00:00:00
Okay, we are gathered here today
00:00:03
in holy unity, brothers and sisters, to
00:00:08
convene and discuss on this most holy
00:00:12
day, the day the All-In podcast drops
00:00:15
many topics. AI data centers, China,
00:00:20
justice, human dignity,
00:00:22
Daario unwinding these SPVS hasn't been
00:00:25
good for the Vatican. We got in at 20
00:00:27
billion. That was a 50 bagger for us.
00:00:31
So, let's get started.
00:00:32
>> Jason, I'm pretty sure you believed you
00:00:34
were the vicer of God before the
00:00:36
encyclical. So, this is nothing new for
00:00:38
you.
00:00:41
>> Let your winners ride.
00:00:49
>> We open sourced it to the fans and
00:00:51
they've just gone crazy with it.
00:00:55
I'm going all in.
00:00:57
>> The smoke has risen from Tratad's pool
00:01:00
house
00:01:02
and from the poker room.
00:01:04
>> He's staying in my pool house. He's been
00:01:05
there for the last 3 days.
00:01:07
>> It's been magnificent. He didn't know.
00:01:09
You know what? I understand where OJ was
00:01:12
coming from. You know, you put Ko Kalin
00:01:14
in your house for long enough, you just
00:01:16
lose your [ __ ] At some at some point,
00:01:18
somebody's getting whacked. All right,
00:01:21
enough with the shenanigans. Uh, but
00:01:23
it's been great staying at the house
00:01:24
because there's actually Chimant is not
00:01:27
aware of this. There's an iPad in the
00:01:28
kitchen and that's logged in to Uber
00:01:31
Eats, Door Dash, Instacart, Amazon,
00:01:34
Laura Piana. Come on, stop.
00:01:36
>> No, there is. It's literally every
00:01:38
single service. And I told the house
00:01:41
manager like, listen, any packages that
00:01:44
come in the next 72 hours, right to the
00:01:47
pool house, if it says JCAL, right to
00:01:49
the pool house. So, all these packages
00:01:51
have been coming. Then I relabeled them,
00:01:53
gave them back, sent them to the ranch,
00:01:55
and now the house manager is sending
00:01:57
that stuff to the ranch.
00:01:58
>> Laura Piana wants to know why my inseam
00:02:00
went from 36 to 12.
00:02:04
>> Your waist size went from 32 to 36.
00:02:08
All right, welcome to the program,
00:02:09
everybody. David Saxs is here. How you
00:02:11
doing, David?
00:02:11
>> I'm good.
00:02:12
>> Chimoth Poly Hopetia is back at the 8090
00:02:17
office. I was at the 8090 office the
00:02:19
last couple days and it's a vibe. It's a
00:02:22
vibe. There's like a great culture going
00:02:24
on.
00:02:25
>> If you're a bestie and you show up at
00:02:26
that office though, everybody there is a
00:02:29
huge fan of the pod. So I was like it
00:02:32
was like being royalty. It's stopped by
00:02:34
everybody. Hey, I'm a developer here.
00:02:35
I'm a big fan of the show. Thank you for
00:02:37
giving it to Chimath. We can't give it
00:02:39
to him because he pays our mortgage and
00:02:41
everything, but every time you stick it
00:02:43
to Chimath, we love it. We're cheering
00:02:44
for you in the secret slack room. Um,
00:02:47
and
00:02:48
>> there's a secret slack room.
00:02:50
>> There is. There is definitely a secret
00:02:52
slack room going on.
00:02:54
>> No, but it was great. The vibes were
00:02:55
awesome. You're building a lot of
00:02:56
software. A lot of young talent. I don't
00:02:59
want to say that where your secret
00:03:00
source is, but there's a secret source
00:03:02
of talent you have. And uh, man, those
00:03:04
are some smart kids.
00:03:07
>> I'm happy to say it. Look, when I was at
00:03:09
Facebook, we became the most aggressive
00:03:11
recruiter of Waterlue co-ops. And so, I
00:03:13
went back to the well. Yeah. We recruit
00:03:17
more interns every quarter than we have
00:03:19
full-time engineers, which we do on
00:03:21
purpose because it puts a ton of
00:03:23
pressure on the product actually being
00:03:25
good.
00:03:27
>> We had 400 people apply this quarter for
00:03:29
internships.
00:03:30
>> Wow. It's very interesting. Uh with us
00:03:32
sitting in for Freedberg, who's busy
00:03:35
with some potatoes seed this week doing
00:03:38
great stuff at Ohio, the one, the only
00:03:42
Bill Gurley is here. He's been running
00:03:44
down a dream. If you haven't bought the
00:03:45
book, get the book. It's incredible. And
00:03:48
you're off book tour, so now you have
00:03:49
time for us. Yeah.
00:03:50
>> Yes. And I, you know, I had told you if
00:03:53
you ever talk about the Pope, I'd love
00:03:55
to hop on. And so
00:03:57
>> yes, you were like,
00:03:59
>> well, you know, he Bill's an
00:04:01
evangelical. I'm a Catholic. So, we do
00:04:03
have some common ground here. Do you you
00:04:06
when's the last time you were at church,
00:04:07
Bill? Have you have you thought now that
00:04:09
you're, you know,
00:04:10
>> when Jal gets sacriiggious?
00:04:15
I got to come on there and make sure JCL
00:04:18
doesn't get out of line with the Pope.
00:04:21
>> Listen, the Pope is God's messenger on
00:04:24
earth. We should give him a base level
00:04:26
of respect.
00:04:28
By the way, that's us imitating Bill
00:04:29
Gurley. This not actually Bill Gurley.
00:04:31
Just for those of you listening, you
00:04:33
think they were confused.
00:04:34
>> They were literally they were confused.
00:04:36
We don't want to put words in your
00:04:38
mouth, but just point of clarification.
00:04:40
But hey, everybody knows you were uh you
00:04:44
know, you uh handed the baton over at uh
00:04:46
Benchmark after a very successful couple
00:04:48
of decades in venture capital. You wrote
00:04:50
the book. You've now got a nonprofit.
00:04:53
You're doing your own spin, I think, on
00:04:56
uh maybe what Peter Teal does with his
00:04:58
fellowship. You started your own running
00:05:00
down a dream fellowship, I understand.
00:05:01
>> Yeah, it's uh it's targeted at a
00:05:03
different demographic. It's called uh
00:05:05
rdad.org runningdownadream.org.
00:05:08
There's a website we can give a link to.
00:05:10
We're going to do $5,000 grants to
00:05:13
people who want to chase their dreams
00:05:15
but need some help. And so there is an
00:05:17
application process like Teal Fellows
00:05:20
and other programs and we've been out
00:05:22
talking to those people. Um but we're
00:05:24
we're we're actually live. We went live
00:05:26
last week for the application. So if you
00:05:29
know people who have read the book and
00:05:31
are inspired and need some help, have
00:05:33
them apply.
00:05:34
>> Yeah. All right, folks. And uh Good for
00:05:36
you, BG. Yeah, great two two other So, I
00:05:39
did a TED talk which will come out soon
00:05:42
that's related to the book. Um I I uh
00:05:45
there's a professor in Miami that's
00:05:47
built a course around the book which I'm
00:05:49
excited about and he's he's he's doing
00:05:51
it in a kind of an open source way so
00:05:53
that other people can borrow that as
00:05:55
well. And so if there anyone out there
00:05:57
I'd love to uh help them do that.
00:05:59
>> What's your take on all of this
00:06:01
dumerism? Like if you're a young person
00:06:03
and you're in college or you're in high
00:06:06
school, is this is this much to do about
00:06:08
nothing or how do you run down a dream
00:06:10
in the face of something like that?
00:06:11
>> Yeah. Well, I you know, I started the
00:06:13
book before this happened and I've been
00:06:15
asked the question a lot and it it came
00:06:17
up in the TED talk. I I fear that the a
00:06:22
lot of people are in jobs they actually
00:06:23
don't care about that much. And um
00:06:26
there's a Gallup poll that backs this
00:06:28
up. They came up put that word quiet
00:06:30
quitters. They're like 59% of the people
00:06:32
they surveyed are kind of ambivalent
00:06:35
about their job. And when you're
00:06:36
ambivalent about your job, you're not
00:06:38
high agency. And so you don't lean in.
00:06:40
You know, if you if you look at how
00:06:43
Jason talks about all how they
00:06:44
implemented AI and all of his his
00:06:47
different working groups, you hear that
00:06:50
enthusiasm and that high agency and then
00:06:54
you want to go try these things. And I
00:06:56
think the best way to protect yourself
00:06:59
from AI is to be the most AI enabled
00:07:01
version of yourself you can be. But if
00:07:04
you're ambivalent about your job, you're
00:07:06
probably not doing that and you could
00:07:08
be, you know, a sitting duck. So I think
00:07:10
it's the mindset that's the problem.
00:07:12
>> I created an associate in training
00:07:14
program for my firm cuz we want to like
00:07:16
help people get into venture capital and
00:07:18
we gave them a choice of assignments.
00:07:21
One of them was to write coverage of one
00:07:22
of our portfolio companies that's
00:07:24
breaking out. Micro one is the name of
00:07:25
it and just give us like hey here's a
00:07:27
competitive landscape basically write a
00:07:28
deal memo and coverage of that company
00:07:30
and then we gave them another option to
00:07:33
vibe code a very specific project I've
00:07:35
wanted to have for our venture firm for
00:07:37
a long time you know on competitive
00:07:40
intelligence and I would say I think
00:07:42
like maybe 80% of the students applying
00:07:45
and we had four or 500 people apply for
00:07:47
six positions 80% of them did the vibe
00:07:50
coding and I was shocked I thought it
00:07:53
would be the exact opposite Anybody can
00:07:54
write, anybody can throw in JBT and get
00:07:57
some output. But they they actually
00:07:59
built software and th that's the scary
00:08:01
thing. The students who graduated, at
00:08:04
least this is my perception, Chimoth,
00:08:06
the students who graduated like 5 10
00:08:08
years ago before AI, they're not AI
00:08:11
first. They feel lost in a drift. They
00:08:12
don't have agency. But the group coming
00:08:15
out of college right now that cheated
00:08:16
their way through school using Chad GBT,
00:08:19
doing their assignments, like using
00:08:21
those tools, I'm joking, cheating, but I
00:08:23
mean hacking. I I agree with that.
00:08:24
>> So they were totally
00:08:26
>> Yeah. And and they're just like, I know
00:08:28
how to use these tools to get through my
00:08:30
finals. Yeah.
00:08:31
>> Girly, I think you're saying something
00:08:32
super important. I said this last week,
00:08:34
which is
00:08:35
>> nobody asks the warehouse worker at
00:08:37
Amazon whether they actually want that
00:08:38
job. And so to your point, job
00:08:41
satisfaction isn't some external person
00:08:44
judging your job to be valid and saying
00:08:46
you must be able to have it. I think it
00:08:48
should be asking the person that does
00:08:50
the job, do you like it and do you want
00:08:51
to keep it? Those are two very different
00:08:53
questions. And
00:08:54
>> yeah,
00:08:55
>> I think that the all of this AI doom and
00:08:57
gloom was a lot and too much frankly of
00:09:00
the former and not enough of the latter.
00:09:02
And now this whole lie is kind of
00:09:06
getting undone. I think Saxs posted
00:09:08
about it this week as well. The Goldman
00:09:09
Sachs CEO said it. And now in this crazy
00:09:11
twist of fate now that we need to have
00:09:13
trillion dollar IPOs, the entire
00:09:15
Frontier Labs are all like, "Wow, it's
00:09:17
going to be a bonanza of jobs." And
00:09:19
>> Mark Cuban had had a great quote. He
00:09:21
said there are two types of people in
00:09:22
the world. Those that use AI to learn
00:09:25
faster than they ever could before and
00:09:27
those that use AI to avoid learning
00:09:29
altogether.
00:09:30
>> And I I think it's this notion of high
00:09:32
agency or not. That's pretty good.
00:09:34
>> Are you leaning in and using this stuff
00:09:37
to be ever more powerful in what you try
00:09:40
and accomplish or are you using it, you
00:09:42
know, as a cheat code? And if you're in
00:09:44
the latter, yeah, you're probably at
00:09:46
risk. You get asked a lot about how to
00:09:49
educate yourself if you're a parent of
00:09:50
kids
00:09:52
so that you can put them on a path to
00:09:54
launch and do well and chase their
00:09:56
dreams.
00:09:57
>> You have a good answer for that
00:09:58
question.
00:09:58
>> I I mean the the second chapter of the
00:10:00
book's all about lifetime learning and
00:10:03
it's kind of a requirement that you're
00:10:05
following your fascination because the
00:10:06
lifetime learning comes for free if
00:10:09
you're fascinated with something like
00:10:10
you just constantly soak up and devour
00:10:13
new information. And I I do think that a
00:10:16
lot of kids get exhausted because we've
00:10:19
made high school and college such a
00:10:21
grind that they think the learning ends
00:10:23
the day they they walk out with their
00:10:25
diploma. And as we all know, the best
00:10:28
and brightest in all of our fields are
00:10:31
on a constant learning journey. And when
00:10:33
something new comes out, they dive in
00:10:35
and try and figure it out, right? And so
00:10:38
and every every single uh person in the
00:10:41
book that we profiled has that kind of
00:10:43
attitude about their craft, you know, in
00:10:45
every day. And so I I think the real
00:10:48
test is if you're not proactively
00:10:51
self-learning, then you're probably not
00:10:52
tilting against something that you
00:10:54
really adore and are fascinated by.
00:10:56
>> Sax, you wanted to jump in there. With
00:10:58
respect to new college grads, I was
00:11:01
going to say that I think the single
00:11:02
most marketable skill in the economy
00:11:05
right now has got to be proficiency in
00:11:07
Claude. If you're going into a firm
00:11:10
right now and you're the only one who
00:11:11
knows Claude, it would be like you're
00:11:14
the only one who knows how to work a
00:11:16
spreadsheet, you know, or word
00:11:17
processor, the advantage would be
00:11:19
enormous. Now, I think that that's
00:11:21
probably a short-term arbitrage because
00:11:23
eventually everyone's going to have to
00:11:24
figure out how to use these tools. But
00:11:27
as a young college graduate right now,
00:11:28
you have such an advantage if you're an
00:11:30
AI native uh just knowing how to use
00:11:33
these tools. And this thought uh
00:11:36
partially occurred to me when I saw what
00:11:38
our producer Nick has been doing with
00:11:41
using Claude for you know he's been
00:11:42
creating this daily briefing document.
00:11:45
>> We've been doing it we've been doing it
00:11:47
for three months actually.
00:11:49
>> I just ran it for the first time
00:11:51
apparently. I didn't you've been busy.
00:11:53
>> Yeah. Well, I just, you know, I thought
00:11:55
it would just be AI slop and it would
00:11:57
just kind of give me a roundup of news
00:12:00
that I was getting in my X feed anyway.
00:12:02
But actually the thing that was really
00:12:04
impressive about it was that it
00:12:07
predicted topics that I would
00:12:09
specifically be interested in based on
00:12:11
my previous comments on the pod and also
00:12:15
it went back and looked at previous
00:12:18
transcripts and what I had said and then
00:12:22
had updates to those topics based on
00:12:25
specific things I had said. So again it
00:12:27
was highly highly contextual. But then I
00:12:30
asked Nick, how did you generate that?
00:12:31
and he showed me the the custom prompt
00:12:34
that he designed for Claude and then the
00:12:36
skills document and they were very long
00:12:39
and detailed documents. They weren't
00:12:40
written in code but they were very
00:12:42
technical and I just realized looking at
00:12:44
that that the average person is not
00:12:45
going to be able to generate this. I
00:12:46
mean this is why this idea that you're
00:12:48
just going to be able to like throw AI
00:12:50
into an organization and it's just
00:12:52
magically going to generate value is not
00:12:53
true. You have to know how to
00:12:55
>> get value out of it. I mean maybe we
00:12:57
could just show these documents on the
00:12:58
screen.
00:12:59
>> Yeah. I mean the the the interesting
00:13:01
thing uh Sachs is you can you just have
00:13:04
to ask your AI you ask claude or chatgpt
00:13:08
or whatever you're using hey I want you
00:13:10
to make me a mega prompt and you like a
00:13:13
mega pint like give me a mega prompt of
00:13:16
you're an producer of a podcast these
00:13:19
are the four characters on the podcast
00:13:21
what would be a great prompt for me and
00:13:22
it will actually suggest a prompt
00:13:24
>> and then you can refine the prompt so
00:13:26
you actually have a dialogue about a
00:13:28
prompt as opposed writing the prompt
00:13:30
yourself
00:13:34
like
00:13:34
>> well Nick can you show on the screen to
00:13:36
scroll through the training rules and
00:13:39
then also there's the skills document
00:13:41
that was written on how to be a producer
00:13:43
>> for this podcast which I thought was
00:13:46
really impressive
00:13:48
>> by the way David what you said I think
00:13:50
is true of almost every single job type
00:13:53
like it's not just tech or programming
00:13:55
if you're in marketing if you're in
00:13:57
legal if you're in accounting
00:13:59
Like any role you might have at a firm
00:14:01
sales, if you're the most AI savvy
00:14:06
person of all your peers,
00:14:08
>> you are golden. Like you are golden like
00:14:12
in your
00:14:12
>> 10x more valuable than the next person
00:14:14
who's not basically.
00:14:15
>> Yes. Yes. And and and I think that I
00:14:19
don't think it goes away because I think
00:14:21
you learn how to get better at it over
00:14:23
time. So having an early advantage I
00:14:26
think will extend for a while.
00:14:28
because you can learn more more and more
00:14:30
things you can accomplish.
00:14:32
>> Should we let producer Nick describe
00:14:34
what we were just looking at there?
00:14:35
>> Yeah, go ahead, produce. What? Producer
00:14:36
Nick, explain the process.
00:14:38
>> Yeah, once we got access to Claude
00:14:40
Co-work and it had that like further
00:14:43
expanded memory access. I thought it
00:14:45
would be interesting to just start
00:14:46
feeding every transcript into it and
00:14:47
seeing if it could actually
00:14:49
contextualize new stories that were
00:14:51
coming out based on past things that you
00:14:53
guys have said. And I gave it like a
00:14:56
general prompt of what I wanted and I
00:14:57
said, "How would you write a skills file
00:14:59
or some training rules for this?" And it
00:15:01
wrote all of it for me.
00:15:03
>> Yeah.
00:15:03
>> Oh, so you were less good than I
00:15:05
thought.
00:15:07
>> I thought you were
00:15:08
>> It's a hack. It's a hack. You use AI to
00:15:10
make the skills. Yeah. And
00:15:12
>> but you've been updating that over time,
00:15:13
right? As you've been iterating and
00:15:14
learning
00:15:15
>> every single day and every single day
00:15:16
gets smarter and better.
00:15:17
>> The recursiveness of this is incredible.
00:15:19
>> So you need someone to manage that
00:15:21
process, right? because the four besties
00:15:23
are not going to do that. So you need a
00:15:25
producer of the show to do that. This is
00:15:26
why people think, oh, it's just going to
00:15:28
wipe out all the jobs. No, someone still
00:15:29
has to supervise, iterate, validate, you
00:15:33
know, all those kinds of things.
00:15:35
>> Yeah. And it's
00:15:38
it's really interesting that the the
00:15:40
people who are coming into the workforce
00:15:42
right now are super aware of this and
00:15:44
they're putting the tools to work and
00:15:46
it's much easier for them to get a job.
00:15:48
I mean, I I literally looked at the top
00:15:50
nine candidates for this associate and
00:15:52
training program I have, and we're going
00:15:54
to do it every year. Every summer, we
00:15:56
start it. We do it for a year. We pay
00:15:57
you to learn. And um it it it was just
00:16:01
extraordinary how you could tell
00:16:04
immediately if the person had systems
00:16:06
thinking Sachs like they understood the
00:16:08
the process of venture capital that
00:16:10
there was a structure to it. You had to
00:16:12
source deals. You had to make decisions
00:16:13
on which ones to invest in. You had to
00:16:15
do diligence. you, you know, you had to
00:16:17
double down on investments, they they
00:16:19
just understood the process and then
00:16:22
when if you just talk to one of these
00:16:25
LLMs, it will tell you what to do. So,
00:16:27
you can say, I don't know what I'm
00:16:28
doing. What should I do next? And then
00:16:30
it actually tells you what to do next.
00:16:32
So, for people who are intimidated about
00:16:33
this and uh maybe think like, I it I'm
00:16:37
I'm already too far behind, I encourage
00:16:39
you to pop up Claude, go into co-work,
00:16:41
and say, "What can I do to be better at
00:16:43
my job?" and just start talking and
00:16:46
literally if the more you talk and you
00:16:48
can use voice uh you know text to voice
00:16:51
I use whisperflow is a really cool
00:16:52
program for this and I have a foot pedal
00:16:54
to do it you just ramble and ramble and
00:16:56
ramble and keep adding stuff you don't
00:16:58
have to be structured it will build the
00:17:00
structure around the two or three
00:17:03
paragraphs that you give it as
00:17:05
instructions that's the thing people are
00:17:07
getting caught up on now is Bill they
00:17:09
they think they have to type when in
00:17:12
fact if you just blather on and on a
00:17:15
scale I have a unique uh ability to do.
00:17:18
You just blather on. It's a superpower.
00:17:20
You blather on and the thing makes sense
00:17:22
of it. It is unbelievable what the
00:17:24
blatheron prompt can get in terms of
00:17:27
output. Thanks for coming uh to uh my
00:17:32
TED talk. All right, let's get started.
00:17:33
There's a lot to talk about and uh we
00:17:35
got a big docket today. We're going to
00:17:38
start with the Pope. The Pope is dope.
00:17:41
And uh the Pope Leo, he's the 14th,
00:17:44
released his first encyclical encyclical
00:17:48
on AI. And it was long. 235 pages over
00:17:53
42,000 words. Just to give you an idea,
00:17:56
Bill Gar,
00:17:57
>> when did he write it, do you think? When
00:17:58
did he put that together?
00:17:59
>> Well, no, no, I think he used chat GBT.
00:18:01
That's what it says here um in the
00:18:03
notes. No, I mean I I'm guessing
00:18:05
>> how long did it take for him to write
00:18:06
this in between all of his other tasks?
00:18:08
I think it's a six-month process to do
00:18:10
this, but I'm sure he had collaborators.
00:18:12
>> Bill, your book, I'm assuming, was
00:18:14
>> write it.
00:18:15
>> I'm sure there was a team that wrote it.
00:18:17
But Bill, your book's 60 70,000 words,
00:18:19
I'm guessing. So, this is almost a
00:18:21
literal book, right?
00:18:24
>> In terms of how long it is, and it's
00:18:27
called Magnifica Humanitas or
00:18:30
Magnificent Humanity. In it, he warns
00:18:32
business leaders to safeguard humanity
00:18:35
from AI. His core argument is AI is not
00:18:38
inherently evil, but technology is never
00:18:41
neutral and that technology takes on the
00:18:43
characteristics, wait for it, of those
00:18:45
who build, finance, and control it. And
00:18:48
I don't think he thinks super highly of
00:18:50
that group of people. The Pope called
00:18:52
for regulation of AI companies.
00:18:54
Obviously, we're going to have that
00:18:55
debate here. Some of the things he uh
00:18:58
called for I think are not very
00:19:01
debatable and there's a lot of consensus
00:19:03
around worker retrainment, safety uh for
00:19:07
children and guard rails, a ban on
00:19:09
autonomous weapons. That's the uh Skynet
00:19:12
rule. Don't build terminators with your
00:19:14
AI. But he was joined by anthropic
00:19:17
co-founder Chris Ola. I don't know how
00:19:19
many co-founders there are of this
00:19:21
company, uh but apparently there's
00:19:23
dozens. And Ola is not Catholic.
00:19:26
According to a Vanity Fair profile, he
00:19:28
was raised evangelical and now he's an
00:19:30
atheist. The folks at Amazon, Google,
00:19:32
and Meta lobbyed the Vatican on April
00:19:35
29th to soften the language in his
00:19:37
missive and uh he was not swayed.
00:19:41
His central question, Saxs, is will AI
00:19:44
be used to concentrate power in the
00:19:46
hands of a few or will it serve
00:19:48
everyone? Something you brought up when
00:19:51
you mentioned monopolies, duopolies,
00:19:54
etc. two weeks ago on this very podcast.
00:19:57
What's your take on the Pope and his
00:20:00
interest and his missives on AI and
00:20:04
promoting a bit of AI regulation?
00:20:07
>> Well, I very much agree with the Pope
00:20:10
that the biggest risk of AI is a
00:20:12
centralization of power and then its
00:20:15
misuse against us um in some Orwellian
00:20:18
way. I think it's government that's
00:20:20
going to do that. Um, not necessarily an
00:20:23
individual actor because it's
00:20:24
governments that ultimately have the
00:20:26
power. So, I do worry about the
00:20:28
potential for AI to be used to surveil
00:20:31
us, censor us, control us as Orwell
00:20:35
described in 1984. So, if that's where
00:20:37
the Pope is going with this, I very much
00:20:39
agree with him.
00:20:41
The maybe where we end up in different
00:20:43
places is he thinks that government
00:20:45
regulation is the way to prevent this.
00:20:48
And I would just say that we have to be
00:20:49
careful not to empower government too
00:20:51
much because if you give government the
00:20:54
power to regulate or approve AI
00:20:58
development,
00:20:59
if you create say an FDA for AI as many
00:21:02
people are calling on, that will give
00:21:05
government the power to approve models
00:21:08
and and therefore give notes to model
00:21:10
developers. And very soon this
00:21:13
definition of safety will expand because
00:21:16
the government always takes an expansive
00:21:18
view of its powers. And we saw this
00:21:19
during the social media wars where the
00:21:22
definition of trust and safety expanded
00:21:24
to issues like psychological safety,
00:21:28
microaggressions,
00:21:30
disinformation,
00:21:31
transphobia, and so on. That, you know,
00:21:33
again, these social media companies were
00:21:36
told that they had to stamp out all of
00:21:38
those threats to safety. And it ended up
00:21:41
becoming a censorship agenda. So I get
00:21:43
very worried about you know what if some
00:21:45
government agency can give notes to the
00:21:48
model developers and they start telling
00:21:49
the model developers that your
00:21:51
definition of safety is not expansive
00:21:53
enough. You have to again protect the
00:21:55
public from disinformation or you know
00:21:58
psychological harms. So again, I think
00:22:01
we just have to be careful not to
00:22:03
arandise government because that's going
00:22:05
to be the most likely culprit in terms
00:22:07
of the centralization of of power. And
00:22:10
um I know the um the Vatican likes
00:22:12
Latin. This is a problem of political
00:22:14
philosophy that goes all the way back to
00:22:16
Socrates. It's called quis custodos
00:22:20
custodes which is who will guard the
00:22:22
guardians. In other words, if we entrust
00:22:25
a set of guardians to protect us from a
00:22:28
bunch of threats, what's to stop them
00:22:31
from becoming tyrannical and from
00:22:33
becoming the new threat against us? And
00:22:36
I mean, this is a the central dilemma of
00:22:38
political power.
00:22:39
>> Who watches?
00:22:41
>> Yeah. Who watches the watchers? Who
00:22:42
guards the guardians? Meaning, who's
00:22:44
going to protect us against our
00:22:45
guardians if they turn against us? The
00:22:48
genius of the American founding by the
00:22:50
way is that it was a second order
00:22:52
solution to this question. The founders
00:22:54
of America very much understood this and
00:22:56
what they came up with is we have to
00:22:59
have the guardians guard against each
00:23:00
other. And so they came up with the idea
00:23:02
of separation of powers. We'd have
00:23:04
separation of federal and state. We'd
00:23:06
have um the three branches of of the
00:23:08
government. Even within the legislative
00:23:10
branch, it was a biccameal legislature.
00:23:12
So they divided up the powers in a way
00:23:14
that hopefully the guardians would check
00:23:17
against each other as opposed to
00:23:18
becoming tyrannical against us. And that
00:23:22
that is kind of my view on AI is that
00:23:24
ultimately we have to have a solution of
00:23:27
checks and balances. If the AI market
00:23:30
becomes monopolized and falls into the
00:23:32
hands of one or two companies, I would
00:23:34
use antitrust law very aggressively to
00:23:37
as a check and balance against their
00:23:39
power. Right now we have a very
00:23:41
competitive market. You know, we have
00:23:42
five frontier labs competing very
00:23:44
aggressively. As long as the market is
00:23:47
competitive, I would use that because I
00:23:50
think competition generates the best
00:23:51
outcomes. It helps us win against China.
00:23:54
But it also protects the population
00:23:56
because these companies, you know, if
00:23:57
they get out of line, there's some
00:23:59
competitor that can offer something
00:24:00
better.
00:24:01
>> Consumers can opt out of it. If they
00:24:02
don't trust GPT, they can use Anthropic
00:24:05
or if they don't trust Anthropic, they
00:24:06
can go to Grock. Bill, you had the
00:24:08
number one rated talk at the All-In
00:24:12
Summit in history, 2,851
00:24:14
miles. You have been famously against
00:24:17
regulatory capture. In light of the
00:24:20
Pope's comments of, hey, regulating,
00:24:22
what do you think is common sense?
00:24:24
Because AI is everything. AI can help
00:24:27
people make boweapons. It can also help
00:24:29
people get their term paper in or do you
00:24:32
know uh be a better salesperson at you
00:24:35
know Oracle like we're talking about
00:24:37
paper like we're talking about oxygen
00:24:39
here this is like a fundamental
00:24:41
horizontal technology so where do you
00:24:44
think there is a case to regulating AI
00:24:47
if at all and where do you think yeah
00:24:50
free market will figure it out
00:24:51
>> well I have I have two takes one on the
00:24:53
pope and one on anthropic so your
00:24:56
questions your question powerful let's
00:24:58
with the more let's go with the more
00:25:00
powerful entity. We go you want to go in
00:25:02
reverse the least powerful of the two
00:25:04
go.
00:25:04
>> So so this pope said that and I have to
00:25:08
learn how to pronounce all these Latin
00:25:09
words like you that this encyclical was
00:25:12
u was mirrored after one done by Leo I
00:25:15
13th in 1891 and he invoked that he even
00:25:19
said he chose the name because he's so
00:25:21
enamored with Leo I 13th. Leo the 13th
00:25:24
encyclical warned that the industrial
00:25:26
revolution was going to be bad for
00:25:28
people. So let me tell you what happened
00:25:31
from 1891 till today. The work week went
00:25:35
from over 60 hours to 34 hours globally.
00:25:38
Real wages went up 8 to 10x adjusted for
00:25:41
inflation. The medium worker now earns
00:25:43
more than a doctor did in 19 in 1891.
00:25:46
Global GDP per capita went from 1500 to
00:25:49
20K. Child labor in the US went from 18%
00:25:52
to zero. Workplace deaths fell by 40x.
00:25:57
Life expectancy went up 60%. And global
00:26:01
poverty went from 75% of humanity to
00:26:03
under 10%. All those things happened
00:26:06
because of technology, innovation, and
00:26:09
capitalism, which is exactly what Leo
00:26:11
the 13th was warning against. So he got
00:26:14
it dead wrong. He got the whole thing
00:26:16
precisely wrong. So it's an interesting
00:26:19
thing to say you're borrowing from.
00:26:22
>> Yeah. Uh so now on to
00:26:26
>> Yeah. Anthropic and just common sense
00:26:28
around do you think there how would you
00:26:32
regulate and or protect against maybe
00:26:34
we'll broaden the term here protect
00:26:36
against nefarious uses of the
00:26:38
technology. Obviously we all want
00:26:40
children to be protected. We want to
00:26:42
have truth uh and honesty in terms of
00:26:45
facts and and all of us sharing some
00:26:47
some basic truths and we obviously don't
00:26:50
want people using this technology for
00:26:52
boweapons and the Terminator scenario.
00:26:54
>> I have to tell you that that Anthropic
00:26:56
is a mystery to me. I've never ever seen
00:26:59
a company that is both leading their
00:27:03
field and the most negatively outspoken
00:27:07
commenter on what they do. I I've just
00:27:10
never seen it. And my initial theory was
00:27:12
the regulatory capture theory that they
00:27:15
just want to ensure there's regulation.
00:27:18
And quite frankly, I think they're, you
00:27:21
know, very close to achieving that. Like
00:27:23
they have stirred up, you know, a
00:27:25
frantic position, especially in America.
00:27:29
American consumers are definitely afraid
00:27:31
of AI. Um, I think I've talked to you
00:27:34
guys in the past about, you know, the
00:27:36
book that Jonathan Heights written about
00:27:38
social media and there's a whole bunch
00:27:40
of state legislators that think we
00:27:42
should have regulated social media and
00:27:44
so now they're destined to want to get
00:27:47
in front of it. And we know that
00:27:50
Anthropics, one of the most aggressive
00:27:52
lobbying company startups of all time.
00:27:55
You know, the the the amount of effort
00:27:57
that they're putting in, the amount of
00:27:58
money at a statebystate basis. So that
00:28:02
was always my first theory, but then
00:28:03
they just they got so loud that I I've
00:28:06
literally in the past 30 days read
00:28:10
everything I can about anthropic and
00:28:12
I've come up with a new theory. This
00:28:13
this
00:28:14
>> okay new breaking theory.
00:28:15
>> This I call it the Dr. Frankenstein
00:28:18
theory. Um you remember when Elon had
00:28:21
that conversation with Larry Paige where
00:28:23
Larry called literally sitting next to
00:28:25
him when he called?
00:28:27
>> Explain the story real quick. while we
00:28:29
were at uh a birthday party and and you
00:28:33
know Elon was like listen humanity needs
00:28:35
to be protected from the stuff at
00:28:36
DeepMind because at DeepMind they had an
00:28:39
example of the AI having tried to break
00:28:42
out to jailbreak out of its computer and
00:28:45
not be turned off and you know had some
00:28:48
sentience or some you know inkling of
00:28:50
sentience and he said you know we have
00:28:51
to protect the human species and he said
00:28:53
well Larry said well what do you think
00:28:55
that's species because you care about
00:28:57
the human species over AI. This is at
00:28:59
least 15 years ago.
00:29:01
>> No, this is right before Elon co-founded
00:29:03
uh Open AI, right? Back in 2015 or
00:29:06
something.
00:29:07
>> The actual story here is Elon Hon and
00:29:11
Google had backed Deis and the team at
00:29:14
DeepMind when they were an independent
00:29:16
company. Then Elon was like, "Oh my god,
00:29:19
Google's going to buy this." And I
00:29:20
remember having the conversation with
00:29:21
Elon about this. We have to figure out a
00:29:23
way for DeepMind not to go to Google. We
00:29:26
have to block this somehow. But he
00:29:27
begged those folks to not sell to Google
00:29:30
because Google was running the table on
00:29:32
everything and he wanted this technology
00:29:34
to be independent and he was on the
00:29:36
board of the company
00:29:37
>> and he also said this was his motivation
00:29:39
to launch open AAI as a nonprofit.
00:29:41
>> Google got it. He just said we we this
00:29:43
is this technology is too powerful for
00:29:45
any one person. So like once again you
00:29:47
got to give Elon a lot of credit. He saw
00:29:48
the writing on the wall if one person
00:29:50
can and he saw it 15 20 years ago and
00:29:52
him and Sam Harris used to debate this
00:29:54
over dinner. You know what happens if
00:29:56
somebody controls this and they run away
00:29:58
with it. It would be extremely
00:30:00
dangerous. It has to be available to all
00:30:02
the people. Essentially the pope's
00:30:03
position. It has to be in the service of
00:30:05
humanity, not ruled by one person. It's
00:30:08
far too powerful.
00:30:09
>> So the reason I call this the Dr.
00:30:10
Frankenstein theory is the more I dig,
00:30:13
I've met people who I who I dare say
00:30:16
think it's their responsibility and
00:30:19
they're excited about building a species
00:30:23
that's that's superior to humans. And I
00:30:26
would just encourage people to read, you
00:30:29
know, as much as they can about
00:30:30
anthropic. Chris Ola worked on this
00:30:33
thing called the Constitution. It's
00:30:35
about 80 pages. It's hard to get
00:30:37
through, but I would encourage you to
00:30:38
read it. Amanda Ascll who is the chief
00:30:41
philosopher has started doing podcasts.
00:30:43
I would encourage you to listen to them
00:30:45
and listen to her language. And then
00:30:48
Daario wrote this blog post called
00:30:51
Machines of Loving Grace.
00:30:53
>> Loving grace. I read it
00:30:54
>> and it it was based on a poem and the
00:30:57
poem is kind of weird. I we should put a
00:30:59
link to the poem. It's quite short. But
00:31:02
the last the last stanza of the poem
00:31:05
says, "I like to think of a cybernetic
00:31:07
ecology where we are free of our labors
00:31:11
and join back to nature. Return to our
00:31:13
mammal brothers and sisters." I don't
00:31:15
know what that means. Like we're going
00:31:16
to go live in the fields where the
00:31:18
mammals live. I I And then the kicker
00:31:21
and all watched over by machines of
00:31:24
loving grace. Sounds like overlord to
00:31:26
me. And then in Daario's post he says he
00:31:31
near the end and it's very long. You
00:31:33
read it Jamal. I mean machines of love
00:31:35
and grace is very long but he's he's
00:31:37
talking about in the future what are
00:31:39
humans going to do because he believes
00:31:41
in the massive abundance and UBI and
00:31:44
that we won't have to work. I don't
00:31:45
believe in any of those things but he
00:31:47
does. And then he says it could be a
00:31:49
capitalist economy of AI systems which
00:31:52
then give out resources to humans based
00:31:56
on some secondary economy of what the AI
00:31:59
systems think makes sense to reward in
00:32:02
humans. So So that's envisioning a a
00:32:06
deity of sorts that's going to break
00:32:08
ties and discern decide what humans
00:32:11
>> it's a it's a computational reward
00:32:13
function for humans. It decides how much
00:32:16
you're worth.
00:32:16
>> Yeah. So, I don't think they think
00:32:18
they're writing software. I think
00:32:19
they're midwifing a deity here. And and
00:32:24
I don't know which one I'm more afraid
00:32:25
of, the regulatory capture or or or or
00:32:28
this second theory I call the Dr.
00:32:30
Frankenstein theory. It it's more it's
00:32:33
more scary to me. I think the second
00:32:35
thing
00:32:35
>> these are delusions of grandeur. Let's
00:32:37
call it what it is. They believe that
00:32:39
they are so intelligent. I know some of
00:32:41
these folks, the Burning Man sort of
00:32:43
offshoot of it, transhumanism. They
00:32:45
believe that they're so powerful, these
00:32:49
individuals, that they can create God
00:32:51
and that by creating God, they are like
00:32:54
this Prometheus kind of species. It
00:32:57
literally is the ultimate level of
00:33:00
narcissism and delusion of grandeur to
00:33:03
think you can create God and that then
00:33:05
the god you create like you're saying
00:33:07
Bill is going to be so benevolent and
00:33:11
perfect that you create constructed the
00:33:12
perfect God that will give you your
00:33:14
pellet will give you your little
00:33:16
scenarian you know
00:33:17
>> I just would correct you I didn't say it
00:33:19
Daario said it
00:33:20
>> right but no to but to your point of
00:33:21
like just taking them at their word they
00:33:24
actually believe that they can create
00:33:25
God and that they'll create a god so
00:33:27
good that it's better than humanity.
00:33:29
Saxs, your thoughts.
00:33:30
>> Well, I guess the question then is why
00:33:33
are they pushing for the let's call it
00:33:35
red capture agenda where
00:33:38
>> I know why.
00:33:38
>> Go ahead, Jamal. Go ahead.
00:33:41
>> That is very reductive game theory. So,
00:33:45
if you want to be unexploitable, I think
00:33:47
the best thing that you could do if
00:33:48
you're trying to build a super god is
00:33:50
have three or four entities in a room,
00:33:52
close the door behind you, and then
00:33:54
dominate those other three or four
00:33:56
entities, and then you set the rules.
00:33:58
And because your counterparty is unable
00:34:01
to track at the level of technical
00:34:04
capability that you would have, you
00:34:06
create this massive asymmetry that
00:34:08
allows you to exploit them. That's just
00:34:09
simple game theory optimization. And you
00:34:12
know what Bill said is so powerful. I've
00:34:14
read these things and it's laborious and
00:34:15
it takes time, but every time they put
00:34:17
these things out, just take the time to
00:34:19
read it. And what I have said before,
00:34:22
Bill, I don't know your point of view on
00:34:24
this, but I initially thought that this
00:34:26
was mostly game theory, that a lot of
00:34:30
their reactions I thought were less
00:34:32
rooted in their dogmatic beliefs and
00:34:35
more rooted in a GTO approach to either
00:34:38
raising capital or putting pressure on
00:34:40
competitors. Either way, both could be
00:34:43
true. What your framing is and my
00:34:45
framing, although mine's more tactical
00:34:47
than yours to be fair, because I've
00:34:49
always thought that these moves make
00:34:51
sense through that lens. How do you
00:34:53
absorb most of the capital? How then do
00:34:56
you make sure that you are in a position
00:34:58
to disproportionately affect the rules?
00:35:02
And how do you create an oversight body
00:35:05
that is less capable and intellectually
00:35:09
aware as you are about the actual
00:35:12
details because
00:35:13
>> the referees don't understand the game.
00:35:14
Right?
00:35:15
>> If the refs don't understand the game,
00:35:17
you'll run over the game. Yeah.
00:35:18
>> By the way, by the way, one thing they
00:35:20
have achieved by doing this is I think
00:35:23
that if you pulled the, let's just call
00:35:26
it the intellectual elites, so everyone
00:35:28
in the media and whatnot and the
00:35:30
professors and all those, and they were
00:35:33
to rank the different AI players by who
00:35:36
they think is most caring, I think
00:35:39
they'd probably put Anthropic first
00:35:41
because they've been out with the
00:35:43
doomerism talk. And so it's given them a
00:35:46
halo with the people that may matter to
00:35:49
what they want to accomplish. It it's
00:35:51
simultaneous
00:35:53
creating a lot of trouble like with the
00:35:56
data centers and whatnot. Like there's
00:35:57
there's negative ramifications.
00:35:58
>> What you're saying is so important
00:36:00
because on the one hand they create
00:36:02
empathy
00:36:03
and then they write these documents that
00:36:05
expose what they think and nobody
00:36:07
actually connects the dots.
00:36:08
>> Yeah. To steelman their position for a
00:36:10
second. I mean, I think probably the way
00:36:12
they think about it is that they are
00:36:14
creating something very powerful,
00:36:15
something godlike, and therefore it
00:36:18
needs to be safe
00:36:20
and that they care the most about that
00:36:23
out of everybody. Nobody else takes this
00:36:25
seriously. Remember that Enthropic was
00:36:27
basically a spin out of open AI and they
00:36:30
felt that Sam and the company leadership
00:36:33
weren't taking their point of view
00:36:34
seriously enough.
00:36:35
>> It was the most woke portion of Open AI.
00:36:39
>> We're steel Manning. So the most for so
00:36:42
they they see the power of it. They're
00:36:44
the ones who are concerned about safety
00:36:47
>> and they care the most and therefore
00:36:49
they're in the best position to do that.
00:36:51
>> Now I think the issue is just you can
00:36:55
see how this can lead to red capture,
00:36:57
right? Which is if you brand yourself as
00:37:00
the safe AI company and then try to
00:37:03
characterize everybody else as a
00:37:05
reckless player and reckless AI needs to
00:37:08
be stopped. you can see how this would
00:37:11
basically further your monopolistic
00:37:14
control over this industry. And if you
00:37:17
see AI through the lens that you know
00:37:20
that really frankly the pope and I see
00:37:21
it which is centralization versus
00:37:23
decentralization I do think that is you
00:37:26
know one of the key lenses we should
00:37:28
have on the technology is whether you
00:37:30
want this to be a centralized or
00:37:32
decentralized technology. This way of
00:37:34
viewing the world leads to more
00:37:36
centralization and I think that's
00:37:38
dangerous. I mean, if AI is this very
00:37:41
powerful technology, I think it needs to
00:37:44
be decentralized so that all of us can
00:37:46
protect ourselves to some degree, right?
00:37:48
We need to be able to run we need to be
00:37:50
able to run the AI ourselves on our own
00:37:53
hardware if we so choose, so we're not
00:37:56
beholden to a single company that might
00:37:59
be in bed with a deep state.
00:38:00
>> Let's say it very pointedly. If benefits
00:38:03
and compensation and economic support
00:38:07
were all of a sudden tied to some
00:38:09
algorithmic decision, this is a
00:38:11
dystopian episode of Black Mirror that
00:38:14
we're dealing with. And to your point,
00:38:15
Saxs, you want 100 or 1,000 or 100,000
00:38:19
versions of what that answer is so that
00:38:21
there's actually a way to refute
00:38:24
a singular answer. A singular answer to
00:38:26
these kinds of questions, which is
00:38:28
effectively what some folks would want,
00:38:30
is incredibly dangerous.
00:38:32
>> And this is something that is in
00:38:34
control, I think, of humanity. I've been
00:38:37
talking about AI sovereignty here for a
00:38:39
bit just in terms of how much more cost
00:38:43
effective it is and how you're not
00:38:45
training other people's AIs with your
00:38:47
knowledge and your insights. This is why
00:38:49
it's super important that open- source,
00:38:51
open- source agents and local hardware
00:38:53
be able to run these models and that
00:38:56
consumers and companies learn how to
00:38:58
roll their own language models, how to
00:39:00
make a small language model, an SML, a
00:39:03
VSSML, a verticalized one and run it on
00:39:06
your Apple hardware because Apple
00:39:08
actually has taken a principled approach
00:39:10
historically to your sovereignty for
00:39:12
your data. Data sovereignty now is
00:39:14
privacy.
00:39:15
>> Yes. And now it's intelligence
00:39:16
sovereignty. the the intelligent
00:39:18
sovereignty is different than privacy.
00:39:19
Privacy is, oh, you can't see my photos.
00:39:22
You can't, you know, peek into my notes
00:39:25
app and what I wrote there in my
00:39:26
journal. Now, intelligent sovereignty is
00:39:29
you can't tell me what to think. You
00:39:31
can't use your AI to analyze my photos,
00:39:34
to analyze my emails, to analyze my
00:39:36
messages, and tell me how to interpret
00:39:38
the world. That's actually going to be
00:39:39
the next key piece here. This is why I
00:39:41
think Apple is just the dark horse in
00:39:44
this entire race. there is an
00:39:45
open-source product that can run on this
00:39:48
hardware, the M5s, the, you know, 48
00:39:50
gigs, 128 gigs, the stu new Mac Studio
00:39:53
coming out with supposedly a terabyte.
00:39:55
That changes the whole game. And this is
00:39:58
so paradoxical, Bill, that our
00:40:00
adversary, the Chinese of all people,
00:40:04
the Communist Party is leading the
00:40:06
open-source movement and the United
00:40:08
States is centralizing.
00:40:10
>> They're leading the openweight movement.
00:40:11
It's not open source. Just the
00:40:14
distinction is important.
00:40:15
>> Yeah. Yeah.
00:40:16
>> Look, I Jacob, I agree with you about
00:40:18
the importance of open source because
00:40:20
open source means software freedom. You
00:40:22
can run the program yourself on your own
00:40:24
hardware. You don't have to share. You
00:40:26
don't have to give up your data
00:40:27
sovereignty. You don't have to give up
00:40:28
your privacy to again to some monopolist
00:40:31
who's going to be, you know, in bed with
00:40:33
the government or the deep state, right?
00:40:34
So that's the thing we're all afraid of.
00:40:36
And if that's the only AI that's
00:40:38
available is from the, you know,
00:40:39
monopoly or duopoly,
00:40:42
then your choices are to live off the
00:40:44
grid and not participate in the modern
00:40:45
economy or give up control, right, to
00:40:49
some social credit system. So I think
00:40:51
the open source is really important. And
00:40:53
by the way, that was Elon's instinct in
00:40:54
creating open AI. He was afraid that
00:40:56
Google was going to monopolize AI. So
00:40:59
he's like, let's create open AI so that
00:41:03
it's not dominated by a single company.
00:41:05
But that that is I think the right
00:41:07
answer here is I know people want to I
00:41:10
think their instinct to the idea of
00:41:12
powerful AI is to clamp down and just
00:41:14
control it. But actually you have to
00:41:17
have multiple players. That's the only
00:41:18
way you're going to be protected is is
00:41:20
to have multiple players.
00:41:21
>> This next wave of the market evolution I
00:41:25
think is going to be extremely high
00:41:27
stakes and messy. Nick, just throw this
00:41:28
up because I just want these guys to
00:41:30
react to it. So this is a company that I
00:41:33
just ran into on X called Rogo. And what
00:41:36
they did was they created a test bench
00:41:38
and a set of evals to be a financial
00:41:41
analyst essentially and tested all of
00:41:43
the frontier models. And it was so
00:41:46
interesting because they summarized I
00:41:48
read their paper that they published and
00:41:50
I quoted the most interesting part
00:41:51
because I see it everywhere now across
00:41:54
all EVALs which is this one phrase there
00:41:57
is no single best model anymore. at the
00:41:59
top of the leaderboard. Opus 47, GPT55,
00:42:02
Sonnet 46 appear almost
00:42:05
indistinguishable,
00:42:07
separated by less than, in this case,
00:42:09
you know, 3/10en of a percentage point
00:42:10
overall. Read superficially, the results
00:42:12
suggest convergence. Three frontier
00:42:14
systems reaching roughly the same level
00:42:16
of capability. Okay, why is that
00:42:18
interesting? Well, you got trillions of
00:42:20
dollars going into each of these guys to
00:42:22
trying to create these next superb
00:42:24
brain,
00:42:26
but increasingly our existing set of
00:42:28
evals and our existing capabilities
00:42:31
when applied on these models roughly
00:42:33
produce the same thing which
00:42:35
theoretically says that these things are
00:42:36
getting commoditized way too quickly.
00:42:37
And then you'd say, well, what's the ROI
00:42:39
on all this incremental spend, which is
00:42:41
a very interesting economic and
00:42:42
investment question. So I don't know
00:42:44
like gurley what do you think happens if
00:42:47
these evals continue to asmtote and we
00:42:51
need more and more and more money for
00:42:52
training
00:42:54
>> some of the smarter people in the open
00:42:56
source community have suggested to me
00:42:59
that we need more open-source connectors
00:43:02
of types so MCP uh is actually run by
00:43:06
the Linux Foundation and if you think
00:43:08
about any surface area where a model
00:43:11
might interact with other software the
00:43:13
more of those connectors that can be
00:43:16
open sourced and commoditized, it would
00:43:18
lower. This is what Google did with
00:43:20
Kubernetes uh to to try and commoditize
00:43:24
where workflows live off of AWS and to
00:43:28
make it easy to migrate. And so the more
00:43:31
you can create systems that make that
00:43:33
type of exchange you just described
00:43:35
super easy so that you can plug and play
00:43:37
the model and you have to worry about
00:43:39
things like context and how does context
00:43:42
come in and and data and you know stuff
00:43:44
that like glean and and data bricks do
00:43:47
but how anyway if you can do that if you
00:43:49
can create more of those connectors like
00:43:51
that then the models become swappable
00:43:54
and certainly with the both the model
00:43:56
companies trying to move up the stack
00:43:58
you have massive
00:43:59
desire from the app layer players to try
00:44:02
and figure this out and we already you
00:44:04
know watched what cursor is doing and
00:44:06
playing with their own model and being
00:44:07
forced to kind of reckon with the fact
00:44:10
that they're coming up the stack fast.
00:44:11
So I think that's a really good insight
00:44:14
that this gentleman shared with me and I
00:44:16
think we the founders and developers
00:44:19
that are out there should work on more
00:44:21
of these interfaces and throw them into
00:44:23
the open source world just to make it
00:44:25
more exchangeable, swappable. Is there
00:44:27
an issue right now with we don't have a
00:44:29
good harness for open source? I mean the
00:44:32
way that like claude is a harness for
00:44:36
>> Yeah. There's people making open-source
00:44:37
versions of this or building companies
00:44:39
around harnessing and building the
00:44:42
integrations into it. But open source is
00:44:44
always like the last to build the fit
00:44:46
and finish around the product. They
00:44:48
focus on the core of the product, right?
00:44:50
So like Linux for your desktop never
00:44:53
really took off because the interface
00:44:54
was never polished. The UI was never
00:44:56
like perfect, but there are companies
00:44:58
building that and I'll just I'll show
00:44:59
you one company that we invested in.
00:45:01
This is a company called Abacus and they
00:45:04
had a very simple idea. They came up
00:45:06
with their own hardware stack. They came
00:45:08
up with their own uh platform and now
00:45:10
they are sold out of these boxes that
00:45:13
they're building for insurance,
00:45:15
healthcare and everybody wants to run AI
00:45:18
inside their organization and then start
00:45:20
building their own models. We actually
00:45:23
incubated this in our incubator and you
00:45:25
can check it out goabacus.co.
00:45:28
They're just basically saying and
00:45:30
organizations cannot get enough of this
00:45:32
product. It is crazy how
00:45:36
savvy these organizations are getting
00:45:38
and Chamat you're doing it with 8090 as
00:45:39
well I think where they're just like we
00:45:41
have to build headless products so that
00:45:44
we don't get locked into
00:45:47
any one provider. Whenever we go into
00:45:49
the Fortune 1000, we never compete with
00:45:51
OpenAI or Anthropic, they'll have a
00:45:54
preference sometimes of what they want
00:45:55
to see under the hood. So, our control
00:45:57
plane can basically hot swap, as Bill
00:46:00
said, between one or the other. We've
00:46:02
also started to lay the seeds for open
00:46:04
source and open weights.
00:46:06
But the reason is because they don't
00:46:08
want to be tied into one of these
00:46:10
critical frontier labs. They want to be
00:46:12
able to ride the wave of innovation, but
00:46:14
they're afraid of two things. They're
00:46:16
afraid that one technology leaprogs the
00:46:18
other too quickly for them to
00:46:19
participate and they pick the wrong one.
00:46:21
And the second thing that they're
00:46:22
increasingly afraid of is terms of
00:46:24
service and being at the sake of a
00:46:28
frontier lab in a political philosophy
00:46:30
that they may be in the crosshairs of
00:46:32
accidentally. Right? So you're a
00:46:34
hospital system in Canada. You support
00:46:36
the euthanasia laws in Canada, but this
00:46:38
frontier model in America says, "No,
00:46:40
can't do it. So now we shut you off."
00:46:42
Right? That's an an example. I'm not
00:46:44
saying one is right or wrong. It's just
00:46:46
to illustrate the case. So a lot of the
00:46:48
folks that we see now in the fortune
00:46:50
1000 and increasingly the global 1000.
00:46:53
They want as Gurley said abstraction
00:46:55
above it. They want to sit as Sach said
00:46:57
in a control plane. They want to see be
00:46:59
at this level and they want to have the
00:47:01
flexibility because they don't know how
00:47:03
it's going to shake out. They see all
00:47:04
the money being invested at the model
00:47:06
layers but they see the model quality
00:47:07
asmtote. So they're like, "Wait a
00:47:09
minute, what are we supposed to do just
00:47:10
from a risk perspective?"
00:47:12
>> And and regula regulated industries are
00:47:15
particularly sensitive to these kind of
00:47:17
issues you're bringing up.
00:47:18
>> Hugely hugely sensitive and regul.
00:47:20
>> So if you just follow what finance,
00:47:22
healthcare, you know, and and those kind
00:47:24
of folks are doing, they're just like
00:47:26
this has to be onrem and they're very
00:47:28
concerned about a data leak and they're
00:47:30
very concerned about HIPPA compliance.
00:47:31
They're very concerned about training a
00:47:33
model. Like what if you know all of a
00:47:36
sudden somebody does you know a query or
00:47:39
or writes a prompt and it pulls some
00:47:41
information from that Canadian
00:47:43
healthcare system and all of a sudden
00:47:45
somebody gets a result and that sounds
00:47:47
farical. Remember stable diffusion
00:47:51
built themselves on Getty on Getty
00:47:53
images
00:47:54
>> and they all of a sudden the Getty image
00:47:56
watermark was in the output like system
00:47:59
you see anthropic and open AI in all of
00:48:01
these Fortune 1000s at the developer
00:48:04
layer cuz most of the developers have
00:48:06
their own credit cards they're allowed
00:48:07
to sign up for them you eventually wrap
00:48:09
them in an enterprise license so it's a
00:48:12
typical PLG-led market motion like we
00:48:14
saw in Slack we've seen it everywhere
00:48:16
the interesting thing is not that but
00:48:18
it's the unwind that happens then when
00:48:20
you have these huge licenses you have
00:48:22
these huge buckets of spend you can't
00:48:24
really tick and tie it together the CEOs
00:48:27
then wake up and are told by the CFO hey
00:48:29
FYI here's where we are Uber was one
00:48:32
example a second I don't know Nick if
00:48:35
you have this tweet but from Vivecarpali
00:48:37
the founder of Clover yeah this was just
00:48:39
yesterday overheard from a fortune 20
00:48:41
company CEO asked for a billion in AI
00:48:44
generated OPEX savings at the beginning
00:48:46
of the year so we're 6 months in the
00:48:48
team has spent $200 million on tokens
00:48:51
and with minimal results.
00:48:54
And so now they're in this weird motion
00:48:57
now where the CEO is pulling the budget
00:48:59
back and now you're having to cut the
00:49:00
licenses. You just saw Microsoft
00:49:02
announce that they're killing the claude
00:49:03
licenses.
00:49:05
It's a super dynamic market right now
00:49:07
and I don't think we know what the
00:49:09
terminal solution looks like. And by the
00:49:11
way, I would I would add Claude is
00:49:13
really good at product. Like Claude for
00:49:15
Excel is better than Copilot by not by a
00:49:19
little, by a lot.
00:49:20
>> And so, you know, anyone that's going to
00:49:23
run against them, they're they are a a
00:49:26
worthy foe, I should say.
00:49:28
>> Yeah. I think I think Claude is
00:49:29
exceptional, by the way. I mean, I use
00:49:31
it every day. Yesterday, I hit my token
00:49:33
limit on my pro plan. I had to put on my
00:49:35
credit card, spent another couple
00:49:37
thousand bucks, and I'm like, I was so
00:49:38
angry, but I did it because it's so
00:49:40
good.
00:49:40
>> Yeah.
00:49:41
>> Yeah. Go ahead, Sax. Wrap us up here.
00:49:44
Yeah.
00:49:44
>> Yeah. So, well, just to wrap up, let me
00:49:46
just connect a couple ideas. So, one is
00:49:49
that in terms of the the red capture
00:49:52
agenda that you're seeing in Washington,
00:49:55
I think where it's all leading to is an
00:49:56
effort to ban open source models or open
00:50:00
weight models.
00:50:01
>> There's a lot of breadcrumbs leading
00:50:02
here. I think people who want this are
00:50:04
being a little bit circumspect. They
00:50:05
don't feel like they're quite there in
00:50:07
terms of being able to justify it yet.
00:50:09
>> Can you explain it?
00:50:10
>> Sure. You look at you look at a lot of
00:50:11
the rhetoric around how models need to
00:50:14
have guard rails and that with open
00:50:16
source models, the guardrails can be
00:50:18
removed and therefore they're dangerous.
00:50:20
You see this rhetoric already in
00:50:22
anthropics blog posts. So, you know, any
00:50:24
threat that they describe, they kind of
00:50:27
go out of their way to take that shot at
00:50:30
open source models. you saw it with
00:50:31
respect to cyber for example or with
00:50:34
respect to bio threats things like that
00:50:36
I mean I've seen that type of language
00:50:38
repeatedly that open models lack guard
00:50:41
rails or the guardrails can be taken off
00:50:43
and therefore it's a problem and I think
00:50:46
again they're trying to create ideas or
00:50:50
put predicate facts in the public record
00:50:53
to justify an action later on and I
00:50:57
think it's just a matter of time before
00:51:00
they feel like they're at a position
00:51:01
where maybe they can push for that type
00:51:04
of ban directly. They're not quite there
00:51:07
yet.
00:51:07
>> But what does that do then to the rest
00:51:09
of the market? Like let's just say
00:51:10
America bans open source and open
00:51:12
weight. Okay. Well, what about the rest
00:51:14
of the world? I mean,
00:51:16
>> it sure it'll put
00:51:17
>> they're going to leap frog us.
00:51:19
>> Sure. You'll put the US on an island.
00:51:20
Well, first of all, as we all know, what
00:51:22
does it mean to ban a openweight model?
00:51:24
It's a file. It's a bunch of numbers,
00:51:26
you know, that you can run on your your
00:51:28
laptop.
00:51:29
>> Yeah. But what it will do is you think
00:51:32
about like all the cloud service
00:51:34
providers who run open models like they
00:51:36
will stop doing that because they got to
00:51:38
comply with the law and so all this
00:51:40
infrastructure that's been built up it
00:51:43
will get much harder to use open models
00:51:45
in the United States now the rest of the
00:51:47
world will continue to benefit from them
00:51:49
because there's a tremendous benefit in
00:51:51
terms of cost and customization and
00:51:54
control that you get with an open model
00:51:56
>> and we're on a completely different
00:51:58
price curve. And we haven't talked about
00:52:00
this yet. There was an economic and
00:52:02
capital mode to training that is going
00:52:05
away. It's going away in two ways. One
00:52:07
is because we're getting these domain
00:52:08
specific architectures at the silicon
00:52:10
layer. And then second, we're rebuilding
00:52:12
all of the core components. I don't know
00:52:14
if you guys saw yesterday, but Elon was
00:52:16
like, we've rewritten the entire
00:52:18
training complex in C and it's an order
00:52:20
of magnitude increase and we can run it
00:52:22
on 220,000 GPUs. at the scale of what
00:52:25
they're trying to do. Those kinds of
00:52:27
innovations are going to make the cost
00:52:29
of model training so much cheaper that
00:52:32
it's like, why would we stick to the $10
00:52:35
billion training runs when we can have
00:52:37
the $10 million training runs?
00:52:38
>> Well, if it got 1% better, just as a
00:52:41
thought experiment, Nick, could you find
00:52:43
Elon?
00:52:45
If okay, if it got 1% better, that's the
00:52:48
equivalent of 2,000 GPUs, which is the
00:52:51
equivalent of hundreds of millions of
00:52:53
dollars in compute. So every 1% equals
00:52:56
hundreds of millions in compute. If he
00:52:59
gets 10% 20% more efficient every
00:53:01
quarter, every
00:53:02
>> look at this speed improvement, the
00:53:05
speed improvement versus jacks for for
00:53:08
training runs is now an order of
00:53:09
magnitude. When you think about then the
00:53:11
capex buildout, the opex, the power,
00:53:16
the cabling, the copper, all of it.
00:53:21
And now this is a closed source model,
00:53:23
but I'm pretty sure that just that tweet
00:53:26
is going to get read by enough people
00:53:28
where there's going to be five or six
00:53:29
open- source stacks for training that
00:53:32
are rebuilt closest to the bare metal as
00:53:35
possible.
00:53:35
>> Yeah.
00:53:36
>> Why wouldn't you do that now? And so to
00:53:38
your point, Sax, cutting that off so
00:53:40
that we lose that kind of innovation
00:53:41
makes no sense to me.
00:53:43
>> I agree. And and like I said, I don't
00:53:45
know that the forces who want to ban
00:53:46
open source are strong enough or have
00:53:49
made the case or created the predicate
00:53:52
facts necessary yet to ban open source,
00:53:55
but I do think it is on the agenda and
00:53:57
it's where all the breadcrumb trails are
00:53:59
leading. So just watch out for that. I I
00:54:01
agree totally with with what David just
00:54:04
said and I wrote a blog post recently on
00:54:06
open source and and made the exact same
00:54:09
point.
00:54:09
>> I read that too. That was a good one too
00:54:12
on above the crowd.
00:54:13
>> No, it's not.
00:54:17
It was on the Santa Fe Institute.
00:54:20
>> Oh, no, it's it's the P3 Institute which
00:54:23
is my new my new institute. Anyway, um I
00:54:26
same same exact conclusion which is rest
00:54:29
of the world ends up running on Chinese
00:54:32
models if if if they're able to succeed
00:54:35
at what you just said.
00:54:36
>> And if you want to know the canary in
00:54:38
the coal mine sacks obviously the place
00:54:40
they love regulation most is the EU. So
00:54:43
EU has already done volley after volley
00:54:46
of proposed regulation for AI and open
00:54:50
uh and open source is particularly in
00:54:52
the crosshairs there because nobody's in
00:54:54
charge of it. So are you going to get a
00:54:56
bunch of open source contributors having
00:54:58
to vet their model with the EU
00:55:00
regulators like that's obviously not
00:55:02
going to happen. Nobody's in charge of
00:55:03
it. There just a bunch of contributors.
00:55:06
But open source is the solution I think
00:55:08
to
00:55:09
>> Yes, I agree.
00:55:10
>> It is the back stop. It is the backs
00:55:12
stop. I mean, unless you want to live
00:55:13
off the grid. I mean, if you want to
00:55:14
participate in the modern economy, it is
00:55:16
the backs stop. And let me just make one
00:55:18
other final point. It kind of maybe
00:55:19
leads into our next topic is I do think
00:55:21
that there is the potential for the
00:55:24
monopolization of this market to a
00:55:26
greater degree than people may be
00:55:28
pricing in right now. First of all,
00:55:30
we've seen that every other major tech
00:55:31
category has led to a monopoly or
00:55:33
duopoly situation. Seems to be the the
00:55:36
way that these things work out. But also
00:55:39
if you look at the growth rates right
00:55:41
now, Enthropic does seem to be pulling
00:55:43
away. There's a article in the
00:55:46
information showing the latest numbers
00:55:47
where I think Anthropic's now at they
00:55:49
seem to have pulled away from open AI,
00:55:51
which is not surprising and something I
00:55:53
I predicted. Look, if you have one
00:55:54
company that's growing at 10x
00:55:56
year-over-year and another company
00:55:58
that's growing at 3x year-over-year,
00:56:00
within 2 years, the first company will
00:56:03
have 90% market share. This is the power
00:56:05
of compounding, right? is just do the
00:56:07
math on it. 10 * 10 is 100. 3 * 3 is 9.
00:56:12
So again, if you just are able to
00:56:13
outgrow your competitor at that rate for
00:56:15
2 years, you will achieve monopoly
00:56:18
market share. Now there are reasons to
00:56:20
believe that anthropic cannot continue
00:56:22
that growth rate for 2 years. There's
00:56:24
going to be a competitive response. It's
00:56:26
already happened. Also, there may not be
00:56:28
enough compute to support that kind of
00:56:30
growth. There may be physical
00:56:31
constraints, but you'd always rather be
00:56:34
the company that has that inertia that's
00:56:36
on that totally trajectory than the one
00:56:38
that has to do something different to
00:56:41
then knock that leader off its current
00:56:43
trajectory.
00:56:44
>> Did you guys see what just hit the wire?
00:56:45
Nick, can you throw it up from Poly
00:56:47
Market? This is insanity. Poly Market
00:56:49
puts out there that an AI consultant
00:56:52
revealed that one of their clients
00:56:54
accidentally spent half a billion
00:56:56
dollars in a single month after failing
00:56:59
to set employee limits on clock usage.
00:57:03
>> What?
00:57:03
>> Oh my god, look at this. Look, the 16.6
00:57:08
million per day, almost 700,000 per
00:57:11
hour. Oh my god. Well, there seems to be
00:57:15
there seems to be a new like meme taking
00:57:17
shape that somehow like all this token
00:57:19
spend is is wasteful and basically
00:57:20
useless. And you know, we're constantly
00:57:23
oscillating between narratives like AI
00:57:25
is going to put everyone out of work to
00:57:27
like AI is useless and it's a bubble.
00:57:29
The doomers can't seem to make up their
00:57:30
minds whether AI is going to be our new
00:57:32
god or whether it's basically a total
00:57:34
waste of money and it's going to lead to
00:57:36
a bust. But in any event, yeah, I think
00:57:38
you know the there there's no question
00:57:40
that token efficiency is going to be a
00:57:41
big theme over the next year because the
00:57:44
spend has been ramping up way faster
00:57:46
than enterprise customers thought and
00:57:48
there's going to be a drive for
00:57:50
efficiency. Does that fundamentally
00:57:51
change the dynamics? I don't think so.
00:57:55
But it it might, you know, it might
00:57:56
temper the growth to some degree. Well,
00:57:58
and they've done a tremendous job making
00:58:01
people believe that tokens are free by
00:58:04
giving them these crazy deals like $20 a
00:58:06
month, you can do whatever you want.
00:58:07
$200 a month, you can do whatever you
00:58:08
want. And it's like everybody's leaving
00:58:10
the hose on, everybody's watering, and
00:58:13
then
00:58:13
>> you get a photo that says you've hit
00:58:15
your usage and it's like, "Come back at
00:58:16
230." I'm like, "230? It's 10:30. I
00:58:19
can't do anything between 10:30 and
00:58:20
2:30." And then it says, "Well, you can
00:58:22
put in your credit card." And so I did.
00:58:24
>> Yeah. But I mean it's it's literally
00:58:26
like the first the first 10,000 gallons
00:58:30
of water are free basically and then all
00:58:32
of a sudden it's like okay it's a penny
00:58:34
a gallon and then everybody in the
00:58:36
organization and this has literally
00:58:38
happened in our organization. One person
00:58:40
built like an interface for the founder
00:58:42
university program. Another person built
00:58:44
one. Then another person was like,
00:58:45
"Well, those two people got credit at
00:58:46
the management team meeting, so I'm
00:58:48
going to build an interface." And the
00:58:49
next person builds an interface. And
00:58:50
then everybody shipping like interfaces
00:58:53
and I literally had three different
00:58:55
people on the team make three different
00:58:56
versions of like a founder university
00:58:58
portal and I'm like, "We don't need
00:59:00
three. Can we get coordinated here?" And
00:59:02
it didn't get to the point of like
00:59:04
spending thousands of dollars, but it
00:59:05
certainly got to the point of spending
00:59:07
hundreds of dollars and it would have
00:59:08
gotten to tens of thousands.
00:59:10
>> Are we still on the first topic? What
00:59:11
are we doing? Well, no, we kind of
00:59:12
merged like two or three of them
00:59:13
together.
00:59:14
>> Oh, we did? Okay.
00:59:15
>> And it's super interesting. Trust me,
00:59:16
>> it's super interesting. I think what
00:59:17
Gurley said is one of the most
00:59:19
interesting things I have heard
00:59:22
>> in a long time.
00:59:23
>> Take people by their word. And if you
00:59:26
read their words,
00:59:27
>> if you just read their words and you can
00:59:30
understand what they're saying, you
00:59:32
don't have to guess about why they want
00:59:34
to have a digital guide. Well, now I'm
00:59:37
not the sharpest I'm not the sharpest
00:59:39
arrow in the quiver, but I can take down
00:59:41
a buck. And I can tell you that this
00:59:44
don't make a lot of sense to me. Even
00:59:46
the dullest arrow can take a buck down.
00:59:49
>> All right, let's get back to
00:59:52
It's so great having you here, Bill. We
00:59:53
missed you.
00:59:54
>> I got you. I got you.
00:59:55
>> We missed you, brother.
00:59:57
>> We're going to transition to the next
00:59:58
topic. There is some evidence that
01:00:00
Daario is mitigating his dumer rhetoric.
01:00:03
Did you see this?
01:00:04
>> Let me get to it. Yeah. Yeah, I got to
01:00:06
it here. All right, we we're going to
01:00:07
have to talk for the 16th time in the
01:00:10
last 18 months about AI's impact on
01:00:12
labor because again this chaotic
01:00:15
schizophrenic
01:00:17
interpretation of the data continues.
01:00:19
Cloudfare as we talked about last week,
01:00:21
shout out Matt Prince
01:00:24
Shimath's favorite CEO of the year.
01:00:27
Letter of the year
01:00:28
>> letter of the year. He cut 20%
01:00:30
>> award for the letter of the year
01:00:32
>> making friends every week.
01:00:34
here on the program. So they both blamed
01:00:37
AI spec explicitly and specifically and
01:00:40
Zuck then paired his 8,000
01:00:45
cuts at meta with the fact that he has
01:00:48
put uh spyw wear on everybody's laptop
01:00:51
to study every employee to make their
01:00:54
training data better. That got leaked
01:00:56
and people thought, hey, that's a Black
01:00:58
Mirror episode. We're we're working at
01:00:59
Meta in order to, you know, get our
01:01:02
two-year severance package. But on the
01:01:05
other side of the table, Goldman Sachs's
01:01:07
uh CEO, David Solomon, wrote an op-ed in
01:01:11
the New York Times. I'm the CEO of
01:01:13
Goldman Sachs. Period. The AI job
01:01:16
apocalypse is overblown. Period.
01:01:18
Obviously, he might be fighting for that
01:01:21
anthropic or open AI IPO in the coming
01:01:24
months, or maybe is doing it right now.
01:01:26
He made three points. AI won't eliminate
01:01:29
25% of jobs. It's going to automate 25%
01:01:32
of work hours and workers will fill that
01:01:34
time with higher level tasks. Obviously,
01:01:36
that didn't happen in the case of
01:01:37
Zuckerberg's layoffs. Just because a job
01:01:40
can be replaced doesn't mean it will be.
01:01:41
Bank tellers increased after ATMs. Live
01:01:45
entertainment became more popular after
01:01:46
TV. And the US labor market creates and
01:01:49
destroys 25 to 35 million jobs annually.
01:01:52
And the gross churn dwarfs net losses.
01:01:56
New categories like agentic AI
01:01:57
management are already hiring yada yada
01:02:00
yada. Uh a publication called Fortune is
01:02:02
apparently still publishing AI slop and
01:02:04
they say both Sam Wman and Daario have
01:02:06
walked back their AI job apocalypse
01:02:10
predictions as they gear up for an IPO
01:02:13
sax have at it. You know you've been
01:02:15
saying uh and your prediction was you
01:02:19
took the other side hey we're going to
01:02:20
create more jobs. There was a a recent
01:02:24
one of the job boards put out some stats
01:02:26
that the number of software jobs is
01:02:27
going up, the number of listings of
01:02:29
other jobs going down. So, I guess
01:02:31
you're probably in the camp of creative
01:02:33
destruction and churn at this point,
01:02:35
Sax.
01:02:37
>> Well, I mean, I think you should be
01:02:38
giving me more credit than that cuz my
01:02:40
most contrarian take back in January on
01:02:43
our prediction show is that AI would
01:02:46
lead to job gains, not job loss. And
01:02:49
over the past week, you've now seen the
01:02:52
narrative shift, I would say, almost
01:02:54
completely towards that position. So,
01:02:56
you have the CEO of Goldman Sachs right
01:02:57
in this in the New York Times. You know,
01:03:00
I don't think he'd be doing that if he
01:03:01
felt like he was completely stepping out
01:03:03
on a limb. Maybe even more importantly,
01:03:05
you had Sam and even Daario now walking
01:03:09
back their claims of massive job loss.
01:03:12
And they explained why Daario said, it's
01:03:15
kind of like the 25% of work hours
01:03:17
thing. He said that AI might automate
01:03:19
away 90% of someone's task, but the
01:03:22
other 10% will expand to do a whole
01:03:25
bunch of new new tasks and new things,
01:03:28
which is very similar to the the types
01:03:30
of of arguments that people like me have
01:03:33
been saying and actually that Jensen's
01:03:35
been saying that just because you
01:03:36
automate away some task doesn't mean
01:03:39
that you automate away the purpose of a
01:03:40
job. But now the worker is freed up to
01:03:43
do new things, to do the higher
01:03:45
complexity tasks that David Solomon, the
01:03:47
Goldman CEO, is talking about. So the
01:03:50
fact that Daario is now walking this
01:03:52
back and coming around to my position,
01:03:57
I think that that's kind of amazing. And
01:04:00
uh where do I go to get my apology? You
01:04:02
know,
01:04:02
>> well, we're going to have an ap we're
01:04:04
going to have an official apology form
01:04:06
that you can fill out. It's got check
01:04:08
boxes. I was wrong. I mean, some
01:04:11
mornings I woke up thinking, why am I
01:04:13
going out defending these guys? You
01:04:15
know, these idiots. I mean, they're
01:04:17
scaring the public with all these dire
01:04:20
predictions about an apocalyptic future.
01:04:23
There was no data to support that. I
01:04:25
mean, we can all debate what's going to
01:04:26
happen in the future, and we probably
01:04:27
should be humble about what is going to
01:04:29
happen in the future because we don't
01:04:30
completely know, and this industry is
01:04:32
very dynamic. But you have to look at
01:04:34
what is the data that we have so far in
01:04:36
the current situation. And we do not see
01:04:39
data that supports massive job loss. You
01:04:41
can cite this layoff or that layoff Jcal
01:04:44
those are anecdotes and the plural of
01:04:46
anecdotes is not data. If you look at
01:04:48
the actual data like Yale Budget Lab did
01:04:50
they said no discernable disruption in
01:04:52
the labor market in the last 3 years due
01:04:55
to AI they've done a comprehensive
01:04:57
study. You look at job postings for
01:04:59
software engineers. It's up 15%
01:05:02
year-over-year. Their job postings for
01:05:04
software developers have hit a new
01:05:06
three-year high despite the fact that
01:05:10
coding is the single breakout use case
01:05:12
of AI this year. So if AI has not caused
01:05:16
job elimination for software developers,
01:05:18
what category has it caused? I mean code
01:05:21
is now the number one use case I think
01:05:24
of AI in the enterprise.
01:05:26
>> Okay,
01:05:26
>> let's be honest. Over the last five or
01:05:28
10 years, a lot of companies overhired.
01:05:31
They mishhired. These CEOs did not have
01:05:34
a good handle on it. Their opex budgets
01:05:37
completely got bloated, inflated,
01:05:41
and they need to sort of get back to
01:05:43
where they were, get back to a fighting
01:05:45
weight. And it's this old adage of never
01:05:49
>> never waste a crisis.
01:05:50
>> Never let a good crisis go to waste.
01:05:52
Exactly. And so they point to this
01:05:53
thing. It's very simple to say. It's AI.
01:05:56
It's two letters. And say we're going to
01:05:57
fire people. But underneath that is not
01:06:00
AI because we know this. It hasn't done
01:06:02
anything measurable yet at the end
01:06:04
consumption of these tokens. Nobody is
01:06:07
standing there and saying look at my
01:06:09
filing
01:06:10
here is the lift that I have gotten.
01:06:12
Nobody has said that yet. That's very
01:06:14
important to observe. And so instead
01:06:16
what people are doing is realizing okay
01:06:18
I have this cover now to go and clean up
01:06:20
what was very poor management and
01:06:23
mismanagement over the last 5 and 10
01:06:25
years where I overhired and I mishhired.
01:06:28
That's what's happening today.
01:06:29
>> Okay, Bill Gurley, I'm going to let you
01:06:31
chime in here. You've got two besties
01:06:33
saying, "Hey, this is all hogwash. It's
01:06:36
AI washing. These jobs were just, you
01:06:39
know, the strategy obviously in Silicon
01:06:41
Valley was
01:06:41
>> they need a scapegoat. They need a
01:06:43
scapegoat.
01:06:43
>> They're hired two years ahead of time.
01:06:45
Build for the future and it was a vanity
01:06:47
metric and you were blocking talent from
01:06:49
working on other startups or
01:06:50
competitors. The Google strategy."
01:06:52
>> Hold on. Wait, wait, wait. You just said
01:06:54
the critical thing. That is exactly why
01:06:55
they did it.
01:06:56
>> Yes. That was the explicit strategy from
01:06:58
>> the actual strategy. These guys were a
01:07:00
wash in cash. And so part of it is you
01:07:02
were just hoarding talent or what you
01:07:04
thought was talent.
01:07:06
>> Yes. And just keeping them off the
01:07:07
market.
01:07:07
>> And now you're jettisoning it because
01:07:09
the reality is as companies get bigger,
01:07:11
their growth rates monotonically
01:07:12
decrease and you get to like a GDP plus
01:07:14
some number and your valuation
01:07:16
frameworks change and there's nothing
01:07:18
you can do to fight that law of gravity
01:07:20
in the public markets. And so as each of
01:07:22
these CEOs who at some point thought
01:07:24
they were different and the rules didn't
01:07:26
apply to them are now realizing you're
01:07:29
just like everybody else. Okay, we have
01:07:31
to stay humble as Sax said, but Bill
01:07:33
Gurley, would you like to apologize for
01:07:34
Sachs andor give him credit for his
01:07:36
incredible non-conensus?
01:07:38
>> He wasn't the Hold on. He wasn't the one
01:07:40
promoting the jobs apocalypse.
01:07:42
>> It was
01:07:44
you. I will give my thoughts in a
01:07:46
moment.
01:07:49
You're the for the mainream media.
01:07:51
>> I'll give mine.
01:07:54
>> You always represent the legacy media on
01:07:55
our show. Jay Cal, you have been in the
01:07:57
fourth.
01:07:57
>> I represent the legacy media represent.
01:08:00
You're the New York bluehaired.
01:08:03
>> I'm just giving you the statistics,
01:08:05
guys. I'm just presenting the numbers.
01:08:07
Now, let's remember anecdotes.
01:08:09
>> Let's remember.
01:08:09
>> Actually, let me give you an important
01:08:10
statistic. Let me give you a very No,
01:08:12
no. This is really important.
01:08:14
>> We have to let Bill Girly comment. Then
01:08:15
you can really important. Do you use
01:08:18
ketamine? I don't use ketamine. That's
01:08:20
the terrible drug. Do not use ketamine,
01:08:22
folks.
01:08:23
>> Bill Gurley, you have the floor.
01:08:26
>> I would just touch on two things that I
01:08:28
already said earlier. One, you know,
01:08:30
historically innovation has led to more
01:08:33
prosperity for humans. And I gave those
01:08:36
numbers from 1891 to today. I see no
01:08:39
reason why that won't happen here. In
01:08:41
the short run, from a bottomup
01:08:43
perspective, every human that wants to
01:08:45
protect themselves needs to be the most
01:08:47
AI enabled version of themselves they
01:08:50
can be. And the people that might be a
01:08:52
threat of job loss are someone who like
01:08:55
stands hard, fast, and refuses to use
01:08:58
AI. And I would just say that's simply
01:09:02
like saying, "I'm not going to use
01:09:03
email. I'm not going to use a
01:09:05
spreadsheet. I'm not going to use a
01:09:06
computer." And and you know, you
01:09:09
probably are at risk.
01:09:10
>> Yeah. Yeah. The paradigm will shift to
01:09:12
give you actually my position which is
01:09:15
>> Would you like me to give my position or
01:09:16
just want to jump?
01:09:17
>> Yeah, I do but I I never got to finish
01:09:18
that point. So, but I can do it after
01:09:20
you.
01:09:20
>> Yeah. Yeah. So, I I will give my
01:09:22
position on this which is there and it's
01:09:25
always been the same which is there's
01:09:26
going to be a massive job displacement
01:09:28
that occurs and that massive job
01:09:30
displacement is going to come because
01:09:33
CEOs in many cases believe that this
01:09:36
technology is going to make people more
01:09:38
efficient. they can do more with less
01:09:40
and they will be rewarded by the public
01:09:43
market by just having higher earnings
01:09:45
and we see that for every single
01:09:46
company. Now I fully concur it was
01:09:49
because of bloating and I gave my
01:09:50
position there. I know specifically that
01:09:52
Sergey and Larry took that strategy of
01:09:54
taking talent off the market so there
01:09:56
wasn't a Google competitor that was
01:09:58
literally explained to me by those
01:10:00
individuals. We hire people and then we
01:10:02
figure out what to do with them later.
01:10:04
That strategy permit just became the
01:10:07
standard in Silicon Valley and now it's
01:10:09
being reversed.
01:10:11
Now there will be wholesale jobs that
01:10:13
will be retired. If you look at
01:10:16
self-driving that's obviously happening
01:10:18
with Whimo with 3,000 vehicles and and
01:10:20
there'll be many more on the roads. That
01:10:22
job will be eliminated. We will be
01:10:23
sitting here in but 5 10 years and the
01:10:26
idea of somebody driving a taxi is going
01:10:29
to seem silly and dangerous. We will see
01:10:32
the same exact thing happen with
01:10:33
Optimist. You may have seen the figure
01:10:35
robot sorting packages. All those
01:10:37
sorting jobs at Amazon factories are
01:10:39
going away. Amazon themselves, these are
01:10:42
the savviest people in the world said,
01:10:44
"We are going to eliminate 600,000
01:10:46
future positions and we are going to cut
01:10:49
positions." And Andy Jasse said, "This
01:10:51
is going to be a reoccurring theme. As
01:10:53
we deploy AI, we will do more with
01:10:55
less." You will see the headcount at all
01:10:57
these big companies dramatically
01:10:59
decrease or stay the same as earnings
01:11:01
massively increase. And you can take the
01:11:03
position, Saxs, that oh my god, the
01:11:05
numbers are in my favor. They're not.
01:11:07
The numbers are in my favor. The job
01:11:09
loss is tremendous. And there are
01:11:11
numbers associated with that. 8,000
01:11:13
people at Meta after 20,000 before that.
01:11:16
And if you look at the steady state of
01:11:17
these companies, they has nothing to do
01:11:19
with AI. Let me finish.
01:11:21
>> They overhired.
01:11:22
>> No, no, no. We are beyond that. We are
01:11:24
beyond that. They are now getting rid of
01:11:25
people. When they say they're getting
01:11:27
rid of measurers, you can take them at
01:11:29
their word. When they say they're
01:11:30
getting rid of middle managers, you can
01:11:32
take them on their way and
01:11:33
>> scapegoating. That is you've given your
01:11:35
position already. I'm giving mine. My
01:11:36
position is they are obsessed with this
01:11:39
technology, they're obsessed with
01:11:40
earnings and they will continue that.
01:11:41
Now on the other side of the ledger, I
01:11:43
believe we'll have a Cambrian explosion
01:11:45
in startups and all these this talent if
01:11:48
they embrace the tools to Bill Gurley's
01:11:49
point are going to be able to solve more
01:11:51
problems and create small companies of
01:11:53
five or 10 people who are laid off from
01:11:55
Amazon or Meta and make double their
01:11:59
salary or have a better job that they
01:12:01
control. I believe that is going to be
01:12:02
the ultimate solution. But that
01:12:03
transition is going to be extremely
01:12:05
painful and we should have some humility
01:12:07
on this [ __ ] podcast for the people
01:12:09
impacted. Every cab driver is losing
01:12:11
their job. Every truck driver is losing
01:12:13
their job in the next 10 years. Anybody
01:12:15
sorting packages losing a job. Now you
01:12:17
can say all you want. You can say all
01:12:19
No, let me finish my thought. You can
01:12:21
say all you want, Chimov, that those
01:12:23
people don't want those jobs. But they
01:12:25
may need those jobs
01:12:27
>> and you are an elitist by definition. We
01:12:29
are all elitists on this program. We are
01:12:31
elite performers.
01:12:33
>> And these people are gonna lose their
01:12:34
jobs and they may not get a job very
01:12:36
quickly. By being able to call something
01:12:39
what we think it is
01:12:41
is not being elitist. It's actually
01:12:43
telling the truth. Meta over hired.
01:12:47
Okay? You could have stopped the company
01:12:49
at 3,000 people when I left and it would
01:12:51
not have changed the outcome of that
01:12:52
company. There was no need to go to
01:12:54
90,000 people and burn $50 billion on
01:12:56
VR. They did it because they had the
01:12:58
freedom to do it. That's allowed. It's
01:13:00
capitalism. Okay? They're coming back to
01:13:03
realize that there's a more efficient
01:13:05
version of what they are. That has
01:13:07
nothing to do with AI. That's the only
01:13:09
point I'm trying to make. All you have
01:13:11
to do is just say that.
01:13:13
>> I think you're wrong. And let me explain
01:13:15
to you why you're wrong. I believe
01:13:16
you're wrong. I believe Zuckerberg is
01:13:19
putting that software on people's
01:13:20
computers in order to to find more jobs
01:13:24
to eliminate to increase it. And the
01:13:26
surface area of problems in the world is
01:13:28
not decreasing, but what is uh
01:13:31
decreasing is the number of humans to
01:13:33
take on the next opportunity. And that's
01:13:36
going to continue. And I think the
01:13:37
companies that will be rewarded and
01:13:38
their stock prices will be rewarded are
01:13:40
the ones who do much more with much
01:13:42
less. And they're going to keep
01:13:45
eliminating these jobs. And I take them
01:13:47
at their word.
01:13:48
>> You don't have to explain everything
01:13:49
with conspiracy. Maybe they just
01:13:51
mismanaged for a period and they could
01:13:54
agreed on that. I I think that explains
01:13:57
the postcoid two or three years. I think
01:13:59
what we're seeing this year is actually
01:14:01
the tools working. the tools are working
01:14:04
and there are jobs that are no longer
01:14:06
needed. The measurers as Matthew Prince
01:14:08
pointed out or product managers or
01:14:10
designers, those have all been
01:14:12
consolidated into one job. Somebody who
01:14:14
ships a product and it doesn't require
01:14:17
12 people. It requires two people now. I
01:14:19
know
01:14:19
>> I don't think that's been consolidated.
01:14:21
I see it in Fortune 1000 companies all
01:14:23
the time. I don't think what you're
01:14:24
saying adopters. You're talking about
01:14:25
the slowest adopters. I'm on the front
01:14:27
line with
01:14:28
>> startup. These are where all the jobs
01:14:30
are. But I'm sorry, but a startup is not
01:14:31
going to go and enter a regulated market
01:14:33
and put JP Morgan out of business. Not
01:14:35
gonna happen.
01:14:36
>> They will eventually uh displace those
01:14:39
companies. It happens all the time.
01:14:41
>> Not going to happen.
01:14:42
>> We're going to agree to disagree.
01:14:43
>> Good luck to the startup trying to
01:14:44
disrupt Boeing. Good luck. I'm going to
01:14:46
take Boeing.
01:14:48
>> Okay. Well, some people might take
01:14:49
SpaceX. So,
01:14:50
>> good luck making drugs out of an Excel
01:14:53
spreadsheet. I'll take the regulated
01:14:54
pharma company. Good luck.
01:14:56
>> Sure. Listen, there are some industries
01:14:58
that are so much to one and you're going
01:15:00
to show up at the FDA and like, okay,
01:15:02
where's your team? Oh, it's just me. I
01:15:03
do it all.
01:15:05
>> Me and my model. Look at this.
01:15:06
>> You joke. Somebody just did that in
01:15:08
>> It's not a joke. It's not a joke. And
01:15:09
it's not going to happen because that's
01:15:11
not the way society wants safety,
01:15:13
predictability, governance, auditability
01:15:15
to work.
01:15:16
>> Yeah, I there's a distinct difference
01:15:19
between, you know, drugs and software
01:15:21
and services in the world. I think we
01:15:23
can agree on that. And listen, a regula
01:15:25
with truck driving is one of the most
01:15:27
regulated industries out there. So is
01:15:28
cab driving and taxis as Bill and I well
01:15:30
know and those jobs are being
01:15:33
eliminated. Bill, I'm gonna give you the
01:15:34
final word, then Sax, I'll give you the
01:15:35
final word.
01:15:36
>> A chance to respond.
01:15:37
>> Let's do Sax.
01:15:38
>> Okay, Sax, then Bill, go.
01:15:39
>> Well, first of all, Jake, you remind me
01:15:41
of the Troskyite who when confronted
01:15:44
with the fact that none of Trosky's
01:15:45
predictions had come true that simply
01:15:48
proved how far-sighted Trosky was.
01:15:50
>> I didn't go to graduate school. You're
01:15:52
going to need another reference.
01:15:53
>> In other words, none of your predictions
01:15:55
about job loss have come true. In fact,
01:15:56
the data
01:15:57
>> none zero data
01:15:59
>> except for what M just did last week.
01:16:01
But go ahead.
01:16:01
>> That's an anecdote. That is not
01:16:03
>> It's not an You're calling 8,000 people
01:16:06
losing their jobs an anecdote.
01:16:07
>> You don't hear yourself?
01:16:09
>> Hold on.
01:16:10
>> Do you hear yourself? It's not an
01:16:11
anecdote. 8,000 people lost their jobs.
01:16:13
>> Can I make my case? I heard you about
01:16:15
the the meta data point. First of all,
01:16:18
those jobs that job loss was not
01:16:20
directly attributable to AI. It just
01:16:22
wasn't. That's something you've invented
01:16:24
and put in the data.
01:16:25
>> Something Zuckerberg said. No, they they
01:16:27
clarified that. Okay. Okay. Sure.
01:16:29
>> He said it was related to they were
01:16:31
trying to balance additional spending
01:16:33
capex, but it was not directly related
01:16:35
to AI. But even if it were even if it
01:16:37
were 100% the case that was due to AI,
01:16:40
you're not netting those jobs against
01:16:42
all the other jobs that are being
01:16:43
created because of AI and all the new
01:16:46
companies that are being created right
01:16:48
now because of AI. So, you're just
01:16:50
cherrypicking one statistic. You're
01:16:53
attributing 100% of that to AI and then
01:16:55
you're not basically netting it and pre
01:16:57
presenting a balance. You've got
01:16:59
>> I'm not cherrypicking it. I am reading
01:17:01
the news and I'm describing what the CEO
01:17:04
said. Jack at Block said he's doing this
01:17:07
because of AI. Matthew Prince said it's
01:17:09
AI. Zuckerberg said it's AI.
01:17:12
>> I'm just taking them on their word.
01:17:14
>> Yes. Exactly. So Jack Dorsey came out
01:17:16
and said that he was going to do a 50%
01:17:18
elimination because of AI. Okay. And
01:17:21
within 24 hours, all the financial
01:17:24
analysts on X said that Jack was AI
01:17:26
washing and that block had horribly
01:17:29
overstaffed during COVID. It was running
01:17:31
much more inefficiently than all of its
01:17:32
other peers in this category. And
01:17:35
they've needed to do a 50% job cut for a
01:17:37
long time. So pretty much everyone
01:17:39
thought that was pure AI washing. In
01:17:41
fact, you've just proven my point. And
01:17:44
what exactly are the efficiencies that
01:17:46
Jack is getting? I mean, this is the
01:17:47
most handwavy thing ever that, oh, we're
01:17:49
just magically going to be able to
01:17:50
eliminate half our cost structure right
01:17:52
now.
01:17:53
>> Okay. So, Jack, Matthew Prince,
01:17:55
Zuckerberg, and Andy Jasse are all lying
01:17:58
and doing awashing.
01:18:01
This was due to AI. He just did it.
01:18:03
>> That's that's your reading of it. But
01:18:05
like I said, even if you attribute those
01:18:07
specific job losses to AI, which is
01:18:10
questionable, you're not netting it
01:18:11
against all the job creation that's
01:18:13
happening and also the new company
01:18:17
creation. Furthermore, let me just give
01:18:19
you some I specifically attributed that
01:18:22
the future and the new jobs will come
01:18:23
from startups. So don't misrepresent my
01:18:24
point. Thank you.
01:18:25
>> Okay, we currently have a 4.3%
01:18:28
unemployment rate in the economy.
01:18:30
Economists consider 5% to be full
01:18:33
employment. So basically unemployment is
01:18:35
at or near record lows right now despite
01:18:38
the of our lifetime despite the fact
01:18:40
that we're over three years into this AI
01:18:42
wave. Second and again this is the point
01:18:45
I wanted to make earlier with respect to
01:18:47
coding. Coding is the single job
01:18:49
category most impacted by AI right now.
01:18:52
We are at the point where AI is writing
01:18:54
most of the code. We have almost
01:18:56
complete automation of codew writing.
01:18:59
You would think that if you could look
01:19:02
at this in a simple Malthusian way, all
01:19:04
the software developers would be getting
01:19:06
laid off right now. Is that happening?
01:19:08
No. No. Software developers are not
01:19:10
being laid off on net. In fact, job
01:19:13
postings, job wrecks for software
01:19:15
developers are at a three-year high,
01:19:18
growing 15% year-over-year. Now, why is
01:19:21
this? I think the explanation is really,
01:19:23
really important. Okay, you look at code
01:19:26
commits on GitHub, which is the leading
01:19:27
code repository. There were 1 billion
01:19:30
code commits last year. In the past
01:19:33
month, there's been 1.1 billion. So, in
01:19:35
other words,
01:19:36
>> make something easier, more people do
01:19:39
it,
01:19:39
>> right? We have basically a 14x
01:19:42
year-over-year increase in code
01:19:43
generation. That code has to be managed
01:19:45
by somebody. You still need humans to
01:19:48
look under the hood. And when the amount
01:19:50
of code explodes and you get 10x or 100x
01:19:53
more code, the complexity also rises as
01:19:56
well. So look, we're not hiring 10 times
01:19:58
more engineers, but you do need more
01:20:00
engineers now to manage all of that
01:20:03
code. The other thing that's happening
01:20:05
is that there's been an explosion of the
01:20:08
use of code across the economy by
01:20:10
different businesses, different
01:20:11
applications, and different use cases.
01:20:13
I'm hearing from people who are now
01:20:15
hiring software engineers who never
01:20:17
would have hired them before. I was
01:20:18
talking to a fund manager, and he said
01:20:20
that his next two hires were not going
01:20:22
to be data analysts. they were going to
01:20:24
be software developers because they're
01:20:25
now deploying code for the first time in
01:20:28
ways that they were not before. This
01:20:30
goes back to my point about claude
01:20:32
proficiency being the most marketable
01:20:34
skill right now in the economy. People
01:20:37
are using these tools in entirely new
01:20:39
ways. I think that we're at the outset
01:20:42
of a boom right now caused by bespoke
01:20:45
software proliferating throughout the
01:20:47
economy and being used by firms that
01:20:49
never thought of themselves as tech
01:20:51
firms before. All of which is leading to
01:20:54
more productivity and that leads to a
01:20:56
healthier economy and that leads to more
01:20:57
job creation. And you're seeing that
01:20:59
again in the aggregate numbers and that
01:21:01
doesn't even include the bluecollar boom
01:21:04
that's happening right now with the
01:21:06
development of all this infrastructure,
01:21:07
the data centers and the new energy and
01:21:09
power generation. We are seeing hundreds
01:21:11
of thousands of new construction jobs
01:21:13
being created among bluecollars. Jan,
01:21:16
I'm sure you don't want them losing
01:21:17
their jobs by turning this boom off. So
01:21:20
again,
01:21:20
>> no I I never advocated you misconrue you
01:21:24
like to misconrue my position. I am very
01:21:26
clear there's job displacement going on
01:21:29
and the job displacement is related to
01:21:30
AI but net I do think the economy will
01:21:32
grow. Bill Gurley
01:21:34
>> maybe at some point in the future you'll
01:21:36
be right like the Trosky eye communism
01:21:38
has never been tried. Maybe it'll work.
01:21:40
>> Nobody knows your Trosky references.
01:21:42
Okay you lose you lost 95% of the
01:21:44
audience. Just speak like a normal
01:21:45
person.
01:21:45
>> Chimath laughed. Chimath understood it.
01:21:47
>> Okay great. I know the artist is smarter
01:21:50
smarter than you give them credit for.
01:21:51
>> Okay. No, I just think you're just
01:21:53
making these deep polls to try to sound
01:21:55
smarter than you actually are. Uh the
01:21:56
reality is the reality. These people are
01:21:59
being laid off because of AI. Bill
01:22:00
Gurley, of 20 million people in the
01:22:02
United States driving cabs and trucks
01:22:04
and doing that as a job right now, how
01:22:07
many of those do you think will lose
01:22:08
their jobs to self-driving in the next
01:22:10
decade or two based on just being in
01:22:12
there? And I'm not trying to lead the
01:22:13
witness here in any way. Obviously, some
01:22:15
people prefer a human driver, but what
01:22:17
what's your take on on that specific
01:22:18
part of the economy?
01:22:20
>> I think it's impossible to go with a
01:22:22
100% automated uh solution
01:22:26
because the the economics don't work
01:22:28
well. And so, I think like some of the
01:22:31
other examples that were given, ATMs and
01:22:33
whatnot, I I think the the use of
01:22:36
nonownership
01:22:38
cars is going to go way up. So, it's
01:22:40
going to keep growing through this and
01:22:43
humans are going to be used for like 50%
01:22:45
of it instead of a hundred. And so, I
01:22:48
might not be surprised if the number
01:22:50
actually stays the same or grows. And
01:22:54
let's remember these jobs didn't exist
01:22:56
before because regulation had limited
01:23:00
what the taxi market was capable of and
01:23:02
and getting around that actually led to
01:23:05
job creation. And so I I'm not a big fan
01:23:08
of the doomerism because around jobs,
01:23:11
you know, there's a word lite that that
01:23:13
kind of is used to to talk about it. And
01:23:17
I don't have high confidence in any
01:23:19
government program for skills
01:23:21
retraining. So it's not clear to me what
01:23:25
okay yes it's happened. What do we do
01:23:27
now? It's not clear to me. I think the
01:23:29
thing you can do the most one we already
01:23:32
talked about use the new tools. Know
01:23:33
what it's capable of in your field. like
01:23:35
get out there. And then two, if your job
01:23:38
is going to go away and maybe it's a job
01:23:39
you don't care about, start thinking
01:23:41
about where there are opportunities.
01:23:44
Everyone's talking about it. The skilled
01:23:46
trades are like we're we're short of
01:23:49
people.
01:23:50
>> Shortage for plumbing, electricians,
01:23:52
HVAC, all that. Yeah.
01:23:53
>> It's amazing how Jay uses facts that
01:23:55
haven't happened yet as like support for
01:23:57
his argument. Like you just state that,
01:24:00
oh, all the truck drivers are losing
01:24:01
their jobs. All the drivers are losing
01:24:03
their jobs. And then you say that this
01:24:05
proves my take. Yeah, I know it's your
01:24:07
belief, but that is not proof. Do you
01:24:09
understand?
01:24:10
>> The proof I gave was Amazon and Andy
01:24:11
Jasse, Shopify,
01:24:14
they were doing that before Mark Benny
01:24:16
off automated Amazon. Let me ask Bill
01:24:20
everybody knows our Amazon car being
01:24:23
delivered so you cherry pick anecdotes
01:24:25
and then misattribute them to AI.
01:24:27
>> They literally have a self-driving
01:24:28
division. It's called Zuks.
01:24:29
>> You're the biggest AI washer there is.
01:24:31
They are the largest
01:24:33
user of robotics in the world. So yes,
01:24:36
Chimath, they are pursuing robotics
01:24:39
massively more than anybody and they are
01:24:41
pursuing self-driving.
01:24:42
>> You just you just like all these words
01:24:44
together. At one point it's a warehouse
01:24:46
worker, then it's a driver, then it's
01:24:47
Amazon.
01:24:49
It's just
01:24:49
>> it's not it's not you can you can
01:24:51
personally attack me all you want. The
01:24:53
the issue here is self-driving is going
01:24:56
to take away I believe the majority of
01:25:00
>> Okay, that's the key word. Great. Let's
01:25:02
put it there as a belief. Who knows? You
01:25:04
don't know and I don't know.
01:25:05
>> Okay. And I think the same robotics. But
01:25:07
I will take people I will take people at
01:25:10
their word. I'm curious, Bill, your take
01:25:12
on
01:25:12
>> these large enterprises. You've heard
01:25:14
two positions here.
01:25:15
>> I have a question for you.
01:25:17
>> I have a legit question for you.
01:25:18
>> Can I just let the guest be involved,
01:25:20
please? Sax your monopoly. actually
01:25:22
engaging with your with your
01:25:23
perspective. Explain to me. No, no. This
01:25:25
let me let me truly ask you.
01:25:27
>> Okay.
01:25:27
>> Explain to me why job postings for
01:25:30
software engineers are up 15%
01:25:31
year-over-year despite the fact that
01:25:33
code has now been fully automated.
01:25:35
>> Oh, I think there's a Cambrian explosion
01:25:37
in uh software. You're absolutely
01:25:40
correct. And I believe people who know
01:25:41
how to vibe code or to like who are
01:25:44
non-developers are making bespoke
01:25:46
software. I've said that a hundred times
01:25:47
on this podcast over the last few years
01:25:49
and I predicted it. So absolutely I
01:25:51
believe that will be an area of job
01:25:52
growth. I believe the positions that are
01:25:55
being removed or I just I know based on
01:25:57
what we're hearing is product managers,
01:26:00
middle managers, what Matthew Prince
01:26:02
call measurers, what other people call
01:26:04
mid management. Everybody believes that
01:26:06
the recording and this daily standups
01:26:10
and the uh zoom calls all of that is
01:26:13
turning into people management is being
01:26:15
done better by AI and people are more
01:26:18
self-directed and then the stack of
01:26:20
people to build products is being
01:26:21
consolidated right it's like the the
01:26:24
typical designer can now vibe code the
01:26:26
developer can do front-end design and UX
01:26:29
and they can project manage themselves
01:26:30
so I there there are a series of jobs
01:26:33
that will increase and a series of jobs
01:26:35
that will be eliminated just like the
01:26:37
mail room got eliminated and mess bike
01:26:40
messengers got eliminated
01:26:42
that
01:26:43
>> got on net on net do you think there'll
01:26:45
be mass
01:26:46
>> do you think there'll be on well your
01:26:47
position is shifting a little bit do you
01:26:49
think on net do you think on net
01:26:51
there'll be mass job loss
01:26:54
>> uh I think there is a chance that we're
01:26:56
going to see uh job loss increase in the
01:26:59
short to midterm and then eventually
01:27:02
the displaced people are going to have
01:27:03
to learn or leave the workforce, which
01:27:05
is what happened during other
01:27:07
revolutions like this. Some people went
01:27:09
with the paradigm and adapted and some
01:27:11
people didn't and just retired. That I I
01:27:13
saw that firsthand in the PC revolution
01:27:15
as but one example. Some lawyers just
01:27:17
would never use these tools and they
01:27:19
just retired at 55 65 and they moved on.
01:27:22
And then other attorneys were PC first
01:27:25
and they just took that.
01:27:25
>> By the way, did you guys see the news
01:27:26
that Kirkland Ellis is going to spend
01:27:29
half a billion dollars to roll their own
01:27:31
Frontier model?
01:27:32
>> Makes total sense. That was like to our
01:27:33
earlier point today is that people are
01:27:34
doing on prem and going to make their
01:27:36
own models. Bill, I have one specific
01:27:37
question for you and thank you for the
01:27:39
good engagement there, Sax. It was it
01:27:41
lacked the ad homonym that usually uh
01:27:43
starts every conversation we have.
01:27:45
>> I don't usually call you an idiot.
01:27:46
>> That's because it's in our minds. Okay,
01:27:48
we're thinking about it. We're just not
01:27:50
saying it.
01:27:50
>> Good. I like it better. I like it
01:27:52
better. Um Bill, specifically when Andy
01:27:55
Jasse, you know, uh last spring said,
01:27:57
"Hey, we're going to do more with less.
01:27:59
We're going to be AI first." and they
01:28:00
said, "We're not going to hire these
01:28:01
600,000 jobs." When you see Tubby Lucky,
01:28:04
he say, "You have to do AI first before
01:28:06
you ask for a headcount and prove to me
01:28:07
that you tried AI first before hiring
01:28:09
somebody." Do you think this is
01:28:13
a sign that these organizations are AI
01:28:16
washing or do you think these recent
01:28:18
ones are more, hey, we're going to do
01:28:20
more with less and and the size of these
01:28:22
companies will be smaller uh because of
01:28:24
AI? One thing that that I think that
01:28:27
last question misses and I think a lot
01:28:30
of the the AI dumerism stuff misses is
01:28:34
that competition exists. And so if you I
01:28:38
don't think there's any scenario where
01:28:40
you just do more for less and all of a
01:28:42
sudden everyone has 70% operating
01:28:45
margins. That's not going to happen.
01:28:48
Someone else is going to come along and
01:28:50
do more for less and lower the price.
01:28:52
And so the thing that could happen is we
01:28:55
could have a productivity boom from
01:28:57
lowerpriced goods and services and the
01:28:59
basket of goods that humans are able to
01:29:02
buy gets cheaper and cheaper and cheaper
01:29:04
and that's been true in many categories.
01:29:06
Unfortunately, it's offset by what
01:29:08
happens in healthcare and other
01:29:10
regulated industries.
01:29:11
>> Yeah. Education. But but yeah, so I
01:29:14
expect products to get cheaper and
01:29:17
people to be able to create more with
01:29:19
less. But I don't think it leads to
01:29:22
obscene
01:29:23
profits
01:29:25
>> because it'll be whittleled away in
01:29:26
competition.
01:29:27
>> Okay.
01:29:28
>> By the way, just just on this AI washing
01:29:30
point, there's a a trial lawyer named
01:29:33
Donnie King. He's a securities
01:29:34
litigation partner at a firm called
01:29:36
Acriman. He and his colleagues have
01:29:40
started to warn that we could start
01:29:42
seeing shareholder lawsuits against
01:29:45
companies that engage in this type of AI
01:29:47
washing cuz he thinks it's a type of
01:29:49
puffery, right? Because essentially what
01:29:51
the company is doing is attributing
01:29:53
their own non-performance or their
01:29:55
operational issues to AI when in fact
01:29:58
there are real problems in the business
01:30:00
and therefore it could be a form of
01:30:02
securities fraud.
01:30:03
>> Wait, securities fraud? Yeah.
01:30:05
>> I want to double click on this. Did you
01:30:06
see the CEO of Whisk today on his note
01:30:08
about layoffs?
01:30:09
>> No. Who's Wisk?
01:30:11
>> Find me the AI washing in there. Wix.
01:30:13
>> Oh, yeah. Those that's the website
01:30:14
builder. Yeah. I mean, you can build
01:30:16
websites with Claude. Yeah. The whole
01:30:19
website business is challenged. Yeah.
01:30:21
>> Interesting that he just he just talked
01:30:23
about operational details.
01:30:25
>> Did he? I I didn't read the note.
01:30:26
>> Of course, you didn't read the note.
01:30:28
>> Well, I mean, you said it just happened.
01:30:31
>> I I will read it. Uh, this is I'm just
01:30:35
this broke at 9:25 this morning.
01:30:37
>> I think it's really interesting that
01:30:39
this lawyer thinks there's so much AI
01:30:41
washing going on that he thinks it could
01:30:43
constitute securities fraud and he's
01:30:44
warning clients not to engage in it. But
01:30:46
look, Jake Al, you're like the last
01:30:48
person who hasn't gotten the memo on
01:30:49
this. There was a huge narrative shift
01:30:51
this week. Sam is backing off this. Even
01:30:53
Daario's walking it back. You got the
01:30:55
Goldman Sachs CEO.
01:30:58
You got the explosion in job postings.
01:31:00
Everyone's coming around to the idea
01:31:03
that the job apocalypse is massively
01:31:06
overblown.
01:31:07
>> I mean, it could be overblown looking at
01:31:09
the stats.
01:31:10
>> Your apology, I'm happy to accept.
01:31:12
>> No need for an apology. I mean, my
01:31:13
position has always been apologize.
01:31:15
>> It's displacement. You're going to have
01:31:16
some displaced in the short to midterm
01:31:18
and then eventually there'll be more
01:31:20
problems to solve and people will have
01:31:21
to reallocate. I do think we're being uh
01:31:24
uh, you know, I think the tech industry
01:31:26
itself doesn't have enough empathy or
01:31:29
enough thoughtfulness when discussing
01:31:31
this because these are real people
01:31:32
losing real jobs and you can point at
01:31:35
statistics uh and you think they're
01:31:37
spinning, but these are real people
01:31:39
losing real jobs who may not make the
01:31:41
transition.
01:31:42
>> Jason, I got I got
01:31:43
>> Do you think it's more empathetic to be
01:31:45
scaring the be Jesus out of people that
01:31:47
they're going to lose?
01:31:47
>> I'm not in the scaring camp. I'm not in
01:31:49
the scaring camp. I I I am in the
01:31:51
enabling camp. That's why I keep saying
01:31:52
if you've been laid off, you should
01:31:54
start a company and you should embrace
01:31:56
the tools. So, I I'm I'm all about
01:31:57
empowering people. I think they if they
01:31:59
learn the tools, they'll have 10 job
01:32:00
offers. Uh and they'll start their own
01:32:02
companies. So, I do think there's a
01:32:04
solution to it. I just think we're going
01:32:05
to go through, you know, low millions of
01:32:08
jobs being lost or being retired and
01:32:11
transitioned out over this next couple
01:32:13
years.
01:32:14
>> I was just going to be empathetic and
01:32:15
offer some solutions. So we talked about
01:32:18
the skills trade deficit of of people
01:32:22
working in that. Micro has a foundation
01:32:24
called Micro Works where they fund they
01:32:28
funded $16 million 2600 people get a
01:32:32
free scholarship for to be become a
01:32:34
plumber, welder or electrician. So go
01:32:37
check that out if you want to reskill. I
01:32:38
think
01:32:39
>> generation tool belt. Yeah.
01:32:40
>> I think it's better than you know having
01:32:42
the government fix things. And then, you
01:32:44
know, as as we started and talked about,
01:32:46
I've got a new uh grant program myself
01:32:48
to help people
01:32:50
>> Yes.
01:32:50
>> kind of tilt their career in a different
01:32:52
direction. Go do something you love and
01:32:53
and apply and maybe I can help fund your
01:32:57
you moving in that direction.
01:32:58
>> And and as to your vibe shift, I think
01:33:00
it's because candidly people's houses
01:33:04
have been Molotov cocktailed because
01:33:05
they're doomerism. And people
01:33:07
specifically are citing that when they
01:33:08
shoot at their houses and throw Molotov
01:33:10
cocktails at them twice in the same
01:33:12
week. And if you're IPOing and you're
01:33:15
coming out saying, "Hey, jobs are going
01:33:16
away. Jobs are going away." That's just
01:33:18
a really bad look.
01:33:19
>> And or it's because we called it out and
01:33:21
they got cut and so now they were
01:33:23
telling the truth.
01:33:24
>> All right, everybody. This has been
01:33:26
another amazing episode of the AllIn
01:33:28
podcast. Thanks for coming.
01:33:30
>> No, no, no. Sorry. I need to do one
01:33:32
thing
01:33:34
>> just doing at the end of this.
01:33:35
>> Yes.
01:33:36
>> She's a friend of ours. I just want to
01:33:38
just give a huge shout out to Tulsy
01:33:39
Gabbard and specifically her husband
01:33:41
Abraham. is tragic.
01:33:42
>> Going through some really tough stuff
01:33:44
with cancer. He is going to kick its
01:33:47
ass. I just wanted to say we love you.
01:33:50
>> Yeah. Uh Tulsi is great.
01:33:52
>> Cheers everybody. Uh that's episode 275
01:33:54
in the can. We'll see you next time.
01:33:56
Bye-bye. Love you Blues.
01:33:58
>> We'll let your winners ride.
01:34:01
>> Rainman David.
01:34:06
>> We open sourced it to the fans and
01:34:08
they've just gone crazy with it.
01:34:10
>> Love you. Queen of Kino.
01:34:19
>> Besties are gone.
01:34:21
>> That is my dog taking notice in your
01:34:23
driveways.
01:34:26
>> Oh man, my appetiter will be.
01:34:29
>> We should all just get a room and just
01:34:30
have one big huge orgy cuz they're all
01:34:32
just useless. It's like this like sexual
01:34:34
tension that we just need to release
01:34:35
somehow.
01:34:37
>> Wet your feet. Wet your feet. her feet.
01:34:42
>> We need to get mer.
01:34:52
I'm going all in.
