 3: And then I think in the next six to twelve
35:50
Speaker 3: months we'll be doing our first implants for vision where
35:54
Speaker 3: even if somebody is completely blind, could we can write
35:57
Speaker 3: directly to the visual cortex. And we've had that working
36:02
Speaker 3: in monkeys. Actually, I think one of our monkeys now
36:05
Speaker 3: has had a visual implant for three years. And at
36:09
Speaker 3: first it'll be relatively fairly low resolution, but long term
36:12
Speaker 3: you would have very high resolution and be able to
36:14
Speaker 3: see multi spectral wavelengths, so you could see an infrared
36:19
Speaker 3: ultra violet radar like a superpower situation. But like at
36:23
Speaker 3: some point the sabernetic implants would would not simply be
36:26
Speaker 3: correcting things that went wrong, but augmenting human capabilities dramatically,
36:31
Speaker 3: augmenting augmenting intelligence and senses and bandwidth dramatically. And that's
36:36
Speaker 3: that's going to happen at some point, but digital superintelligence
36:39
Speaker 3: will happen well before that. At least if we have
36:41
Speaker 3: a neural link, we'll be able to appreciate the AI better.
36:46
Speaker 1: I guess one of the limiting reagents to all of
36:49
Speaker 1: your efforts across all of these different domains is access
36:52
Speaker 1: to the smartest possible people. But you know, sort of
36:56
Speaker 1: simultaneous to that, we have you know, the rocks can
36:59
Speaker 1: talk and reason, and then you know they're maybe one
37:01
Speaker 1: hundred and thirty IQ now and they're probably going to
37:03
Speaker 1: be super intelligent soon. How do you reconcile those two things,
37:07
Speaker 1: like what's going to happen in you know, five ten
37:10
Speaker 1: years and what should the people in this room do
37:12
Speaker 1: to make sure that you know they're the ones who
37:14
Speaker 1: are creating instead of maybe below the API line.
37:17
Speaker 3: Well, they called it the singularity for a reason, Because
37:20
Speaker 3: we don't know what's going to happen in the not
37:22
Speaker 3: that far future. The percentage of intelligence that is human
37:25
Speaker 3: will be quite small. At some point, the collective some
37:28
Speaker 3: of human intelligence will be less than one percent of
37:31
Speaker 3: all intelligence. And if things get to a cult a
37:34
Speaker 3: ship level too, we're talking about human intelligence even assuming
37:39
Speaker 3: a significant increase in human population and intelligence augmentation, like
37:43
Speaker 3: massive intelligence augmentation where like everyone has an IQ of
37:46
Speaker 3: one thousand type of thing. Even in that circumstance, collective
37:50
Speaker 3: human intelligence will be probably one billion that of digital intelligence. Anyway,
37:55
Speaker 3: we're the biological bootbootloader for digital superintelligence.
37:58
Speaker 1: I guess just to end off.
38:01
Speaker 2: Was that it was like was I a good bootload.
38:03
Speaker 1: Where do we go? How do we go from here?
38:05
Speaker 1: I mean, I mean all of this is pretty wild
38:07
Speaker 1: sci fi stuff that also could be built by the
38:10
Speaker 1: people in this room. You know, if you do, you
38:12
Speaker 1: have a closing thought for the smartest technical people of
38:15
Speaker 1: this generation right now? What should they be doing? What
38:18
Speaker 1: should they what should they be working on? What should
38:19
Speaker 1: they be thinking about, you know, tonight as they go
38:22
Speaker 1: to dinner.
38:22
Speaker 3: Well, as I saw Rolfe with, I think if you're
38:25
Speaker 3: doing something useful, that's great. If you just just try
38:28
Speaker 3: to be as useful as possible to your fellow human
38:31
Speaker 3: beings and that that then you're doing something good. I
38:34
Speaker 3: keep harping on this, like focus on super truthful AI.
38:37
Speaker 3: That that's the most important thing for AI safety. You know, obviously,
38:40
Speaker 3: if you know anyone's interested in working at x AI mean,
38:44
Speaker 3: please please let us know. We're aiming to make rock
38:48
Speaker 3: the maximumly truth seeking AI and uh, I think that's
38:52
Speaker 3: a very important thing.
38:53
Speaker 2: Hopefully we can understand the nature of the universe.
38:55
Speaker 3: That that's really I guess what AI can hopefully tell
38:58
Speaker 3: us maybe AI AI and maybe tell us where are
39:00
Speaker 3: the aliens? And you know, how did the universe really start,
39:03
Speaker 3: how will it end? What are the questions that we
39:05
Speaker 3: don't know that we should ask, And are we in
39:08
Speaker 3: a simulation or what level of similation are we in?
39:11
Speaker 1: Well, I think we're going to find out.
39:13
Speaker 2: An MPC.
39:15
Speaker 1: Elon, thank you so much for joining us. Everyone, please
39:18
Speaker 1: give it up for Elon Musk Basically,