Latest / Core Memory / An AI Engineer Is Here, and It Might Redesign the Physical World - EP 20 Paul Eremenko

Transcript

An AI Engineer Is Here, and It Might Redesign the Physical World - EP 20 Paul Eremenko · Core Memory

0:00
SPEAKER_02: Is it Lee, like L-E-E?
0:03
SPEAKER_02: It's L-I.
0:04
SPEAKER_02: L-I.
0:04
SPEAKER_02: Where does... Long Island.
0:06
SPEAKER_02: Long Island.
0:07
SPEAKER_02: We adopted him in Long Island.
0:09
SPEAKER_02: Oh, okay.
0:09
SPEAKER_02: That was the only reason.
0:10
SPEAKER_02: No other affiliation to Long Island.
0:11
SPEAKER_02: And Lee goes with you everywhere.
0:13
SPEAKER_02: Lee goes with me everywhere.
0:14
SPEAKER_02: He was a shelter dog.
0:16
SPEAKER_02: And he had separation anxiety.
0:18
SPEAKER_02: There he is.
0:19
SPEAKER_02: And we got started working on his separation anxiety.
0:23
We adopted him like fall of 2020 and started working on his
0:28
separation anxiety.
0:29
SPEAKER_02: And then the pandemic started.
0:30
SPEAKER_00: Okay.
0:30
SPEAKER_02: And then we gave up working on his separation anxiety.
0:33
SPEAKER_02: And by the end of the pandemic, I had separation anxiety.
0:35
Well,
0:35
yeah,
0:35
I was going to say the pandemic was probably good for his
0:38
separation anxiety.
0:38
SPEAKER_02: Well, in the sense that we were never separate.
0:40
SPEAKER_00: Right.
0:40
SPEAKER_02: And so,
0:41
SPEAKER_02: so it was indulgent of his separation anxiety,
0:43
SPEAKER_02: but as a result,
0:44
SPEAKER_02: it never,
0:45
SPEAKER_02: it never cured and I developed it as well.
0:47
SPEAKER_02: So now we have mutual separation anxiety.
0:49
SPEAKER_02: When I go without him, I have like phantom dog syndrome.
0:53
SPEAKER_00: Well,
0:53
SPEAKER_00: so if people see Lee cruising around,
0:55
SPEAKER_00: he's hanging out with us,
0:58
SPEAKER_00: you,
0:58
SPEAKER_00: oh,
0:59
SPEAKER_00: we are like on your,
1:00
SPEAKER_00: well,
1:00
SPEAKER_00: okay.
1:00
SPEAKER_00: I've got two questions.
1:01
SPEAKER_00: I mean, one is, and I am going to do a intro here in a second, but the
1:07
SPEAKER_00: You fly.
1:08
SPEAKER_00: You flew in your own plane, right?
1:10
SPEAKER_00: I did not.
1:11
SPEAKER_00: No, no.
1:12
SPEAKER_02: I flew JSX, which is almost as good.
1:16
SPEAKER_00: I thought when we were emailing,
1:18
SPEAKER_00: it sounded like you were,
1:19
SPEAKER_00: I didn't know if you were just hopping in on your own plane.
1:22
SPEAKER_02: Okay.
1:23
SPEAKER_02: No, but I do fly.
1:23
SPEAKER_02: I do fly.
1:24
SPEAKER_02: I enjoy it very much.
1:26
SPEAKER_02: And it's been a lifelong, lifelong thing.
1:28
SPEAKER_00: Okay.
1:29
SPEAKER_00: We'll get into that.
1:31
SPEAKER_00: All right.
1:32
SPEAKER_00: I'm going to kick us off with a proper intro.
1:34
SPEAKER_00: We'll try to.
1:36
SPEAKER_00: Let's do it.
1:51
SPEAKER_00: All right, Paul Aromenko.
1:53
SPEAKER_00: Thank you so much for being here.
1:55
SPEAKER_00: I guess in some ways this is vaguely home turf for people who don't know,
1:59
SPEAKER_00: we record this at the playground global offices.
2:02
SPEAKER_00: They were kind enough to give me some space to build my podcast studio.
2:06
I know they've,
2:07
they've invested in your,
2:09
at least one of your companies in the past.
2:11
SPEAKER_00: Yeah.
2:11
SPEAKER_00: So you've spent time in here.
2:13
SPEAKER_02: Some of my favorite people in Silicon Valley are here.
2:16
SPEAKER_00: Okay.
2:16
SPEAKER_00: And real quick housekeeping item before I do the proper, proper intro.
2:20
SPEAKER_00: This is all made possible by E1 Ventures.
2:24
SPEAKER_00: Kind enough to lend some support to the podcast.
2:29
SPEAKER_00: So here we go.
2:31
SPEAKER_00: This one's exciting.
2:32
SPEAKER_00: I've known you for a while now.
2:34
SPEAKER_00: You have this tremendous background,
2:35
which I'm going to give to people so they know a little context
2:39
of where you're
2:40
SPEAKER_00: coming from.
2:40
SPEAKER_00: I mean...
2:42
SPEAKER_00: It's hard to find.
2:43
SPEAKER_00: There's a lot of impressive stuff on here.
2:46
You're born in Ukraine,
2:47
moved to the United States when you were about 11.
2:49
SPEAKER_00: That's right.
2:50
SPEAKER_00: You have this very impressive family.
2:52
SPEAKER_00: I learned this week,
2:54
SPEAKER_00: going back to some Russian generals who were...
2:57
SPEAKER_00: Marshal of the Soviet Union.
2:58
SPEAKER_00: Of the Soviet Union.
2:59
SPEAKER_00: Your father, I think, is a quite famous mathematician.
3:02
SPEAKER_00: That's right.
3:03
SPEAKER_00: You...
3:05
SPEAKER_00: Got astronautics degrees from MIT and Caltech.
3:10
SPEAKER_00: And then also a law degree from Georgetown, I think.
3:13
SPEAKER_00: Yep, that is true.
3:14
SPEAKER_00: Okay, so I'm doing all right so far.
3:16
SPEAKER_00: And then you were the CTO at Airbus.
3:20
SPEAKER_00: I think you became CTO when you were about 35.
3:23
SPEAKER_00: You're one of the youngest.
3:24
SPEAKER_00: That might be true.
3:26
SPEAKER_00: I'm not sure.
3:27
SPEAKER_02: I've never claimed that, but if somebody did the math, then it's true.
3:31
SPEAKER_00: The internet says this.
3:34
You were also the CEO of Airbus's venture capital,
3:38
like sort of innovation.
3:40
SPEAKER_02: Yeah, sort of the Silicon Valley outpost called A-cubed.
3:42
SPEAKER_00: That's how I got into Airbus.
3:44
SPEAKER_00: You were also the CTO of United Technologies Corporation.
3:48
SPEAKER_00: You worked at DARPA.
3:50
SPEAKER_00: You worked at Google on...
3:52
SPEAKER_00: A lot of this stuff is aerospace,
3:53
SPEAKER_00: but Google was really a lot of projects,
3:55
SPEAKER_00: but phones and trying to rethink hardware more broadly,
3:59
SPEAKER_00: I think.
3:59
SPEAKER_00: Is that fair to say?
4:00
SPEAKER_00: That's right.
4:00
Yeah,
4:00
that was my only sort of major detour out of airspace up until
4:05
the current start.
4:06
SPEAKER_00: And then we met while you were running Universal Hydrogen.
4:09
SPEAKER_00: The last time we were together was at your hangar that you had.
4:15
SPEAKER_00: I think your neighbors were Harrison Ford and SpaceX, Elon.
4:19
SPEAKER_00: That's right.
4:19
SPEAKER_00: Right.
4:19
SPEAKER_00: So that's quite the pairing to be surrounded by.
4:22
SPEAKER_00: And then now you've got a new startup.
4:27
SPEAKER_00: We'll get to the startup in a bit.
4:29
SPEAKER_00: Why don't you give us like the elevator pitch on what you guys are doing?
4:32
SPEAKER_02: Sure.
4:33
SPEAKER_02: Yeah.
4:33
SPEAKER_02: We're trying to build an AI engineer for physical systems.
4:38
So for those of your listeners who are familiar with maybe
4:41
Devon in the software realm.
4:42
SPEAKER_02: I wrote the first story on Devon.
4:44
SPEAKER_02: Yes.
4:47
SPEAKER_02: We are doing Devon for physical product or physical system engineering.
4:53
SPEAKER_00: And the name of the company, it's P1AI.
4:55
SPEAKER_00: P1AI.
4:56
SPEAKER_00: P1AI.
4:57
SPEAKER_00: And the agent is called Archie, I think.
5:00
SPEAKER_00: And you guys launched basically this year.
5:03
SPEAKER_02: So we founded the company at the end of last summer.
5:07
SPEAKER_02: We raised a small pre-seed kind of last August,
5:10
started recruiting the team in September and really got going
5:12
in October of 24th.
5:15
SPEAKER_02: And just last month,
5:16
SPEAKER_02: or I guess now in April,
5:18
SPEAKER_02: two months ago,
5:19
we raised a fairly sizable seed round and are kind of off to
5:25
the races and
5:26
de-stealthed and significantly growing the team and actually
5:29
launching it into the
5:30
SPEAKER_02: wild pretty soon here.
5:31
SPEAKER_00: OK,
5:31
SPEAKER_00: we will get into that because,
5:33
SPEAKER_00: as you know,
5:33
SPEAKER_00: I'm fascinated by engineering and I'm covering AI all the time.
5:37
SPEAKER_00: And I love the idea of an AI for the physical world.
5:41
SPEAKER_00: But I want to start.
5:42
SPEAKER_00: I mean, you've lived such an interesting life.
5:46
SPEAKER_00: When you go through your background, one part that I was
5:50
that just grabbed me moving from ukraine to the us at 11 11 is
5:54
an interesting age
5:56
to do that you know if you come when you're much i moved from
5:59
south africa when i
6:00
was three i mean i feel more or less american i guess even
6:03
though my family's from
6:05
SPEAKER_00: a little all over the place but um
6:08
SPEAKER_00: Yeah.
6:08
SPEAKER_00: I don't know.
6:08
I mean,
6:08
that must've been,
6:09
did you end up in New York or where did you end up?
6:12
SPEAKER_00: No,
6:12
SPEAKER_02: actually we ended,
6:13
SPEAKER_02: uh,
6:14
SPEAKER_02: we ended up in Lexington,
6:15
SPEAKER_02: Kentucky of all places,
6:16
SPEAKER_02: which probably is why I have no accent,
6:19
SPEAKER_00: right?
6:19
SPEAKER_02: Is there was, there was really no Russian Ukrainian diaspora.
6:22
SPEAKER_00: I was going to ask about the accent.
6:24
SPEAKER_00: I mean,
6:24
SPEAKER_00: cause you're,
6:24
SPEAKER_00: you would,
6:25
SPEAKER_00: you would usually there's some hits you would just never,
6:28
SPEAKER_00: never know.
6:29
SPEAKER_02: Yeah.
6:29
SPEAKER_02: Yeah.
6:29
No,
6:30
I plunged plunged straight into American culture,
6:32
assimilated as quickly as possible.
6:34
SPEAKER_02: And yeah, we initially landed.
6:37
SPEAKER_02: My dad is a mathematician, as you mentioned.
6:38
And so initially he had a visiting appointment at University of
6:41
Kentucky at
6:41
SPEAKER_02: Lexington for about a year.
6:43
SPEAKER_02: And during that year we left.
6:45
SPEAKER_02: It was the Soviet Union.
6:46
SPEAKER_02: And during that year in 1990, the Soviet Union disintegrated.
6:49
SPEAKER_02: And so we had passports from a nonexistent country.
6:52
which led to some interesting sort of legal machinations that
6:55
ensued over the next years.
6:57
SPEAKER_02: But ultimately, my dad landed at Purdue, where he still is.
7:02
And I grew up like did high school and flight training and
7:06
things like that in Indiana.
7:08
SPEAKER_00: And I mean, I ticked off a couple of things from your family.
7:11
SPEAKER_00: I don't know.
7:12
SPEAKER_00: I guess this is a hard question to answer.
7:13
It seems like a very prestigious family with a long line of
7:17
exceptional people.
7:18
SPEAKER_00: Yeah.
7:19
SPEAKER_02: So I think both on my dad's and on my mom's side,
7:24
SPEAKER_02: it's an intelligentsia family,
7:27
SPEAKER_02: I guess I would put it that way.
7:30
SPEAKER_02: A lot of military folks,
7:32
SPEAKER_02: again,
7:32
SPEAKER_02: on both sides,
7:34
some of whom did some fairly impressive prominent things during
7:38
World War II.
7:40
SPEAKER_02: And then doctors, a lot of doctors.
7:46
SPEAKER_02: And then my father, of course, is a mathematician.
7:48
SPEAKER_02: And I describe myself as a failed mathematician.
7:51
SPEAKER_02: I think he had high hopes that we would be one of those father-son pairs.
7:54
SPEAKER_02: And I disappointed sorely my freshman year at MIT in that regard.
7:58
SPEAKER_00: But he's like next level.
7:59
SPEAKER_00: I forget which ward he won about.
8:01
SPEAKER_00: Humble prize.
8:02
SPEAKER_00: Right.
8:02
SPEAKER_00: I mean, was he recruited?
8:05
SPEAKER_00: The Soviet Union was falling apart.
8:07
SPEAKER_00: I mean, maybe he just wanted to get out for a better life for his family.
8:11
SPEAKER_00: Or was he was he pulled over here by the U.S. ?
8:15
SPEAKER_02: I think it was a combination of the opportunity being there.
8:18
SPEAKER_02: And like I said, we left fully intending to return after a year.
8:21
SPEAKER_02: It was a visiting gig for a year hosted by University of Kentucky.
8:26
SPEAKER_02: And then during that year,
8:27
SPEAKER_02: things became so complicated back home,
8:30
SPEAKER_02: back in the motherland,
8:32
SPEAKER_02: that it just made sense to stay.
8:34
SPEAKER_02: It wasn't clear whether there was going to be a civil war or what...
8:39
SPEAKER_02: If you recall that period, it was pretty exciting.
8:43
SPEAKER_00: Yes.
8:43
SPEAKER_00: Yes.
8:43
SPEAKER_00: I lived through that wild times.
8:46
SPEAKER_00: Okay.
8:47
SPEAKER_00: So your dad wanted you to be a mathematician.
8:49
I assume based on your academic record,
8:52
you did pretty well in high school.
8:55
SPEAKER_00: Were you a sciencey, mathy kind of kid?
8:58
SPEAKER_02: I was a sciencey, mathy kind of kid.
9:00
SPEAKER_02: I was also a little bit of sort of, I guess, a nonconformist.
9:04
SPEAKER_00: Okay.
9:05
SPEAKER_02: So I didn't really play by the rules very much.
9:07
SPEAKER_02: And so I ended up graduating from a very good high school,
9:10
SPEAKER_02: but not,
9:11
SPEAKER_02: I think I wasn't,
9:12
SPEAKER_02: I didn't make the top quarter of the class.
9:14
SPEAKER_02: Okay.
9:15
SPEAKER_02: But nonetheless,
9:18
SPEAKER_02: MIT and a couple other schools looked at the holistic,
9:22
SPEAKER_02: I suppose,
9:22
SPEAKER_02: at the holistic application.
9:24
SPEAKER_02: and let me in.
9:26
SPEAKER_02: Yeah, I mean, what stood out?
9:28
SPEAKER_00: You were flying, I think, from a pretty early age, weren't you?
9:31
SPEAKER_02: Yeah, I started flying, I think, when I was 14.
9:34
SPEAKER_02: You could solo when you were 16,
9:36
SPEAKER_02: and so I soloed on my 16th birthday,
9:38
SPEAKER_02: and you can get your pilot's license at 17,
9:40
SPEAKER_02: and I got it on my 17th birthday.
9:42
SPEAKER_02: I probably shouldn't have,
9:43
SPEAKER_02: and in fact,
9:45
SPEAKER_02: I screwed up a landing during my chuck ride,
9:48
SPEAKER_02: and the examiner was like,
9:49
SPEAKER_02: well,
9:49
SPEAKER_02: try again.
9:50
SPEAKER_02: And the second one was a little bit better,
9:52
SPEAKER_02: not super great,
9:54
SPEAKER_02: and we taxied in,
9:55
SPEAKER_02: and I was like,
9:55
SPEAKER_02: ah,
9:56
SPEAKER_02: crap,
9:57
SPEAKER_02: I failed for sure.
9:58
SPEAKER_02: And he turned to me and said, you know, I know your instructor.
10:02
SPEAKER_02: Go do another another another couple hours of landings practice with her.
10:06
SPEAKER_02: But happy birthday.
10:07
SPEAKER_02: OK.
10:10
I hate to make generalizations like this,
10:13
but I feel like this one holds true.
10:15
I think most people I know who got a pilot's license in their
10:19
teenage years,
10:20
SPEAKER_00: they do tend to be pretty
10:23
SPEAKER_00: extraordinary people that accomplish quite a lot in their lives.
10:27
I guess this is some early indicator of drive and or just
10:35
ability to get things
10:36
SPEAKER_00: done and a fascination with,
10:39
SPEAKER_00: I don't know,
10:39
SPEAKER_00: I guess like adventure and science and it makes sense,
10:42
SPEAKER_00: I suppose.
10:44
Yeah,
10:44
I mean,
10:44
I think in all honesty,
10:45
I was and probably still am a space cadet.
10:48
But if you're a 14 year old kid in Indiana,
10:51
there's not that much you can do with space.
10:53
SPEAKER_02: Right.
10:53
SPEAKER_02: So,
10:54
SPEAKER_02: you know,
10:54
SPEAKER_02: I was fascinated with telescopes and stuff like that,
10:57
SPEAKER_02: but that's not nearly as not nearly as cool.
10:59
And so I think flying was about as close to doing spacey stuff
11:03
as you could do.
11:04
SPEAKER_02: Yeah.
11:04
SPEAKER_02: So it was an early manifestation of that.
11:06
SPEAKER_00: And so how does this, clearly you're getting these aeronautics degrees.
11:12
This is sticking with you as you're going through school,
11:14
which you do get a law degree as well.
11:17
SPEAKER_00: As you're going through all the schooling,
11:19
SPEAKER_00: what are you thinking you're going to end up doing in life?
11:23
SPEAKER_02: Well,
11:24
SPEAKER_02: I did also have a very early fascination with,
11:26
SPEAKER_02: and I promise this is not revisionist history,
11:29
SPEAKER_02: with AI.
11:31
SPEAKER_02: And when I came to MIT,
11:32
MIT has a pretty active undergraduate research opportunity
11:36
program called Europe.
11:38
SPEAKER_02: And and I my first stop was at the A.I.
11:42
SPEAKER_02: lab and I was like, I want to do something at the A.I.
11:45
SPEAKER_02: lab.
11:46
SPEAKER_02: And what year is this?
11:47
SPEAKER_02: This was 1997.
11:48
So,
11:49
OK,
11:50
probably in some kind of winter,
11:52
perhaps,
11:52
or no,
11:53
maybe that's like
11:55
SPEAKER_02: Yes, it was certainly winter.
11:56
SPEAKER_02: AI Lab was doing a lot of robotics.
11:59
SPEAKER_02: Rodney Brooks,
12:00
SPEAKER_02: by the way,
12:01
SPEAKER_02: who's also part of the Playground family here,
12:04
SPEAKER_02: was,
12:04
SPEAKER_02: I think,
12:04
SPEAKER_02: the head of the AI Lab at the time.
12:06
SPEAKER_02: And he was very much focused on mobile robotics, as I guess he still is.
12:13
But the folks that I approached at the AI lab were not terribly
12:15
welcoming.
12:16
SPEAKER_02: And so my second stop was at the aerospace department.
12:19
And I was very interested in aerial robotics and joined the MIT
12:25
aerial robotics team.
12:26
SPEAKER_02: And then by sophomore year,
12:27
SPEAKER_02: I think I was leading the MIT aerial robotics team,
12:29
SPEAKER_02: which were autonomous helicopters to perform various kinds of missions,
12:33
SPEAKER_02: the kinds of things that we have drones doing today.
12:36
Um,
12:37
this was the very,
12:38
very early days of that in,
12:39
in kind of the late nineties.
12:41
SPEAKER_00: Okay.
12:42
SPEAKER_00: Okay.
12:42
SPEAKER_00: And it, okay.
12:43
So you're into all this different stuff and,
12:45
and the AI point is well taken.
12:47
SPEAKER_00: I actually,
12:47
SPEAKER_00: it was a little after the time that you were describing,
12:50
SPEAKER_00: but I found some like PBS interview with you.
12:53
And I want to say it was like 2013 and someone is proposing
12:57
this idea of,
12:58
uh,
12:59
quite sophisticated AIs and they're like,
13:02
you know,
13:02
wouldn't that be that'll never happen?
13:05
SPEAKER_00: I don't know.
13:06
SPEAKER_00: You were already, you know, promoting this idea back then.
13:10
SPEAKER_00: That was.
13:10
SPEAKER_02: Yes, that was already my DARPA days.
13:12
SPEAKER_02: I think I think the episode, if I recall, was called Rise of the Drones.
13:15
SPEAKER_00: Yeah.
13:15
SPEAKER_02: Right.
13:18
SPEAKER_02: And yeah,
13:19
SPEAKER_02: I mean,
13:19
SPEAKER_02: I've always been a proponent of autonomy and very interested in autonomy
13:25
SPEAKER_02: technologies.
13:26
SPEAKER_02: So you're not a doer.
13:28
Uh,
13:29
well,
13:30
let's talk about,
13:33
how deep do you want to go down that rabbit hole?
13:37
SPEAKER_02: I mean,
13:37
SPEAKER_02: look,
13:38
SPEAKER_02: I think,
13:39
SPEAKER_02: uh,
13:39
SPEAKER_02: I think,
13:41
SPEAKER_02: I think we are on the cusp of what Verna Vinci called the singularity.
13:45
SPEAKER_02: Right.
13:45
And I am super grateful to be alive at this time because I
13:50
think it's probably one
13:51
SPEAKER_02: of the most interesting times in human history.
13:54
But I do think that Vinji's sort of description of the
13:57
singularity as a point at
13:59
which technological change progresses so quickly that we can't
14:02
really see past it.
14:03
SPEAKER_02: So kind of like an event horizon of astrophysics singularity.
14:08
SPEAKER_02: Um, I think, I think that point is very valid.
14:10
SPEAKER_02: Um,
14:11
SPEAKER_02: and so I don't think we know what,
14:13
SPEAKER_02: what is on the other side of the singularity,
14:14
SPEAKER_02: but I think it's,
14:16
SPEAKER_02: it is nigh.
14:17
SPEAKER_00: So we don't know and you're excited,
14:20
SPEAKER_00: but there's your,
14:21
SPEAKER_00: so you're,
14:22
SPEAKER_00: I mean,
14:22
SPEAKER_00: it sounds like you're not a necessarily a doomer,
14:25
SPEAKER_00: but you're not discounting the fact that this could go awry.
14:27
SPEAKER_00: I guess that's what you're saying.
14:29
I think it absolutely could,
14:30
but I think we have the opportunity to shape it.
14:31
SPEAKER_02: And that's one of the reasons that I really wanted,
14:34
SPEAKER_02: I felt the urgency to dive into the AI world in the last couple of years,
14:38
SPEAKER_02: sort of like head first,
14:40
SPEAKER_02: is because I do think that there is an opportunity.
14:43
SPEAKER_02: It's not out of our control.
14:46
SPEAKER_02: It's not fate.
14:47
SPEAKER_02: It's our future to shape and build.
14:50
SPEAKER_02: And I want to be here for that.
14:51
SPEAKER_00: So you do have this kind of 15,
14:53
SPEAKER_00: 20 year period where you're very heavily,
14:55
SPEAKER_00: I mean,
14:56
SPEAKER_00: you couldn't be more in the aerospace world in many ways.
15:00
SPEAKER_00: Was that what you expected to do coming out of college?
15:05
SPEAKER_02: Yes, I think it was.
15:07
I don't think I could have forecast the specific sort of
15:09
sequence of events,
15:11
SPEAKER_02: but I remember even in high school being asked,
15:14
SPEAKER_02: like,
15:14
SPEAKER_02: what do you want to be?
15:15
SPEAKER_02: What's your dream job when you grow up?
15:17
SPEAKER_02: And my answer was, I want to be the chief engineer at Boeing.
15:21
SPEAKER_02: And so I missed the target a little bit.
15:23
SPEAKER_02: Right.
15:23
I ended up I ended up at Airbus,
15:26
but kind of kind of check that off the bucket list.
15:30
SPEAKER_02: So I was pretty, pretty goal or mission oriented.
15:34
SPEAKER_02: Right.
15:34
SPEAKER_02: Now,
15:35
the fact that I ultimately landed in that particular role at
15:38
that particular
15:38
SPEAKER_02: company was probably a series of fortuitous coincidences.
15:42
Well,
15:42
I mean,
15:43
to get there at 35 as ostensibly an American,
15:49
how did that happen?
15:50
SPEAKER_00: I don't think anybody has ever called me ostensibly an American.
15:57
SPEAKER_00: What did they see?
15:58
SPEAKER_00: What had you done?
15:59
SPEAKER_00: What had you been doing right before that that,
16:03
SPEAKER_00: you know,
16:03
SPEAKER_00: made you attractive to Airbus and to take on such a big role?
16:08
So initially I was recruited to Airbus to kind of found their
16:12
Silicon Valley
16:12
SPEAKER_02: presence,
16:14
SPEAKER_02: which is this innovation center called Acubed.
16:17
SPEAKER_02: And I also helped a little bit with Airbus Ventures.
16:21
SPEAKER_02: And that went well.
16:24
SPEAKER_02: I came out of Google to do that, but I had an aerospace background.
16:28
I was a DARPA mixed with kind of Silicon Valley exposure
16:33
through my times at
16:34
SPEAKER_02: Motorola and at Google.
16:36
SPEAKER_02: And and it went well.
16:39
SPEAKER_02: Right.
16:39
SPEAKER_02: A cubed.
16:40
SPEAKER_02: We launched a number of very exciting project.
16:42
I think we launched launched one of the very first TV tolls
16:45
that I remember that
16:46
SPEAKER_00: that actually flew the Vahana.
16:48
SPEAKER_00: Vahana.
16:48
SPEAKER_02: Right.
16:48
SPEAKER_02: Yep.
16:50
SPEAKER_02: We we launched the Uber Uberization of helicopters as a service.
16:57
SPEAKER_02: Right.
16:57
SPEAKER_02: So you could do sort of one click dispatch of a helicopter.
17:00
SPEAKER_02: Yeah.
17:00
SPEAKER_02: And maybe a ride to the helipad and a ride from the helipad.
17:03
SPEAKER_02: Right.
17:03
SPEAKER_02: So kind of make it part of a one click point to point.
17:05
SPEAKER_02: transit.
17:07
SPEAKER_02: We launched a bunch of autonomy work,
17:09
which ultimately led to the first fully autonomous flight of an
17:13
A350 gate to gate,
17:15
which was a program that we continued when I became CTO and
17:20
came to fruition a
17:21
SPEAKER_02: couple of years later.
17:22
SPEAKER_02: So we did a lot of exciting stuff.
17:24
SPEAKER_02: And I think Tom Enders,
17:26
SPEAKER_02: who was the CEO of Airbus at the time,
17:28
SPEAKER_02: saw that and said,
17:29
this is what Airbus engineering and the Airbus technology
17:32
organization should look
17:33
SPEAKER_02: like.
17:34
SPEAKER_02: and invited me to Europe.
17:36
And I think that my American-ness was a positive in that
17:42
particular context,
17:43
SPEAKER_02: but it was a very narrow window in which it was a positive.
17:45
SPEAKER_02: In general, it probably is not a positive for success at Airbus.
17:48
SPEAKER_02: Just culturally or?
17:50
SPEAKER_02: Culturally,
17:50
it is a very Franco-German company and there's a lot of tug of
17:54
war between the
17:55
SPEAKER_02: French and the Germans.
17:56
SPEAKER_02: And because Tom's sort of mandate for his new CTO was,
18:00
SPEAKER_02: I really want a drastic change in rethinking of the way we do technology.
18:06
He wanted somebody who wasn't French or German because if there
18:09
was gonna be
18:09
SPEAKER_02: restructuring,
18:10
SPEAKER_02: if there were gonna be shifts of where we do things and job losses,
18:15
SPEAKER_02: job creation,
18:15
SPEAKER_02: et cetera,
18:16
SPEAKER_02: he didn't want somebody who would be accused of one-sidedism,
18:19
SPEAKER_02: right?
18:20
SPEAKER_00: But the system rejected some of his desires or it went okay?
18:25
relatively okay i think it went i i'm pleased to say that a lot
18:29
of the changes that
18:30
we made and we did make some pretty pretty drastic uh
18:33
rethinking of how airbus does
18:35
uh technology development most of the things have persisted to
18:39
this day okay this
18:41
SPEAKER_00: is one thing i wanted to get into i don't know
18:43
where you're at in terms of being able to publicly speak about
18:47
Airbus and Boeing or
18:48
SPEAKER_00: your,
18:49
SPEAKER_00: I know we've had chats,
18:51
SPEAKER_00: sort of private chats,
18:52
SPEAKER_00: you have opinions on these things.
18:54
SPEAKER_00: So anyway, I'll go there and we'll see where we end up.
19:00
SPEAKER_00: Like what did you find with the Airbus culture when you got there?
19:07
SPEAKER_00: How would you describe it to people?
19:09
SPEAKER_02: So it is very mission focused.
19:12
SPEAKER_02: I think the people that come to Airbus really love airplanes.
19:16
SPEAKER_02: They really love flying.
19:18
SPEAKER_02: They really love Airbus.
19:20
SPEAKER_02: It is the national pride of France, at least.
19:23
SPEAKER_02: I think in Germany,
19:24
it ranks maybe a little lower on the on the list of most
19:27
desirable places for an
19:28
SPEAKER_02: engineer to work.
19:29
SPEAKER_02: Right.
19:29
SPEAKER_02: Because there's just a lot of automotive companies and things like that.
19:32
SPEAKER_02: But at Airbus, it is for engineering grads from top schools.
19:36
SPEAKER_02: It is the number one employer of choice, right?
19:38
SPEAKER_02: So it is really getting the kind of the kind of of engineering talent.
19:43
And both the French and German education systems are very,
19:46
very good,
19:47
right?
19:47
SPEAKER_02: So the quality of engineering talent was extraordinary at Airbus.
19:52
And people were really passionate and dedicated to the mission,
19:55
I would say.
19:56
A bit more on the on the negative side,
19:59
it is also hyper political organization.
20:03
It was originally created as a government formed entity with
20:09
government ownership
20:10
SPEAKER_02: on across four governments,
20:12
SPEAKER_02: French,
20:12
SPEAKER_02: German,
20:14
SPEAKER_02: British and Spanish.
20:15
SPEAKER_02: Spanish share is relatively small.
20:16
SPEAKER_02: The British exited their share.
20:19
SPEAKER_02: But there is still very,
20:20
very strong undercurrent of sort of national identity politics
20:27
that makes things
20:28
SPEAKER_02: difficult.
20:30
SPEAKER_00: Yeah,
20:30
SPEAKER_00: and the thing that I know better is Arianespace,
20:34
SPEAKER_00: the European sort of vaguely similar structure in a lot of ways.
20:39
SPEAKER_02: And his Airbus is a partial subsidiary, right?
20:45
SPEAKER_00: Exactly.
20:45
SPEAKER_00: Part of Airbus and kind of French-led, different countries chipping in.
20:51
SPEAKER_00: But, you know, I mean, Arianespace, oh, man, you know, just a rough...
20:55
sort of miscalculated on the spacex of it all quite quite
20:59
dramatically as did ula
21:00
right to be fair i probably like as did everybody yeah the
21:04
universe um and and you
21:06
know to spacex's credit sort of pulled off things people
21:09
weren't expecting but i
21:11
SPEAKER_00: like when i look now at the
21:13
european space program in particular it just it feels um
21:16
there's part of me it
21:17
thinks it's like a bit of a lost cause you know you've missed
21:21
reusable rockets it's
21:23
like still barely even talking about they have a couple small
21:27
rocket startups um
21:29
yeah i mean this part is i was just in europe last month and
21:33
people are like asking
21:34
me how this is gonna go and you don't want to just be some it's
21:37
not like i'm not
21:38
SPEAKER_00: even like pro uh
21:40
SPEAKER_00: I'm not, I'm not super nationalistic person.
21:42
SPEAKER_00: You know, this isn't like a rah-rah us thing.
21:44
It's just kind of like,
21:45
man,
21:46
I mean,
21:46
maybe you guys want to focus on something else.
21:49
SPEAKER_00: I don't know if this is going to pan out for you very well.
21:52
SPEAKER_00: Um, that when you compare,
21:57
SPEAKER_00: Airbus, you said you wanted to work at Boeing as a kid.
22:00
Can you where do you have a sense as to which one you feel is
22:06
more innovative
22:08
between Airbus and Boeing and has its act together at this
22:11
moment in time?
22:12
SPEAKER_02: Well, I think so.
22:13
SPEAKER_02: Look, I think that both companies have big company syndrome, right?
22:17
That's just a function of being a very large organization
22:20
that's existed for decades.
22:23
SPEAKER_02: Boeing is much older, obviously, right?
22:25
SPEAKER_02: So it probably has more ossification.
22:29
But on the other hand,
22:31
Boeing's internal politics are probably somewhat simpler.
22:34
SPEAKER_02: I did spend a little bit of time in Boeing, actually, in my career.
22:37
I did a summer internship between undergrad and grad school
22:41
working for a guy named
22:41
SPEAKER_02: Bob Liebeck,
22:42
who was the father of the blended wing body down here in
22:46
California in Long Beach.
22:49
And so I got a little and I,
22:50
of course,
22:51
I know a lot of a lot of Boeing Boeing execs.
22:54
SPEAKER_02: So I think, you know, I think both are phenomenal companies.
22:58
SPEAKER_02: Right.
22:59
That that on the commercial aviation side have produced
23:02
remarkable products.
23:03
SPEAKER_02: Right.
23:04
SPEAKER_02: That have changed the course of human history.
23:06
SPEAKER_02: Right.
23:07
The fact that we have like the modern commercial air
23:09
transportation system,
23:11
SPEAKER_02: which was largely built by by those two companies.
23:14
SPEAKER_02: is is a marvel.
23:15
SPEAKER_02: Right.
23:15
The level of safety and and cost efficiency in that the and the
23:21
resulting sort of
23:21
SPEAKER_02: globalization and interchange of cultural interchange,
23:26
SPEAKER_02: commerce,
23:26
SPEAKER_02: et cetera,
23:27
SPEAKER_02: that has that it's precipitated probably rivals the Internet.
23:29
SPEAKER_02: Right.
23:29
SPEAKER_02: In terms of impact on 20th century human history.
23:32
SPEAKER_00: Sure.
23:32
SPEAKER_00: Yeah.
23:33
SPEAKER_02: So so I have deep, deep respect for both companies.
23:37
SPEAKER_02: I think both have had challenges throughout their history.
23:41
SPEAKER_02: I think Boeing is going through a challenging period.
23:47
SPEAKER_02: Kelly Orberg, by the way, was a colleague of mine at United Technologies.
23:50
He was the head of Collins Aerospace,
23:52
which was one of the one of the four businesses at UTC.
23:55
SPEAKER_02: and is I have huge, huge respect for Kelly.
24:00
SPEAKER_02: I think he is a he's a very strong engineer,
24:02
SPEAKER_02: but he's also just a phenomenal leader and a nice guy.
24:04
And I I think he'll do great things at Boeing if if given if
24:10
given the opportunity
24:11
SPEAKER_02: for the board,
24:11
SPEAKER_02: if the board supports him.
24:13
SPEAKER_00: Yeah.
24:15
SPEAKER_00: Total tangent.
24:17
SPEAKER_00: When I was looking you up and I was reading about United Technologies,
24:20
SPEAKER_00: it traces its roots all the way back to Litton Industries,
24:23
SPEAKER_00: the Litton brothers that lived up here in Silicon Valley.
24:26
SPEAKER_00: This is wild to me.
24:30
SPEAKER_00: California sometimes will surprise you, you know?
24:34
SPEAKER_02: Well,
24:35
SPEAKER_02: you know,
24:35
so so the origin of UTC is actually United Aircraft and
24:38
Technology Corporation,
24:39
SPEAKER_02: which was early 20th century,
24:41
SPEAKER_02: 1920s,
24:41
SPEAKER_02: 1930s.
24:41
SPEAKER_02: This is sort of the era I'm talking about.
24:45
SPEAKER_00: Yeah, that's right.
24:45
And it used to own it was a it was a it was a full stack
24:50
aerospace company.
24:51
SPEAKER_02: Right.
24:51
SPEAKER_02: So it owned Boeing.
24:53
SPEAKER_02: And it owned United Airlines in the early days.
24:55
SPEAKER_02: Yeah.
24:56
SPEAKER_02: And it was one of the first applications of the of the antitrust,
25:02
SPEAKER_02: the big antitrust act and movement.
25:06
SPEAKER_02: And,
25:06
SPEAKER_02: um,
25:08
SPEAKER_02: uh,
25:08
SPEAKER_02: and,
25:08
SPEAKER_02: and it was,
25:09
SPEAKER_02: uh,
25:10
SPEAKER_02: it was forced to divest,
25:12
SPEAKER_02: uh,
25:12
several of these businesses and the decision was made along the
25:15
Mississippi,
25:15
SPEAKER_02: right?
25:16
SPEAKER_02: There was not, not more sophisticated than that.
25:17
SPEAKER_02: So you,
25:18
SPEAKER_02: uh,
25:18
you ATC as it was called at the time was allowed to keep
25:20
everything east of the
25:21
SPEAKER_02: Mississippi and had to divest,
25:23
SPEAKER_02: uh,
25:24
SPEAKER_02: Boeing and United airlines.
25:25
SPEAKER_00: Okay.
25:25
SPEAKER_00: I did not know that.
25:27
SPEAKER_00: Um, if you, okay.
25:30
SPEAKER_00: If you have to,
25:33
SPEAKER_00: When you look at the future of commercial aviation,
25:36
SPEAKER_00: I don't know,
25:39
SPEAKER_00: what do you see happening between sort of Airbus,
25:41
SPEAKER_00: Boeing,
25:42
SPEAKER_00: China?
25:43
SPEAKER_00: Like, what does the next 10, 20 years look like in your head?
25:47
SPEAKER_02: Well, so I could give you a pre-AI and a post-AI answer, right?
25:52
And it is entirely possible that everything about the
25:56
engineered world is going to
25:58
SPEAKER_02: change in the next several years.
25:59
SPEAKER_02: And frankly,
26:02
the reason that I founded the startup that I did is to help
26:07
make that change and
26:08
SPEAKER_02: help steer that change.
26:10
And so it is possible that we're talking about a completely
26:14
different engineered
26:15
SPEAKER_02: world 10 to 20 years from now.
26:17
SPEAKER_02: Right.
26:18
And so so so my predictions that I would have made a few years
26:22
ago or certainly
26:23
when I was at Airbus about what is the future of commercial air
26:26
transport,
26:28
SPEAKER_02: as you suggested,
26:28
SPEAKER_02: are would be like decadal type predictions because it takes,
26:31
SPEAKER_02: you know,
26:32
like a new a new generation of airplanes materializes roughly
26:35
every decade.
26:37
And so I could have probably told you what the next generation
26:39
of airplane looks
26:39
SPEAKER_02: like and maybe kind of guessed a little bit at the one after that.
26:44
But I'm not sure that in a post-AI world,
26:48
all of that's going to hold anymore.
26:49
SPEAKER_00: OK, well, I definitely want to get into that.
26:51
SPEAKER_00: All right.
26:51
SPEAKER_00: Just a couple of real quick questions before we did.
26:53
SPEAKER_00: Now I'm extra intrigued.
26:57
SPEAKER_00: Just on this new generation of aircraft point,
26:59
SPEAKER_00: I mean,
26:59
SPEAKER_00: when I was writing about universal hydrogen,
27:02
SPEAKER_00: you guys were trying to change the fueling infrastructure for aircraft.
27:07
SPEAKER_00: And some of this would hinge on what the next...
27:11
sort of workhorses of commercial aviation would look like the
27:14
next generation
27:15
SPEAKER_00: planes from companies like Airbus and Boeing.
27:19
SPEAKER_00: I hope I'm not like outing you somehow.
27:21
SPEAKER_00: I mean,
27:22
SPEAKER_00: if I remember right,
27:22
you weren't super bullish on these companies' abilities to kind
27:27
of create these
27:29
next gen planes and to be innovative in how they were pushing
27:33
things forward.
27:33
SPEAKER_00: Is that fair?
27:35
SPEAKER_02: Yeah, maybe.
27:37
SPEAKER_02: So let me pull on a couple threads in your question.
27:39
So first of all,
27:40
the next airplane is probably coming in the next five,
27:45
six,
27:45
seven years.
27:46
I think right now Boeing doesn't have the capital and Airbus
27:50
doesn't have the
27:50
incentive to launch a new airplane program because Airbus is
27:53
selling the A321
27:54
SPEAKER_02: family like hotcakes and it's doing very well for the company.
27:58
SPEAKER_02: Why would you cannibalize that?
27:59
SPEAKER_02: Yeah.
28:00
And Boeing has a lot of sort of structural cultural issues that
28:05
it needs to work
28:05
SPEAKER_02: through over the next few years.
28:06
SPEAKER_02: But probably early 2030s, there will be a new airplane.
28:10
SPEAKER_02: Three,
28:11
SPEAKER_02: four,
28:11
SPEAKER_02: five years ago,
28:11
SPEAKER_02: if you had asked me,
28:12
I would have said that there's a reasonable possibility that
28:16
the airplane could be
28:17
SPEAKER_02: a hydrogen airplane.
28:19
SPEAKER_02: I no longer think that not as a function of Airbus or Boeing,
28:23
SPEAKER_02: but actually as a function of hydrogen cost.
28:25
And we can talk about why hydrogen cost didn't follow the curve
28:28
that that we
28:30
SPEAKER_02: thought it would in in 2020,
28:32
SPEAKER_02: 2019 and 2020.
28:34
But I don't think that hydrogen is going to be a competitive
28:37
competitive aviation fuel anymore.
28:38
SPEAKER_00: Okay.
28:39
SPEAKER_00: So the goal of universal hydrogen,
28:41
SPEAKER_00: it's been a while since I wrote about it,
28:42
SPEAKER_00: so I might get some of this wrong.
28:43
But I remember you guys were,
28:46
you wanted to prove it out as a viable fuel source for planes.
28:51
And you were going to create an infrastructure where this could
28:54
be delivered to airports.
28:57
SPEAKER_00: You were trying to make it easy on the industry to switch over, right?
29:00
SPEAKER_00: So you were doing a lot.
29:01
SPEAKER_00: You were...
29:02
You were kind of retrofitting some planes to show how they
29:04
could fly with hydrogen
29:07
SPEAKER_00: as a fuel source.
29:07
And then you were also working on this infrastructure piece of
29:10
actually delivering
29:11
SPEAKER_00: the hydrogen safely,
29:13
effectively to airports and making it easy for the planes to
29:17
refuel and swap out
29:18
SPEAKER_00: some of this infrastructure.
29:19
SPEAKER_00: Is that...
29:20
SPEAKER_00: Yeah, that's exactly right.
29:22
SPEAKER_00: Yep, exactly.
29:23
SPEAKER_00: And so you guys were going at it.
29:25
SPEAKER_00: As I mentioned, it was pretty cool.
29:27
SPEAKER_00: You were in Hawthorne.
29:28
SPEAKER_00: You had stuff going on in France and in California.
29:31
SPEAKER_00: And you were in Hawthorne right next door to Harrison Ford,
29:35
SPEAKER_00: which probably apparently is like a...
29:38
possibly dangerous hangar to be next to if the internet is to
29:43
be believed.
29:44
SPEAKER_00: So it didn't work out.
29:47
SPEAKER_00: Yeah, I mean, let's get into it.
29:50
SPEAKER_00: What happened?
29:51
SPEAKER_00: Because I know you were so excited.
29:52
SPEAKER_00: I remember when I was reporting on you guys, you were so excited.
29:56
SPEAKER_00: I felt like you explained it really well.
29:58
SPEAKER_00: When I sat back,
29:59
SPEAKER_00: I was like,
29:59
SPEAKER_00: man,
30:00
this is going to be a journey because there are a lot of things
30:03
that have to fall
30:04
SPEAKER_00: into place to make this work.
30:06
So why don't you tell us like how far you got and then what
30:09
ended up happening?
30:11
SPEAKER_02: Yeah.
30:11
SPEAKER_02: So so as you correctly summarized,
30:14
the goal was first to develop a hydrogen delivery
30:16
infrastructure that would be
30:19
easier than sort of rebuilding the current jet fuel delivery
30:22
infrastructure.
30:22
SPEAKER_02: Right.
30:22
SPEAKER_02: So the idea was we modularize hydrogen,
30:24
SPEAKER_02: put it in these like Nespresso style pods,
30:26
SPEAKER_02: much bigger,
30:27
SPEAKER_02: but but modular pods.
30:29
And use intermodal freight,
30:30
the existing intermodal freight network to deliver it,
30:32
right?
30:32
SPEAKER_02: So any airport that can take intermodal cargo,
30:35
SPEAKER_02: which is pretty much every one of them,
30:37
could take hydrogen and kind of plug it into the airplane as a
30:40
fuel source.
30:41
SPEAKER_02: And hydrogen is a phenomenal energy storage mechanism.
30:45
It's not really a true fuel in the sense of you don't extract
30:48
it from the environment.
30:48
SPEAKER_02: Well,
30:49
there actually are some pushes to try to get hydrogen out of
30:52
the environment,
30:53
SPEAKER_02: but typically you create it.
30:54
SPEAKER_02: So it's created.
30:58
SPEAKER_02: It takes a lot of energy to create hydrogen.
31:00
SPEAKER_02: Um,
31:01
SPEAKER_02: but it is a very,
31:02
SPEAKER_02: uh,
31:02
SPEAKER_02: very efficient storage mechanism,
31:04
SPEAKER_02: energy storage mechanism,
31:05
SPEAKER_02: which is why it's great,
31:06
SPEAKER_02: great for aerospace.
31:07
SPEAKER_02: And so we were creating these modular modular pods and then,
31:10
SPEAKER_02: and then to sort of prime demand for our modular pods,
31:14
we converted a regional airliner to fly on hydrogen and flew
31:18
the sort of world's
31:19
SPEAKER_02: biggest airplane,
31:20
SPEAKER_02: uh,
31:20
SPEAKER_02: on fuel cells.
31:21
SPEAKER_02: I remember that.
31:22
SPEAKER_02: Yeah.
31:23
Awesome proof of,
31:24
uh,
31:25
not just concept,
31:26
but,
31:26
uh,
31:26
yeah,
31:27
getting the work done.
31:28
SPEAKER_02: Yeah.
31:28
So product wise and technology wise,
31:30
the company was a huge success,
31:32
right?
31:32
SPEAKER_02: Like we delivered on every every product milestone.
31:35
SPEAKER_02: We had a pretty clear certification path with the FAA for all of this.
31:41
SPEAKER_02: We flew in record time of,
31:42
SPEAKER_02: I think,
31:42
two years from from getting first first first financed and and
31:47
kind of everything
31:48
SPEAKER_02: was was humming on schedule,
31:50
SPEAKER_02: on budget.
31:50
SPEAKER_02: We had an incredibly talented team, which I'm still very, very proud of.
31:54
What all of this was predicated on was that the cost of
31:58
hydrogen production would
32:00
SPEAKER_02: drop very,
32:01
SPEAKER_02: very rapidly.
32:02
SPEAKER_02: Right.
32:02
So when we were sitting in 2019,
32:05
2020,
32:05
conceiving the company,
32:07
hydrogen was pretty expensive.
32:09
And the expectation was that it would,
32:11
by 2024 or five,
32:14
it would reach parity with jet fuel.
32:15
SPEAKER_02: And you might ask why.
32:16
SPEAKER_02: I was going to ask why.
32:19
SPEAKER_02: Right.
32:19
SPEAKER_02: Because it's not obvious.
32:20
SPEAKER_02: It's not obvious that it should.
32:21
SPEAKER_02: Right.
32:22
SPEAKER_00: I made a fan with Lee over here.
32:26
SPEAKER_02: He'll be a regular.
32:29
SPEAKER_02: And so so so how do you make hydrogen?
32:31
SPEAKER_02: You make hydrogen by electrolyzing water.
32:33
SPEAKER_02: At least that's how you make green hydrogen.
32:34
SPEAKER_02: So hydrogen.
32:35
You can also crack hydrocarbons to extract the hydrogen,
32:38
but then you have the carbon leftover.
32:39
SPEAKER_02: And so it's not a particularly clean process.
32:41
SPEAKER_02: The clean way is you take renewable electricity.
32:44
SPEAKER_02: And you electrolyze water, so you break it into hydrogen and oxygen.
32:48
Oxygen you can release,
32:50
it's useful,
32:50
or you can sell it for other applications.
32:53
SPEAKER_02: I hear it's good.
32:54
SPEAKER_02: And then it is good.
32:56
SPEAKER_02: And then hydrogen you can use to power airplanes or other things.
33:00
And so the cost of hydrogen is principally the cost of
33:02
renewable electricity.
33:03
SPEAKER_02: And the thesis,
33:04
SPEAKER_02: again,
33:05
SPEAKER_02: five,
33:05
SPEAKER_02: six years ago,
33:06
SPEAKER_02: was that there is this huge growth in renewable production.
33:10
SPEAKER_02: And a lot of renewables you size for peak grid consumption, right?
33:14
So you will you build a wind farm to power a nearby city or
33:18
region and and you size
33:20
SPEAKER_02: it for peak peak consumption,
33:22
SPEAKER_02: you know,
33:23
SPEAKER_02: evening hours,
33:23
SPEAKER_02: right,
33:24
SPEAKER_02: or midday heat or whatnot.
33:26
SPEAKER_02: And then at night, wind still blows, not as much, but still blows.
33:30
You're still producing electricity and typically you would just
33:32
dump it because
33:32
SPEAKER_02: there's no massive grid scale storage,
33:35
SPEAKER_02: right?
33:36
And so you could then take that off peak electricity and
33:39
produce hydrogen very,
33:40
very cheaply.
33:42
SPEAKER_02: So that was the thesis.
33:43
And most of the hydrogen cost projections were based on very
33:47
affordable off-peak
33:48
SPEAKER_02: renewables being scavenged at near zero electrical costs,
33:52
SPEAKER_02: right?
33:53
SPEAKER_02: Because otherwise it was getting dumped for hydrogen.
33:56
SPEAKER_02: And then data centers happened.
33:59
SPEAKER_02: Oh, right.
33:59
SPEAKER_02: So yeah.
34:01
SPEAKER_02: Yeah.
34:01
SPEAKER_02: Yeah.
34:02
SPEAKER_02: And so,
34:02
so that thesis did not pan out and the cost of hydrogen
34:05
production of green
34:06
SPEAKER_02: hydrogen production today is nowhere near that of,
34:08
SPEAKER_02: of,
34:09
SPEAKER_02: of jet fuel.
34:10
SPEAKER_00: Just because we went on some mad energy consumption dash.
34:16
SPEAKER_02: Yes.
34:16
SPEAKER_02: And it is also, I mean, there's more nuance to it than that, right?
34:20
SPEAKER_02: Also the mix of renewables.
34:22
SPEAKER_02: You have solar, for instance, right?
34:23
SPEAKER_02: And solar then actually doesn't get produced at night.
34:27
SPEAKER_02: So you don't necessarily get as much off peak.
34:29
SPEAKER_02: So there's also a balance of what kinds of renewables go into it.
34:34
But the bottom line is electricity prices for hydrogen
34:39
production did not follow the curve.
34:41
SPEAKER_00: Okay.
34:41
SPEAKER_00: And so,
34:42
SPEAKER_00: OK,
34:43
so we're not going to have a hydrogen plane in some massive
34:47
hydrogen fuel aviation
34:49
SPEAKER_00: industry anytime soon.
34:51
SPEAKER_00: I don't think so.
34:53
SPEAKER_00: Yeah.
34:54
SPEAKER_00: And you were not.
34:54
SPEAKER_00: If I remember, you weren't like super bullish on the SAFs either.
34:58
SPEAKER_00: These these alternative fuels.
35:01
SPEAKER_02: Yeah, SAFs are a little bit of a carbon shell game, right?
35:05
SPEAKER_02: Right.
35:06
SPEAKER_00: Yeah, so what does that mean?
35:07
SPEAKER_00: I mean, that means for what?
35:09
SPEAKER_00: The next 10, 12?
35:10
SPEAKER_00: Okay, well, we'll get into it.
35:11
Yeah,
35:11
we can come back to the question of what happens to the
35:13
aviation industry.
35:14
SPEAKER_00: If things were holding normal ways...
35:20
SPEAKER_00: It sounds like the future you're talking about would probably be just a,
35:24
SPEAKER_00: you know,
35:24
quite a more still quite traditional commercial aviation
35:29
industry for the next 10,
35:30
SPEAKER_00: 20,
35:30
30 years.
35:30
SPEAKER_02: I think there will be innovation.
35:31
SPEAKER_02: I think there'll be innovation in other aspects of it, right?
35:34
SPEAKER_02: But yes, I do think that it'll be hydrocarbon powered.
35:37
SPEAKER_02: I think some of those hydrocarbons will be SAFs.
35:39
I think a lot of SAFs are kind of a collective delusion in
35:44
terms of their cleanliness,
35:45
right?
35:46
SPEAKER_02: There's no, it's an offset scheme, right?
35:48
SPEAKER_00: Yeah.
35:48
SPEAKER_00: I don't want to go nuts on this point,
35:49
SPEAKER_00: but I mean,
35:50
SPEAKER_00: they get,
35:50
SPEAKER_00: they sort of get like a free pass as far as I can tell in the press.
35:54
SPEAKER_00: And you were, you were kind of like, this is the math does not check out.
35:57
SPEAKER_00: No,
35:58
SPEAKER_02: the math certainly does not,
36:00
SPEAKER_02: does not check out,
36:01
SPEAKER_02: um,
36:01
SPEAKER_02: for,
36:02
SPEAKER_02: for,
36:02
SPEAKER_02: for SAFs,
36:03
SPEAKER_02: but,
36:03
SPEAKER_02: but it makes,
36:03
SPEAKER_02: you know,
36:04
SPEAKER_02: to the extent that it makes people feel better.
36:05
SPEAKER_02: Great.
36:06
SPEAKER_02: Right.
36:07
SPEAKER_00: Um, is it great?
36:07
SPEAKER_00: This is what I'm wondering.
36:08
SPEAKER_00: I feel like,
36:09
SPEAKER_00: I feel like this is the,
36:10
SPEAKER_00: the,
36:10
SPEAKER_00: you know,
36:10
SPEAKER_00: anytime the aviation industry is like,
36:12
SPEAKER_00: oh,
36:12
SPEAKER_00: well,
36:12
SPEAKER_00: we're going to be,
36:13
SPEAKER_00: um,
36:14
more climate friendly it's it's it all if we're not doing
36:17
hydrogen it all revolves
36:19
around saps and then you're not the only person i talk to who's
36:22
like this is mostly
36:23
bullshit so um i don't i don't know where that that leaves i
36:26
feel like that leaves
36:27
us at a pretty um you know pretty much the same maybe the
36:31
planes get more efficient
36:32
SPEAKER_00: and things like that but um otherwise
36:35
SPEAKER_02: I mean,
36:35
SPEAKER_02: I think the energy discourse has changed,
36:38
like the global energy discourse has changed drastically over
36:41
the last couple of
36:41
SPEAKER_02: years,
36:42
SPEAKER_02: again,
36:42
SPEAKER_02: with the advent of AI and data center energy demand and things like that.
36:48
So I think you don't hear people talk about the Paris Agreement
36:52
as much anymore and
36:54
SPEAKER_02: about actually trying to hit net zero or true zero by 2030.
36:59
SPEAKER_02: Because it clearly isn't gonna happen.
37:00
SPEAKER_02: Because we're in an arms race.
37:02
SPEAKER_02: It clearly isn't gonna happen.
37:03
And I think the conversation around climate change has veered
37:07
much more towards mitigation.
37:09
SPEAKER_02: Right.
37:09
SPEAKER_02: Resiliency and mitigation.
37:11
Um,
37:12
and,
37:13
uh,
37:13
and so that's the,
37:14
that's,
37:14
that's probably the reality we live in.
37:16
SPEAKER_00: Okay.
37:17
SPEAKER_00: Okay.
37:18
SPEAKER_00: All right.
37:18
SPEAKER_00: So, so universal hydrogen is, is winding down.
37:21
SPEAKER_00: Um, you've been thinking about AI for many, many years.
37:25
SPEAKER_00: You're a hardware head.
37:27
SPEAKER_00: You're an engineering head.
37:29
SPEAKER_00: I mean,
37:30
SPEAKER_00: I guess I can see in my own head,
37:31
SPEAKER_00: you know,
37:32
SPEAKER_00: how this idea began to formulate,
37:33
SPEAKER_00: but why don't you tell me the origin story of P1?
37:38
SPEAKER_02: Yeah, so the origin story goes way, way back.
37:41
SPEAKER_02: I mean,
37:42
SPEAKER_02: I grew up on hard sci-fi and reading the hard sci-fi,
37:49
SPEAKER_02: I was promised Dyson spheres and starships,
37:51
SPEAKER_02: right?
37:52
And so when the AI revolution started to really take off in the
37:57
early 2020s,
38:01
SPEAKER_02: I was like, I'm working on hydrogen airplanes.
38:02
SPEAKER_02: Great.
38:02
SPEAKER_02: You know, I'm doing my thing.
38:05
SPEAKER_02: But I am I'm excited that clearly,
38:07
SPEAKER_02: clearly something is really rapidly,
38:10
SPEAKER_02: you know,
38:10
SPEAKER_02: there's been a step change.
38:12
SPEAKER_02: It's going to rapidly accelerate.
38:14
And and so clearly the biggest impact of AI is going to be in
38:19
designing starships
38:20
SPEAKER_02: and Dyson spheres or at least the built world.
38:22
SPEAKER_02: Right.
38:23
SPEAKER_02: As a stepping stone to that.
38:24
SPEAKER_02: And I want to meet the people who are doing that.
38:27
SPEAKER_02: And I tried,
38:29
SPEAKER_02: except there aren't any people,
38:30
SPEAKER_02: and there weren't any people,
38:31
SPEAKER_02: and there still aren't very many who are doing that.
38:34
Even though you'd come from this world of DARPA and Google X
38:39
and all these places
38:40
SPEAKER_00: that would be populated with like-minded type folks,
38:44
SPEAKER_00: I would assume.
38:45
SPEAKER_02: That's right.
38:47
But the AI world was very much working on sort of chatbots and
38:52
better cat video generators.
38:54
SPEAKER_02: I'm pleased to say that over the last couple of years,
38:56
it's evolved much more towards medical diagnostics and drug
38:59
discovery and those
39:00
SPEAKER_02: kinds of things,
39:01
SPEAKER_02: software engineering,
39:02
but still very seldom do any of the frontier labs talk about
39:07
physical product
39:08
SPEAKER_02: engineering.
39:09
SPEAKER_02: So I started poking around and sort of being like, why?
39:12
I mean,
39:12
robotics is kind of the closest they seem to be getting on the
39:16
frontier lab side.
39:17
SPEAKER_02: Sure.
39:17
SPEAKER_02: Yes.
39:18
Robotics has has has been there and remains and maybe is the is
39:23
the immediate next
39:25
SPEAKER_02: next big focus area.
39:27
SPEAKER_02: But robotics doesn't help you build the built world.
39:29
SPEAKER_02: It helps you interact with the with the existing built world.
39:32
SPEAKER_02: Right.
39:33
SPEAKER_02: But it is it is a it is a different thing in kind.
39:36
And so I asked the question of,
39:37
okay,
39:38
well,
39:38
like,
39:38
why isn't anybody working on this?
39:40
SPEAKER_02: Like,
39:41
clearly I'm not the only one that thinks that that is
39:43
potentially the biggest sort
39:44
of impact that AI is going to have on the human experiences
39:49
through the built
39:49
SPEAKER_02: world.
39:50
SPEAKER_02: And the answer was training data.
39:53
Like if you want an AI,
39:57
the enabler of this revolution in AI is very large training
40:02
data sets.
40:03
And the reason that we were getting chat bots and cat videos is
40:05
because there's a
40:06
SPEAKER_02: lot of written language and a lot of cat videos out there,
40:12
SPEAKER_02: right?
40:13
SPEAKER_02: And if you want an AI that can help you design an airplane,
40:17
SPEAKER_02: it needs to see a million airplane designs.
40:19
And there just haven't been a million airplane designs since
40:22
the Wright brothers.
40:22
SPEAKER_02: Yeah.
40:23
SPEAKER_00: Yeah, no.
40:26
SPEAKER_00: These were going to be my questions.
40:27
SPEAKER_00: I mean,
40:28
SPEAKER_00: so if you,
40:30
SPEAKER_00: the two things,
40:31
SPEAKER_00: as you mentioned,
40:32
SPEAKER_00: that have taken off first,
40:33
SPEAKER_00: obviously we kind of mathematically solve language.
40:36
SPEAKER_00: We spent the last 40 years vomiting our every thought onto the internet,
40:41
SPEAKER_00: you know,
40:42
SPEAKER_00: along with just these corpuses of knowledge that existed and then code,
40:47
SPEAKER_00: similar thing.
40:47
SPEAKER_00: We've got these huge repositories to lean on.
40:50
SPEAKER_00: These have been the things,
40:52
SPEAKER_00: that have worked so far.
40:53
SPEAKER_00: I was poking around you guys, your website also.
40:56
SPEAKER_00: Okay.
40:56
SPEAKER_00: I mean, there's not a lot on the website yet.
40:58
SPEAKER_00: It's awesome.
40:59
SPEAKER_00: It's like command line interface.
41:02
SPEAKER_02: I wrote it.
41:02
SPEAKER_02: I hand coded it lovingly over a weekend.
41:04
SPEAKER_00: Yeah.
41:05
SPEAKER_00: So at the moment, it's kind of bare bones.
41:07
SPEAKER_00: It's just a little information about the company,
41:09
but you do point to a couple of videos where you can see some
41:12
of these early Archie
41:14
SPEAKER_00: demos.
41:15
SPEAKER_00: And,
41:16
SPEAKER_00: um,
41:16
SPEAKER_00: yeah,
41:16
SPEAKER_00: I mean,
41:16
SPEAKER_00: so that as I was watching,
41:18
SPEAKER_00: going through them and I heard you talking about this a bit,
41:20
SPEAKER_00: um,
41:21
SPEAKER_00: I'm still not clear on where you would get the training data.
41:24
SPEAKER_00: So if you, if I'm going to, let me back this up for one second.
41:29
SPEAKER_00: So.
41:31
SPEAKER_00: and I will out myself as knowing some things, but not enough.
41:34
SPEAKER_00: I mean,
41:35
SPEAKER_00: today,
41:35
SPEAKER_00: if I wanna go design an airplane,
41:38
SPEAKER_00: I am firing up something like ANSYS or something like that,
41:41
SPEAKER_00: right?
41:42
SPEAKER_00: I mean, I'm trying to build like a computer model of this thing.
41:45
SPEAKER_00: I wanna get the materials that I'm gonna use.
41:49
SPEAKER_00: And we do have software that gives us,
41:52
SPEAKER_00: some of the properties of those materials.
41:54
And then obviously we get to computer simulations and we can
41:57
get kind of far,
41:58
right?
41:58
And so what I was trying to understand is if you AI all this,
42:03
what the next step looks like.
42:05
SPEAKER_00: Is that one right way to think about this?
42:09
SPEAKER_02: So I think you have to conceptualize physical product development,
42:14
SPEAKER_02: whether it's an airplane or some other piece of industrial hardware,
42:19
SPEAKER_02: as more of a socio-technical system.
42:21
SPEAKER_02: So there are humans and there are tools.
42:23
And there are things that humans are good at and there are
42:25
things that humans rely on tools for.
42:27
SPEAKER_02: So what are humans good at?
42:28
Humans are very good at taking massive requirements and
42:32
figuring out what are
42:33
SPEAKER_02: actually the key design drivers among those requirements.
42:37
postulating some some solutions or some approaches to meeting
42:41
those key design drivers.
42:43
SPEAKER_02: Right.
42:43
And then doing kind of a first order,
42:45
like here's what the answer is going to look like.
42:48
SPEAKER_02: And then knowing what tools are available,
42:50
what is the range of validity of those tools and what are the
42:53
inputs to those
42:54
SPEAKER_02: tools?
42:54
SPEAKER_02: Like how do you drive the tools?
42:56
SPEAKER_02: Right.
42:56
SPEAKER_02: Yeah.
42:57
SPEAKER_02: And so we are
43:00
SPEAKER_02: focus squarely on that cognitive automation.
43:03
SPEAKER_02: So what I just described on doing what the human does,
43:06
SPEAKER_02: because there's no point in competing with the tools.
43:09
SPEAKER_02: The tools are there.
43:10
SPEAKER_02: The human uses the tools.
43:11
SPEAKER_02: Archie should use the tools.
43:13
SPEAKER_02: So we don't try to replace any tool.
43:15
SPEAKER_02: We don't try to encompass, supplant, compete with, make any tool better.
43:20
SPEAKER_02: The tools are what they are.
43:22
SPEAKER_02: There are a lot of them.
43:23
SPEAKER_02: So at Airbus,
43:24
SPEAKER_02: I think we had seven to 800 active engineering tools in a new program,
43:29
SPEAKER_02: and the whole ecosystem was several thousand engineering tools.
43:34
SPEAKER_02: So it is a very complex ecosystem.
43:37
SPEAKER_02: Some of those tools are Fortran that goes back decades.
43:40
SPEAKER_02: Some of them are very complex, modern,
43:42
SPEAKER_02: sort of design software.
43:46
SPEAKER_02: And so the best thing that the agent can do,
43:49
SPEAKER_02: the AI agent is learn what are the tools,
43:52
SPEAKER_02: when to apply them and how to apply them.
43:54
And then focus on that upstream reasoning piece that we still
43:59
require thousands and
44:01
SPEAKER_02: tens of thousands of humans in a large industrial OEM.
44:06
SPEAKER_02: uh, to do.
44:07
SPEAKER_02: So, okay.
44:08
SPEAKER_00: So if I'm, I'm, I want to be a, and I'm picking something totally random.
44:13
SPEAKER_00: So, you know, this may or may not be a good place to start.
44:15
SPEAKER_00: I, I am a rocket startup, huge debate inside the company.
44:20
Are we going to build this thing out of carbon fiber or are we
44:23
going to build it
44:24
SPEAKER_00: out of,
44:25
SPEAKER_00: um,
44:26
SPEAKER_00: aluminum,
44:27
SPEAKER_00: stainless steel?
44:28
SPEAKER_00: Um,
44:31
SPEAKER_00: At some point, yeah, you've got all these trade-offs in these materials.
44:34
SPEAKER_00: I mean, this is like an active debate that goes on today.
44:37
SPEAKER_00: And it is something that is not solved.
44:40
And people put their thesis forward as to why one course is
44:43
better than the other.
44:44
SPEAKER_00: So where would you plug in to this part of the process?
44:49
SPEAKER_02: So initially we probably wouldn't, right?
44:52
That's probably an architectural decision that a chief engineer
44:57
at the company
44:58
SPEAKER_02: would ultimately make,
44:59
SPEAKER_02: right?
44:59
SPEAKER_02: And some of it may end up being data-driven,
45:01
SPEAKER_02: some of it may end up being a judgment call or human bias,
45:05
SPEAKER_02: right?
45:06
SPEAKER_02: That isn't the low-hanging fruit for Archie.
45:10
Eventually,
45:11
of course,
45:11
we do expect and want to develop,
45:14
sort of move up the
45:16
the engineering expertise scale,
45:18
but the low hanging fruit is doing what junior engineers do.
45:20
SPEAKER_02: OK.
45:21
And junior engineers do a pretty rote sort of application of
45:26
the process that I
45:27
SPEAKER_02: just described,
45:27
SPEAKER_02: right,
45:28
SPEAKER_02: which is requirements,
45:29
SPEAKER_02: pick out key design drivers,
45:30
SPEAKER_02: do a first order sizing.
45:32
And that sizing may not be of like the whole airplane or the
45:34
whole rocket engine.
45:35
SPEAKER_02: You may be sizing a left bracket kind of thing.
45:38
SPEAKER_02: Right.
45:38
SPEAKER_02: Or you may be sizing a pump or something else.
45:42
SPEAKER_02: Right.
45:42
SPEAKER_02: So so it's kind of a fractal.
45:43
SPEAKER_02: Everybody does those things.
45:44
SPEAKER_02: Right.
45:45
SPEAKER_02: But but it's kind of fractal.
45:46
So you may be doing it at a different different level of
45:48
abstraction within the
45:50
SPEAKER_02: system,
45:51
SPEAKER_02: the system design tree.
45:53
SPEAKER_02: And and then and then you you you have postulate a couple of solutions.
45:58
SPEAKER_02: Right.
45:59
SPEAKER_02: You do a first order sizing and then you go do you refine the the design.
46:04
SPEAKER_00: So I'm picking a pump for a certain job and Archie is helping me
46:10
Argy's helping me through this problem to size this thing right
46:14
and pick the right product.
46:17
SPEAKER_02: That's right.
46:18
SPEAKER_02: And can go do a bunch of design trade-offs and say,
46:20
SPEAKER_02: OK,
46:20
SPEAKER_02: actually,
46:21
SPEAKER_02: can you build me a matrix and try varying this versus that?
46:25
SPEAKER_02: And oh,
46:26
SPEAKER_02: by the way,
46:26
SPEAKER_02: we have a supplier that we are in a contract dispute with,
46:29
SPEAKER_02: so can we not use that supplier?
46:30
SPEAKER_02: Make sure that doesn't factor in and give me the minimum cost version,
46:34
SPEAKER_02: give me the minimum schedule version.
46:36
Those kinds of design explorations that you would typically go
46:39
assign to a
46:42
SPEAKER_02: relatively junior engineer
46:43
SPEAKER_02: That's what we're targeting as initial capability.
46:48
SPEAKER_02: And there's a huge market for that.
46:49
SPEAKER_00: Okay.
46:49
SPEAKER_00: Yeah.
46:49
SPEAKER_00: No, I saw the example on the website.
46:51
SPEAKER_00: I mean, you have an HVAC system.
46:53
SPEAKER_00: It has...
46:54
six seven core components is like we're gonna we're gonna
46:57
change something here
46:58
what does that do down here we want a certain amount of
47:01
performance out of it how
47:02
do we get that without um you know what are the changes we can
47:05
make to get that
47:06
performance without without um changing the motor or something
47:09
like that and and it
47:10
seemed to be advising on these these various trade-offs and the
47:13
cost benefits it
47:15
SPEAKER_00: was pretty cool i mean you could even like you said i mean you could
47:19
SPEAKER_00: get the actual parts list,
47:20
SPEAKER_00: see who supplied it,
47:21
SPEAKER_00: how much it costs,
47:23
SPEAKER_00: and then your options of what else you could buy.
47:26
SPEAKER_02: That's right.
47:26
And that was the toy,
47:27
sort of the toy problem that we set for ourselves in the
47:29
pre-seed.
47:30
SPEAKER_02: It wasn't really tied to any particular customer use case.
47:33
SPEAKER_02: We just picked a domain that was pretty multi-physics, right?
47:36
So you have refrigerant flows and air flows and thermal
47:38
interactions and electrical,
47:39
right?
47:40
It kind of has all the physics that a much more complex system
47:42
like an airplane
47:43
SPEAKER_02: would have,
47:44
SPEAKER_02: but at the,
47:45
SPEAKER_02: you know,
47:46
SPEAKER_02: 10,
47:46
SPEAKER_02: 15 component level,
47:47
SPEAKER_02: and the physics are pretty linearizable, right?
47:51
SPEAKER_02: So you're talking about like,
47:52
SPEAKER_02: if you're talking about airflow,
47:53
it's a one dimensional flow through a duct rather than a three
47:56
dimensional flow
47:56
SPEAKER_02: over a wing.
47:57
SPEAKER_02: So much simpler to analyze.
47:59
SPEAKER_02: So that was our toy problem.
48:01
SPEAKER_02: The first application domain for us is data center cooling systems.
48:04
SPEAKER_02: Oh, okay.
48:05
SPEAKER_02: Which there's a certain sort of poetic elegance, right?
48:08
SPEAKER_02: To AI helping design its own data center, right?
48:11
SPEAKER_00: Okay, all right, wait a minute.
48:12
SPEAKER_00: So you've gone from fighting,
48:16
SPEAKER_00: The good fight for climate change.
48:19
SPEAKER_00: So I guess you'll make us more efficient data centers.
48:23
SPEAKER_02: For sure, you can make more efficient data centers.
48:27
SPEAKER_02: Again, look, I think AI changes everything.
48:31
SPEAKER_02: It is really a technological discontinuity.
48:35
And so I think our approach to the whole climate conversation
48:38
needs to change.
48:39
SPEAKER_02: And appropriately, I think, is changing.
48:41
SPEAKER_00: So, okay.
48:42
SPEAKER_00: So if data center design is kind of the first step,
48:46
how do you actually get training data to make something like
48:49
that affect it?
48:51
SPEAKER_02: Yeah,
48:51
SPEAKER_02: so this is sort of our,
48:54
the novelty that we as P1AI bring to the table is the creation
48:59
of these large
48:59
synthetic training data sets for physical product domains that
49:04
are physics-based
49:05
SPEAKER_02: and supply chain informed.
49:07
SPEAKER_02: And the idea behind this goes back 15 years to my DARPA days, actually.
49:13
And so I joined DARPA right at the beginning of the Obama I
49:18
administration.
49:20
And this was a couple of years,
49:23
basically just before the Joint Strike Fighter was fielded.
49:27
And we were going through kind of all of the teething pains and
49:32
the long tail of
49:33
SPEAKER_02: the cost and schedule overruns that the JSF experienced.
49:37
SPEAKER_02: And the big issue at the Pentagon was breaking the cost curve, right?
49:41
SPEAKER_02: Because if you plot the cost of like the F-16,
49:44
SPEAKER_02: the F-22,
49:45
SPEAKER_02: the F-35,
49:46
SPEAKER_02: it's a very steep exponential.
49:48
SPEAKER_02: And the question was, how do you afford the next one?
49:50
SPEAKER_02: Like, what does this mean?
49:52
SPEAKER_02: JSF was the most expensive thing that the human species has ever built.
49:57
SPEAKER_02: Right.
49:58
SPEAKER_02: And was massive overrun, both in terms of schedule and budget.
50:02
SPEAKER_02: And so I came to DARPA with a very specific proposal,
50:06
which was we have like clearly the growth and cost is largely a
50:11
function of product
50:12
SPEAKER_02: complexity.
50:13
SPEAKER_02: Right.
50:13
SPEAKER_02: We're just
50:14
SPEAKER_02: putting a lot more and compactifying it in a very small space in terms of
50:19
SPEAKER_02: electronics,
50:20
SPEAKER_02: engines that perform at higher temperature points,
50:22
SPEAKER_02: right?
50:23
SPEAKER_02: There's just a lot more.
50:24
We expect more performance and that drives a significant growth
50:26
in complexity.
50:28
SPEAKER_02: And as a result of that complexity, you know,
50:31
SPEAKER_02: Historically,
50:31
the way we do physical systems is we build an engineering
50:35
organization that mirrors
50:36
SPEAKER_02: the product.
50:38
And if the product is more complex,
50:39
the engineering organization is bigger.
50:41
SPEAKER_02: There's a lot more interactions in the product.
50:43
SPEAKER_02: There's a lot more interactions in the engineering organization,
50:46
SPEAKER_02: and it doesn't scale well,
50:47
SPEAKER_02: right?
50:48
SPEAKER_02: Um, and, but where, where, where has it scaled?
50:52
SPEAKER_02: What's, what's our existence proof that it can scale better than that.
50:55
SPEAKER_02: And the existence proof for me was semiconductors.
50:57
SPEAKER_02: Okay.
50:58
SPEAKER_02: Is the advent of VLSI design and EDA electronic design automation,
51:02
SPEAKER_02: uh,
51:03
SPEAKER_02: in the eighties led that whole industry.
51:05
SPEAKER_02: Uh,
51:06
SPEAKER_02: and this was kind of from the 80,
51:08
SPEAKER_02: 88 to the three 86,
51:09
SPEAKER_02: you know,
51:09
SPEAKER_02: going by Intel,
51:10
SPEAKER_02: Intel processor numbers.
51:11
SPEAKER_02: Yeah.
51:12
SPEAKER_00: This is what this is like Carver meat and Lynn Conway.
51:14
SPEAKER_00: Yeah.
51:15
SPEAKER_02: Exactly.
51:15
SPEAKER_02: Yep.
51:16
SPEAKER_02: Conway and Mead VLSI design.
51:19
And the idea was that you can raise the level of abstraction on
51:22
the design process,
51:23
right?
51:23
SPEAKER_02: You can model the fab,
51:26
SPEAKER_02: right,
51:26
SPEAKER_02: with a physics library of like what could go,
51:28
SPEAKER_02: what can the fab make,
51:29
SPEAKER_02: what could go into your system.
51:31
And you have these rich multi-physics models of the constituent
51:34
components.
51:35
SPEAKER_02: And then you can assemble your system from that physics library.
51:39
And you would have a computable abstraction of your physical
51:43
hardware so that an
51:44
individual designer could deal with that abstraction rather
51:47
than dealing at the
51:48
SPEAKER_02: gate level with the hardware.
51:50
SPEAKER_00: So this is like Cadence for all of the hardware.
51:55
SPEAKER_00: That's right.
51:56
SPEAKER_02: And in fact,
51:57
SPEAKER_02: the founder of Cadence and Synopsis,
51:59
SPEAKER_02: who's a Berkeley professor,
52:01
SPEAKER_02: Alberto San Giovanni Vincentelli,
52:03
SPEAKER_02: was an advisor to the DARPA program that I started.
52:06
SPEAKER_02: But I came to DARPA with this idea of like,
52:07
SPEAKER_02: hey,
52:08
SPEAKER_02: let's take VLSI and try to apply it to a fighter jet.
52:12
SPEAKER_02: So make it very multi-physics,
52:15
SPEAKER_02: much larger systems,
52:16
SPEAKER_02: much more non-linear physics,
52:18
SPEAKER_02: much tighter coupling to the environment than you get in a semiconductor.
52:21
SPEAKER_00: Did we do that?
52:23
SPEAKER_02: So we did.
52:24
SPEAKER_02: I would say that the program was maybe 60, 70% successful.
52:27
SPEAKER_02: If a DARPA program is 100% successful, you didn't try hard enough, right?
52:32
SPEAKER_02: Like you weren't shooting high enough.
52:36
SPEAKER_02: And a lot of the stuff that we developed permeated the industry.
52:39
SPEAKER_02: A lot of it was open source.
52:40
SPEAKER_02: A lot of it permeated the industry.
52:42
SPEAKER_02: One of the more bizarre sort of surreal moments in my Airbus tenure
52:46
SPEAKER_02: was from the Airbus Business Intelligence Organization,
52:50
which collects like open source intelligence on what's
52:52
happening competitors,
52:53
SPEAKER_02: showing me a slide that Boeing briefed at a conference.
52:58
SPEAKER_02: And within it was like,
53:00
SPEAKER_02: verbatim copies of my my DARPA work.
53:03
SPEAKER_02: Yeah.
53:04
SPEAKER_02: And I was like, huh, this is very meta, right?
53:06
SPEAKER_02: Like,
53:07
SPEAKER_02: but it's but it really permeated,
53:08
SPEAKER_02: I think permeated permeated the industry and probably,
53:12
SPEAKER_02: frankly,
53:12
SPEAKER_02: you know,
53:12
SPEAKER_02: like I did a lot of cool stuff at Airbus,
53:15
but probably the longest lasting legacy was transitioning
53:20
Airbus engineering into a
53:21
SPEAKER_02: model based enterprise for the next airplane.
53:22
SPEAKER_02: OK.
53:23
SPEAKER_02: Right.
53:23
And using this kind of like compositional model based synthesis
53:26
to design to design
53:27
SPEAKER_02: the next airplane.
53:29
SPEAKER_02: So, okay.
53:29
SPEAKER_00: This helps me understand a lot.
53:31
SPEAKER_00: Um, how, okay.
53:33
SPEAKER_00: Well,
53:33
but then how on earth do you use synthetic data to get to
53:36
somewhere accurate on
53:38
SPEAKER_00: something like this?
53:39
SPEAKER_02: Right.
53:39
SPEAKER_02: So,
53:39
SPEAKER_02: so back to like,
53:40
SPEAKER_02: so,
53:41
SPEAKER_02: so,
53:41
SPEAKER_02: sorry,
53:41
SPEAKER_02: I told you the dark story,
53:42
SPEAKER_00: but I didn't really help me wrap my head around it.
53:45
SPEAKER_00: Yeah.
53:45
SPEAKER_02: Um, so that was 15 years ago, right?
53:47
SPEAKER_02: Like I said, those techniques, um,
53:49
SPEAKER_02: have permeated the industry to a significant extent,
53:54
SPEAKER_02: particularly sort of at the bleeding edge of complexity.
53:56
SPEAKER_02: So more aerospace, less industrial chillers.
54:01
And so when the question arose like,
54:03
hey,
54:03
so I need a million airplane designs.
54:05
SPEAKER_02: There haven't been a million airplane designs ever made.
54:08
SPEAKER_02: to train my AI model.
54:10
SPEAKER_02: I was like,
54:10
SPEAKER_02: well,
54:11
I kind of know how to make a million airplane designs that are
54:14
going to be physics
54:14
based and supply chain informed because if the physics library
54:19
that you're
54:19
composing your system out of is tied to what's available in the
54:23
supply chain,
54:24
then every design that you compose will sort of by definition
54:28
be realizable.
54:30
SPEAKER_02: It may never be realized.
54:31
SPEAKER_02: It may never have been realized like historically,
54:34
SPEAKER_02: but it should be theoretically if you wanted to go build it,
54:36
SPEAKER_02: you could build it.
54:38
SPEAKER_02: Now,
54:38
SPEAKER_02: some of those designs to train an AI model,
54:41
you actually want to sample densely around the area where most
54:46
of the work is going
54:47
SPEAKER_02: to be happening.
54:47
And most of the work that engineers do are incremental changes
54:50
to existing designs.
54:52
SPEAKER_02: So you want to sample very densely around those kind of dominant designs,
54:57
but you do also want to sample sparsely in the corners and
55:00
edges of the design
55:01
SPEAKER_02: space because you learned something.
55:03
SPEAKER_02: It's sort of like,
55:05
SPEAKER_02: you know,
55:05
SPEAKER_02: we train college students when when they study engineering,
55:09
SPEAKER_02: we really train them on synthetic,
55:10
SPEAKER_02: mostly on synthetic data.
55:11
SPEAKER_02: Like you may go build a one airplane,
55:15
SPEAKER_02: right,
55:15
SPEAKER_02: model airplane of some sort as part of your undergraduate education.
55:20
But you do a bunch of designs,
55:22
you do a bunch of practice designs,
55:23
and that's synthetic data.
55:24
And some of those designs might be wacky and intentionally
55:27
wacky because that
55:28
teaches you about a corner of the design space that may be a
55:31
terrible idea.
55:32
SPEAKER_02: Right.
55:32
SPEAKER_02: But it teaches you why it's a terrible idea.
55:34
SPEAKER_02: And so you get sort of like the gradients of of why it's a terrible idea.
55:37
And so so sampling like which million airplane designs you you
55:42
choose to make using
55:43
these compositional model based synthesis tools is actually
55:46
pretty tricky,
55:47
SPEAKER_02: pretty tricky questions.
55:47
SPEAKER_02: And that's a lot of what we are
55:50
SPEAKER_02: we're working on at P1AI.
55:52
SPEAKER_02: So you have smart humans guiding that process.
55:56
SPEAKER_02: Sure.
55:56
The synthetic data generation process has a fair degree of
56:00
human involvement.
56:01
We're working on trying to automate some of that human
56:04
involvement,
56:05
but that's not a bad place.
56:06
SPEAKER_02: Like we're not shy to have human involvement in that process.
56:09
SPEAKER_02: Think of it as a professor guiding a student.
56:12
SPEAKER_00: I didn't mean it as a knocker.
56:15
SPEAKER_00: I'm trying to understand how you get good synthetic data.
56:18
SPEAKER_00: Yeah.
56:18
SPEAKER_00: Because it feels like you would want an expert in the loop.
56:23
SPEAKER_00: Absolutely.
56:24
And so we have a number of sort of domain subject matter
56:28
experts who are helping us
56:31
SPEAKER_02: figure out how do you sample the design space?
56:34
SPEAKER_02: Are we sampling it the right way?
56:36
SPEAKER_02: And sanity checking a lot of the designs that we produce.
56:38
SPEAKER_02: They can't sanity check like a million designs or 10 million designs,
56:42
SPEAKER_02: but they can sanity check some of the seed designs in the space.
56:45
And then there's like a reinforcement learning aspect to the
56:48
synthetic data on all
56:49
SPEAKER_00: those designs.
56:50
SPEAKER_00: Is that?
56:51
SPEAKER_02: So it's a good question.
56:54
SPEAKER_02: So the synthetic data itself, right, is kind of AI agnostic, right?
56:58
So it's just a data set of a million designs and a performance
57:01
vector for each design.
57:03
SPEAKER_02: Right.
57:03
SPEAKER_02: And by the way, the performance vector.
57:04
SPEAKER_02: So we simulate each design to make sure that it's a feasible.
57:07
SPEAKER_02: It is, in fact, a physically realizable design.
57:10
SPEAKER_02: And we we have a performance vector so you can train on a design and the
57:13
SPEAKER_02: performance of that design.
57:14
SPEAKER_02: Right.
57:15
Because the design process is a mapping from the design space
57:18
to the performance
57:19
SPEAKER_02: space or from the performance space to the design space.
57:23
SPEAKER_02: And and the good news,
57:24
SPEAKER_02: by the way,
57:24
SPEAKER_02: is that you can use fairly low fidelity simulation for generating this,
57:29
SPEAKER_02: because,
57:29
SPEAKER_02: again,
57:29
SPEAKER_02: we're trying to replicate
57:32
SPEAKER_02: we're trying to do cognitive automation,
57:34
SPEAKER_02: replicate the human cognition,
57:36
SPEAKER_02: and that is low fidelity.
57:37
SPEAKER_02: So a human does not solve the Navier-Stokes equations
57:41
SPEAKER_02: to model the flow over a wing,
57:43
SPEAKER_02: the airflow over a wing,
57:44
humans have a fairly low order surrogate model for how to
57:48
estimate lift on a wing
57:49
SPEAKER_02: or drag on a wing.
57:51
And so we only need to be at that level because we know that
57:54
there will then be a
57:55
SPEAKER_02: Navier-Stokes solver to do the detailed design.
57:58
SPEAKER_02: So we produce this synthetic data set.
58:00
SPEAKER_02: And so to your question about reinforcement learning,
58:03
SPEAKER_02: then the question is,
58:03
SPEAKER_02: what do you do with it?
58:05
SPEAKER_02: If you just throw a million airplane designs at an LLM,
58:09
SPEAKER_02: it does not magically become a good engineer.
58:12
And so the AI architecture that you train is today a federated
58:19
set of models that
58:21
SPEAKER_02: are good at individual parts of what comprises engineering reasoning,
58:25
SPEAKER_02: right?
58:25
SPEAKER_02: Because there is spatial reasoning, there is performance estimation.
58:29
SPEAKER_02: There is the multi-physics aspect of it.
58:31
SPEAKER_02: And so we have dedicated models for all of that.
58:33
SPEAKER_02: And then we do use a big LLM as a reasoner,
58:36
SPEAKER_02: kind of as a front end that knows about all of those other neural models,
58:41
SPEAKER_02: as well as all of the non-neural tools that a human would use.
58:45
And you're basing this off of a like an open source
58:49
foundational model for that part.
58:51
SPEAKER_02: That's right.
58:51
We do use open weights models because our customers don't
58:54
generally want their
58:56
proprietary because eventually so so we build synthetic data
58:58
sets to to to to
59:01
SPEAKER_02: initially train the model.
59:02
SPEAKER_02: But of course,
59:02
it gets better as it encounters real data from from a customer
59:07
and the customer
59:08
SPEAKER_02: data we want the customer to have complete control over.
59:11
SPEAKER_02: And so for that reason, we do use open weight models that we self host.
59:15
And I,
59:15
and I'm sure,
59:17
so you're starting kind of with this low level engineer type
59:20
jobs.
59:21
SPEAKER_00: I'm sure there's,
59:22
SPEAKER_00: you know,
59:22
SPEAKER_00: a giant path where this could go on the,
59:25
SPEAKER_00: the,
59:26
SPEAKER_00: if I was going to go in a data center design is part of this.
59:29
It sounds to me like part of this is,
59:31
is it's helping with tasks much in the same way.
59:37
It sounds like with coding,
59:39
um,
59:39
you kind of having the AI take on some tasks that a human
59:43
normally does and then that's useful for all sorts of reasons
59:47
and the humans can
59:48
SPEAKER_00: focus on some other stuff.
59:49
SPEAKER_00: Is it also like an iteration thing?
59:51
SPEAKER_00: I mean,
59:52
if I'm thinking about building the new Stargate and I just want
59:56
this in the most
59:57
SPEAKER_00: innovative way that this AI can,
1:00:00
SPEAKER_00: because you've trained it on all these models,
1:00:02
SPEAKER_00: that it can just run through a lot of design choices,
1:00:04
a human doesn't even have time or the inclination to think of
1:00:08
or are we not,
1:00:09
SPEAKER_00: is that further down the line?
1:00:10
SPEAKER_02: No, for sure.
1:00:11
SPEAKER_02: Right.
1:00:11
SPEAKER_02: And it is faster than a human.
1:00:14
SPEAKER_02: We what we haven't quantified is the error rate.
1:00:17
SPEAKER_02: Right.
1:00:17
SPEAKER_02: And that's next for us is we're going to put this out in the wild.
1:00:20
SPEAKER_02: And by in the wild,
1:00:21
I mean in an actual engineering workforce with one of the
1:00:25
industrial OEMs that does
1:00:27
SPEAKER_02: data center cooling systems.
1:00:28
SPEAKER_02: So these are industrial chillers, basically, and let them use it.
1:00:32
SPEAKER_02: And we want to quantify the error rate.
1:00:34
SPEAKER_02: Okay.
1:00:35
SPEAKER_00: Because that's my big it's
1:00:36
SPEAKER_00: Still my massive question around AI,
1:00:38
and I don't want this to sound as cynical as it does because I
1:00:42
sort of feel like
1:00:42
SPEAKER_00: we're in such early innings with this technology.
1:00:45
SPEAKER_00: But it's like, you know, the AI can write.
1:00:47
SPEAKER_00: I mean, sometimes Claude writes really nice things.
1:00:49
I've seen Sam Altman demoing a new version of OpenAI's writing
1:00:56
tools that seems
1:00:57
SPEAKER_00: much more sophisticated.
1:00:58
But,
1:00:58
you know,
1:00:59
when I ask it to write something,
1:01:00
for me,
1:01:01
like on one hand,
1:01:02
it's miraculous.
1:01:03
SPEAKER_00: It's like, oh my God, you can do...
1:01:07
SPEAKER_00: asking you to combine Shakespeare and Hunter S.
1:01:09
SPEAKER_00: Thompson and describe this bar of chocolate and you can do it,
1:01:12
SPEAKER_00: right?
1:01:13
SPEAKER_00: So hats off.
1:01:13
SPEAKER_00: When I ask it to write something for me, I wouldn't publish it, right?
1:01:18
SPEAKER_00: I'm still like, no, that's not, I feel like I write better than that.
1:01:22
SPEAKER_00: It sort of lets you down.
1:01:23
Robotics,
1:01:24
it's like does awesome stuff,
1:01:25
kind of lets you down a little bit.
1:01:26
SPEAKER_00: Coding does awesome stuff.
1:01:29
But if you talk to the world's best coders,
1:01:31
they're like,
1:01:31
no,
1:01:32
it's not replicating what I do.
1:01:34
SPEAKER_00: And I know you said where you're starting,
1:01:37
It feels like we're early days with this technology and there's
1:01:41
just a lot of
1:01:41
SPEAKER_00: belief that we should keep throwing GPUs,
1:01:45
SPEAKER_00: we keep improving this technology,
1:01:46
we do get to some new level where it isn't making these
1:01:51
mistakes and feeling still
1:01:52
SPEAKER_00: a bit rudimentary.
1:01:54
SPEAKER_02: I think coding is the best analog, right?
1:01:57
SPEAKER_02: And and sure,
1:01:58
SPEAKER_02: you know,
1:01:58
I don't I don't think anybody claims that the AI co-pilots or
1:02:02
agent or the more
1:02:03
autonomous agents like Devon fully replace every every kind of
1:02:07
software developer.
1:02:09
SPEAKER_02: But it is it is seems incontrovertible,
1:02:11
SPEAKER_02: right,
1:02:11
SPEAKER_02: that they are certainly replacing the low end,
1:02:14
SPEAKER_02: the low end of that profession.
1:02:16
SPEAKER_02: And so if and if you think about it, you know,
1:02:19
SPEAKER_02: So developing a semiconductor is like coding in VHDL or Verilog, right?
1:02:23
Those are languages where you represent the hardware and the
1:02:26
physics of what
1:02:27
SPEAKER_02: happens in the microprocessor.
1:02:29
SPEAKER_02: And in our case,
1:02:30
we have similarly sort of a language programmatic
1:02:33
representation of a physical
1:02:34
SPEAKER_02: system with all of the different physics domains.
1:02:36
SPEAKER_02: Again, air flows, fluid flows, electrical, thermal.
1:02:40
SPEAKER_02: et cetera.
1:02:41
SPEAKER_02: And,
1:02:41
SPEAKER_02: and so it,
1:02:42
it doesn't seem all that crazy that it should be able to write
1:02:46
programmatic
1:02:46
SPEAKER_02: representations of physical systems as well as writing,
1:02:49
SPEAKER_02: you know,
1:02:49
SPEAKER_02: programmatic representations of algorithms.
1:02:51
SPEAKER_02: Okay.
1:02:51
SPEAKER_02: Right.
1:02:52
SPEAKER_02: It doesn't seem like such a great leap.
1:02:53
SPEAKER_00: Okay.
1:02:54
SPEAKER_00: Okay.
1:02:54
SPEAKER_00: Let's get weird.
1:02:55
SPEAKER_00: I mean,
1:02:55
SPEAKER_00: you were saying,
1:02:57
SPEAKER_00: you know,
1:02:58
SPEAKER_00: when I asked you about the airplane question of,
1:03:00
SPEAKER_00: of what these look like,
1:03:01
if it sounded like where you were going to go was that do you
1:03:05
think if these tools
1:03:06
SPEAKER_00: get good enough,
1:03:06
SPEAKER_00: they start to,
1:03:07
SPEAKER_00: uh,
1:03:08
think of things we really haven't thought of and start pushing
1:03:11
the field.
1:03:11
SPEAKER_00: I mean, expand that.
1:03:13
SPEAKER_00: I know you, you were talking about starship.
1:03:14
SPEAKER_00: You've been, this has been like quest since your childhood.
1:03:18
SPEAKER_00: I mean,
1:03:18
SPEAKER_00: you,
1:03:18
you wanted a craft where we would be exploring through the
1:03:22
solar system and the
1:03:23
SPEAKER_00: universe.
1:03:23
SPEAKER_00: Right.
1:03:23
SPEAKER_00: I mean, this, this is like at your, at your core.
1:03:26
SPEAKER_02: Yeah,
1:03:26
SPEAKER_02: I mean,
1:03:27
SPEAKER_02: like I said,
1:03:28
I grew up on hard sci-fi and I'm just excited that I think that
1:03:34
starships are
1:03:34
SPEAKER_02: within the realm of attainability during my lifetime.
1:03:38
SPEAKER_00: Yeah, I mean, so let's do the thought exercise.
1:03:40
SPEAKER_00: I mean,
1:03:40
SPEAKER_00: it sounded,
1:03:41
SPEAKER_00: tell me when I was asking that question and you're like,
1:03:43
SPEAKER_00: I could give you two answers.
1:03:44
One is the non-AI answer and one is the AI answer of where
1:03:49
someone like a Boeing
1:03:50
SPEAKER_00: and Airbus is going.
1:03:52
SPEAKER_00: What does that AI answer look like to you?
1:03:56
SPEAKER_02: Yeah, by the way, I think I undersold the non-AI answer.
1:03:59
SPEAKER_02: I do think that there will be cool things in the next airplane.
1:04:03
SPEAKER_02: I think it'll be single pilot capable, for instance.
1:04:05
SPEAKER_02: I think there'll be incremental degrees of autonomy,
1:04:10
SPEAKER_02: better cabin experiences,
1:04:12
SPEAKER_02: more efficient engines,
1:04:13
SPEAKER_02: et cetera.
1:04:14
SPEAKER_02: Just for the record.
1:04:17
But yeah,
1:04:18
I think there could be there could be a complete discontinuity.
1:04:20
SPEAKER_02: I think it's difficult to see past that discontinuity.
1:04:25
SPEAKER_02: But but it does seem plausible,
1:04:29
SPEAKER_02: right,
1:04:30
that you get you get a super intelligence that can help create
1:04:36
things that we
1:04:37
SPEAKER_02: cannot create today.
1:04:39
SPEAKER_02: Right.
1:04:39
SPEAKER_02: Design engineering systems that we cannot design today.
1:04:43
SPEAKER_02: And that's the promise, right?
1:04:44
SPEAKER_02: That's the exciting part.
1:04:45
SPEAKER_00: You want me to tell you what they look like?
1:04:47
SPEAKER_00: No, no, like how confident are you on that?
1:04:50
SPEAKER_00: Okay, let's take rockets, right?
1:04:51
SPEAKER_00: I mean,
1:04:53
SPEAKER_00: SpaceX has done amazing things,
1:04:54
but not so much really because it rethought how the rocket is
1:04:58
designed in mass,
1:04:59
SPEAKER_00: I mean,
1:04:59
SPEAKER_00: reusability,
1:05:00
SPEAKER_00: yes.
1:05:00
SPEAKER_00: But otherwise,
1:05:01
SPEAKER_00: you know,
1:05:01
the physics of when the Falcon 9 first came out is like a lot
1:05:05
of rockets that came
1:05:06
SPEAKER_00: before it because you have this tiny margin to play with.
1:05:09
SPEAKER_00: We know what happens when you put...
1:05:12
SPEAKER_00: liquid oxygen and kerosene together.
1:05:15
SPEAKER_00: And there's only so much you can do, right?
1:05:17
SPEAKER_00: The tube has to be, the tube has to do certain things.
1:05:21
SPEAKER_00: The fuel can only do so much.
1:05:22
SPEAKER_00: And so an AI comes along.
1:05:25
SPEAKER_00: I mean,
1:05:25
is it that much smarter that then it's got like a rocket design
1:05:30
that we just never
1:05:31
SPEAKER_00: even thought of before?
1:05:32
SPEAKER_00: Or aren't we kind of constrained by physics on a lot of this?
1:05:37
SPEAKER_00: I don't know.
1:05:40
SPEAKER_00: What do you think?
1:05:40
I think that there are other ways of getting to orbit than
1:05:44
chemical reactive propulsion.
1:05:46
SPEAKER_00: Okay, so I shouldn't fixate on the rocket.
1:05:47
SPEAKER_00: For instance, right?
1:05:48
SPEAKER_02: I mean, I'm giving you examples.
1:05:49
SPEAKER_02: Obviously, I am not super intelligent, right?
1:05:51
SPEAKER_02: I am just moderately intelligent.
1:05:53
SPEAKER_02: So I can only conjecture, right?
1:05:57
SPEAKER_02: But one direction,
1:05:59
and I know you and a previous guest talked a little bit about
1:06:03
non-traditional
1:06:04
SPEAKER_02: propulsion technologies,
1:06:06
SPEAKER_00: right?
1:06:07
SPEAKER_02: The kinetic stuff to me doesn't compile.
1:06:09
SPEAKER_02: We looked at a lot of this at DARPA
1:06:11
SPEAKER_02: in my time at DARPA.
1:06:13
SPEAKER_02: The kinetic stuff doesn't quite compile for me, right?
1:06:16
Because you basically have to be at orbital velocity,
1:06:18
but like at sea level.
1:06:21
SPEAKER_02: And so you get, first of all, you have very high Gs, right?
1:06:24
SPEAKER_02: To even get to orbital velocity in a very short period.
1:06:27
SPEAKER_02: But then you're moving at Mach 20.
1:06:30
SPEAKER_02: in like sea level atmosphere.
1:06:33
SPEAKER_02: So that's not really great thermally.
1:06:35
SPEAKER_02: And so,
1:06:36
and then you have to circularize or you have to have an
1:06:38
injection burn also unclear
1:06:40
SPEAKER_02: how you do that.
1:06:40
SPEAKER_02: So you have a little motor,
1:06:42
SPEAKER_02: but then your rocket motor has to survive some unfathomable Gs,
1:06:45
SPEAKER_02: right?
1:06:45
SPEAKER_02: There's a lot of issue unless you're throwing slugs into space, right?
1:06:49
SPEAKER_02: There are a lot of issues with that, but beamed propulsion
1:06:53
I think does have a lot of promise and and hasn't had a huge
1:06:56
amount of investment into it.
1:06:59
So we're talking about a laser beam or a microwave beam where
1:07:01
you can accelerate.
1:07:03
SPEAKER_02: You can have a gradual acceleration profile.
1:07:05
SPEAKER_02: Right.
1:07:06
There are questions about focusing the beam and atmospherics
1:07:09
and birds and all
1:07:11
SPEAKER_02: sorts of other things.
1:07:11
SPEAKER_02: Right.
1:07:12
SPEAKER_02: So there are a lot of engineering challenges, which is why it hasn't been
1:07:16
SPEAKER_02: scaled up, but conceptually it makes a lot of sense, right?
1:07:20
And any propulsion mechanism where you can leave the big energy
1:07:23
source on the
1:07:24
SPEAKER_02: ground,
1:07:25
SPEAKER_02: right,
1:07:26
SPEAKER_02: for sure is appealing.
1:07:29
SPEAKER_02: And so that could be a direction, right?
1:07:32
Certainly you could imagine fission and fusion as alternative
1:07:36
energy sources for
1:07:37
SPEAKER_02: propulsion,
1:07:38
SPEAKER_02: right?
1:07:38
So I think there are many directions that have been limited by
1:07:41
just our capacity
1:07:43
SPEAKER_02: for engineering as a species.
1:07:45
SPEAKER_00: Right.
1:07:45
SPEAKER_00: Okay.
1:07:45
SPEAKER_00: So I was kind of approaching this from the wrong direction.
1:07:48
SPEAKER_00: It's, it's not do the same thing better.
1:07:50
SPEAKER_00: It's do new, new things.
1:07:52
SPEAKER_00: Um, okay.
1:07:54
SPEAKER_00: Give me your,
1:07:54
SPEAKER_00: like things go as well as they possibly could,
1:07:58
SPEAKER_00: you know,
1:07:58
SPEAKER_00: both for,
1:07:59
SPEAKER_00: for you and your vision with this company and also AI.
1:08:04
SPEAKER_00: I mean, this is total this is going back to your sci fi roots.
1:08:07
SPEAKER_00: I mean, seriously, like, what do you think 20 years from now?
1:08:10
SPEAKER_00: I don't know.
1:08:11
SPEAKER_00: Pick your number 20, 40 years.
1:08:12
SPEAKER_00: I don't care.
1:08:13
SPEAKER_00: I think it's much less than that, much less than that.
1:08:15
SPEAKER_00: And what and what the material world is is drastically changing.
1:08:21
SPEAKER_02: I think it's drastically different.
1:08:23
SPEAKER_02: Yeah, I think.
1:08:25
SPEAKER_02: And look,
1:08:26
I think the the key to it ultimately is going to be a recursive
1:08:29
self-improvement of
1:08:29
SPEAKER_02: these of these models.
1:08:32
SPEAKER_02: And there are some early and promising results.
1:08:36
And there are big questions about how do you steer that
1:08:38
recursive self-improvement
1:08:39
so that the models don't get away from you and learn Asimov's
1:08:45
laws of robotics,
1:08:47
SPEAKER_02: which I think are important and are well thought out in their time.
1:08:52
SPEAKER_02: But I see no theoretical barrier to an AI that can self-improve.
1:08:59
SPEAKER_02: And once it can do that,
1:09:00
SPEAKER_02: you will hit a very steep,
1:09:02
SPEAKER_02: like a much steeper exponential than the one we are currently on,
1:09:05
SPEAKER_02: right,
1:09:05
SPEAKER_02: in terms of the capabilities of those models.
1:09:07
SPEAKER_02: And you think it's much sooner?
1:09:09
SPEAKER_02: It's hard to predict,
1:09:10
SPEAKER_02: but I think at the pace,
1:09:13
SPEAKER_02: at the current sort of human-driven pace of model self-improvement,
1:09:16
SPEAKER_02: maybe it's five,
1:09:17
SPEAKER_02: 10,
1:09:18
SPEAKER_02: 15 years away,
1:09:19
SPEAKER_02: right?
1:09:20
SPEAKER_02: But once you hit that self-improvement moment,
1:09:24
SPEAKER_02: I think the slope of your exponential changes drastically.
1:09:28
so our previous guest was zach abrams and he comes from the
1:09:31
fintech world and he
1:09:32
was talking about i don't know we could be on the verge of the
1:09:35
wealthiest being on
1:09:37
SPEAKER_00: planet earth being an ai that's started a company yeah i mean
1:09:43
SPEAKER_00: Unless I'm wrong.
1:09:44
SPEAKER_00: I mean, I'm kind of here.
1:09:45
SPEAKER_00: It sounds like you think,
1:09:46
SPEAKER_00: I mean,
1:09:47
SPEAKER_00: maybe the next great airplane company is started and run by an AI almost.
1:09:53
SPEAKER_00: I mean, it could be, right?
1:09:56
SPEAKER_02: I think the,
1:09:57
SPEAKER_02: you know,
1:09:58
SPEAKER_02: look,
1:09:59
SPEAKER_02: again,
1:09:59
I think that there will be this moment that Vinji called the
1:10:03
singularity beyond
1:10:04
which it's almost impossible because it is going to be so
1:10:08
sensitive to sort of
1:10:09
SPEAKER_02: initial conditions,
1:10:10
SPEAKER_02: right?
1:10:10
SPEAKER_02: It's almost impossible to predict what is on the other side.
1:10:14
SPEAKER_02: But but I think we're going to live through it.
1:10:16
And I think that there is an opportunity for us for sure to try
1:10:19
and I think it's inevitable.
1:10:20
SPEAKER_02: Right.
1:10:20
SPEAKER_02: The question is, when when does it happen?
1:10:24
And and can we make fine course adjustments as we go through it
1:10:28
to ensure that what
1:10:29
SPEAKER_02: emerges on the other side is,
1:10:30
SPEAKER_02: in fact,
1:10:31
SPEAKER_02: a cornucopia utopia?
1:10:34
SPEAKER_02: versus we all get turned into paperclips.
1:10:37
SPEAKER_00: So I don't ask everyone this,
1:10:38
SPEAKER_00: but you seem like the kind of person that has thought about these things.
1:10:42
SPEAKER_00: What is the ultimate goal of all this?
1:10:45
SPEAKER_00: When you were thinking about Starship as a kid,
1:10:48
I suspect you were thinking probably more about human
1:10:52
exploration of the solar
1:10:54
SPEAKER_00: system and things like that.
1:10:56
SPEAKER_00: Is that the goal of all this, or is it just to push...
1:11:01
SPEAKER_00: some intelligence that we helped create out into the universe?
1:11:06
SPEAKER_00: I don't know.
1:11:06
SPEAKER_00: What would you?
1:11:06
SPEAKER_02: It's a deep, deep philosophical question, right?
1:11:12
SPEAKER_02: Well, first of all, I am a big fan of von Neumann probes also.
1:11:15
SPEAKER_02: Right.
1:11:15
SPEAKER_02: So, yes, I love I love the vision of generation starships.
1:11:19
SPEAKER_02: Right.
1:11:19
SPEAKER_00: What is what's a von Neumann probe?
1:11:21
SPEAKER_02: A von Neumann probe is a self-replicating spacecraft, right?
1:11:25
SPEAKER_02: That can sort of perpetuate itself, travel, right?
1:11:29
SPEAKER_02: Then reproduce and build newer versions of itself.
1:11:32
SPEAKER_00: And I assume von Neumann actually came up with that.
1:11:35
SPEAKER_00: He came up with a lot of things.
1:11:37
SPEAKER_01: Okay.
1:11:38
SPEAKER_01: Okay.
1:11:39
SPEAKER_01: Okay.
1:11:39
SPEAKER_00: Okay.
1:11:39
SPEAKER_00: Wait, so sorry, I cut you off.
1:11:41
SPEAKER_00: So you're into that.
1:11:42
SPEAKER_00: So that would be the sort of spreading intelligence through the...
1:11:45
sure and is that human intelligence right or is it is it
1:11:48
intelligence that humans
1:11:49
help build yeah it's an interesting question yeah well i've
1:11:56
this is a philosophical
1:11:57
i mean we could sit here with beers for many hours talking
1:12:00
about this stuff i i
1:12:01
feel like you know the last i don't know i feel like most of
1:12:05
the people who maybe
1:12:06
SPEAKER_00: grew up watching um
1:12:08
SPEAKER_00: the Apollo program and it all got into space and NASA and all these,
1:12:12
SPEAKER_00: you know,
1:12:12
SPEAKER_00: when I was growing up,
1:12:13
SPEAKER_00: it tended to be human exploration of space.
1:12:16
SPEAKER_00: And that was the,
1:12:19
spirit of all this it was it was that part of us that wants to
1:12:22
explore and and and
1:12:24
i'm sure some of that still exists but i can feel it in my
1:12:26
bones right now that
1:12:27
we're kind of seeding you know mostly like can we can we send
1:12:33
some intelligence
1:12:33
that we at least had some part in um off in into the universe
1:12:38
that seems like the
1:12:38
SPEAKER_00: central quest to me that we're gravitating to
1:12:42
Sure,
1:12:42
and I think there's also a question of what does it mean to be
1:12:44
human,
1:12:45
right?
1:12:46
And perhaps an intelligence that's trained fundamentally on the
1:12:51
human experience,
1:12:51
SPEAKER_02: interacting with humans and on the human body of creative work,
1:12:55
SPEAKER_02: right?
1:12:57
SPEAKER_02: Maybe it is, yeah, maybe 4chan, right?
1:13:00
SPEAKER_02: Maybe that is a human intelligence, right?
1:13:02
SPEAKER_02: It's just non-biological.
1:13:03
SPEAKER_02: Or maybe with technologies like Neuralink,
1:13:07
SPEAKER_02: Right.
1:13:07
Maybe there is a hybrid of biological and non-biological human
1:13:12
intelligence.
1:13:12
SPEAKER_02: You're pro or con transhumanism?
1:13:14
SPEAKER_02: I'm pro almost all technology.
1:13:16
SPEAKER_00: Yeah.
1:13:16
SPEAKER_00: Yeah.
1:13:17
Yeah.
1:13:17
So you would,
1:13:18
if it got good enough,
1:13:19
you would consumer elective surgery at Neuralink?
1:13:23
SPEAKER_00: A hundred percent.
1:13:24
SPEAKER_00: Yeah.
1:13:24
SPEAKER_02: A hundred percent.
1:13:25
SPEAKER_00: What would you want to do with it?
1:13:27
SPEAKER_02: Well, I think I, like, I, I, I think
1:13:32
SPEAKER_02: I think the idea of of an uploaded intelligence is is very interesting.
1:13:36
SPEAKER_02: Right.
1:13:37
I think the idea of of potentially merging multiple
1:13:41
intelligences is also also very interesting.
1:13:45
SPEAKER_02: And I don't know, I would be I would be excited to explore.
1:13:48
SPEAKER_00: Yeah.
1:13:49
SPEAKER_00: What do you there's that Ilya quote where he's like,
1:13:52
SPEAKER_00: you know,
1:13:53
if you valued intelligence as the main as one of the main
1:13:57
properties of humans that
1:13:59
SPEAKER_00: got you excited those days.
1:14:01
SPEAKER_00: are done.
1:14:02
SPEAKER_00: Um, yeah, I find this one.
1:14:05
SPEAKER_00: Do you, do you get the spirit of what I'm saying?
1:14:07
SPEAKER_02: Well, I think so.
1:14:08
SPEAKER_02: But again,
1:14:09
SPEAKER_02: I think that,
1:14:10
SPEAKER_02: uh,
1:14:11
SPEAKER_02: that the,
1:14:12
that the artificial intelligence is said that we are creating
1:14:14
are not substantially
1:14:16
SPEAKER_02: less human,
1:14:16
SPEAKER_02: right?
1:14:16
Because they're trained on the same,
1:14:18
on the same things that we're trained on.
1:14:20
SPEAKER_00: No, no, no, yes.
1:14:20
SPEAKER_00: And if we're talking about some sort of merge,
1:14:22
SPEAKER_00: I mean,
1:14:22
SPEAKER_00: this is a little to the side of what I was thinking.
1:14:24
SPEAKER_02: But even if there's a bifurcation, right?
1:14:26
SPEAKER_02: You just have a biological and a non-biological version, right?
1:14:28
SPEAKER_02: But it's trained fundamentally on the same data set.
1:14:31
SPEAKER_00: I do struggle with this.
1:14:34
SPEAKER_00: I've got two teenage boys,
1:14:35
SPEAKER_00: and it's like,
1:14:36
SPEAKER_00: oh,
1:14:36
SPEAKER_00: you want them to do well in the world,
1:14:38
SPEAKER_00: and you get fooled into these.
1:14:40
SPEAKER_00: It's like, oh, he did well on the SAT.
1:14:43
SPEAKER_00: Good for you.
1:14:43
SPEAKER_00: And then I do totally feel like we're entering somewhere where it's like,
1:14:49
SPEAKER_00: Good luck.
1:14:49
SPEAKER_00: I mean,
1:14:52
SPEAKER_00: like intelligence seems like the last thing,
1:14:54
SPEAKER_00: one of the last things that you can sort of be openly.
1:15:00
SPEAKER_00: It's like one of those human,
1:15:01
SPEAKER_00: most of our human traits,
1:15:02
SPEAKER_00: you're really not supposed to like hold against each other these days,
1:15:05
SPEAKER_00: you know,
1:15:06
SPEAKER_00: and but intelligence is like OK to be smarter than so.
1:15:08
It's like,
1:15:08
oh,
1:15:08
I hired this person because they're smarter than the other one.
1:15:11
SPEAKER_00: You can you can hold that against people.
1:15:15
SPEAKER_00: I feel like all these little judgment calls are about to go away.
1:15:18
SPEAKER_02: What are you, what are you urging your kids to study?
1:15:21
SPEAKER_00: I battle with this a lot,
1:15:22
SPEAKER_00: you know,
1:15:23
SPEAKER_00: especially cause I'm in the oddly,
1:15:25
I'm in the fortunate position of getting to interview all these
1:15:28
people like you and
1:15:29
SPEAKER_00: all the people at the frontier labs.
1:15:31
SPEAKER_00: And, um, and we live in Silicon Valley.
1:15:34
SPEAKER_00: And so hopefully God, if anyone has some insight into it,
1:15:38
SPEAKER_00: But I'm at a loss.
1:15:39
SPEAKER_00: I mean,
1:15:39
he he's my oldest gravitated towards computer science and I'm
1:15:45
not going to stop him
1:15:46
SPEAKER_00: because it's great on so many levels.
1:15:48
SPEAKER_00: He's thinking through things.
1:15:49
SPEAKER_00: He's learning this stuff.
1:15:50
SPEAKER_00: But then I'm like, man, where does that?
1:15:52
SPEAKER_00: I don't know where that ends up for you.
1:15:54
SPEAKER_00: I feel like
1:15:58
I don't think I've ever heard anyone give satisfying advice on
1:16:00
this because it
1:16:01
SPEAKER_00: usually chalks up to you have to be adaptable.
1:16:03
SPEAKER_00: You've just got to go with the flow.
1:16:06
SPEAKER_00: Or human creativity, which AI will never replicate, which I don't buy.
1:16:10
SPEAKER_00: And some of this might be true and it's great to say,
1:16:13
SPEAKER_00: but it's like,
1:16:13
SPEAKER_00: how do you,
1:16:14
SPEAKER_00: I don't know,
1:16:15
SPEAKER_00: how do you actually impart that?
1:16:16
I mean,
1:16:17
you know,
1:16:17
and then for a while I was like,
1:16:18
Jesus,
1:16:19
you should just...
1:16:20
let's go farm you know and then and then now everybody's now
1:16:23
i'm seeing optimus and
1:16:24
some of these androids and i'm like oh that's gonna get pretty
1:16:28
good pretty quick i
1:16:29
think i don't know i thought our hands would would hold out a
1:16:33
little bit longer
1:16:34
than i think they're going to what would i mean sorry it's such
1:16:37
a stupid question
1:16:38
SPEAKER_00: like yeah for what would you tell a kid to do
1:16:42
SPEAKER_02: I don't think I have any good advice, right?
1:16:43
SPEAKER_00: I mean,
1:16:44
SPEAKER_00: I probably would say,
1:16:44
SPEAKER_02: frankly,
1:16:45
I probably would still at this juncture say computer science so
1:16:47
that you understand
1:16:48
SPEAKER_02: how these systems are built.
1:16:49
SPEAKER_00: Yeah.
1:16:49
SPEAKER_00: Right?
1:16:50
SPEAKER_00: Yeah.
1:16:50
SPEAKER_00: Yeah.
1:16:51
SPEAKER_00: Yeah.
1:16:51
SPEAKER_00: God, it's wild.
1:16:53
SPEAKER_00: Well, Paul, thank you.
1:16:54
SPEAKER_00: I want to be mindful of your time.
1:16:56
SPEAKER_00: You've been so...
1:16:57
SPEAKER_00: Generous.
1:16:57
SPEAKER_00: And so just to cap things off where you are with P1,
1:17:01
SPEAKER_00: it sounds like this year the goal is to get out with a customer,
1:17:04
SPEAKER_00: start testing the software,
1:17:07
and you're going to go data center chillers first and then just
1:17:12
keep pushing on.
1:17:13
yeah move on to other verticals right and eventually we want to
1:17:16
end up with
1:17:17
aerospace and defense and then starships and dyson spheres okay
1:17:20
okay well i'm
1:17:21
excited um i get the first ride on any this is part of the
1:17:25
contract you signed i
1:17:28
get to get off this place um thank you paul so much for your
1:17:31
time i appreciate it
1:17:32
SPEAKER_00: thank you lee
1:17:34
SPEAKER_00: I made a friend.
1:17:35
SPEAKER_00: You sure did.
1:17:36
SPEAKER_00: Is this like normal or did I do really well?
1:17:39
SPEAKER_00: He likes smart and kind people.
1:17:42
SPEAKER_00: Oh, well, there you go.
1:17:42
SPEAKER_00: That was the right answer.
1:17:44
SPEAKER_00: Thank you, Paul, so much for coming by.
1:17:45
SPEAKER_00: I appreciate it.
1:17:46
SPEAKER_00: My pleasure.
1:17:47
SPEAKER_00: Thanks, Ashley.
1:17:51
core memory is hosted by me ashley vance the show is produced
1:17:56
by david nicholson
1:17:58
and me our theme music is by james mercer and john sortland and
1:18:03
the show is edited
1:18:05
by the john sortland thank you all for listening as mentioned
1:18:10
the show is sponsored
1:18:12
SPEAKER_00: by you on ventures thanks