00:00:00
How many PRs you think are going to get
00:00:01
pushed to the core structural internet
00:00:03
in 100 days? What's the overunder
00:00:05
number? Cuz I'll give you a number.
00:00:06
>> You're going to say zero. My my answer
00:00:08
to that is
00:00:09
>> I'll say like 10,000. But it's going to
00:00:11
be immediately
00:00:11
>> if it prevents your browser history from
00:00:13
being released to everybody in the
00:00:15
world, Chamath, that may be something
00:00:17
that you're willing to, you know, let
00:00:18
100 days pass on.
00:00:19
>> I think you got Chimat's attention when
00:00:20
you said browser history.
00:00:21
>> What about the dickpicks?
00:00:25
>> Chamat is he's going to release them
00:00:27
himself.
00:00:30
We'll let your winners ride.
00:00:38
>> We open source it to the fans and
00:00:40
they've just gone crazy with it. Love
00:00:41
you.
00:00:46
>> All right, everybody. Welcome back to
00:00:47
the number one podcast in the world.
00:00:48
David Freeberg is out this week. But in
00:00:51
his place, the one, the only,
00:00:54
our fifth bestie, Brad Gersonner. I
00:00:56
mean, why don't you ever give me puts a
00:00:58
little namaste in your payday anymore?
00:01:00
You used to be
00:01:02
>> I'm going to bring back the greatest
00:01:03
moderator, but now it's just kind of You
00:01:06
know what? These guys beat me up. They
00:01:08
beat me up and they just beat the the
00:01:10
joy out of me doing this program.
00:01:13
>> It's because you're a Roana apologist
00:01:15
now.
00:01:16
>> No, I We'll get into it. Okay. Save it
00:01:18
for the [ __ ] Roana apologist. just
00:01:21
because I said like, "Hey, they've
00:01:23
stopped [ __ ] maxing and they've
00:01:25
started doing like some logical things."
00:01:28
Uh, yeah. Okay, here we go.
00:01:30
>> It's great to be here. Great to be here.
00:01:31
>> Good to have you. Good to have you here.
00:01:33
And of course, uh, we have David Saxs is
00:01:37
back. Everybody wants to hear from David
00:01:38
Sax. We missed you last week, bestie.
00:01:40
>> We didn't beat the joy out of you. We
00:01:42
just try to beat some of the hot air.
00:01:44
Turn
00:01:45
>> any any fluff that you can put on the
00:01:47
show that just involves you talking and
00:01:50
saying nothing is
00:01:52
>> that's the stuff we got.
00:01:53
>> Turn up. Yeah. Turn up. Okay. Yeah,
00:01:55
we'll cut it right out. Um we'll cut it
00:01:58
out and we'll just put a promo in for
00:01:59
the syndicate.com. Thank you. Also with
00:02:01
us, Jamalet is here.
00:02:04
>> How's your [ __ ] maxing going since
00:02:06
last week? Did you have a a a [ __ ]
00:02:08
maxing full weekend? Did you have a good
00:02:10
full weekend of just smoking cigars in
00:02:12
the back deck and not ruminating about
00:02:14
all the chaos you've caused in the last
00:02:16
20 years?
00:02:17
>> I think I've done generally more good
00:02:20
>> than than not.
00:02:22
>> Oh, you have. But there's been some
00:02:24
chaotic moments. Don't think about it.
00:02:26
>> You can't, bro. You can't have ups
00:02:28
without downs, man. It's like, what are
00:02:30
you there to do? Just like plate
00:02:31
everybody and be a loser? Are you there
00:02:33
to be a winner? Yes, you're in the
00:02:35
arena, but have you stopped going to
00:02:37
therapy after realizing ruminating?
00:02:40
>> What's up with this uh sudden interest
00:02:42
in [ __ ] maxing? Are you like the
00:02:44
clavvicular for [ __ ] maxing?
00:02:47
>> No, the world finally caught up with me.
00:02:49
That's it. What do you I mean, I've been
00:02:50
[ __ ] maxing this whole time. THEY JUST
00:02:52
DIDN'T HAVE A name for it, guys.
00:02:54
>> Wow. Okay. Eli's videos are really good.
00:02:56
I watched two more this week.
00:02:58
What take us through what's so appealing
00:03:01
about not ruminating, smoking a cigar,
00:03:03
and just living your life?
00:03:05
>> Because what he says actually works at
00:03:07
every level of society and every sort of
00:03:10
thing that you may want to achieve. Even
00:03:12
if you're trying to like climb the
00:03:14
rungs,
00:03:17
you very quickly learn that the more you
00:03:19
want something, the less you're going to
00:03:20
get it. And I think that's like his real
00:03:23
message is let go, live life, and just
00:03:27
try stuff or don't try stuff. And I
00:03:29
think that that detachment is really
00:03:31
healthy for people. I like it. I like it
00:03:34
a lot.
00:03:35
>> Who's the guy who says this? I actually
00:03:37
didn't know.
00:03:37
>> Elisha Long. Well, Eli, I think, is how
00:03:40
he goes by.
00:03:41
>> But he's fantastic. He Mark YouTube
00:03:43
channel.
00:03:44
>> Mark Andre found him
00:03:46
>> and he's like, "This is this guy is the
00:03:48
new guy.
00:03:49
>> Modernday philosopher. He gives you a
00:03:51
road map for how to live your life,
00:03:52
right? A new age sage.
00:03:54
>> What's the name of the guy? The
00:03:55
character's name from Dune.
00:03:57
>> I was into girls books. I was dating
00:04:00
girls.
00:04:01
>> He's the Lison Algib of the modern
00:04:03
internet.
00:04:04
>> This is why we need Freeberg here is to
00:04:06
explain these deep holes. All right,
00:04:09
listen. We got a lot to get to. Don't
00:04:10
The basic point is build something and
00:04:12
don't ruminate. Okay, ruminating is just
00:04:14
not worth it. Just everybody go for it.
00:04:16
>> No, just do stuff. Stop blathering in
00:04:18
your own head. Just do stuff.
00:04:19
Absolutely. All right. Listen, speaking
00:04:21
of doing so, Anthropic is withholding
00:04:23
its newest model, Mythos. I'm using the
00:04:26
Greek uh pronunciation, its newest
00:04:28
model, Mythos, uh saying it is far too
00:04:31
dangerous for any of us to have access
00:04:33
to it. According to the company, the
00:04:35
model autonomously found thousands of
00:04:37
vulnerabilities, including bugs in every
00:04:39
major operating system and web browser.
00:04:42
This uh little study they did included
00:04:45
20 year old exploits that had been
00:04:47
missed by security audits for decades.
00:04:49
Uh some examples, they found a 27y old
00:04:51
vulnerability in OpenBSD used in
00:04:54
firewalls and critical infrastructure.
00:04:55
They found a 16-year-old bug in FFmpeg
00:04:59
that was missed by automated tools after
00:05:01
5 million scans. The Linux kernel, all
00:05:05
kinds of uh bugs they found. They
00:05:07
released a hype video hyping up why they
00:05:11
were not going to share this model.
00:05:13
Here's Dario. Come on the program
00:05:15
anytime, brother.
00:05:16
>> But as a side effect of being good at
00:05:17
code, it's also good at cyber.
00:05:19
>> The model that we're experimenting with
00:05:22
is by and large as good as a
00:05:25
professional human at identifying bugs.
00:05:28
It's good for us because we can find
00:05:30
more vulnerabilities sooner and we can
00:05:32
fix them.
00:05:33
>> It has the ability to chain together
00:05:35
vulnerabilities. So what this means is
00:05:37
you find two vulnerabilities, either of
00:05:39
which doesn't really get you very much
00:05:40
independently, but this model is able to
00:05:43
create exploits out of three, four,
00:05:45
sometimes five vulnerabilities that in
00:05:47
sequence give you some kind of very
00:05:49
sophisticated end outcome.
00:05:50
>> All right, Brad, uh, by the way, that
00:05:51
set they're using there, that's the same
00:05:53
room those guys play Dungeons and
00:05:54
Dragons in every Sunday. Brad, you're
00:05:58
Brad, you're an investor in this
00:06:00
company. Is this virtue signaling or is
00:06:03
it reality? Is this a good move by them
00:06:06
to not release this model and be
00:06:08
thoughtful, give it to a handful of
00:06:09
people and just find all the bugs it can
00:06:12
before releasing it to the public? And
00:06:14
we've got a lot more issues to discuss.
00:06:16
>> I I actually think they deserve a ton of
00:06:18
credit here and let me walk you through
00:06:20
why, right? They the company could have
00:06:22
just released Mythos, broken a lot of
00:06:24
core things on the internet. Often times
00:06:26
in Silicon Valley, we say move fast and
00:06:28
break things. In this case, it means
00:06:30
just releasing the model to move further
00:06:31
ahead of your competition. But here the
00:06:33
company realized it would wreak havoc.
00:06:35
They ran their own vulnerability
00:06:37
testing. They saw that it would allow
00:06:39
offensive hacking and people to expose
00:06:41
browsers and browser history, expose
00:06:43
credit cards, you know, on the internet.
00:06:46
So, you know what I like about this is
00:06:48
they didn't need government to hold
00:06:50
their hand on this. We have plenty of
00:06:52
government regulations. They know it's
00:06:54
in the best long-term interest of the
00:06:56
company and the industry, you know. So,
00:06:58
they set up Project Glass Wing. It's an
00:07:00
AIdriven, you know, kind of cyber
00:07:02
coalition. Apple, Microsoft, Google,
00:07:05
Amazon, JP Morgan, 40 of the most
00:07:07
important companies. And their goal is
00:07:09
very simple. Let's spend a 100 days use
00:07:11
advanced AI to find and to fix and to
00:07:14
harden these software vulnerabilities
00:07:17
before hackers exploit them. Now, what I
00:07:20
think this represents, Jason, is a
00:07:22
threshold that we're crossing. Mythos
00:07:25
and Spud, which is going to be out from
00:07:27
OpenAI any day now, which is the first
00:07:30
Blackwell trained model at OpenAI. They
00:07:33
represent the beginning of what I would
00:07:35
call AGI models. These are models with
00:07:38
massive step function improvements and
00:07:40
intelligence. Um, and they're just too
00:07:43
smart to be released immediately,
00:07:45
you know. And by the way, there was
00:07:47
nothing that said that every time you
00:07:49
you finish a model, you got to
00:07:51
immediately release it GA. So they set
00:07:53
up this idea of sandboxing, building
00:07:55
defensive alliances,
00:07:57
you know, in order to move away from
00:07:59
that regime. I I think it shows, and
00:08:01
Saxon and I have talked about this a
00:08:03
lot, so I'm interested to hear what he
00:08:04
thinks. It shows you can trust the
00:08:07
industry and market forces in
00:08:10
coordination with the government. They
00:08:11
were talking to the government about
00:08:12
this. But they're not relying on some
00:08:15
top- down regulation in order to do
00:08:18
this. They laid out a blueprint that
00:08:19
seems to me very pragmatic that now that
00:08:22
we're at this threshold, we're going to
00:08:24
sandbox these things. I think that open
00:08:26
AAI will end up doing the same thing. I
00:08:28
think Google will end up doing the same
00:08:30
thing. It's an aggressive way to keep
00:08:32
the RA, you know, the pressure on and
00:08:34
and win the race at AI while making the
00:08:37
tradeoffs to protect safety. So, you
00:08:39
know, I think you're always going to
00:08:40
have to make these trade-offs. I think
00:08:42
in this case, it was a great move by
00:08:44
Dario and team and I think they deserve
00:08:46
a lot of credit. Sachs, when you look at
00:08:48
this, we had Emil Michael on the program
00:08:50
a couple weeks ago. It might have been
00:08:51
four or five weeks ago, and we had a
00:08:53
very thoughtful discussion about, hey,
00:08:55
if the government is going to have these
00:08:57
tools, you know, an anthropic wants to
00:08:59
withhold them and, you know, what is the
00:09:02
proper relationship there, you have to
00:09:04
think that the government, and I know
00:09:06
you don't speak for all parts of the
00:09:08
government. If you were just going to
00:09:10
run through the game theory, they must
00:09:11
have gone to the government and said,
00:09:12
"Listen, this thing is so powerful, it
00:09:14
can put together two or three hacks,
00:09:16
create a novel attack vector, and this
00:09:19
is incredibly dangerous. What if China
00:09:21
has it? And if this thing is as powerful
00:09:23
as Daario says it is, then this is an
00:09:26
offensive weapon as well for us to take
00:09:28
out, let's just pick, you know, uh, a
00:09:31
pressing issue, the North Korea's
00:09:34
ballistic missile program. This is
00:09:36
equivalent the way it's being described
00:09:38
as the Manhattan project perhaps. So
00:09:41
what are the chances two-part question
00:09:43
for you Sax that China already has this
00:09:45
and is using it and do you think Daario
00:09:48
is doing the right thing by regulating
00:09:50
themselves?
00:09:51
>> I think Anthropic has proven that it's
00:09:54
very good at two things. One is product
00:09:57
releases. The second is scaring people.
00:10:00
And we've seen a pattern in their
00:10:02
previous releases of at the same time
00:10:05
they roll out a new model or new model
00:10:07
card, something like that. They also
00:10:09
roll out some study showing really the
00:10:12
worst possible implication of where the
00:10:15
technology could lead. We saw this last
00:10:17
year about a year ago. They rolled out
00:10:19
this blackmail study where supposedly
00:10:22
the new model could blackmail users.
00:10:25
There's been a whole bunch of these
00:10:26
things. Actually, I went back to Grock
00:10:28
and I just asked, "Hey, give me examples
00:10:30
where Antropic has basically used scare
00:10:32
tactics and it's it's a pattern." Okay,
00:10:35
it's a pattern.
00:10:36
>> Okay,
00:10:37
>> these guys, I'm not saying it's not
00:10:40
sincere, but they have a proven pattern
00:10:42
of using fear as a way to market their
00:10:45
new products. And if you think back to,
00:10:48
again, my favorite example is this
00:10:50
blackmail study where they prompted the
00:10:53
model over 200 times to get the result
00:10:55
they wanted. And that result was was
00:10:58
clearly reverse engineered and it got
00:11:00
them the headlines they wanted. And I
00:11:02
would say the proof that it's reverse
00:11:04
engineered is we're now a year later.
00:11:06
There's a bunch of open- source models
00:11:08
out there that have the same level of
00:11:10
capability that that anthropic model
00:11:13
had. And have you seen any examples of
00:11:15
blackmail in the wild? I don't think so.
00:11:18
So in other words, if that study were
00:11:22
true in the sense of being a likely
00:11:23
outcome of that model, I think you would
00:11:25
see examples in the wild of that
00:11:27
behavior. And we haven't seen any of
00:11:29
that in the past year. Now, let's talk
00:11:31
about this specific example with cyber
00:11:33
hacking.
00:11:34
>> I actually think that this one is more
00:11:37
on the legitimate side. I mean, look,
00:11:39
the reason why I bring this up is
00:11:41
anytime Anthropic is scaring people, you
00:11:42
have to ask, is this a tactic? Is this
00:11:45
part of their Chicken Little routine? or
00:11:47
is it real? You know, are they crying
00:11:49
wolf or not? I actually would give them
00:11:50
credit in this case and say this is more
00:11:53
on the the real side. It just makes
00:11:55
sense, right? So that as the coding
00:11:58
models become more and more capable,
00:11:59
they're more capable of finding bugs.
00:12:01
That means they're more capable of
00:12:02
finding vulnerabilities. And like one of
00:12:04
their engineers said, that means they're
00:12:05
more capable of stringing together
00:12:06
multiple vulnerabilities and creating an
00:12:08
exploit. And so I do think that over say
00:12:11
the next 6 months we're going to have
00:12:13
this call it one-time period of catching
00:12:16
up where AIdriven cyber is going to be
00:12:20
able to detect a whole range of of bugs
00:12:22
that maybe have been dormant over the
00:12:24
past 20 years across a wide range of
00:12:27
systems. And so I do think that there is
00:12:30
real risk here. And I do think therefore
00:12:33
that having this pre-release period
00:12:35
makes a lot of sense where they're
00:12:36
giving the capability to all these
00:12:39
software companies that have existing
00:12:40
code bases to use the tool to detect the
00:12:43
vulnerabilities for themselves so they
00:12:45
can patch them before these capabilities
00:12:47
are widely available. And by the way, it
00:12:49
won't just be anthropic that makes these
00:12:52
capabilities available. We know that
00:12:54
like let's say the Chinese open source
00:12:55
models like Kimmy K2, it's about 6
00:12:58
months behind. So we have a window here
00:13:00
of maybe 6 months where we're still in
00:13:03
this pre-release period where I think
00:13:06
companies that have large code bases can
00:13:09
get advanced access to this model and uh
00:13:12
I guess open AI is going to release a
00:13:14
similar thing in the next few weeks. I
00:13:16
do think that every company or IT
00:13:18
department or CISO that is managing code
00:13:23
bases should take this seriously and use
00:13:26
the next few months to detect any again
00:13:29
like dormant bugs or vulnerabilities and
00:13:32
and roll out patches. If everybody does
00:13:34
their job and reacts the right way, then
00:13:36
I do not think it will be the doomsday
00:13:39
scenario that Anthropic is sort of
00:13:41
portraying. But it's one of these things
00:13:43
where the fear might end up being a good
00:13:45
thing in order to drive people to in
00:13:49
order to drive the correct behavior. So
00:13:51
>> sure,
00:13:51
>> I ultimately think this is going to work
00:13:52
out fine, but you do need everyone to
00:13:55
kind of pay attention, use the
00:13:56
capabilities,
00:13:58
>> fix the bugs, then we're going to get
00:13:59
into a big arms race between AI being
00:14:02
used for cyber offense and AI being used
00:14:04
for cyber defense, but it'll be a more
00:14:06
normal sort of of period. Chimath, we
00:14:09
have uh Daario and uh you know a number
00:14:12
of the participants here are taking this
00:14:14
super seriously. They're making a big
00:14:16
statement. Zach's very nuanced uh I
00:14:19
think take there. What's your take on
00:14:21
how do these companies have it both
00:14:23
ways? Hey, this is shouldn't be
00:14:26
regulated. This should be regulated. If
00:14:28
this is in fact a cataclysmic, oh my
00:14:31
god, they're going to hack everything.
00:14:33
What if the Chinese have this right now?
00:14:35
That would speak to more government
00:14:37
either coordination, regulation, or some
00:14:39
kind of relationship between the CIA,
00:14:42
the FBI for domestic stuff, and these
00:14:45
companies because there it is a non-zero
00:14:48
chance that the Chinese have an equal
00:14:50
capability here. We're assuming they're
00:14:52
behind, but who knows what they're doing
00:14:54
behind closed doors. So, what's your
00:14:55
take on this? Is it uh The Boy Who Cried
00:14:58
Wolf, or is this the real deal? Now,
00:15:00
>> I think it's mostly theater.
00:15:03
>> Okay.
00:15:04
In February of 2019,
00:15:07
when Daario was still at OpenAI,
00:15:10
they did the same thing with GPT2.
00:15:14
That was a 1.5 billion parameter model,
00:15:17
which sounds like a total fart in the
00:15:19
wind in 2026. But at that time, this 1.5
00:15:24
billion parameter model was supposed to
00:15:26
be the end of days. And it was supposed
00:15:28
to unleash this torrent of spam and
00:15:30
misinformation. And that was the big
00:15:32
bugaboo at the time. And so what
00:15:34
happened? They went through this
00:15:35
methodical roll out over six or nine
00:15:37
months. They started releasing the
00:15:39
smaller parameter models and then they
00:15:41
scaled up to the big 1.5 billion
00:15:42
parameter model. And at the end of it,
00:15:44
it was a huge nothing burger.
00:15:47
If you actually think that Mythos is
00:15:49
capable of doing what it says it can do,
00:15:52
two things are true. One is a very
00:15:55
sophisticated hacker can probably do
00:15:57
those things right now with Opus.
00:16:00
And two, if these exploits
00:16:04
are this easy to find,
00:16:07
whether you use Opus or whether you use
00:16:09
Mythos, the reality is you'd have to
00:16:10
shut down the internet for about 5 years
00:16:12
to patch them all. So when you see like
00:16:15
a large multi- trillion dollar gang,
00:16:20
it's a bit of theater. Why? What do you
00:16:22
think they can actually accomplish in 2
00:16:24
months? Do you actually think that if
00:16:26
there's these vulnerabilities, it's all
00:16:28
going to get fixed? Let's give them six
00:16:30
months. Let's give them nine months. But
00:16:33
the reality is that capitalism moves
00:16:35
forward, the funding needs moves
00:16:37
forward, and the need for these guys to
00:16:40
build adoption moves forward. And that's
00:16:43
going to supersede
00:16:45
what this is. So I do think that Sax is
00:16:47
right that they have figured out a very
00:16:50
clever go-to market muscle here and a go
00:16:54
to market motion that activates
00:16:57
hyper attention and hyper usage and so I
00:17:00
give them tremendous credit and I'll
00:17:02
maintain what I've maintained before.
00:17:04
Anthropic is shooting the lights out
00:17:06
right now. This is like Steph Curry
00:17:08
going bananas from every everywhere on
00:17:10
the court. These guys are hunking
00:17:12
threes.
00:17:13
>> It's all in that. Okay. So huge kudos to
00:17:16
Anthropic,
00:17:18
but we've seen it before. We saw it when
00:17:21
these folks were the principal
00:17:23
architects at OpenAI who are now seeing
00:17:25
the same playbook here. I think we'll
00:17:27
look back and I think what we'll say are
00:17:30
these two things. One is if we're really
00:17:31
going to patch all these security holes,
00:17:33
we need to shut down the internet
00:17:35
for some number of years, honestly,
00:17:37
literally years. And the second is an
00:17:40
advanced hacker can probably do this
00:17:41
today with Opus if they really wanted
00:17:44
to.
00:17:46
>> Okay. Hey Brad, I gota I'll get you in
00:17:47
here for the for the last word. I I'm
00:17:49
going to go with Yeah, maybe they did uh
00:17:52
Crywolf before, but based on what I see
00:17:55
with these models advancing and using
00:17:57
them and I'm using a lot of the open
00:17:58
source ones right now from China. I
00:18:00
think that this is like code red kind of
00:18:03
moment. This is Defcon. like we should
00:18:05
be taking this deadly seriously and I
00:18:07
think these companies got to coordinate
00:18:08
with the CIA and this is uh equally a
00:18:11
defensive as offensive opportunity. Do
00:18:14
you think this
00:18:15
>> you're asking for the nationalization of
00:18:16
AI now?
00:18:17
>> No, I actually I I I don't think it
00:18:20
should be nationalized. Um although I
00:18:21
did see people sort of insinuating that.
00:18:24
I think these companies need to build a
00:18:26
group Brad that work and coordinate with
00:18:28
the CIA. I assume that they're already
00:18:30
doing this. I'm assuming you Emil
00:18:32
Michael and uh you know Trump and
00:18:35
everybody have these people in a room
00:18:36
and that they've given the defcon and
00:18:39
said hey how can our government use this
00:18:41
to stop bad actors and this is already
00:18:44
being coordinated with the CIA and the
00:18:45
FBI. I am 100% certain of that that
00:18:48
Dario went to them and said look what we
00:18:50
found. This is the real deal. I'll give
00:18:52
you the last word on this Brad since
00:18:53
you're an investor in both companies and
00:18:55
you know them quite well. the frontier
00:18:56
model forum which was which was put
00:18:58
together in 23 um is cooperating on
00:19:01
anti- and adversarial distillation stuff
00:19:04
as we speak right they don't want to
00:19:06
make it easy on you know so Google and
00:19:08
and open AAI and and anthropic they're
00:19:11
coordinating on this stuff you know
00:19:13
there are times where I've pushed back
00:19:14
on anthropic because I thought it was
00:19:16
you know perhaps regulatory capture or
00:19:18
something else this is very different in
00:19:20
my mind right he could have easily Dario
00:19:22
could have easily come out and said oh
00:19:24
my god we passed a threshold we need to
00:19:25
have a government moratorum. Remember,
00:19:28
even our friend Elon called for a
00:19:29
six-month moratorium in 2023 because of
00:19:32
civilization risk. This guy didn't do
00:19:34
that. Instead, he said, "Okay, what what
00:19:37
should we do? I'm going to get 40 of the
00:19:38
leading companies together. We're going
00:19:40
to spend a 100 days sandboxing,
00:19:42
hardening the systems, and then we're
00:19:43
we're we're going to keep pushing
00:19:44
forward."
00:19:45
>> What do you honestly think is going to
00:19:46
get accomplished in a 100 days? How many
00:19:48
PRs you think are going to get pushed to
00:19:50
the core structural internet in 100
00:19:52
days? What's the overunder number? Cuz
00:19:54
I'll give you a number. You're gonna say
00:19:55
zero. My my answer to that is
00:19:57
>> I'll say like 10,000, but it's going to
00:19:59
be immediate.
00:20:00
>> But if it prevents your browser history
00:20:02
from being released to everybody in the
00:20:03
world, Chimat, that may be something
00:20:05
that you're willing to, you know, let a
00:20:07
100 days pass on.
00:20:08
>> I think you got Chimat's attention when
00:20:09
you said browser history.
00:20:10
>> What about the dickpicks?
00:20:14
>> As Chimat is, he's going to release them
00:20:16
himself right now. CHIMAT'S LIKE, "HEY,
00:20:18
CHINESE HACKERS, HERE ARE MY DICKPICKS.
00:20:20
Please put them out."
00:20:21
>> Oh my god. we have to be out there
00:20:24
complimenting when they're doing the
00:20:25
right things or relying on the market
00:20:26
rather than running to the nanny state
00:20:28
and saying do more of this. So this to
00:20:30
me was just an example of of a of a good
00:20:32
balance. I'm sure we're going to have
00:20:34
plenty of debates about this in the
00:20:35
future. But you know this is one I would
00:20:37
like to see more of.
00:20:38
>> This is why to use your word Jake I
00:20:40
tried to have a more nuanced take is
00:20:42
because we have no choice but to take
00:20:44
this seriously. Whether it's total
00:20:46
theater, whether it's fear-mongering,
00:20:49
and they do have a pattern around this,
00:20:51
we can't take the risk, right? And it
00:20:53
does logically make sense that as these
00:20:56
models become more and more capable at
00:20:58
coding, they're going to get better at
00:20:59
cyber. And there's going to be that one
00:21:02
time period where you're moving from
00:21:04
preAI to post AAI, and you need a patch
00:21:06
for that. So, my guess is we're going to
00:21:08
see a lot of patches over the next few
00:21:10
months. I think that that will resolve
00:21:12
the problem.
00:21:14
I think this is a case where I'm going
00:21:16
to give them the benefit of the doubt. I
00:21:18
I think that, you know, I've criticized
00:21:20
him in the past. I think that blackmail
00:21:22
study was embarrassing to the level of
00:21:25
being a hoax, but I think in this case,
00:21:28
I'm going to give him credit and say
00:21:29
that I think that it's legit.
00:21:31
>> So, it's not the anthropic hoax. This
00:21:33
could be legit. I, you know, looking at
00:21:36
>> we have no choice but to treat it that
00:21:37
way.
00:21:38
>> Of course. Yeah. I mean, even if two
00:21:41
things could be true at the same time,
00:21:42
Saxs, they could have used this tactic
00:21:44
before. It could be performative, like
00:21:46
the video with the dramatic music in the
00:21:48
background. It does have a little bit of
00:21:50
drama to it, and the way they presented
00:21:53
it is very dramatic, but it does make
00:21:55
logical sense that the one company that
00:21:59
made the bet on code bigger than anybody
00:22:02
else would be the one who would discover
00:22:04
this quickest. And you know in a 100
00:22:05
days that's a pretty good um that's a
00:22:08
pretty big advantage versus the hackers.
00:22:10
But let me think one more point there
00:22:11
Jimat
00:22:12
>> the most important thing that people
00:22:14
haven't talked about here is the amount
00:22:17
of code being pushed right now because
00:22:18
of these tools is 10x 100x in most
00:22:21
organizations. So we need to have this
00:22:24
type of security embedded in these new
00:22:26
coding tools to do it in real time.
00:22:29
That's the opportunity. There should be
00:22:31
real time correcting of this. If this
00:22:33
was real, they picked the wrong
00:22:34
companies. Meaning, there are energy
00:22:38
companies, folks that control nuclear
00:22:41
reactors. There are airplane companies
00:22:44
that are flying hundreds of thousands of
00:22:45
people in essentially manufactured
00:22:48
missiles of like streaming gas going at
00:22:52
500 miles an hour. None of those
00:22:54
companies were the ones that were
00:22:55
included in this. And so I think if you
00:22:58
really thought that this was end of
00:23:00
days,
00:23:02
at a minimum we can agree maybe we
00:23:04
should have expanded the circle a touch.
00:23:07
Well, maybe those are customers of the
00:23:09
ones they're including here. Anyway, uh
00:23:11
this is a really important story. We'll
00:23:12
obviously track it in the coming weeks
00:23:14
to see what turns out to be reality. And
00:23:17
uh Daario, do come on the program at
00:23:19
some point. Hey uh Brad, will you get
00:23:20
Dario to come on the program? I've
00:23:21
invited him like three times. I got his
00:23:23
phone number. He's ghosted me. I don't
00:23:24
know why.
00:23:24
>> Wait, he he's ignored you? I get
00:23:27
>> I literally got an introduction from the
00:23:29
number like one of the number one
00:23:30
venture capitalists in the world. He's
00:23:31
on the cap table very early. He just
00:23:33
won't respond. I don't know why.
00:23:35
>> I would tell you Daario's podcast with
00:23:38
Dwarkish who I think is an excellent
00:23:39
podcaster. I've listened to that three
00:23:41
or four times taken notes every time. It
00:23:44
is a really exceptional piece really
00:23:46
exceptional piece of work by by by them.
00:23:48
>> All right, let's keep moving. We got a
00:23:50
lot on the job.
00:23:50
>> You may once again be tarred with your
00:23:52
affiliation with us.
00:23:54
>> Poor you. I mean, I don't care.
00:23:56
Literally, I I've got friends on both
00:23:58
sides of the aisle. I have friends
00:24:00
>> of course you do.
00:24:01
>> Even JCAL.
00:24:02
>> Even JCAL has friends everywhere.
00:24:04
>> Let me ask Brad a question here just
00:24:06
while we're on the topic of anthropic.
00:24:07
There was a really interesting story or
00:24:10
tweet I guess you could say by the
00:24:11
founder of OpenClaw
00:24:13
that
00:24:14
>> Peter.
00:24:14
>> Peter. Yeah. What's his name? Peter
00:24:16
Steinber.
00:24:17
>> Steinberger. Steinberger.
00:24:18
>> Steinberger. Yeah.
00:24:19
renowned coder created openclaw which is
00:24:22
kind of the thing that launched this
00:24:24
whole agent era now you I guess you
00:24:25
could say any event he said that
00:24:28
anthropic was cutting off his access to
00:24:31
was it was to claw is that the next
00:24:33
topic
00:24:34
>> this is on the docket it's a little bit
00:24:36
nuanced
00:24:37
everybody using openclaw would take
00:24:39
their $200 a month subscription to
00:24:41
anthropic which was essentially like a
00:24:43
people were using more tokens and it's
00:24:45
an average the people from openclaw it
00:24:48
is very verbose and those people are
00:24:51
100x the usage of the average
00:24:53
subscriber. So he said you can't use
00:24:56
your 200, you have to use the API. You
00:24:58
move from the $200 plan to the API, add
00:25:00
a zero to your token use. So or more.
00:25:03
And so they essentially anchored
00:25:06
Open Claw and then 10 days later or less
00:25:11
they released or announced their new
00:25:12
agent technology which is according to
00:25:14
them a safer better version of OpenClaw.
00:25:17
So, hey, all fair in love and war and
00:25:20
they have basically shot a huge cannon
00:25:23
across the bow of openclaw.
00:25:25
>> Wait, can you just explain that exactly?
00:25:27
So, so I think you're right that they
00:25:29
systematically copied feature by feature
00:25:32
>> of open claw, incorporated that into
00:25:34
clawed and then the coupross was
00:25:36
basically cutting off open
00:25:38
>> oxygen. Can you just explain exactly
00:25:40
what they did? Okay, very simply, when
00:25:43
you buy a subscription to these
00:25:44
services, they have blended your usage
00:25:48
across many users. So there's, you know,
00:25:50
nine out of 10 users use less than the
00:25:52
tokens they're paying for and the top
00:25:54
10% use much more. When OpenClaw became
00:25:58
a phenomenon, the number one open source
00:26:00
project in history on GitHub with all of
00:26:02
this usage, people went crazy. And you
00:26:05
heard me talking about how crazy I went
00:26:06
for it. those people with the $200
00:26:08
subscriptions were using $2,000 $20,000
00:26:11
worth of tokens. So they said you can no
00:26:14
longer use your subscription to, you
00:26:16
know, either your professional or
00:26:18
enterprise subscription at $200
00:26:20
>> and plug that into your open claw. You
00:26:22
now have to go to the API and pay per
00:26:24
usage. So no more like
00:26:27
>> unlimited. If you use Anthropic's own
00:26:29
agent harness, are you part of the
00:26:31
bundled flat rate? You can assume that
00:26:34
that's what they'll do, which if you
00:26:35
were thinking on an antirust level might
00:26:38
be token dumping or price dumping. I'm
00:26:40
not saying like I'm ratting them.
00:26:42
>> No, it's like bundling, isn't it?
00:26:43
>> Well, price dumping or bundling. When
00:26:45
you price something under the market
00:26:47
price in antitrust, that would be price
00:26:49
dumping, right? And if you were to
00:26:51
bundle, it would be like the bundling
00:26:53
issue.
00:26:54
>> Critically important. You can use
00:26:56
openclaw via claw API and every company
00:26:58
has a right to set the price for its
00:27:00
products. It's just saying that you were
00:27:02
for under their current regime, they
00:27:04
were selling dollars for 10 cents via
00:27:06
OpenClaw because these were such power
00:27:08
users and now they're just saying we
00:27:10
have to price this rationally, but we're
00:27:11
happy to have you guys use the API. So,
00:27:13
>> okay. Okay. But Brad, when you use the
00:27:16
OpenClaw competitor that Anthropic now
00:27:18
offers,
00:27:19
>> correct?
00:27:20
>> Are they subsidizing that? Are you
00:27:22
paying?
00:27:22
>> We don't know yet because it's in closed
00:27:24
beta. So in other words, what I'm saying
00:27:25
is if they charge for API usage, right,
00:27:29
their own first party agent harness or
00:27:32
system, then that would be apples to
00:27:33
apples. But if
00:27:35
>> if they end up charging the bundled flat
00:27:38
rate, let's say, for their stuff, but
00:27:41
then charge the metered rate for third
00:27:43
party stuff, you could make a bundling
00:27:45
argument.
00:27:46
>> Sure. Sure. And you could say it's
00:27:47
anti-competitive assuming that Anthropic
00:27:50
has dominant market share in coding,
00:27:52
which I think most people would say they
00:27:53
do at this point.
00:27:54
>> And assuming that it's the same product,
00:27:56
I mean, the reason most enterprises will
00:27:58
probably use the Anthropic uh version of
00:28:03
this agentic product is because it meets
00:28:05
all of your security parameters, right?
00:28:07
So, Altimter runs, you know, a lot of
00:28:10
stuff on Enthropic. They're already
00:28:12
integrated within our our data
00:28:13
warehouse, our data lake, things of that
00:28:15
nature. So just letting openclaw loose
00:28:18
on the uh altimeter you know data set
00:28:20
would not be wise and so it's a
00:28:22
different fundamental product.
00:28:24
>> No I get that and I think that anthropic
00:28:26
has a huge advantage let's say cloning
00:28:28
open claw and just building it into
00:28:30
claude. I'm not denying that to me that
00:28:33
would be the reason why they don't need
00:28:34
to do price discrimination is because
00:28:36
there's already a very good reason
00:28:38
>> to use the let's call it the bundled
00:28:40
offering on a featured basis. But the
00:28:42
question I'm specifically asking is
00:28:44
whether they're giving themselves a
00:28:46
price advantage because
00:28:48
>> I think Brad is giving the the most
00:28:50
generous interpretation. You're taking a
00:28:52
more cynical one. I'm with you, Saxs.
00:28:53
I'm 100% on the cynical side. Open Claw
00:28:56
is so powerful. It's got so much
00:28:58
momentum that not only is anthropic
00:29:02
trying to ankle it. I believe when Sam
00:29:04
Waltman bought it, it was uh and he
00:29:06
didn't buy OpenClaw itself, he hired
00:29:08
Aqua hired Peter. I believe it was to
00:29:10
subvert the open- source project to get
00:29:12
Peter's next set of genius ideas inside
00:29:15
of OpenAI as opposed to letting them go
00:29:17
there. People are going to say I'm a
00:29:19
conspiracy theorist, but this is the
00:29:21
number one focus and let me just give
00:29:23
you a list of who is trying to kill
00:29:25
OpenClaw/compete with them. Obviously,
00:29:28
you have Anthropic, but also Perplexity
00:29:31
Computer launched. It's awesome. I've
00:29:33
been using it. Anthropic has this clawed
00:29:36
managed agents. They dropped that on
00:29:38
Wednesday, April 8th. uh yesterday uh
00:29:40
today's Thursday when we tape you you
00:29:41
guys listen on Fridays and then you have
00:29:44
Hermes agent that was released on
00:29:47
February 25th that's also open source
00:29:49
and very good so that's in the open
00:29:50
source camp Alibab is coming out with
00:29:52
one that's going to be based on their
00:29:54
Quinn model then you have Elon who said
00:29:57
he's got something called rock computer
00:30:00
coming out of macro hard which is a play
00:30:01
on words for Microsoft in addition to
00:30:04
that Amazon and Apple are preparing uh
00:30:06
new releases of their uh [ __ ] maxing
00:30:09
assistants Alexa and Siri that will be
00:30:11
less [ __ ] in this new version and
00:30:14
then nothing out of uh SAT and Microsoft
00:30:16
yet. So the number one goal I believe in
00:30:20
the large language model frontier model
00:30:23
space is to kill this open source
00:30:25
product.
00:30:26
>> No, I mean come on like why they're
00:30:29
building multi-functioning agents that
00:30:31
can move from answering questions to
00:30:33
actually doing something for you. like
00:30:35
you got to do that because that's what
00:30:37
consumers and enterprises wants. It
00:30:39
doesn't mean that it's about killing
00:30:40
OpenClaw. just this is an obvious thing
00:30:43
right to do it
00:30:44
>> but this is a giant movement to stop it
00:30:47
because this is the equivalent of having
00:30:49
an open-source Android like player in
00:30:51
the market and that could be incredibly
00:30:53
disruptive these I believe open source
00:30:55
is going to win the day on the large
00:30:56
language models and take 90% of the
00:30:58
token usage and I think the entire
00:31:00
frontier model space could be undercut
00:31:02
by open source and I think they realize
00:31:04
that SLMs the the smaller language
00:31:06
models that are verticalized now that
00:31:09
will run on you know, desktops and
00:31:11
laptops and is even starting to run on
00:31:13
the top ones. That is their biggest
00:31:15
competitive threat and I hope it
00:31:16
happens. All due respect to your
00:31:18
investments, Brad, I think this
00:31:20
technology and the interface is uh you
00:31:22
know, he placed bets, but I I think it's
00:31:24
imperative that the agent level, which
00:31:26
is essentially your entire life, you
00:31:29
don't give that to Anthropic. You don't
00:31:31
give that to OpenAI. That's your entire
00:31:33
business, your entire life. It is
00:31:34
foolish for you, Brad, to give your
00:31:36
entire business and all the knowledge
00:31:38
you have to anthropic through that.
00:31:40
unless you're just doing it to boost
00:31:41
your um your your investment in those
00:31:43
companies. But I would be very concerned
00:31:45
if I was you with putting all of your
00:31:48
knowledge that you've earned over a
00:31:49
lifetime into any of these large
00:31:51
language models.
00:31:52
>> All right, Jake, let me ask you. Can I
00:31:54
ask a question? Thank you for that
00:31:55
impassioned monologue. Um actually, I
00:31:58
want to ask my TED talk.
00:31:59
>> I Yes, thank you for that TED talk. Um I
00:32:02
have a yes no question for each of you.
00:32:06
>> Do you believe that anthropic has
00:32:07
dominant market share in coding? right
00:32:10
now? Yes. No,
00:32:13
>> no.
00:32:14
>> In in coding,
00:32:15
>> yes,
00:32:15
>> they had the lead, but not that they had
00:32:16
the lead, but not dominating.
00:32:18
>> I think it's a trillion dollar market,
00:32:19
and these guys have less than 10% of it
00:32:21
today. So, it's hard to make a case that
00:32:24
>> What percent of coding tokens do you
00:32:26
think that anthropic is providing the
00:32:28
market right now?
00:32:29
>> Greater than 50%.
00:32:30
>> Yeah, that's true.
00:32:31
>> Okay, that's called dominant market
00:32:32
share.
00:32:33
>> Uh, I don't know about that.
00:32:35
>> More than 50% of the market.
00:32:36
>> You got to look at what you got to look
00:32:37
at what the TAM is.
00:32:39
with the Tan,
00:32:41
>> right? There are a lot of people who
00:32:42
provide, you know, that that are in this
00:32:46
tiebreaker before we move on to the
00:32:47
next.
00:32:47
>> I'm not saying it's a permanent
00:32:48
condition,
00:32:49
>> but if you're telling me that today
00:32:52
>> Anthropic is delivering over half of the
00:32:55
coding tokens, that's clearly a dominant
00:32:58
position in the market for coding. It's
00:32:59
an early market. It could change, but
00:33:01
>> if I were representing them, David, I
00:33:03
would say nine months ago, everybody t
00:33:05
called us uh, you know, out of the game.
00:33:08
We were being destroyed by open AI in
00:33:10
three months. Now people are saying we
00:33:11
have dominant market position. This is
00:33:13
the fastest changing most competitive
00:33:16
market in the world. I think it would be
00:33:17
very hardressed to walk into, you know,
00:33:20
some district court make the case that
00:33:21
these guys have somehow already formed a
00:33:23
monopoly against Amazon, Google,
00:33:26
Microsoft, Open AI, etc.
00:33:29
>> Well, I'm not saying it's a it's already
00:33:30
a permanent monopoly, but I am just
00:33:33
asking about market share. And I do
00:33:34
think you guys all agree that
00:33:36
>> Shimov, go ahead.
00:33:37
>> They probably have 50 to 60% market
00:33:40
share because I think codeex is actually
00:33:42
quite broadly used as well.
00:33:45
>> But that belies the more important point
00:33:47
which is AI enabled coding I think is
00:33:51
still 5% of the broad market. So it's
00:33:53
kind of a nothing burger. Yes, they're
00:33:55
leading but they're leading in something
00:33:57
that isn't that big yet. Now you would
00:33:59
say how could it not be big? And what I
00:34:01
would say is because most of the stuff
00:34:03
that's being written is still white
00:34:05
sheet denovo code. And I think the ugly
00:34:09
truth is I don't care what model you
00:34:11
have, but the long horizon ability for
00:34:14
any of these models to actually build
00:34:16
enterprisegrade software is still
00:34:20
shiit [ __ ] And that's the actual lived
00:34:24
experience. Not for me, but when I call
00:34:26
on our customers, half a trillion dollar
00:34:29
banks, hundred billion dollar insurance
00:34:31
companies, none of these guys are like,
00:34:32
"Wow, it just works out of the box." It
00:34:34
doesn't work. So, most of it is still
00:34:38
handtuned. So, until I can honestly tell
00:34:41
you that we can point a model at this
00:34:44
with the right guard rails, which I
00:34:46
can't today, what I would say is it's a
00:34:48
small market that will become large as
00:34:52
these models become better.
00:34:54
But we are in the world where we have 50
00:34:57
years of accumulated tech debt as a
00:35:00
world. And I suspect when you enumerate
00:35:02
the number of lines that that
00:35:04
represents, it's hundreds of trillions
00:35:06
of lines of just pretty marginal
00:35:08
mediocre code to bad code. On top of
00:35:12
that, we have all these legacy
00:35:13
languages. I'll tell you one of our
00:35:15
customers, they have to go and get
00:35:17
60year-old pensioners to come into the
00:35:19
office to interpret cope. No, I'm not
00:35:22
joking. This is a
00:35:22
>> snowball for trend.
00:35:24
>> This is a hundred billion dollar a year
00:35:26
revenue company and that's how they
00:35:29
solve these problems. It's not opus just
00:35:31
solves it. So I I would just keep in
00:35:34
mind that most of the tech debt in the
00:35:36
world that exists 99% of it is still
00:35:39
poorly addressed by these models. We are
00:35:42
untying this Gordian knot. It's going to
00:35:44
take decades to do it right. So all the
00:35:46
breathlessness about all this other
00:35:48
stuff, I really think it's not where the
00:35:49
money is. It's not the big time stuff.
00:35:51
And you can tell me, "Oh yeah, it's
00:35:53
going to be the future." And I would
00:35:54
say, "Tell this business that's a
00:35:56
hundred billion dollars a year of
00:35:58
revenue and 50 million billing
00:35:59
relationships that all of a sudden
00:36:01
you're going to open claw your way to a
00:36:02
solution." It's [ __ ] Not to say
00:36:05
that you can't have a great chief of
00:36:07
staff, and not to say you can't do some
00:36:09
useful stuff and trickery and, you know,
00:36:11
have a good knowledge base. I'd like
00:36:13
that, too. But the core things that your
00:36:17
lived experience sits on today is a mess
00:36:20
of tech debt that will get very slowly
00:36:22
replaced. And that's just the reality of
00:36:24
life.
00:36:25
>> And there are competitors that are
00:36:27
extremely disruptive. I'll tell you
00:36:28
about one. We talked about Bit Tensor
00:36:30
Tao on this program a couple weeks ago
00:36:32
when we had the um Jensen interview. You
00:36:34
brought it up actually Chimath. There's
00:36:35
a there's a project that's subnet 62.
00:36:38
It's called Ridges AI. And what they're
00:36:41
doing is a competitor that is not only
00:36:44
open- source but anybody can contribute
00:36:46
to it. They spent about a million
00:36:48
dollars in tow like rewards and in 45
00:36:50
days they hit 80% of what Claude 4 is
00:36:54
and they did that in under 45 days. The
00:36:56
way that works is they give rewards for
00:36:58
people who and they can do this
00:37:01
anonymously make that coding product
00:37:03
which is like codeex or claude code
00:37:05
better. that flywheel is racing right
00:37:09
now with participation in the same way
00:37:10
Bitcoin is. So you're going to see a lot
00:37:13
of open- source and these crypto
00:37:15
open-source combinations and uh anybody
00:37:18
who's not investigated this, I highly
00:37:21
recommend you investigate this.
00:37:23
>> I do think you're right about one
00:37:24
specific thing. I would put zero,
00:37:26
literally the probability zero of any
00:37:29
important company worth anything more
00:37:31
than a dollar having and outsourcing
00:37:34
their production code to an open source
00:37:36
project. That'll never happen. However,
00:37:37
what will happen though is when you look
00:37:40
at the cost of training this 10 trillion
00:37:43
parameter model on Blackwell and when
00:37:47
you look in the future let's just say in
00:37:50
six or nine months that a 15 or 20
00:37:52
trillion perm model is going to get
00:37:54
trained on Vera Rubin I think Jason
00:37:56
where you are right I have zero and just
00:37:59
to be clear I have no investments in
00:38:01
this at all I'm
00:38:03
>> to be so super clear
00:38:04
>> I'm just observing because another
00:38:06
project other than Bit Tensor that
00:38:07
someone brought up to me is Venice. The
00:38:09
concept of opensource training and
00:38:12
orchestration
00:38:14
is a hugely disruptive idea which is the
00:38:17
complete orthogonal attack vector to
00:38:20
this idea that you have to raise tens
00:38:22
and tens of billions of dollars to train
00:38:24
your models because if the capital
00:38:26
markets run out of 10 and 20 billion
00:38:29
dollar checks to give people the only
00:38:32
solution is to be totally distributed. I
00:38:34
tend to agree with you Jason that there
00:38:35
is going to be at some point a very
00:38:38
successful open source project for
00:38:40
pre-training.
00:38:42
Absolutely. Will there never ever be an
00:38:45
open- source way where a real company
00:38:47
that has any skin in the game says here
00:38:49
guys re-engineer my codebase as an open
00:38:51
source project. Never going to happen.
00:38:53
>> Yeah, I I think the coding tools will.
00:38:55
And if you look at the history of open
00:38:56
source, Brad, you actually I think had a
00:38:58
lot of bets in this space. Linux,
00:39:00
Kubernetes, Apache, Postgress, like
00:39:03
Terraform, like these open source
00:39:05
projects are deep inside of enterprises.
00:39:07
Deep. And we're sitting here 15, 20
00:39:10
years ago, the same argument was made.
00:39:12
Nobody will ever adopt these inside the
00:39:13
enterprise. You got to go with Oracle,
00:39:15
whatever. And fair enough, many people
00:39:17
do. But I think this is this $29 ridges
00:39:22
um subscription to do this versus 200.
00:39:24
It's starting to take hold inside of
00:39:27
startups. And that's where I always look
00:39:29
at the tip of the spear. Startups love
00:39:30
to, you know, use open source products.
00:39:33
I think this could be the next big
00:39:34
thing. But listen, I I I invest in
00:39:37
things that have a 90% chance of going
00:39:39
to zero. So do your own research. No
00:39:41
crying in the casino.
00:39:43
>> Can I just make a a final few points? So
00:39:46
just just quickly so number one is with
00:39:49
respect to this market for code or code
00:39:51
tokens whatever you want to call it
00:39:53
>> it might be 5% today meaning 5% of the
00:39:56
codes AI generated versus human
00:39:57
generated I think it's going to 95% I
00:40:00
mean I bet any amount of money on that
00:40:02
the only question is when probably over
00:40:04
the next few years so that's point
00:40:06
number one point number two is it's
00:40:08
possible that if you're the early leader
00:40:11
in coding as a AI model company let's
00:40:15
say you have 50 to 60% of market share.
00:40:17
You have the most developers using it.
00:40:19
Therefore, you have the most access to
00:40:21
code bases. You might get the most
00:40:24
training tokens. There is a potential
00:40:26
flywheel there where you can see the
00:40:29
early market leader consolidating its
00:40:30
lead because it's generating the most
00:40:32
code tokens and it's getting access to
00:40:34
the most existing code. Now, I'm not
00:40:36
saying for sure that's going to happen.
00:40:38
is possible that the other guys catch
00:40:39
up, but I think there is a possibility
00:40:42
of a flywheel there and strong, I guess
00:40:45
you'd call it data scale effects, things
00:40:47
like that. So, I do believe that the
00:40:49
market for coding tokens could be
00:40:51
monopolized. Third, Anthropic's revenue
00:40:55
run rate, as based on what I can tell
00:40:56
and what's been publicly released, is
00:40:58
the fastest growing revenue run rate at
00:41:01
scale that I think we've ever seen. Uh,
00:41:03
we
00:41:04
>> perfect segue. It's the next story.
00:41:05
Okay, maybe
00:41:06
>> pull up the the tweets. But this thing
00:41:09
is ramping at a rate we've never seen
00:41:10
before.
00:41:12
>> We can get into that in a second. But
00:41:13
just one last final point
00:41:15
>> is I think it's pretty clear that where
00:41:17
we go from here is agents and coding
00:41:21
gives you a huge step up on agents
00:41:23
because you know one of the main things
00:41:25
that agents need to do is is write code
00:41:27
to be able to enable them to complete
00:41:29
tasks.
00:41:30
>> Correct. And so if it is the case that
00:41:34
coding is this huge market that's going
00:41:36
to be dominated by one or two companies
00:41:40
and then that leads to another huge
00:41:42
market which is agents. My point is just
00:41:44
I think all these companies need to
00:41:46
behave in a very clean way
00:41:49
>> and not engage in tactics that later the
00:41:52
government might say you know what that
00:41:53
was anti-competitive. Everyone should
00:41:55
just I think play fair. Do not engage in
00:41:57
discrimination against other people's
00:41:59
products. engage in fair pricing. I'm
00:42:02
not accusing anyone of breaking any of
00:42:03
the rules, but what I'm saying is that
00:42:05
eventually the government's going to
00:42:07
look at this market with the benefit of
00:42:09
2020 hindsight and I think everyone
00:42:11
should just basically, you know, keep it
00:42:14
>> keep your nose clean.
00:42:15
>> Keep it tight. Keep it tight.
00:42:17
>> Keep it tight. Tight is right. I think
00:42:19
is an excellent point. Let's talk about
00:42:21
the revenue ramp of Anthropic. This is
00:42:24
just unprecedented. Anthropic's revenue
00:42:27
run rate has topped 30 billion with a B.
00:42:31
Early 2023, they turned on revenue. They
00:42:33
started charging for API access. End of
00:42:35
2024, they're at a billion dollar run
00:42:37
rate. February 25, they launched Claw
00:42:40
Code. That was the starter pistol. Mid
00:42:42
2025, $4 billion run rate. End of 2025,
00:42:46
$9 billion run rate. Just a couple of
00:42:49
months later in April, $30 billion run
00:42:52
rate. Yes, that's right. Triple. Uh and
00:42:54
the way they did this is enterprise uh
00:42:57
customers are a major part of the spend.
00:43:00
Dario announced a couple of months ago
00:43:01
that there's over a thousand enterprises
00:43:04
paying over 1 million annually. This is
00:43:07
truly mindboggling when you think about
00:43:09
it because those are the most coveted
00:43:12
customers in the world. These are the
00:43:13
big fish that you just uh when people
00:43:16
are running enterprise software, they
00:43:18
they dream Slack dreamed of getting
00:43:20
these million-dollar customers. Uh
00:43:21
Salesforce dreams of getting these
00:43:23
million-dollar customers. Brad, you're
00:43:24
an investor. I guess uh Sam famously on
00:43:27
BG2 asked you to sell your uh OpenAI
00:43:30
stock back to him. You didn't. You
00:43:32
demired, but you're an investor in both.
00:43:36
How shocking is it to you to place both
00:43:39
of those bets and then see one of them
00:43:42
come from so far behind? You know, Chat
00:43:44
GPT has 900 million users. I don't know
00:43:46
if they've they've passed a billion
00:43:48
officially yet, but they are the Verb,
00:43:50
right? They're the Uber. They're the
00:43:52
Xerox. They're the Polaroid of AI, but
00:43:56
they didn't go after the enterprise.
00:43:58
Daario made that and Daario worked. He
00:44:00
was the co-founder of OpenAI. He left
00:44:03
and according to the New Yorker story
00:44:04
that came out from Ronan Farrell this
00:44:06
week, he was basically left because of
00:44:08
his disgust in working with Sam Alman.
00:44:13
Your thoughts?
00:44:14
>> Well, you know, before we go down the
00:44:15
OpenAI rabbit hole, let's just really
00:44:17
contextualize like what's going on here.
00:44:20
You know, check I I I have this
00:44:21
additional chart. you showed one, you
00:44:23
know, they added 4 billion of revenue in
00:44:25
January, 7 billion in February, 11
00:44:28
billion of annualized run rates, um, or
00:44:31
10 or 11 billion in March, just to put
00:44:33
in perspective, that's data bricks plus
00:44:35
Palanteer combined that they added in a
00:44:38
single month, right? So we started with
00:44:41
everybody at the start of the year
00:44:42
ringing their hands including you know
00:44:45
Gurley and others saying we're in a big
00:44:46
bubble asking whether the AI revenues
00:44:49
would show up to justify all of this
00:44:51
investment and bam you have the largest
00:44:53
revenue explosion in the history of
00:44:55
technology. So the company's plans were
00:44:58
to end the year at about a $30 billion
00:45:01
ex exit run rate. They got there by the
00:45:04
end of March right and I suspect that
00:45:06
it's continuing in April. So you have to
00:45:08
ask what's going on and what's the big
00:45:10
so what the first thing for me is that
00:45:13
model and product capability just hit
00:45:15
this threshold we talked about earlier
00:45:17
near AGI whatever the hell you want to
00:45:19
call it and everybody like alimter said
00:45:22
damn this is so good I have to have it
00:45:24
this is no longer about my IT budget
00:45:26
this is about labor augmentation and
00:45:28
labor replacement and by the way co-work
00:45:31
is growing even faster than Claude go at
00:45:35
the same stage of development
00:45:38
So what it showed is we have a near
00:45:40
infinite TAM. It turns out that the TAM
00:45:42
for intelligence is radically different
00:45:44
than anything that we've seen before.
00:45:47
And I think the best example of this,
00:45:49
right? This is millions of
00:45:51
self-interested parties, consumers,
00:45:54
enterprises, a thousand now over a
00:45:57
million dollars. Right? It's not that
00:45:59
there was some great go to market and
00:46:01
anthropic that all of a sudden, you
00:46:02
know, they snuck up and blew everybody
00:46:04
away. No, it was companies demanding the
00:46:06
product. They're getting throttled on
00:46:08
the product. Why? Because it's so good.
00:46:10
It makes them better at their business.
00:46:12
We are all self-interested actors. And
00:46:14
when millions of those people are all
00:46:16
making the same decision, there's a huge
00:46:18
tell. And the tell here is that the TAM
00:46:21
is as big as Daario and Sam and others
00:46:24
have been saying. We knew intelligence
00:46:26
was going to scale on the exponential.
00:46:28
The question was whether revenue will
00:46:30
scale on the exponential, and that's
00:46:32
what we're seeing. And remember, they're
00:46:33
doing this with only 1 1/2 to 2 gawatt
00:46:37
of compute, right? These guys are
00:46:39
massively compute constrained. They're
00:46:41
each going to be adding 3 GW of compute
00:46:43
this year. And so that will unlock they
00:46:46
would be growing even faster. But for
00:46:49
that, and then Jason, to your point
00:46:50
about the open source models that we all
00:46:53
want to be a part of this solution, I've
00:46:55
talked to a lot of big companies, 65 to
00:46:57
70% of their token consumption is
00:47:00
open-source model, right? are these
00:47:02
cheap Chinese and other tokens. So these
00:47:05
revenue ramps are happening while the
00:47:07
world is already using open source. This
00:47:10
is not frontier only. This is Frontier
00:47:12
plus open source. We're going to see
00:47:14
massive token optimization over the
00:47:16
course of the year. But what happens on
00:47:18
this Jebans paradox is the co the unit
00:47:21
costs right of intelligence is
00:47:23
plummeting. Not the cost of tokens. The
00:47:26
unit cost of intelligence is plummeting
00:47:28
because the capabilities of these models
00:47:30
is so much better. I look at what it
00:47:32
does for Altimeter day in and day out. I
00:47:34
talked to a major uh company yesterday.
00:47:37
They're on a run rate to do a hundred
00:47:39
million of token consumption this year
00:47:41
on about $5 billion in opex. They think
00:47:44
that we're now nearing peak employment
00:47:46
in their company, but that their token
00:47:48
their intelligence consumption, okay,
00:47:50
let's not call it token consumption,
00:47:52
right? because tokens may go up a lot,
00:47:54
but their intelligence consumption is
00:47:56
going to go up, you know, a lot. So, I
00:47:59
would leave you with this. We're early
00:48:01
to Chimas's point. We have low
00:48:04
penetration of the global 2000. We have
00:48:06
low penetration of the use cases. We
00:48:09
have low penetration of of within the
00:48:12
use cases that they're already using.
00:48:14
And the models are only getting better.
00:48:15
So I think when you look out toward the
00:48:17
end of the year, I would not be shocked
00:48:20
if you see Anthropic exiting this year
00:48:23
at 80 to 100 billion in revenue. And by
00:48:26
the way, doing it at the same time that
00:48:28
OpenAI, who is also on the wave, they'll
00:48:30
be releasing an incredible model in the
00:48:32
next imminently. They're going to be on
00:48:35
that wave and you're going to see an
00:48:36
inflection in their revenues as well.
00:48:38
>> Okay, Chimath, question one has been
00:48:41
answered. The question of hey, does this
00:48:43
stuff actually have utility? that went
00:48:45
from a question mark to an exclamation
00:48:46
point. Of course, it's got utility.
00:48:48
People are getting value from it. And it
00:48:49
might be variable. Some people get more
00:48:51
value than others. Number two, the
00:48:52
revenue ramp was a big question. Now,
00:48:54
that's turned into an exclamation point.
00:48:56
The final piece of the puzzle that
00:48:58
you've brought up many times is can this
00:48:59
be profitable? And these companies are
00:49:02
burning through a large amount of cash.
00:49:05
So, what is your take on when these
00:49:07
companies can get out of the J curve? We
00:49:09
talked about this, I think, three
00:49:10
episodes ago. I estimated like we're
00:49:12
going to be looking at $4500 billion in
00:49:15
investment into these data centers at a
00:49:17
minimum and then they have to climb out
00:49:20
of that to get to profitability. So what
00:49:22
are your thoughts on these becoming
00:49:25
profitable companies? Do you remember
00:49:27
the
00:49:29
investor that published this list Jason
00:49:31
where he put all of the terms you talk
00:49:35
about when one of the terms you can't
00:49:36
talk about is profit. It's a list where
00:49:38
it's like if you can't talk about free
00:49:41
cash flow, you talk about IBIDA. When
00:49:43
you can't talk about IBIDA, you talk
00:49:44
about
00:49:46
>> margina.
00:49:47
When you can't talk about that, you talk
00:49:49
about revenue. And then when you can't
00:49:50
talk about revenue, you talk about gross
00:49:53
revenue
00:49:53
>> bookings.
00:49:54
>> So you can kind of figure out,
00:49:58
I think, where we are in any part of any
00:50:01
cycle by just indexing into what does
00:50:04
everybody talk about. I think where we
00:50:07
are is we are between gross revenue and
00:50:11
net revenue. That's where the discussion
00:50:12
is.
00:50:13
>> Okay.
00:50:14
>> There was another article I think today
00:50:16
in I think maybe it was the information
00:50:18
that tried to categorize and distinguish
00:50:21
that anthropic presents gross, open AI
00:50:24
presents net. They're different. We
00:50:27
don't know what the various take rates
00:50:29
are. So they're saying that there's a
00:50:32
difference. If it's not true, there's
00:50:33
been no clarity provided by these
00:50:35
companies. So, at a minimum, you have
00:50:37
this confusion where there's the
00:50:39
breathless talk. Then there's people
00:50:40
that don't even know the difference
00:50:41
between actual recognized revenue and
00:50:44
run rate revenue and how to multi. I
00:50:46
mean, so we're definitely there, okay?
00:50:48
We can quibble about the details, but we
00:50:50
are not at the place where people are
00:50:51
like, "Oh, here's your steadystate, you
00:50:53
know, free cash flow margin, and here's
00:50:55
what your EBA does." We're never we're
00:50:56
we're years from that. They're gonna
00:50:58
have token maxing IBA like IB at the
00:51:02
Wii.
00:51:02
>> The thing that we need to understand is
00:51:04
how gross margin negative is this
00:51:06
revenue growth.
00:51:07
>> We don't know that and at least we don't
00:51:10
as outsiders.
00:51:11
>> Brad might know.
00:51:12
>> Brad may know. I I I I will tell you
00:51:15
think about this. What are their big
00:51:16
cost inputs? The number one cost input
00:51:19
is the cost of compute. Cost of compute.
00:51:21
>> Right? I just told you they only have a
00:51:23
gigawatt and a half of compute. and they
00:51:25
have that gigawatt and a half of compute
00:51:26
whether they have a billion in revenue
00:51:29
or whether they have 80 billion in
00:51:30
revenue. So you might actually expect to
00:51:33
see these companies their gross margins
00:51:35
are exploding higher like the fastest
00:51:37
increase in gross margins I've probably
00:51:39
seen out of any technology company. So
00:51:41
this is not gross margin negative you're
00:51:42
saying?
00:51:43
>> No definitely not gross margin negative.
00:51:45
And what I would tell you
00:51:46
>> so that they must be hugely profitable
00:51:48
then
00:51:48
>> well you may see accidental why I call
00:51:51
it accidental profitability. They may
00:51:53
not be able to spend this revenue fast
00:51:55
enough chamath on compute. And remember
00:51:58
it's only 2500 people. Google crossed
00:52:01
this revenue threshold when they had
00:52:03
120,000 people. These guys have 2500
00:52:07
people. So the only thing you can really
00:52:08
spend money on, right, is compute. And
00:52:10
they can't stand up the compute fast
00:52:12
enough.
00:52:13
>> But none of this foots to me then to be
00:52:15
honest because if you were on a
00:52:17
threshold of 90% plus gross margin,
00:52:20
>> I'm not saying it's there. I'm not
00:52:22
saying it's 90% plus. I'm just saying
00:52:24
it's gone from meaningfully negative 18
00:52:27
months ago to, you know, very very
00:52:29
positive. I've seen rumored out there
00:52:31
50% is what you're saying. The trend is
00:52:34
there.
00:52:34
>> Let me just say this.
00:52:37
I think if you're an incumbent, you want
00:52:40
the cost of compute to go down. I think
00:52:42
if you're not an incumbent, so
00:52:44
specifically, who do I mean? Meta,
00:52:47
Google, and SpaceX.
00:52:51
I think those three people who have all
00:52:53
three of them, well, sorry, Meta and
00:52:55
Google have a fortress balance sheet. I
00:52:57
think by the end of June, SpaceX will
00:52:59
also have a fortress balance sheet. What
00:53:02
they will want to do is they will want
00:53:03
to make this a compute problem because
00:53:05
they will control the the conditions on
00:53:07
the field. You already see this today.
00:53:09
>> Yeah.
00:53:10
>> Meta's models today, what people's
00:53:12
general reviews are it's okay, but the
00:53:15
one thing that people say is it's
00:53:16
incredibly performant. The model quality
00:53:19
is okay, but the performance is great,
00:53:21
which speaks to Meta's huge advantage.
00:53:23
They have a massive compute
00:53:24
infrastructure. So if you're if you're
00:53:25
not open AI and anthropic,
00:53:28
they'll want to make this a capital
00:53:29
problem because then they can win it. If
00:53:31
you're anthropic and open AI, you want
00:53:33
this thing to be as efficient as
00:53:35
possible.
00:53:36
I think where we are is very much in the
00:53:38
early innings. And we're bumbling around
00:53:40
talking about gross margins and you know
00:53:42
revenues. We are not at profitability.
00:53:44
And what is true for Facebook and what
00:53:47
was true for Google was irrespective of
00:53:50
where they got to a billion. Who g
00:53:52
cares? They were profitable by year
00:53:54
three and they never looked back. I was
00:53:57
there. I remember it was glorious.
00:53:59
>> The the cost the cost of building uh you
00:54:03
know AI
00:54:05
totally stipulate is radically higher
00:54:07
than the cost of building retrieval at
00:54:08
Google, right? Like it's just a
00:54:10
fundamentally more expensive problem.
00:54:12
But I will tell you that there's a lot
00:54:14
of FUD out there about negative gross
00:54:16
margins. I mean Jason, you started the
00:54:17
segment by saying they're burning
00:54:19
through large amounts of cash. I think
00:54:21
people are going to be shocked at the
00:54:22
burn how low the burn levels are at
00:54:24
these companies.
00:54:25
>> Anthropic or Open AI.
00:54:27
>> Yes. And and I would say at Open AI as
00:54:28
well like they're if they're on you know
00:54:30
if they do $50 billion this year again
00:54:33
just look at the number of people they
00:54:34
have revenue per people. It's pretty low
00:54:37
and the inference cost is plummeting.
00:54:38
Inference cost is down by 90%
00:54:40
year-over-year. And so just finally I
00:54:43
want to make respond to this point about
00:54:45
gross versus net uh this this tweet that
00:54:48
Chimath was referencing. Okay, so
00:54:50
there's a certain percentage, a
00:54:52
smallalish percentage of Anthropics
00:54:54
revenue, right, that they distribute
00:54:55
through the hyperscalers and like a lot
00:54:57
of arrangements, whether it's Snowflake
00:54:59
or Data Bricks or others, you pay a
00:55:00
commission, right, uh on on that. I will
00:55:04
just tell you that you're talking
00:55:05
singledigit percentage of total revenue
00:55:07
of these companies. So the gross versus
00:55:09
net thing isn't what's being reported.
00:55:11
like the apples for apples is pretty
00:55:13
easy and if you want to be conservative
00:55:14
on it take down Anthropic's revenue by
00:55:17
you know five to 10% which you know
00:55:19
again I don't I think it's better to
00:55:21
gross up OpenAI's revenue but any way
00:55:22
you do it I just think it's a
00:55:24
distraction from what's really what's
00:55:26
really going on here happy to
00:55:27
>> s you have any thoughts on this uh
00:55:29
massive revenue ramp
00:55:31
>> yeah I mean I want to go back to a point
00:55:33
that Brad made because I think it was
00:55:35
just really important and I want to just
00:55:37
underline it consider where we were at
00:55:39
the beginning of the year and What
00:55:41
everybody was saying is that AI was a
00:55:43
big bubble and the evidence they would
00:55:46
point to was the fact that hundreds of
00:55:48
billions of dollars was going into capex
00:55:51
that needed to be spent on these data
00:55:53
centers and there was no evidence of
00:55:55
significant revenue to justify that
00:55:57
spend. Where was the ROI? By the way, as
00:56:00
an aside, the same doomers who were
00:56:02
saying that AI was in a bubble were also
00:56:05
the ones who were saying that AI was so
00:56:06
powerful it's going to put us all out of
00:56:08
work and it's going to, you know, take
00:56:10
over from humanity. I mean, in other
00:56:12
words, they couldn't decide if AI was
00:56:14
too powerful or not powerful enough. But
00:56:17
putting aside that contradiction, they
00:56:19
clearly were making this case that AI
00:56:22
was this big bubble and that there'd be
00:56:23
no payoff or justification for this
00:56:27
massive capex that's being spent. And I
00:56:29
think we're starting to see here there
00:56:31
is justification for it. Uh we're seeing
00:56:33
it just in this one vertical of AI which
00:56:36
is coding. We're again seeing the
00:56:38
fastest revenue growth in history. It's
00:56:40
utterly unprecedented. And this is just
00:56:42
one category or vertical of AI. We know
00:56:47
that agents are coming next and the
00:56:49
enterprise adoption of that is going to
00:56:51
be absolutely massive. So, I guess what
00:56:54
I'm saying is that this is early proof
00:56:56
for I think the thing that makes Silicon
00:57:00
Valley special, which is we're willing
00:57:02
to basically bet on things that just
00:57:05
intuitively on a gut level we know are
00:57:08
the next big thing. We're not that
00:57:09
spreadsheet driven. Actually, Silicon
00:57:11
Valley believes that if you build it,
00:57:13
they will come and is willing to finance
00:57:15
that build out. And that's basically
00:57:17
what's been happening. Again, just the
00:57:19
top four hyperscalers, $350 billion of
00:57:22
expected capex this year on its way, I
00:57:24
think Jensen said 1 trillion by 2030.
00:57:27
So, Silicon Valley, whether it's big
00:57:29
companies, whether it's founders,
00:57:30
they're always willing to bet on this
00:57:32
next big thing. They're not like Wall
00:57:34
Street. They don't need, you know,
00:57:35
specialist to tell them where to go.
00:57:38
They know where the technology is going
00:57:40
and they make their bets based on that.
00:57:42
And I think that there is going to be a
00:57:44
big payoff for this. And I think it's
00:57:47
the thing that's going to make our
00:57:48
economy and the United States in general
00:57:51
remain extremely dynamic and in the lead
00:57:53
on this thing is that we are willing to
00:57:55
make those kinds of bets. And I think
00:57:57
it's going to pay off big time.
00:57:59
>> Yeah, clearly. Hey, um Brad, you didn't
00:58:02
answer my question about the vibes over
00:58:04
at OpenAI versus Quad. Open AI is um I
00:58:09
wouldn't say reeling but there's a lot
00:58:11
of hand ringing going on a lot of
00:58:13
employees leaving a lot of people who
00:58:15
are wondering like is our strategy the
00:58:18
winning strategy of like consumer first
00:58:20
they shut down Sora you know unwinding
00:58:23
the Disney deal and really trying to get
00:58:26
the company focused and it's kind of
00:58:27
like I mean listen the New Yorker story
00:58:29
was a bit of a rehash so I don't think
00:58:30
we have to go into the blowby-blow
00:58:32
because we covered here three years ago
00:58:35
but the truth is a lot of the great
00:58:38
founders, co-founders of OpenAI and a
00:58:40
lot of the great contributors are now at
00:58:43
Anthropic and other large language
00:58:45
models. And in the secondary market,
00:58:48
OpenAI is trading lower than the last
00:58:50
valuation. And Anthropic is trading
00:58:53
significantly above the $380 billion. So
00:58:57
maybe talk a little bit about this
00:58:58
competition, this Microsoft versus
00:59:01
Apple, this Google versus Facebook.
00:59:03
Well, let's let's start with immense
00:59:06
credit where credit is due. Anthropic
00:59:08
was literally counted out of the game
00:59:09
last year. Y,
00:59:10
>> right? And here they come over the last
00:59:12
12 months and and and they've kicked
00:59:14
OpenAI's ass over the last 90 days,
00:59:17
right? And what did Anthropic do?
00:59:19
Anthropic made choices. No multimodal,
00:59:22
no video, no hardware, no chips, no
00:59:24
building data centers. They said, "We're
00:59:26
just going to focus on coding and
00:59:27
co-work. We think that is the path to
00:59:29
AGI and and and and ASI." They executed
00:59:32
their butts off. They took the lead.
00:59:34
2500 people tight pulling on the ore in
00:59:38
the same direction. But I think you
00:59:40
would be seriously foolish to count out
00:59:43
open AI, right? And I think we're we're
00:59:45
we're at peak open AI FUD. And I'll tell
00:59:47
you, it starts with great researchers
00:59:50
and great models. And I think when you
00:59:51
see the Spud model, they're about ready
00:59:53
to release. I think it's going to be an
00:59:55
excellent model. Shows that they're
00:59:57
firmly on the wave. Um, if you look at
01:00:00
what's going on with Codeex, incredible
01:00:02
ramp on Codeex, fastest ramping model
01:00:05
with 5.4, I think 5.5 or Spud, whatever
01:00:08
we're going to call, it's going to be an
01:00:09
even faster ramp.
01:00:10
>> Have you seen Spud? Have you used it?
01:00:12
Have you gotten a preview?
01:00:13
>> People are using Spud, right? So, it it
01:00:16
is being previewed and so
01:00:17
>> So, you're talking to people who've used
01:00:19
it and what are they telling you?
01:00:21
>> They're telling us that it's an
01:00:22
incredible model on par with Mythos,
01:00:24
right? and that it's a a very usable
01:00:27
model in terms of um how it's packaged.
01:00:30
I will say that back to David's point
01:00:34
now this is the most important point I
01:00:35
think anybody can take away here.
01:00:38
This is not zero sum. The TAM of
01:00:42
intelligence is dramatically larger than
01:00:44
any TAM we've ever seen in our investing
01:00:47
careers over the last two decades.
01:00:48
Right? And if you're on the wave, which
01:00:50
Open AAI is, you are going to be selling
01:00:53
into the world's biggest TAM, they are
01:00:55
going to build a very big company. I'm a
01:00:58
buyer of the shares today.
01:00:59
Notwithstanding all of the vibes that
01:01:01
you describe, I think these companies
01:01:03
are firmly on the wave. They are jarred.
01:01:07
They are sitting there saying, "What did
01:01:08
we do wrong? And how do we get our mojo
01:01:10
back?" They want to compete. It is
01:01:12
embarrassing to people on the research
01:01:14
team and the product team over there.
01:01:15
So, I'm not saying there's not a real
01:01:18
awakening occurring there, but I think
01:01:20
that's what the case is. And by the way,
01:01:22
to Chamas's point, do not count out
01:01:24
Meta, right? I think Meta is absolutely
01:01:26
in this game. Google is absolutely in
01:01:28
this game. Elon is absolutely in this
01:01:30
game. And if you're
01:01:31
>> got some stuff dropping shortly that's
01:01:33
going to be very impressive.
01:01:34
>> If you're on team America, the fact that
01:01:36
we have five frontier models competing
01:01:39
against each other and David made sure
01:01:41
they weren't throttled by excessive
01:01:43
government regulation. We have mythos
01:01:45
come out. It's a self-imposed safe
01:01:48
harbor, you know, to harden our system.
01:01:50
It wasn't a call for moratoriums or
01:01:52
getting the government involved. We have
01:01:54
the type of competition that's causing
01:01:55
us to accelerate our lead against the
01:01:58
rest of the world. We can't take our eye
01:02:00
off the prize. We got to stop
01:02:01
adversarial distillation and we need to
01:02:03
make sure that we're distributing our
01:02:05
products around the world. But I view
01:02:07
this as really good for team America.
01:02:09
>> Well said. And here is your poly market
01:02:12
IPOs before 2027. Obviously SpaceX at
01:02:16
95% uh Cerebrus at 94% and uh hey number
01:02:21
five on this list 51% chance that
01:02:24
Anthropic goes out before the end of the
01:02:26
year. 44% chance that OpenAI comes out
01:02:29
before then. All right here is the
01:02:32
closing market cap for Anthropic on Poly
01:02:36
Market only $158,000 in volume. So,
01:02:39
Chimath, when you put in 400K, you're
01:02:42
going to really tilt this market.
01:02:44
78% chance that it's above 600 billion,
01:02:48
19% chance that it doesn't go out. So,
01:02:51
it's looking like this will be a decent
01:02:53
investment for you. Brad, what valuation
01:02:55
did you get into Anthropic at?
01:02:57
>> We first invested in I believe it was
01:02:59
the
01:03:01
uh 30 or $150 billion round.
01:03:04
>> So, this will be a 7x 5x for Altimeter
01:03:07
L, please. Congratulations. I mean, no,
01:03:09
listen. I I I again, there are lots of
01:03:11
people who were there before us and who
01:03:13
are on the board and who are going to do
01:03:14
much better than that. What' you put in?
01:03:16
50.
01:03:17
>> What' you put in?
01:03:18
>> No, we've got billions in both
01:03:19
companies. Uh
01:03:20
>> billions in both companies. Oh my lord.
01:03:24
>> I think there's this existential thing
01:03:26
going on in venture today. David could
01:03:29
talk about it as well. I mean people
01:03:31
can't they're extraordinarily nervous
01:03:33
about you look at the IGV stock index
01:03:36
down 30% year to date down 5% today all
01:03:40
software stocks plummeting right venture
01:03:43
capitalists are terrified to invest
01:03:46
money in anything other than these
01:03:48
frontier models and things like SpaceX
01:03:51
or military modernization finding
01:03:53
something that's out of harm's way of AI
01:03:56
right where you can count on the
01:03:57
terminal value to Chamas insights over
01:04:00
the last few weeks is very difficult to
01:04:02
do. That's why you see this crowding.
01:04:03
So, we've taken a barbell approach,
01:04:05
right? We've got a lot in what we think
01:04:07
are the most important companies that
01:04:09
are on the frontier and then we're
01:04:10
betting with on really small teams that
01:04:12
we think have very defensible businesses
01:04:15
in a world of uh you know, AGI. But it's
01:04:18
>> what happens to all these enterprise
01:04:19
software companies? Do they become PE
01:04:21
takeouts? Do they get consolidated? um
01:04:24
or do they just have to adopt these AI
01:04:27
technologies and and and solve this
01:04:30
problem of hey the frontier model is
01:04:32
just going to solve for whatever these
01:04:34
niche software companies do.
01:04:36
>> I think the market's probably being a
01:04:38
little too pessimistic with respect to
01:04:40
at least some of these software
01:04:41
companies. I mean, obviously, there's
01:04:42
going to be big differences in the
01:04:45
quality of the modes of these companies.
01:04:48
And so, look, software is going to be a
01:04:51
lot cheaper and easier to generate, but
01:04:53
I'm not sure that was the competitive
01:04:55
advantage of a lot of these companies.
01:04:57
So, there's probably a little bit of the
01:04:59
baby being thrown out with the bathwater
01:05:00
right now, and there probably are some
01:05:02
value buys in enterprise software. I
01:05:05
think the interesting question here and
01:05:06
we've been talking about this for a
01:05:08
couple of years in the pod is just where
01:05:10
you see the AI value capture being in
01:05:13
terms of layer of the stack. Remember
01:05:15
where we started it was really just the
01:05:17
chip layer of the stack was where all
01:05:19
the value capture was. It was basically
01:05:20
Nvidia was the first company to be worth
01:05:23
multiple trillions of dollars because of
01:05:25
AI. And for a while it looked like
01:05:27
that's where all the value capture was
01:05:30
going to be because OpenAI for example
01:05:31
was losing so much money and Anthropic
01:05:33
wasn't on the radar as much. Now we're
01:05:36
seeing wait a second um you know it's
01:05:38
not just the chip companies it's also
01:05:40
the hyperscalers are now benefiting and
01:05:42
now we're seeing at the model layer it
01:05:44
looks like Enthropic and Open AI they're
01:05:47
all going to be huge beneficiaries. I
01:05:49
think the next question is at the
01:05:50
application layer of the stack. Okay.
01:05:52
Well, now does all that value capture
01:05:54
just get eaten by the model companies or
01:05:56
are there applications that get
01:05:58
turbocharged? I guess you could say that
01:06:00
Palunteer is already one of them, right?
01:06:02
It's an application company that's been
01:06:04
turbocharged by these model
01:06:06
capabilities. Who else will be a big
01:06:08
beneficiary? Is it again, is it all
01:06:10
going to be at the model layer or will
01:06:12
you see an explosion of value at the
01:06:14
application layer? I'm hoping obviously
01:06:16
that it'll be at all layers of the
01:06:18
stack. PC beneficiaries. But to me,
01:06:21
that's a really interesting question
01:06:22
right now.
01:06:23
>> Yeah. What happens to Salesforce,
01:06:24
HubSpot, you know, Oracle, right down
01:06:26
the line? David, uh, Chimati, your
01:06:28
thoughts here, uh, on the the layers
01:06:31
here and where the value is captured.
01:06:34
>> It's too early to tell.
01:06:35
>> Too early to tell, right? And energy we
01:06:37
kind of put into sort of data center as
01:06:40
well, but that's obviously been a clear
01:06:41
winner. Little housekeeping here.
01:06:43
Liquidity, put a little Tiffany in here.
01:06:45
uh producer Nick D is sold out. There's
01:06:49
a wait list of hundreds of people, but
01:06:51
it is what it is, folks. If you snooze,
01:06:53
you lose and top tier speakers are
01:06:56
coming. Uh it's going to be great. We'll
01:06:58
get a an update from But I think Brad,
01:07:00
you're going to be joining us again.
01:07:01
Yes. For liquidity.
01:07:02
>> I have an update.
01:07:03
>> That's probably not your headliner,
01:07:05
though. I'm probably not your headliner.
01:07:06
>> No, but you always score so high. Every
01:07:08
event you've spoken at, you've been
01:07:09
either number one, two, I don't think
01:07:11
you've ever dropped to three. Go ahead,
01:07:14
Sham. Make your announcement here.
01:07:17
>> Nat sent me an article from Wikipedia
01:07:19
about penile links when you guys are
01:07:21
talking about
01:07:21
>> breaking news.
01:07:22
>> Showing me showing me that I'm in the
01:07:24
large category. Top 5%. She highlighted
01:07:27
it.
01:07:27
>> Top 5%. Okay. And that's with Is that
01:07:30
with Nano Banana or without? Is that
01:07:34
>> She just texted dummy. It's clogged. My
01:07:35
apologies. Clogged.
01:07:36
>> Oh.
01:07:38
>> All right. This is why Jamath isn't
01:07:40
afraid of the cyber is because nothing's
01:07:42
going to come out that's more
01:07:42
embarrassing than what he says himself
01:07:44
on the box.
01:07:44
>> He's like Bezos. When Bezos got hacked,
01:07:46
HE'S LIKE, "GUYS, I GOT HACKED."
01:07:49
>> SO, I saw the agenda for this thing.
01:07:51
It's incredible. Congrats to you guys. I
01:07:52
mean, like the uh like just the fun of
01:07:55
being in Napa, all the poker, all the
01:07:57
the dining experience. This is five star
01:08:00
all the It looks really
01:08:01
>> six-star. It's a man level because
01:08:03
Chimath
01:08:04
>> was, I dare I say, belligerent in his
01:08:08
demands. He said, "This has to be
01:08:10
six-star or I will not show up." Jake
01:08:12
Al. I said, "Okay, boss, get to work."
01:08:15
And uh, Chimath, what do you got any? No
01:08:17
mids. This is all elite. And for the
01:08:20
hundreds of people who are on the wait
01:08:21
list, I am sorry, but we have a capacity
01:08:22
issue. We'll try to get you in for next
01:08:24
year. But Chim, give us some updates
01:08:26
here. You have any updates that you want
01:08:27
to share? because you are running
01:08:28
programming for liquidity 2026 up in
01:08:31
Yon.
01:08:32
>> Look, it's going really well.
01:08:35
Really excited to hear all of these
01:08:36
great folks speak. I think the next two
01:08:38
will release today. Brad Gersonner and
01:08:42
Thomas Leaf of COTU
01:08:44
>> of CO2. That's a great get.
01:08:46
>> We also have I think three people
01:08:47
confirmed for their best ideas pitch.
01:08:49
Really interesting folks. They each run
01:08:52
between one and six or seven billion
01:08:55
>> awesome
01:08:56
>> superstar compounders early in their
01:08:58
career.
01:08:59
>> This is a new zone chamat.
01:09:00
>> It's great. So right now we have Bill
01:09:02
Aman, we have Andre Carpathy, we have
01:09:05
Dan Loe, we have Thomas Lefont, we have
01:09:07
Brad Gersner, we have Sarah Frier and
01:09:10
more to come. We will announce more.
01:09:12
>> There might be one or two surprises. Jay
01:09:13
Cal
01:09:14
>> and a couple and a couple of surprises.
01:09:17
>> Yeah, we we don't announce all the
01:09:18
speakers. Jay Cal's got a couple of
01:09:20
surprises coming. And if you didn't get
01:09:22
in to liquidity, apologies. You're on
01:09:25
the wait list. We are going to be
01:09:26
hosting the fifth
01:09:30
annual all-in summit in Los Angeles
01:09:33
September 13th to the 15th of Sax. You
01:09:36
going to come to that?
01:09:37
>> Allin.com/events.
01:09:40
>> Sax, you should come to that.
01:09:41
>> I've been advised that I can attend
01:09:43
business. I can be in the state for
01:09:44
business reasons.
01:09:46
>> Okay, there you go. Then we'll see you
01:09:47
at liquidity and the summit. Correct.
01:09:49
That's that's big news. Now we just got
01:09:51
a bunch of Sachs stands who are racing.
01:09:53
Uh and now we're going to get Sachs at
01:09:55
This is what happens every year behind
01:09:57
the scenes.
01:09:58
>> Sachs at the last minute says, "Oh, I
01:10:00
have four speakers and I have 72 people
01:10:02
who need tickets and then the whole team
01:10:04
has to like do a fire drill 48 hours
01:10:07
before the event." Okay, here we go,
01:10:08
guys. We're going to go to the third
01:10:09
rail here. We got to catch up on the
01:10:11
Iran war. Here's the latest. Two weeks
01:10:14
into a ceasefire have started just two
01:10:17
days ago at the taping of this VP JD
01:10:20
Vance, friend of the pod is a and some
01:10:23
special consultants Wikoff and friend of
01:10:25
the pod Jared Kushner are headed to
01:10:28
Islamabad, the capital of Pakistan for
01:10:31
talks this very weekend. So while you're
01:10:33
listening to this event, they are going
01:10:34
to be working on the peace deal. Easter
01:10:36
Sunday, Trump posted a truth stating,
01:10:39
"Open the [ __ ] straight, you crazy
01:10:42
bastards, or you're going to be living
01:10:43
in hell. Just watch." Praise be to
01:10:45
Allah. On Tuesday morning, Trump posted
01:10:48
uh a another threat on social media. A
01:10:50
whole civilization will die tonight.
01:10:53
Never to be brought back again. I don't
01:10:55
want that to happen, but it probably
01:10:56
will. Tweets were obviously discussed uh
01:10:59
a lot over the last week. He gave him an
01:11:01
8:00 p.m. deadline.
01:11:03
At 6:30 p.m. POTUS announced on Truth
01:11:05
Social that he had agreed. President
01:11:08
Trump had agreed to a two-week ceasefire
01:11:11
if Iran opens the straight. He also
01:11:13
said, "Hey, listen. We got the straight.
01:11:15
Maybe there'll be a toll booth, but
01:11:16
we'll take the majority of the toll and
01:11:18
we'll split it with Iran." Here's the
01:11:20
quote. We received a 10-point proposal
01:11:22
from Iran, and we believe it's a
01:11:24
workable it is a workable basis on which
01:11:28
to negotiate. And apparently Netanyahu
01:11:31
took the ceasefire to mean level Lebanon
01:11:33
dropping 160 bombs in 10 minutes
01:11:36
yesterday. Saxs, uh, you were out last
01:11:38
week. Everybody wants to know your
01:11:39
position on the war. I'll hand it off to
01:11:42
you. What are your thoughts on how on
01:11:44
the two ceasefire and everything that's
01:11:45
occurred up until this point?
01:11:47
>> Well, look, I have to preface what I'm
01:11:49
about to say, which is I'm not part of
01:11:52
the foreign policy team at the White
01:11:54
House. And the last time I commented on
01:11:57
the war on this show, it somehow made
01:11:59
international headlines that Trump
01:12:01
advisor says XYZ.
01:12:05
And I'm not a Trump adviser on this
01:12:07
issue. I think that'd be a fair headline
01:12:09
to write if it was a technology issue,
01:12:11
but this is not. So whatever I say is
01:12:14
just my personal opinion, but then the
01:12:16
media is going to somehow portray it or
01:12:17
attribute it
01:12:18
>> to the White House or try and create an
01:12:20
issue out of it. So, I feel like I'm
01:12:22
limited in what I can say except that to
01:12:25
say that I think it's terrific that we
01:12:28
have the ceasefire. I think it's great
01:12:30
that there's going to be this meeting in
01:12:33
Islamabad to hammer it out. And I think
01:12:37
what the president's accomplished so far
01:12:38
with the ceasefire is it's a great thing
01:12:41
because what happens with these wars is
01:12:43
they take on a life of their own,
01:12:45
meaning they tend to go up the
01:12:47
escalation ladder, right? And there's a
01:12:48
lot of podcasts that are discussing the
01:12:50
so-called escalation trap and supposedly
01:12:52
there are stages to this based on
01:12:54
historical patterns. And so I think it's
01:12:56
actually very hard to pull out of these
01:12:58
things and I give the president
01:12:59
tremendous credit for negotiating the
01:13:02
ceasefire that we've achieved so far and
01:13:04
then sending the team to hopefully work
01:13:06
this out.
01:13:07
>> Brad, actually my first trip to the
01:13:08
Middle East was when you and I uh maybe
01:13:10
four years ago when Thank you for taking
01:13:12
me. What is your take on where we're at
01:13:14
here? I think we're just wrapped up week
01:13:16
six of this and we're going into week
01:13:17
seven.
01:13:17
>> First, on March 4th, I tweeted the Trump
01:13:20
doctrine in Iran. Massively destroy all
01:13:23
military capa capabilities. Kill the
01:13:25
people building lethal weapons to use
01:13:27
against us and get out. Reserve the
01:13:29
right to do it again if needed. Zero
01:13:31
efforts to build Misonian democracy.
01:13:33
Iran's going to have to build what comes
01:13:35
next. And I think what the market has
01:13:38
said right if you look back at last year
01:13:40
on tariffs Jason the top tobottom draw
01:13:42
down was about 15% on the NASDAQ
01:13:46
intraday is down 22%.
01:13:48
Okay, the draw down in this period over
01:13:51
Iran was only down about 5 to 7% on S&P
01:13:55
and NASDAQ, right? So, the market has
01:13:58
said, listen, re trust Trump at his
01:14:01
words. He said he's not going to get
01:14:02
into an entangled war here. I think he
01:14:05
terrifies the hell out of people with
01:14:06
his tweets about, you know, destroying
01:14:08
civilization and all this other stuff.
01:14:10
But I think people, even though they
01:14:12
don't like to hear it, they've resolved
01:14:14
for themselves that when he says he's
01:14:16
going to get out, he will in fact get
01:14:17
out. Of course, there was a lot of hand
01:14:18
ringing, but if you look at the markets
01:14:20
today, we basically bounced all the way
01:14:22
back from where we were pre-Iran on both
01:14:26
the S&P and the NASDAQ. If in fact we
01:14:29
land the plane, if JD lands the plane,
01:14:31
and by the way, on Lebanon, yes, they
01:14:33
were bombing yesterday, but Netanyahu
01:14:34
has now said that you're going to have
01:14:36
direct government talks between Israel
01:14:37
and Lebanon. So, if the if we land the
01:14:41
plane on these two things, I think it's
01:14:42
off to the races in the market. By the
01:14:44
way, while everybody's focused on Iran,
01:14:47
stay tuned. I think we're getting close
01:14:48
to a deal on Ukraine, Russia, right?
01:14:51
Venezuela is, you know, kind of going
01:14:53
seemingly very well. I think there's
01:14:55
also going to be news on Cuba. You could
01:14:58
envision a world there's risk to the
01:14:59
downside. Certainly, I will stipulate,
01:15:02
but you also have to pay attention to
01:15:03
the risk to the upside. If you land the
01:15:05
plane on those things heading into
01:15:07
America 250 July 4th, the market could
01:15:10
really take off.
01:15:11
>> All right. Well, let's uh maybe uplevel
01:15:13
this a little bit and talk about why
01:15:15
we're in this war to begin with. And
01:15:17
that's the big discussion amongst both
01:15:19
sides of the aisle. On Tuesday, the New
01:15:20
York Times dropped an inside the room
01:15:23
piece on how President Trump made the
01:15:26
decision
01:15:28
according to this report, if it's true.
01:15:30
I know some people don't uh subscribe to
01:15:32
the New York Times anymore or think it's
01:15:33
fake news, but how Trump decided to
01:15:36
basically follow Netanyahu into this
01:15:38
war. On February 11th, Netanyahu met
01:15:40
with Trump at the White House where he
01:15:41
gave him a four-part pitch on attacking
01:15:44
Iran. JD Vance, according to the story,
01:15:46
if it's true, disclaimer, disclaimer,
01:15:48
warned Trump that the war could cause
01:15:50
regional chaos and break apart Trump's
01:15:53
MAGA 2.0, the Trump 2.0 coalition we
01:15:56
talked about here, the big tent. And
01:15:57
that's turned out actually to be true.
01:15:59
There's been a bunch of hand ringing
01:16:00
from Megan, Kelly, Tucker, Carlson,
01:16:02
right on down the line. Rubio was
01:16:04
anti-regime change, but he was largely
01:16:07
ambivalent, according to this story
01:16:08
about the bombing campaign. Susie Wilds,
01:16:12
chief of staff, said she had concerns
01:16:13
about gas prices before the midterms.
01:16:15
Pretty good uh advice there. And General
01:16:18
Dan Kaine, chairman of the Joint Chiefs
01:16:20
of Staff, said this of Netanyahu's
01:16:23
pitch. Quote, "Sir, this is, in my
01:16:24
experience, standard operating procedure
01:16:26
for the Israelis. They oversell and
01:16:28
their plans are not always
01:16:30
well-developed. They know they need us
01:16:31
and that's why they're hard selling. If
01:16:33
you put this together with Rubio's
01:16:35
walked back comments at the start of the
01:16:38
war, we knew, this is quote from Rubio,
01:16:41
we knew there was going to be an Israeli
01:16:44
action. We knew that would precipitate
01:16:47
an attack against American forces and
01:16:49
that's why we did it. I had Josh Shapiro
01:16:52
on the All-In interview show and um uh
01:16:56
he talked a lot about this. There is a
01:16:58
big underpinning here, Chimath, that the
01:17:01
United States foreign policy is being
01:17:03
driven by Netanyahu. Every Jewish
01:17:07
American person I've talked to feels
01:17:08
Netanyahu is not doing Jewish American
01:17:11
and Jewish the Jewish diaspora any
01:17:13
favors here by his approach to these
01:17:15
wars. What are your thoughts on why we
01:17:17
got into this and how we get out of it?
01:17:22
>> I mean, the person that decides is the
01:17:24
president of the United States. some
01:17:26
foreign leader isn't
01:17:29
getting to call the shots in the United
01:17:31
States. I think very practically
01:17:33
speaking,
01:17:35
the markets are effectively
01:17:38
pricing in that this was a small blip
01:17:42
for whatever people think. That's just
01:17:44
what the best prediction market that we
01:17:47
have is telling us. I think that's
01:17:48
important to acknowledge that we're
01:17:51
probably in the endgame here. And the
01:17:53
second thing to acknowledge is if I was
01:17:55
Israel, I would really be concerned
01:17:59
that unless I help find an offramp
01:18:02
quickly, the risk that Israel loses
01:18:05
America as a predictably steadfast ally
01:18:08
could go down. And I think that that's
01:18:09
problematic for Israel
01:18:11
>> far more than is problematic for the
01:18:12
United States.
01:18:13
>> So all of that kind of tells me that we
01:18:16
will find an offramp. A because I think
01:18:19
economically it makes sense and then B
01:18:22
geopolitically I think Israel will want
01:18:23
to make sure that this doesn't burn
01:18:26
a long-standing relationship.
01:18:29
>> Yeah, that that seems to me to be the
01:18:32
major issue here is Americans basically
01:18:34
do not want to be in this war. Americans
01:18:37
do not want a forest policy being
01:18:39
influenced to the extent they believe.
01:18:42
I'm not putting my belief in here.
01:18:43
Americans believe we are being dragged
01:18:46
into this by Israel and that Israel has
01:18:48
too much or Netanyahu specifically has
01:18:50
far too much influence. And then people
01:18:52
believe the anti-semitism that's
01:18:53
occurring here. Josh Shapiro gave me a
01:18:55
lot of push back on this. Uh but all the
01:18:58
Jewish Americans I talked to say
01:18:59
Netanyahu is causing with his actions in
01:19:02
Gaza, Lebanon, Iran. Uh he's gone too
01:19:05
far and it's causing the anti-semitism
01:19:07
we're experiencing uh today. So you can
01:19:09
make your own decisions about that. Any
01:19:11
final thoughts here, Brad, on
01:19:14
the American foreign policy being
01:19:16
influenced too much by Israel?
01:19:18
>> No, it's the discussion. I
01:19:20
>> I mean, listen, um kind of like Sax said
01:19:24
earlier, um I think that we will
01:19:27
ultimately be judged by the outcomes,
01:19:31
right? And that everybody is an armchair
01:19:33
pundit today on, you know, uh the the
01:19:37
the approach that we're taking in these
01:19:39
two different places. I think we could
01:19:41
be on the verge of a massive
01:19:44
transformation of the Gulf States. You
01:19:46
went there with me, Jason. Saudi,
01:19:48
Qataris, Kuwaitis, Emiratis. I've talked
01:19:50
to a lot of them this week. I think
01:19:52
they're very hopeful and optimistic. I
01:19:54
think you could bring Iran into the
01:19:56
fold. But listen, I'm an optimist on all
01:19:58
of this stuff. I I just want to remind
01:20:00
people, doing nothing in Iran had
01:20:04
tremendous risks. Doing nothing in
01:20:07
Venezuela had tremendous risks. So, it's
01:20:10
not as though this was uh, you know,
01:20:13
something that I I I I think wasn't well
01:20:16
calculated, but I think we have to let
01:20:19
the cards be played and and and then let
01:20:21
history be the judge. But I think
01:20:22
there's uh a risk in both directions,
01:20:25
but I'm going to remain optimistic.
01:20:26
>> All right. You uh said in the Gaza
01:20:28
situation, we should have a wide birth
01:20:30
for criticism of Israel and Netanyahu.
01:20:32
What are your thoughts on this belief
01:20:35
here in the United States now in this
01:20:37
discussion that Israel is having far too
01:20:38
much influence over the United States
01:20:40
foreign policy?
01:20:42
Well, I noticed in my feed today that
01:20:44
Naftali Bennett, who's a major Israeli
01:20:47
politician who was a former prime
01:20:49
minister, tweeted polling that showed
01:20:52
that Israel was becoming very unpopular
01:20:55
in the US and he was expressing concern
01:20:57
about that and expressing
01:21:01
the need to to basically address that or
01:21:03
fix that. So, I think you're starting to
01:21:05
see Israeli politicians raising that as
01:21:08
an issue. And I think that's probably a
01:21:11
good thing. Yeah, there it is. And it's
01:21:14
really cool actually how X now just
01:21:16
automatically translates things from
01:21:18
foreign languages, in this case, Hebrew,
01:21:20
and it puts it in your feed. So, yeah.
01:21:22
So, here's Naftali Bennett, former prime
01:21:23
minister, saying, "This is a serious
01:21:24
situation. There's a lot of work ahead
01:21:26
of us to fix everything." Now,
01:21:29
obviously, this is not Netanyahu. this
01:21:31
is one of his um political opponents.
01:21:33
But
01:21:34
yeah, I mean this is something for
01:21:36
Israel to consider and think about and I
01:21:39
think that they would improve their
01:21:41
popularity uh if they got behind the
01:21:43
ceasefire and I have no indication that
01:21:45
they won't but that would certainly be a
01:21:48
good place to start. I have to say just
01:21:50
as an aside, this auto translate feature
01:21:52
has done more for understanding across
01:21:56
borders than anything I've ever seen.
01:21:58
And it is the most impressive tech
01:22:02
feature I I've seen released in years.
01:22:04
Putting AI and large language models
01:22:06
aside for people who don't know what's
01:22:08
happening because of Grock being really
01:22:11
good at doing auto translate. They've
01:22:12
taken the pockets of the best of what's
01:22:16
happening in Japan, what's happening in
01:22:17
Israel, what's happening in France, and
01:22:19
they're surfacing it auto translated.
01:22:21
Then when you reply as an American to
01:22:23
somebody in Japan, they see it autorated
01:22:26
as well, which has led to people who
01:22:28
don't speak the same language engaging
01:22:31
on X in a very nuanced, fun, interesting
01:22:36
way. And that for as a truth mechanism
01:22:40
is just absolutely extraordinary. I
01:22:41
think this is going to have such a
01:22:42
profound effect. Maybe Elon and the X
01:22:45
team should get like a Nobel Peace Prize
01:22:47
award for this. I think it's going to
01:22:48
change. I mean, I hate to be hyperbolic,
01:22:50
but have you been using this feature,
01:22:52
Chamat? Has it been coming up in your
01:22:53
feed? And which language is up in your
01:22:56
feed right now?
01:22:58
>> English. Okay. So, you're not part of
01:23:01
the translation thing. Brad, has this
01:23:02
hit your feed yet? And and which regions
01:23:04
are you?
01:23:05
>> Definitely. Definitely see it in on the
01:23:07
Middle East stuff. Um, and uh, you know,
01:23:10
I've seen on Chinese, I've seen it on on
01:23:12
the Russian Japanese
01:23:14
>> super helpful.
01:23:14
>> Let me tell you, bass Japanese is a
01:23:16
whole another level of beast.
01:23:18
>> Whoa. Man, base Japanese makes like
01:23:22
Fentes and Alex Jones seem tame. They're
01:23:25
like, look at this group of people.
01:23:27
Insert whatever group of immigrants you
01:23:29
like. And they're like, this is
01:23:30
unacceptable behavior. This is not
01:23:32
Japanese culture. These people need to
01:23:34
be get the hell out of Japan. It is
01:23:38
wild, folks. And if you don't have an X
01:23:40
account, you are missing out. Go to
01:23:42
X.com and sign up for this reason only
01:23:45
because you think about the velocity.
01:23:47
Like journalists are not even taking the
01:23:48
time to translate and cover what's going
01:23:50
on in those areas. And this is happening
01:23:52
automatically in real time.
01:23:55
>> So you start thinking about what
01:23:56
happened in Ukraine. If you had people
01:23:58
Russia and Ukraine doing this and having
01:24:00
conversations with each other, it would
01:24:02
be wild.
01:24:03
>> You're like a such a good hype man. The
01:24:05
problem is you hype buttered bread the
01:24:07
same way you hype a nuclear reactor. And
01:24:08
so it's hard to really tell, you know,
01:24:10
what you're really hyping because your
01:24:12
level of excitement, the intonation is
01:24:14
exactly the same.
01:24:15
>> Yo, man, there's nothing better than a
01:24:17
slice of great toast. I mean, if this is
01:24:18
very this in in a way, it is like sliced
01:24:21
bread. It's very simple, but it is so
01:24:23
powerful in the experience.
01:24:25
>> This has been
01:24:26
>> It is true. X is better today than it's
01:24:28
ever been. And remember, they have 70%
01:24:30
fewer employees than they had the day
01:24:33
Elon walked into the building. And so if
01:24:35
there were ever a debate
01:24:37
>> about this, like, and I remember
01:24:38
everybody saying, "Oh, it's going to tip
01:24:40
over. Oh, it's going to be a crappy
01:24:41
experience."
01:24:43
>> The fact of the matter, here's we are a
01:24:45
few years later, 70% fewer employees,
01:24:48
and every other company in Silicon
01:24:49
Valley is looking at that. I think for a
01:24:51
lot of these tech companies, we've hit
01:24:53
peak employment. We're going to create a
01:24:55
tremendous number of new jobs, but for
01:24:58
the existing jobs, these companies are
01:25:00
all realizing they can do more with
01:25:01
less.
01:25:01
>> Nikita Beer just tweeted that they're
01:25:03
about to go ham on these bot accounts
01:25:06
that auto reply.
01:25:07
>> Yes.
01:25:08
>> Those those literally ruined my feed.
01:25:10
>> That's why I went to subscriber mode in
01:25:12
my replies and it's it's worked out
01:25:13
great.
01:25:14
>> Yeah. No, shout out to him and um to
01:25:16
Chris Saka who was in tears at what
01:25:18
happened to Twitter. You
01:25:20
>> It's gonna be okay, Chris. Sorry. No
01:25:23
more tears.
01:25:24
You only let subscribers respond to your
01:25:26
tweets. I
01:25:26
>> I do 50/50. Sometimes I'll just let it
01:25:29
rip and get chaos. And then other times
01:25:31
I have 2,000 paid subscribers. I give
01:25:33
all the money to charity, like 30 grand
01:25:34
a year. And it's just wonderful to get
01:25:36
to know the same 2,000 people out of my
01:25:38
million followers. It's kind of like
01:25:40
having this little subset. So sometimes
01:25:41
I'm like, I don't have time to deal with
01:25:43
a 100red or 200 or 300 replies.
01:25:45
>> You have a million. That's incredible. I
01:25:48
mean, it's just
01:25:49
>> I mean, you have two million. I think
01:25:50
Sax must have a million, right? You have
01:25:52
a million, right, Sax? Only only
01:25:53
>> Brad. How many you have now? You're
01:25:54
getting popular.
01:25:56
>> You built a couple.
01:25:57
>> Got a couple hund.
01:25:59
>> What's your Oh, your alt cap altca.
01:26:01
>> I'm at 1.4 million. What are you at,
01:26:03
Jacob? Have I surpassed you?
01:26:06
>> I think you have. I'm like 1.1.
01:26:07
>> What it cost me to get my real name,
01:26:09
Jason?
01:26:10
>> Uh, I know a guy. Find out.
01:26:13
>> You're 1.1. Yeah, I made it to 1.4. I
01:26:15
don't know how that happened exactly.
01:26:17
>> Just having the number one podcast in
01:26:19
the world. Uh, another amazing episode
01:26:22
of the number one pop and Chimath has
01:26:26
two million, but that's only because he
01:26:28
engages he has just incredible moments
01:26:32
of uh engaging with his haters. Oh my
01:26:35
god, the the the replies that Chimath
01:26:38
sometimes drops are so great. I love
01:26:40
Chimoth goes
01:26:41
>> I light them up. I light them up.
01:26:43
>> He lights them up. Then you had somebody
01:26:45
who was like, "Oh my god, I was in the
01:26:47
casino and you told me to bet black, so
01:26:48
you bet black, so I bet black and I lost
01:26:50
my money." And so you're responsible and
01:26:53
then you paid for the kids college. He
01:26:54
has two young girls and so I I funded
01:26:57
their college accounts.
01:26:58
>> I thought that was hilarious. Just as
01:27:00
>> obviously I'm very happy for him and his
01:27:02
two daughters. I'm even more happy at
01:27:04
how much it'll anger all these other
01:27:06
goofball dorks living in their mom's
01:27:08
basement.
01:27:09
>> Yes.
01:27:09
>> Who'd literally have no take they take
01:27:11
no responsibility for their lives.
01:27:14
And uh they should enjoy those Hot
01:27:16
Pockets. By the way, for those folks in
01:27:17
their mom's basement, the Hot Pockets
01:27:19
and the Fish Sticks are ready and you
01:27:21
get one more hour of Xbox from mom.
01:27:23
>> All right, listen. We missed you,
01:27:24
Freeberg, but this is the best episode
01:27:26
in two years. Uh
01:27:29
>> Freeird at the end of the show.
01:27:32
>> And we will see you all at the liquidity
01:27:34
summit except for the 400 people on the
01:27:36
wait list who aren't going to get in.
01:27:37
>> We got an email from the guys at Athena
01:27:39
because we were just
01:27:40
>> Oh my god. the they they're they're
01:27:43
going to hire like 500 new Athena
01:27:45
assistants.
01:27:46
>> Yes, they had a thousand people after
01:27:48
last week when we mentioned how much we
01:27:50
love Athena.
01:27:51
>> Go to Athena.com.
01:27:52
>> But that's amazing. Those are like 500
01:27:54
hardworking men and women who are like
01:27:56
working
01:27:57
>> in the Philippines.
01:27:59
>> Sax have great jobs.
01:28:00
>> Sax, I'm going to get you a couple
01:28:01
Athena assistants as a birthday present.
01:28:03
That's what I'm going to get.
01:28:04
>> You're going to love this, Sax. H Athena
01:28:06
assistants are the best. Congratulations
01:28:08
to my friends over there. All right,
01:28:10
everybody. We'll see you next time. Love
01:28:11
you boys on
01:28:12
>> tonight favorite podcast.
01:28:19
>> Let your winners ride.
01:28:21
>> Rain
01:28:26
and
01:28:26
>> we open source it to the fans and
01:28:28
they've just gone crazy with it.
01:28:30
>> Love you queen of
01:28:33
winners.
01:28:39
>> Besties are gone.
01:28:41
That is my dog taking out your
01:28:43
driveways.
01:28:46
>> Oh man, my appetiter will meet.
01:28:49
>> We should all just get a room and just
01:28:50
have one big huge orgy cuz they're all
01:28:52
just useless. It's like this like sexual
01:28:54
tension that we just need to release
01:28:55
somehow.
01:29:02
>> That's going to be good. We need to get
01:29:03
merch.
01:29:12
I'm going all in.
