August 4, 2025 • 52 mins
When it comes to tech startups, you often hear about VCs making a ton of money, or founders experiencing life-changing exits. But something is changing in the world of AI. Now it's the engineers themselves getting pay packages that can be in the 9-figure range. Why is this? Why is it happening? How is it changing the culture of Silicon Valley and business more generally? On this episode, we speak with John Coogan and Jordi Hays, the co-hosts of TBPN, a daily show about technology, which covers the industry in a sports-like manner. We talk about the economics of these transactions, why they make sense, and who are the industry's top superstars.
Read more:
Meta Seizes Its Moment to Spend Aggressively in the AI Race
Apple Rebound Looks Elusive as AI Woes Draw Investor Scrutiny
Only http://Bloomberg.com subscribers can get the Odd Lots newsletter in their inbox each week, plus unlimited access to the site and app. Subscribe at bloomberg.com/subscriptions/oddlots
See omnystudio.com/listener for privacy information.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:02):
Bloomberg Audio Studios, Podcasts, radio News.
Speaker 2 (00:18):
Hello and welcome to another episode of The Odd Laws podcast.
Speaker 3 (00:22):
I'm Joe Wisenthal and I'm Tracy Alloway.
Speaker 2 (00:24):
Tracy, I love doing the podcast, but I'm kind of
thinking about becoming a PhD level AI researcher if I
do a career pivot.
Speaker 3 (00:33):
It seems something. Have you seen?
Speaker 2 (00:37):
Have you seen like the salaries supposedly I still don't
actually believe them, But have you seen like the salaries
and comp packages that supposedly Meta and a few others
are paying for it.
Speaker 4 (00:47):
I have seen some of the headlines. I have also
learned a bunch of new words like exploding offer and
acquahireh exploding offer sounds dangerous. What you want is an
explosive offer that you can't turn down.
Speaker 2 (01:01):
I take it the exploding hire is like one that
lasts like five minutes.
Speaker 4 (01:04):
Yeah, so you have to make a decision right away,
and the company that you're currently working for can't counteroffer.
Speaker 2 (01:09):
So I read these headlines about some engineer something getting
paid one hundred million or two hundred and fifty million.
I actually literally don't believe them, Like I actually literally
think that's fake notes. Really, no but I.
Speaker 4 (01:19):
Kind of just think it's probably made up of like
all this weird equity structure.
Speaker 2 (01:23):
But I'm sure it's not all upfront anyway. We sort
of seem to be in a moment and it's actually
not just AI. I would say journalism, finance, et cetera,
where it's like the sportsification of a lot of different industries,
individual talent, the demand for individual superstars, and anything doing
very well.
Speaker 4 (01:44):
I have a bunch of questions when it comes specifically
to AI, such as what makes an AI talent because
I assume if you're Mark Zuckerberg and you're trying to
assemble your AI dream team or something, Okay, sure you
have like a level of technical insight, and maybe you
hear people talk talking about, oh, this one guy is fantastic,
But I assume it's not like you can look at
(02:04):
their like individual levels of code. You can't see what
they're doing on like a daily basis. I don't know.
Speaker 2 (02:11):
I don't know either. I don't know anything about this stuff,
but I'm really excited to say we do have the
perfect guests, A couple of guys who are right in
the middle of all of this and also who have
sort of been ahead of the curve in terms of this,
like I said, the sportification of the industry. We're going
to be speaking with John Coogan and Jordie Hayes. They
are the co hosts of TBPN. It's a live show podcast.
(02:33):
It's become one of my favorite new media properties. It exists,
and I'm really excited we have them both in the
studio here with us today. So John and Jordi, thank
you so much for coming on aut Loots.
Speaker 5 (02:45):
Thanks for having us.
Speaker 3 (02:46):
We're so excited to be here.
Speaker 2 (02:47):
I love by the way you guys like make these
like baseball cards every time, So it's so clever for
those who are not paying attention. Why do you sort
of give the high level overview of like what you
see happening in this like crazy talent for people who
don't know about this, what is actually going on? This
talent war for people who know how to do AI
or train a model or whatever it is.
Speaker 5 (03:09):
Yeah, the League or the mag seven the serious teams
are all worth over a trillion dollars. Tesla's in a
different kind of boat some days, but pretty much there's
multiply seven trillion plus dollar companies now, so there is
an inordinate amount of money to flow around. And if
you think about investing zero point one percent of your
market cap to make your business potentially five percent better,
(03:33):
that's a trade you take all day. The number is staggering,
but now you're seeing companies pay hundreds of millions of dollars.
I think you mentioned that. You know, there's a lot
of debate over whether that was fake news or not.
They seem like they are very real offers and they
are indeed happening, and there's a whole bunch of different
ways that you can kind of underwrite that. But we
could start with going through just a little bit of
(03:55):
the history here of how we got into this place
or what these AI researchers are actually doing. I'm happy
to kind of answer any question.
Speaker 4 (04:03):
Okay, well, why don't we start with the baseball analogy?
Then if I was collecting AI engineer cards, who's the
most valuable? Like who do I actually want to have?
And then secondly, what are the stats that are.
Speaker 6 (04:17):
So taking a little bit of a step back talking
about kind of the sportsification of tech and business, which
has been a huge catalyst for our show. I think
we realized early on people would call TVPN the Sports
Center for Tech or some analogy like that. We were
a little bit it sounded cool. We didn't really know
what that meant. John and I don't watch sports at all,
(04:37):
but for basically, for basically like two decades now, I'm
in my late twenties, John mid thirties, and we've followed
tech in business the way that our college friends follow sports.
So in sports you have players, personalities, coaches, managers, leagues, teams,
and people obsess over all the details. Fully understood that,
(05:01):
I mean, I understood kind of the draw, but it
just was never for me. Whereas John and I would
pick up the newspaper as teenagers and we were tracking
the talent, the CEOs, the companies, the markets, the industries,
and it's just something that we obsessed over. And so
this year has been amazing. As these AI researchers have
(05:21):
been getting these sort of superstar max contracts, which is
what we would call them, we would start joking on
the show and be like, look, this guy just went.
Speaker 3 (05:29):
Over to Meta.
Speaker 6 (05:29):
It's probably you know, four year contracts, you know, one year, Cliff.
And it just got more and more and more real
as the kind of demand for world class AI researchers
completely outstripped the supply, and we actually put out something
a couple of days ago called the Metas List, and
John can go a little bit more into into the
name behind that. It was kind of our take on
(05:51):
the Midas List, which is, we built a list of
top roughly one hundred AI researchers and rank them on
a bunch of different factors. And so when you talk
about what goes into what makes a great AI researcher,
there's a bunch of different factors. One that you can
get right into the numbers with is like citations. So
if somebody is a researcher, they've been publishing, you know, studies, papers,
(06:12):
et cetera for quite a while at this point, and
you can just see quantitatively what their contributions have been
to the industry.
Speaker 3 (06:19):
And so that's like a good starting point to understand.
Speaker 6 (06:21):
And historically Elon even said earlier this week, honestly, it
was like an hour after we posted the Metaalist.
Speaker 3 (06:28):
Probably unrelated, but who knows.
Speaker 6 (06:30):
He's basically saying that AI researchers a and engineers, We're
just going to call them engineers now, and that makes
a lot of sense for someone like Elon to do
as somebody who's always been very engineering focused, less focused
on you know, entirely net new innovation. More so, how
do we make the best possible versions of products that exists.
(06:52):
And it's not to say that he hasn't innovated in
a bunch of different ways, but it is a really
wild moment in time, and it's been a fun moment
because historically you hear about this minas list investor, you know,
making two billion dollars of carry on some deal, or
this founder you know I p O ing. Today we
have you know, the Pigma I p O and uh,
(07:13):
you know there's a bunch of people that are going
to make billions of dollars there, But you don't hear
about the hundredth engineer that was hired, yeah, making you
start a one hundred million dollars signing bonus. And all
of this makes a ton of sense in the context
of what John said, because you look metas up I
checked this morning one hundred and ninety five billion dollars
(07:34):
new market cap and think about how many one hundred
million dollars signing bonuses you can make against that kind
of market cap.
Speaker 3 (07:41):
Should we answer your question, yes, who the who's the bus?
Speaker 5 (07:44):
Who's the whale? To have the who is the light whale?
Speaker 4 (07:51):
Don't say anything about whales or whale products or hunting.
Speaker 2 (07:56):
Large is the way.
Speaker 5 (08:00):
Perhaps he is the Michael Jordan of AI researcher, but
Ilias Sitskiver is really He's at the top of our
list for a variety of reasons. And if you're not
familiar with Ilias Sutzkiver, he is an AI researcher who
was at open AI for a long time, and there's
a few different ways to characterize him. He is both
coming up with new ways to implement AI algorithms the
(08:23):
way you train the model, but he's also very good
at for a long time identifying which the shortest path
in the tech tree. So there are branches of choices
that you need to make as you develop the new
AI models, and he was very early at while he
was at open AI, he identified that the transformer paper
(08:45):
from Google. He didn't invent that it was at Google,
but the transformer technology was extremely important and that it
had the ability to do remarkable things once scaled up massively,
and so he was the driving force between kind of
identifying the transformers as the correct path. Now we look
back on Attention as All you Need, which is the
name of the paper that defines the architecture that is
(09:08):
used in these modern large language models that you use
when you're.
Speaker 3 (09:11):
In chat GPT.
Speaker 5 (09:12):
There were twenty five other potential paths that we could
have gone down. You know, the story goes is that
he really identified that and said, let's go really really
hard on that.
Speaker 4 (09:22):
So he can sort of identify the innovation, what's new,
and then also identify the most efficient path to actually
execute on.
Speaker 5 (09:30):
Yes, and if you're familiar with what the more the
more modern, So the first arc of LMS and these
chatbots scaling up where really this tack the transformer paper,
understand that that's the correct architecture, and then scale it
up really really big. So you need to be able
to not just write the code to implement that that
particular algorithm. You need to marshal the capital to say
(09:52):
we're going really really big and we're going to build
the big data center. We're going to spend a lot
of money, but it's going to be worth it because
we understand the trade offs here. Then the second kind
of innovation or correct call he had was during the
sam Altman, Ouster and return. He was working on a
project that was code named q Star, and there was
a lot of speculation about what q star was. Was
(10:13):
it the secret super intelligence thing?
Speaker 3 (10:15):
What did Ilius see?
Speaker 5 (10:16):
Yes, yeah, because he was going back and forth in
support of Sam Altman and then leaving and then and
then it was back and forth, and there's a lot
of drama there. Always keep them guessing, Always keep them
guessing for sure, and so there was there was a
lot of rumors, but what it wound up being was
just reasoning, which is now what is available in UH
if you use any of the three models in chat GPT,
(10:37):
or you use deep SEKR one and it's it goes
by a number of different names. The project was rebranded
Strawberry and then the O series, And you can think
about this as the test time. Inference is the other buzzword,
but basically the LM, you're using the same foundation model,
but you're running it a ton more to come up
with a bunch of potential answers to questions and then
(10:59):
narrowing that down and essentially applying what's called reinforcement learning
to the transformer architecture and the pre training that's happening,
and so he was very critical in leading that project,
and so that's allowed him to eventually, once he left
open ai, go start a new company called Safe Superintelligence SSI,
and he's gotten that company to what a thirty billion
(11:20):
dollar evaluation, thirty two billion dollar valuation.
Speaker 6 (11:22):
Which ties into the talent war because I think it's
I don't know that it's been officially announced yet, but
people are expecting Daniel Gross to join Meta, who.
Speaker 5 (11:33):
Was the CEO and co founder of that company. Side
and so back to your back to the white dynamic.
Speaker 3 (11:40):
Yeah, yeah, if you guys get together, they start.
Speaker 6 (11:42):
A company within a year, it's worth thirty two billion dollars,
and one of the people on that team decides to
actually leave and join Meta. It sounds insane, probably leaving
billions of dollars on the table, but you can kind
of trace the this talent war actually back to the
open ai founding team. When you think about all the
(12:02):
different players, right, you have Mira Murti who's now with
Thinking Machines, she was a CTO of open Ai. You
have Ilia Sutskiverer with SSI, you have Elon with XAI,
and not to mention, you have these sort of hyperscalers
who are also competing for the same talent. So yeah,
that's really kind of the origin story in the genesis,
And it's been amazing to see open AI's progress despite
(12:24):
you know, the recent reporting is around you know, losing
a bunch of top researchers, which is real, but it
wasn't that long ago that they lost like two very
very very key senior execs in Ilia and Mira.
Speaker 2 (12:53):
So a few stats were recording this. By the way,
July thirty first, Meta came out with earnings last night.
The stock is up like ten percent on stuff like
one hundred and fifty billion more dollars. The other thing is,
and I think this gets ties into the logic behind
the comp so they're going to spend like something like
another seventy billion on capac some stuff like that. So
we know these are incredibly computationally intensive things. They're electricity
(13:18):
intensive things. You mentioned paper citations, but also actually having
the experience of one of these runs, and it does
seem very highly intuitive to me that if you've done
it before, and if you could even cut down the
cost of a training run or set up a big
computer rack for you can pay five very quickly you
(13:39):
have five percent less, right, exactly happened. Instantly pay then
you instantly pay for your salary, I could say, because
and this is different from the B to B SaaS yer, right,
because you can there's these huge kpax costs if you
could just marginally improve the efficiency that that pays that
one hundred million sellar salary.
Speaker 5 (13:57):
And so this this literally just happened right before Mark
Zuckerberg went on the talent acquisition spray that he's been
on for the last few months. Meta's main AI model
is called Lama and they've released a series of versions
of that model, and Lama four was the latest and greatest,
and it didn't go very well. And the rumors about
(14:18):
why it kind of failed to deliver on expectations was
that they kind of went down the wrong path in
the tech tree and they focused a little bit too
much on scale insanely costly. So they've released a part
of Lama four, but they have yet to release Malama
for Behemoth, which is their biggest model, the most the
(14:40):
most aggressive, and they had just little details in the implementation,
little choices of how you chunk the attention in the
transformer model, you know, we're operating in this level of
abstraction up here talking about you know, it's a prediction model,
and then we talk about transformers a little bit. There
are sub algorithms within these systems that you make the
(15:00):
wrong decision and you could get a vastly different.
Speaker 2 (15:03):
Electricity bill goes up by one hundred.
Speaker 5 (15:04):
Many exactly, so all of that money that was spent
on that electricity and then build out of course that
it can absorb that. But when you think about the
cost of getting correct.
Speaker 6 (15:14):
One thing is is a lot of these projects will
be energy constrained too, So efficiency is going to matter
a lot. Oh, it hasn't been the core focus today.
If you talk to anybody at the big labs, they
are focused on maxing out intelligence at the cost of
efficiency because they know they can. You know, you want
to be on the bleeding edge. You want to have
the smartest model.
Speaker 3 (15:34):
You want to be.
Speaker 6 (15:35):
You know, there's debates around which benchmarks actually matter, how
important AI benchmarks really are. But energy is going to
you know, a lot of people are saying Zuck won't
even be able to spend as much as he wants
to spend on data center development purely because of the
energy constraint.
Speaker 4 (15:52):
Can you talk a little bit about what happened with Windsurf,
because this is this is the one that seems to
have like captured everyone's attention, in part because there was
the drama. It almost seemed like it came out of
like a Silicon Valley the show script or something where
all the employees were gathering expecting to hear that they
were going to be bought by Open Ai, and then
they find out that actually their CTO has just been
(16:14):
bought by someone else completely different, and they're sort of
left in the lart ceo.
Speaker 3 (16:18):
Oh is top CEO in top fifty?
Speaker 5 (16:21):
Well, how about I give some prehistory on acquires and
you can take take us through that actual weekend. So
there has been a trend of sort of these zombie
aqua hires. Everyone has different names for this, but effectively,
when a very large company, usually a hyperscaler, wants to
go and acquire a company that has AI talent, they
used to just buy the whole company. And this was
(16:42):
part of the Silicon Valley social contract that even if
I am in operations or sales or finance or HR,
if I join a hot startup and it gets acquired
I'm coming along for the ride to exit and I
get the job at Google at least for a little bit.
And then yeah, if I underperform Google or Meta Amazon,
they might lay me off. But even if I'm somewhat redundant,
(17:03):
I'm coming along for the ride and my cashing out
my shares, and these the reason that you go and
take the risk and take the little bit of the
rougher ride that is a startup. You don't get as
many amenities, but you get the lottery tickets in the
form of stock options that hopefully pay out. But there's
a big question about how much of this is the
acquirer just not wanting to deal with the d duplication
(17:23):
of the back office. There's also the question of the FTC.
A lot of the FTC rulings have made it much
harder to get these these big acquisitions across the finish line.
And even if they even if the FDC does approve,
they can often hold it up for six months. And
we're in a race where if you deliver the best
model today, you're going to make more money. You're going
to be the hot company, you're going to acquire. It's
(17:45):
all this big snowball so companies. This started with the
very few, but character Ai was the big one with
Noam Shazir.
Speaker 6 (17:51):
Who was in a very very interesting situation where so
Nome wanted to go back to Google, which made sense.
Google wanted him there as well researches of a yes,
and so having him go back there made a lot
of sense.
Speaker 3 (18:05):
I think he had built this platform.
Speaker 6 (18:06):
He built like the first at scale AI companionship platform.
If you look at character AI's site traffic today, I
mean it's it's one of the largest sites on the
internet still, which is wild. But I think he realized,
I want to work on AI research. I don't want
to work on AI girlfriends, boyfriends, that that kind of thing.
Speaker 3 (18:24):
And as part of that.
Speaker 6 (18:26):
Deal, character Ai effectively became entirely employee owned and they
had a really strong balance sheet. They had a crazy
amount of users. They weren't monetizing maybe that well yet.
But I think a lot of the employees in that
situation where like this is pretty cool. We're basically running
a co op where we all own a lot of
this company.
Speaker 3 (18:43):
We don't have.
Speaker 6 (18:43):
Investors, we don't have the same kind of pressure to perform,
and that space is competitive, right, even Elon is competing
there now Chat GPT gets used as a companion, but.
Speaker 3 (18:55):
Not competitive like Cogen.
Speaker 6 (18:56):
And so that was the dynamic here that was insane
because Google clearly cares about code generation. They see anthropic adding.
They went from one to four billion dollars in run
rate this year. They're pacing to be somewhere around ten
at the end of the year, and so that's a
Google size market, right. Google is going to ultimately care
about Cogen, So it made sense for them to say, hey,
(19:18):
let's get you know, fifty more hyper talented engineers. This
is winter we're talking about. The issue though, is if
you didn't get brought over to the Google ship, you
were on what I what I was calling a ghost ship.
Speaker 3 (19:31):
John.
Speaker 5 (19:31):
Everyone who was pro the strategy, they would call it
the remain co I was.
Speaker 6 (19:38):
But imagine you're at a company and the dynamic with
Windsurf was fascinating because the company, if you had joined
in August of last year Windsurf, you could have the
product hadn't launched it, so you could have worked and
worked up to the product launch launched it. Seeing this
meteoric growth, the last reported number they had with some
something like eighty million of air R. You have a
term sheet to get acquired from open Ai for three
(20:00):
billion dollars and that falls through, and then all these
employees are looking around and they're like, wait, I didn't
even hit my I haven't even hit my one year cliff.
Yet I don't actually have a right to. I mean,
they might technically own shares or options depending on how
it was structured, but they as I'm sure everyone here
knows you, oftentimes, like sometimes founders would would try to
(20:25):
accelerate their employees, get them compensated as part of a
transaction like that, but there's not necessarily a contractual right
to do that, and it's part of a negotiation. And
so in this case, you have this team that's effectively
split up. And we were covering the whole thing live
because we were hearing that. You know, employees that Friday
were like crying, and there was a ton of confusion
(20:45):
and chaos and everybody was learning facts kind of over
that forty eight hour period. Luckily, our friend Scott Wu
at Cognition, the company that ultimately bought the remain co,
flew to meet the Windsurf team, the new CEO Windsurf
and Basic. They spent the weekend doing this insane deal
and it ended up being a great outcome, I think
for everybody.
Speaker 4 (21:05):
Didn't involve Windsurf have their like marketing department in the
room or something. When they made the announcement, the initial announcement,
and everyone was like because they were expecting to be
bought by open Ai, right, They're like, we're going to
film this for posterity, and then it turns out to
be something.
Speaker 5 (21:21):
Yeah. Yeah, I believe the timeline is that, you know,
Windsurf launches a year ago, is in a knockout, drag
out fight with Cursor. Cursor's doing very well. They're also
going to eighty million or so opening.
Speaker 3 (21:34):
Cursors in the hundreds of millions of revenue.
Speaker 6 (21:37):
Yeah, Windsurf was very clearly the number two player in
the AI id E mark.
Speaker 5 (21:41):
Yeah, and so Cursor is staying independent, but open Ai
wants to continue to get a foothold in this space,
so they make an offer that falls through. The rumor
was because Microsoft would have had ip look through exposure
via the more complex open ai structure.
Speaker 3 (22:00):
Which is ongoing, which is continuous.
Speaker 5 (22:02):
Be ongoing, and then when when Google came through, they
said that they wanted just to buy the team leave
the remain co because it's potentially cleaner from an FTC perspective.
But there's a whole bunch of different reasons and no
one really comments on exactly what happens. But when these
deals happen. This just happened with SCALEAI and Meta. The CEO,
even though there's some sort of amorphous FTC risk, the
(22:26):
CEO if they're going through one of these zombie acquisitions,
does have the ability to kind of set the team
up with certain expectations and say, hey, we're going to
take care of you. Trust me. You're going to get
a check that you would expect based on your ownership.
So if you own point zero one percent of the company,
you would expect that you get x of the headline number,
and yes, it's coming to you. The weird thing about
(22:49):
the Windsurf deal was that that was not messaged and
so there was like this z the zombie ship was
more zombie five.
Speaker 6 (22:57):
So like yeah, and and imagine, imagine you're a you're
an hyper competitive market where you're competing with Google and
Anthropic and Cursor and all these different players, and then
you lose your CEO and your top fifty engineers, and
they're like, and you guys are gonna do against don't worry.
You guys have a strong balance sheet. You guys are
going to do great. And we're also gonna be We're
(23:17):
also gonna be competing with you at at at at Google,
But you guys are gonna be fod luck guys and
so and.
Speaker 2 (23:23):
Thanks, thanks for the help. But this break but like, okay,
cognition came in. It sounds like all the employees will
get something. It's something in the world because there was
cash on the balance but it didn't have to be
that outcome. And so there's two things here that strike me.
One is, again, this does not seem like the B
to B SaaS era, where it's like, if you created
(23:44):
the hottest building product for dentist offices and that was
great gathering traction, no one would like just buy the
talent that built that. So that's really different, right, because
so the the ability to take value out via the
talent channel rather than buying the product itself. But then
also and I'm you know, the knock on effects. Okay, fine,
the Windsurf employees did fine, but going forward, there's no
(24:05):
guarantee that they could have. And so I'm wondering, from
either a VC perspective or a future employees perspective, how
this is going to change the sort of calculation that
anyone makes when participating in a new AI startup, the
fact that the enterprise may not be where the value
actually is.
Speaker 5 (24:20):
So the first question is, like, this kind of is
the example you gave of, like buying the team that
built the dentist B to B SaaS because Windsurf does
not train foundation models and Google is exceptional in training
foundation models. With Gemini and deep Mind, the deep Mind
team is extremely well staffed on AI researchers and continually
(24:42):
seems to push the frontier both the qualitative and quantitative metrics.
And if they were weak anywhere, it was maybe products exactly.
And so this is the like, I don't want to
be like they're just product people. Sure, Windsurf has some
amazing AI researchers, some amazing AI engineers, but what Windsurf
really did. They did not train a frontier model that
was about to disrupt Gemini go deep minds foundation model
(25:07):
they were going after. You know, would you use a
Google product, No, you would use Windsurf on top of
anthropy person or another foundation model.
Speaker 3 (25:15):
And ultimately don't.
Speaker 6 (25:17):
I don't think it's a systemic risk to the social
contract of our industry. And the reason for that is
that when you see a deal like the Google Windsurf
deal get done, it was very concerning. A lot of
people were extremely angry. It ended up being a good outcome.
A lot of employees were looking around, probably concerned about
(25:37):
you know, is the million dollars of stock that I've
been working for for years worth anything at all? I
think that was a good question to ask. But I
think there's two things. One is is AI is a category,
and of course it's touching kind of every category of
venture in the private markets. But we're not seeing You're
not seeing a defense tech founding team or engineering group
(25:58):
get There's no real.
Speaker 3 (25:59):
Act hire value there.
Speaker 6 (26:01):
The acquihires that are getting done in hard tech are hey,
you clearly have good engineering capabilities. We're happy to have
you join the team, but there's no real premium being
placed on that kind of talent. Might be able to
get a great comp packages, but they're certainly not getting
these sort of multi hundred million dollar premiums. The other
factor here is, like we have a set like right now,
(26:23):
if you are one of the top one hundred AI researchers,
you could probably get one hundred million dollar comp package
like within a week if you really wanted it, right,
And there's even people that are at Thinking you know,
the reporting from this week was that and it was
kind of hotly debated, but that people at Thinking Machines,
Miramrati's company, former CTO of Open AI, we're turning down
these sort of one hundred million, multi hundred million dollar
(26:45):
There was a rumor that somebody had turned down a
billion dollar five year contract. And I don't believe that
those deals will be getting done in three years now.
I might be wrong, and there's going to be like
probably some exceptions, but the idea that the thirtieth AI
researcher on your team is going to make more than
a mags Heaven CEO, like that doesn't feel hyper.
Speaker 3 (27:09):
Something has to change either.
Speaker 5 (27:10):
Tim Cook has to make more money.
Speaker 3 (27:13):
Tim Cook is looking.
Speaker 2 (27:15):
For assistances on Wall Street were like a top trader
to consistently make more than or will often make more
than the CEO, absolutely because.
Speaker 4 (27:22):
They're the ones that got the money in the door. Yeah,
can you talk a little bit more about I guess
the fungibility of the IP here in both the practical
(27:46):
and legal sense. So if I'm a VC or some
sort of investor, I invest in an AI company because
I get to own that technology, that intellectual property. But
then let's say the main guy walks out the door,
gets hired by Google or whoever, Like, how much of
what he learned at his original firm or developed is
(28:06):
immediately going to be replicated at Google?
Speaker 3 (28:08):
I bite almost all of it.
Speaker 5 (28:10):
Yeah, I guess this is like, well, that's exactly why
they're paying that much money.
Speaker 4 (28:13):
This is a lot more way of saying how many
lawsuits are we going to get after all these aquhiers?
Speaker 6 (28:19):
So I think one way you can look at some
of Zuck's deals are like unauthorized aquhiers. It's like the
other CEO is not even participating in the deal. But
you know, Zuck is able to come in and be like,
would I pay if this, if these ten people were
working on a company independently, would I pay a billion
dollars to bring them onto my team? One hundred percent? Okay,
let's see the deal. It doesn't matter that you're kind
(28:39):
of piecing it all up from a. I think going
back to like the kind of Wall Street example of
like a top trader who maybe is like putting up
consistently incredible returns with some sort of like differentiated approach,
that person can go and as long as they have
capital at the new company, they're going to be able
to imagine to continue to just follow that same type
of strategy. I do think that AI and large language
(29:02):
models are somewhat different in that you have to look
at what is the AI product that these companies are
going to sell. How much of an impact is that
individual engineer researcher are going to have. And you know,
it's fairly obvious with open Ai their consumer tech company,
right they sell subscriptions to chat GBT, they have some
other use cases. It's obvious that companies like Anthropic, where
(29:23):
they're in the cogeneration business, they don't really have a
consumer business. And at Meta it's a little bit less
clear right now because the sort of ongoing product strategy
is not entirely clear yet. The thing that is obvious
is that if you can make a twenty billion dollars
training around more efficient than you pay for yourself, pretty quickly,
we should.
Speaker 4 (29:42):
Just point out the Wall Street analogy is not perfect, right,
because if you're a high flyer at a Wall Street firm,
either client facing or if you're trading, if you join
someone else, you usually have a non compete agreement in
one form or another. You can't take all your existing
clients with you and then be you're often on an
enforced garden for a really long time. Yeah, I imagine
(30:02):
in the massive race that is AI at the moment,
there's no gardening.
Speaker 5 (30:06):
Also, the nature of the intellectual property I believe is
a little bit different still. I think of this. I
think it's a twenty twelve example from Citadel in Chicago
where a hygh ferquency trader stole some code and they
sent and he had it on his hard drive. You
know this story, Yes, it was really.
Speaker 4 (30:24):
Famous when it happened. He like sent it to himself
for yeah.
Speaker 5 (30:27):
Yeah, So he expltraded some code, some very you know,
definitive code of how to make money in the market
and get an edge, and they he figured out that
they were on his trail, and he threw his hard
drive into the river and they sent scuba divers into
the river and got it back.
Speaker 3 (30:42):
Here.
Speaker 6 (30:43):
The other high profile SV example was the guy going
from Google self driving to yes.
Speaker 5 (30:50):
Yeah and so and so taking specific code.
Speaker 3 (30:53):
That's not happening.
Speaker 5 (30:53):
I mean, there's probably going to be like one example
of this that pops up, but mostly it's just you
get someone who says, I understand that at my previous job,
we scaled up the transformer a little too far and
we didn't focus on reasoning enough, and so we need
to shift this spend to test time inference. And it's
almost like you're leaving you're leaving a company and you're
(31:14):
and you're just you're you're leaving with like a mindset
of like the right balance of capex toopics or something.
Speaker 6 (31:19):
What was the recent Jane Street example that came out
in that lawsuit around it?
Speaker 3 (31:23):
Was it the strategy?
Speaker 4 (31:25):
Yeah?
Speaker 3 (31:26):
It only we we only really all.
Speaker 6 (31:27):
Learned what was happening because there was this lawsuit over
over basically the team bringing over some type of IP
or methodology.
Speaker 5 (31:37):
Yeah, can we talk.
Speaker 2 (31:38):
About later the broader sort of economic world right now,
because there's a bunch of other sort of things going
on and things that you talk about.
Speaker 6 (31:46):
Touch on them last thing, because I think it's pretty interesting.
So we put out the metaslist I think it was Monday,
and we immediately had a ton of inbound from a
lot of these researchers basically like critiquing like the ranking right,
and the ranking was like we got we got people that.
Speaker 3 (32:02):
Work in AI to kind of like rank them. We
the beauty of ranks? Do you get so much the beauty?
Speaker 6 (32:11):
And there's one uh, there was one example that was
interesting where somebody that was ranked fairly high that I
know has gotten one of these nine figure comp packages
and somebody that worked with him basically said this person
was kicked off of every team they worked on and
just like effectively like over multi year period, like consistently.
Speaker 3 (32:31):
Demoted over and over and over.
Speaker 6 (32:34):
And now now like got you know this incredible.
Speaker 4 (32:37):
And soft skills in ranking?
Speaker 3 (32:40):
Well yeah, and it was it was more of like
a technical ability thing.
Speaker 6 (32:44):
It was not even because I think skills are getting
a little bit uh you know, they don't they don't.
You don't care that Lebron James like might get a
little angry at somebody if they underperform, right, No.
Speaker 2 (32:56):
No, no, no, no, there's so this phenomenon of like superstars,
like it's not just it's not just in tech and
like we see it. I mentioned in journalism, where you know,
the sort of like median reporter job in a newsroom.
A lot of that's hollowing out, but you have some
people who become insanely well compensated, either because they're really
smart and they write a great newsletter, or they're really
(33:17):
good looking and they can do front facing video, et cetera,
or you know, like the four of us, you know,
like can do something on video or something like that.
You see it in the range of areas. And then
like the amount of betting that's going on, which of
course adds to this, and so the fact that all
of these events there's some market out there that you
can bet on that. And I'm curious, like from your coast,
(33:39):
you know, you're here in New York, but from your coast,
what does the world look like in terms of just
like the state of this economy.
Speaker 5 (33:45):
Yeah, there's something interesting going on in tech where the
idea of a company going from one hundred billion to
a trillion seemed unfathomable. Yeah, and there's this take that
in fact, that was the easiest ten X of all,
where the hardest ten X was going from zero to one.
From going to zero to a million dollars and getting
these systems, inventing PageRank, and then once you had Google
(34:07):
humming at one hundred billion dollar market cap getting to
a trillion, not to you knock of the work that
they did to get that, but these the numbers have
just kept growing and growing at Internet scale, at Internet speed,
and so the leverage that you're getting from an AI
researcher is ever more increasing, and it just adds ten
x every couple of years. And so ye seeing the
(34:29):
compackages and increase that way, have.
Speaker 6 (34:31):
A couple of things like the Internet is the greatest
distribution engine for information and digital products and apps and
services ever and so at least in venture, that just
means that everything is faster now.
Speaker 3 (34:43):
Even in media too.
Speaker 6 (34:44):
If you're if you are somebody's working at a at
a legacy media company and they set up a sub stack,
they can get a million dollars of ARR on the
first day that they launch. If somebody is working at
a media company and really good at a certain type
of YouTube video, they can launch and immediately the YouTube
algorithm will serve that video to all of their fans
(35:06):
within the first week. And that what that does is
it just changes. You know, the power dynamic between media
companies you see is.
Speaker 5 (35:13):
Leverage come in through financial markets. If you can lever
up or just marshal more capital one idea can generate
billions of dollars in value. Same thing in technology, but
we aren't seeing it everywhere. I don't think we're seeing
it in hard skills would working like unless you get
into true like art territory. But there is this interesting
question about like what's up next, and we were kind
(35:34):
of noodling on like there might be a law firm
in the future that's extremely high leverage in the same
sense that you might have a lawyer who's so good
at what they do and they are so good at
resolving these and the and the connections and everything that
they're doing. They're making billions, but they have a very
very lean team. And so you see a much a
(35:55):
much steeper power law in law. And there's a there's
a number of other professional services that are going through
like a transformation with technology bringing increased leverage to the
profession and that could drive more of that power li
outcome in terms of earning.
Speaker 6 (36:11):
Yeah, the thing you know in the private markets on
the West Coast, which is the dynamic right now that's
fascinating is like, I feel like a lot of people
have like a little bit of PTSD from the twenty
twenty to twenty twenty two era, where many of the
things that in hindsight were incredible top signals are like
all pop popping up again right now, and like we like,
(36:31):
we like trap, we track them just for fun. We
are not in the business of like calling the top
or anything like that. And in many ways it feels like,
you know, there's so many positive indicators. But the interesting
dynamic and in venture is like, as an angel investor,
I've invested in probably sixty five or so different companies
at the pre seed to Series A stage, and there
(36:53):
was a bunch of deals that I did in twenty
twenty one twenty twenty two that ultimately like I paid
too much or just like the team wasn't good enough
to execute against the vision they had. But there was
also just like a handful of deals that I did
that that have been you know, performed so well that
it doesn't matter that I did a bunch of silly deals.
And so venture right now is this interesting kind of
(37:13):
dynamic where everybody knows that it's crazy, Right, you have
hundreds of billions of dollars of value created in the
private markets that there's a lot of real revenue growth,
but there's also hundreds of billions of dollars of value
tied to zero revenue, right. And I think that there's
something that we've been tracking is like this underlying kind
of feeling from people that are maybe under thirty of
(37:34):
there's like this meme that's become very prevalent, and I
think it explains a lot of economic activity today, which
is young people feel that they have two years to
accumulate capital to escape the permanent underclass. And so I
think that drives a lot of investing activity today in
that you know, you'll see young people on the timeline
(37:55):
saying that like you'd be stupid not to use leverage,
you know, and they're and they're just like, you know,
they're not a professional.
Speaker 3 (38:01):
Investor in any capacity. Don't do it.
Speaker 6 (38:05):
And you know, any anytime you have people in venture
making public market stock predictions, we get a little concerned.
Speaker 4 (38:13):
Are we going to get AI talent agents? Not AI agents,
We have those already, but AI talent agents because like.
Speaker 5 (38:19):
We do so we do they're called venture capitalists.
Speaker 4 (38:22):
Oh well, but yes, this was going to be a sentence.
So if the money is no longer in, you know,
I invest in a startup and eventually they get the exit,
they get acquired, or they list or whatever. If the
money is in, eventually the guy that started the startup
gets bought by someone like Meta or whoever. Wouldn't I
invest my money in that person versus in the startup
(38:45):
inventure service?
Speaker 5 (38:46):
Yes, yeah, you can't quite do that, but there are
a ton of roles that are indexed to these high
performance packages. And yes, there are people right now in
Silicon Valley who are effectively talent agents who get a
cut of those big packages and they help me get
siate just.
Speaker 3 (39:01):
Called just talent, called the venture capitalists. No, no, so so
you can.
Speaker 5 (39:05):
You can you find a great researcher and you think that, okay,
maybe they will build a business, but I'm sure that
they are going to be worth hundreds of millions of
dollars a year, and you can just invest in their company.
You will almost certainly see a good return on that investment,
even if the product never gets to scale, because when
they get an ACQUA higher, you will get a.
Speaker 2 (39:25):
Payout, not going there unlike the.
Speaker 6 (39:33):
Are like, oh absolutely no, it's now it's now so
normalized to invest in college dropouts that you actually see
people like investing in high school.
Speaker 5 (39:43):
I'm not kidding about this. So the Iomo Gold Medal
is the math Olympiad, and there are venture capitalists who
will give calls to every single student that performs well
on them on the math Olympiad. And these are high
school students.
Speaker 4 (39:54):
So do they what are these investments? Exactly like, are
you're funding their tuition or no?
Speaker 5 (39:59):
No, no, it's it's it's I am taking ten or twenty
percent of a company of a Delaware c core. Most
likely that you will build something in and who knows
where it goes. Maybe it turns into a great business,
maybe it turns into aquhire. But the downside is extremely
limited because there's always this at least right now, there's
this aqui hire, there's this AQUHIRER on the table where
(40:22):
it used to be.
Speaker 6 (40:23):
That to be clear that it is only an AI
yes only for the people that you would consider to
be the top five hundred yes in there industry.
Speaker 5 (40:32):
So to go back to look at Cursor windsurf thing
like Instagram was the power law winner got the billion
dollar acquisition from Meta Hipstomatic was the second largest photo
filtering app, did not get a billion dollar aquhier from Google. Right,
But now we're in the market where if there's a
leading product with a bunch of AI researchers over here,
and they get a multi billion dollar acquisition for the product,
(40:54):
and the product's working and it's growing and it is
a great business, and you buy it for the value
of the business, then the second best team might get
acquired just for talent, which is a completely different downside protection.
Speaker 6 (41:04):
Yeah, and there's a lot of talent acquisitions traditional acquires
where it's only the talent that benefit, right in the
sense that they get basically a job offer at the
new company, and vcs get some capital back or in
some cases a small amount of money.
Speaker 2 (41:19):
Can I ask a TVPN question This should be like
the ultimate compliment, which is that I got a d
M from some random person I've never seen them the
other day and he said I can build tvpns for X,
So like that's that's deck. That's sort of which means
it's sort of like it's become like Kleenex or one
(41:39):
of these things where it's like a category and where
are you going with it? What are your playing?
Speaker 6 (41:44):
It's so funny that people are calling it TVPN for
X because our show, while it's unique in a variety
of ways, looks very much like traditional is interesting, and so.
Speaker 3 (41:56):
We can't we can't take credit for we held it.
Speaker 5 (42:00):
We invented media, we invented live streaming.
Speaker 2 (42:03):
Of course, course normally it doesn't really I don't know,
why would I watch cable news on.
Speaker 6 (42:09):
I think thing that's exciting is that in the private
markets and venture and tech, if you had a podcast,
there was one format which was a once a week
interview show, and it was a great strategy to do
that ten years ago, around the time that you guys.
Speaker 3 (42:26):
It locked established.
Speaker 5 (42:30):
But if we if we tried to clone this, yeah,
everyone would be like, why is that a knockoff? There's
nothing new about this, nothing fresh about this, Like why
do I want to go on your knockoff? I just
go on the real thing.
Speaker 3 (42:40):
Yeah.
Speaker 6 (42:40):
And so our edge in launching the show at the
beginning of the year is that media is not zero sum.
Speaker 3 (42:48):
Content is not zero s.
Speaker 6 (42:49):
We have a friend that jokes that he's so competitive
he wishes he wishes media zero sum it was truly
zero s. But our edge early was that we just
took it ten times more seriously than anyone else. So
everybody that was creating content for the private markets was
doing it as a part time gig, and we were
happy to compete with a bunch of people that were
part time.
Speaker 2 (43:10):
You know what, Actually, I was wondering in twenty twenty
one or twenty twenty two that craziness. Someone once reached
out to me and they're like, you know what, you
should like, no direct, it's probably the no, but this
was this was the interesting thing. He said, you should
go independent and what you should do is attach a
(43:31):
VC armed to it, and because of the quality of
the guests that you could get, you would get really
good access to deal flow. But it seems like basically
it's very uncomfortable to me because I'm not going to
highlight startup founders and say, oh, you're the perfect guest.
And it's like because like I have like five percent.
But I'm curious about because you mentioned doing Angel the
(43:51):
sort of link of the TVPN at some point, like, yeah,
so an investing.
Speaker 6 (43:57):
Platform Angel investing. I look at it as a hobby.
It's not a great financial activity, right, locking up, I've
heard locking out. I mean, like you can generate great returns,
but it's not the most logical way to invest your time.
But it's really fun. Like we just enjoy supporting founders
early when it's an idea and a team and it's
just it is genuinely addicting. I always joke about people
(44:20):
in San Francisco with Angel. You know, people across the country,
sports betting addictions san Francisco, you know, Angel investing addictions
are I think real. But for us, for us, that was,
you know, as we started having some success a lot
of people, they would probably get a message today, when's
the fun coming, when's the fund coming? And that's been
(44:40):
a way to monetize an audience within tech. If you
have an audience, go raise one hundred million dollar fund,
you get the fee streams do that, you get a
bunch of upside. And we joke because we have the shows.
Every single day we go live at eleven. We have
to eat, we have to work out, we have to
prep the show, we have to talk with partners, we
(45:02):
have to manage our team. There's all these different things
and so somebody might immediately think, Okay, these guys talk
to six founders and investors a day. They have this
like media property that like, I think everybody in ventured
now is going to see our content like once a
week in some form or another if they're if they're online.
But I joke with them, I'm like, so we're live
(45:25):
for three hours every single day during the middle of
the day. And so if you think that we with
a show would would effectively compete at VC is about
winning allocation, right. You can be cool and connected and
have an audience that might get you one hundred K allocation,
but if you have one hundred million dollar fund, you
need to be putting size into deals. And so I
(45:47):
know that if John and Jordi have a VC fund
and we're competing with these other guys that have a.
Speaker 3 (45:51):
Live show that they're live for three hours a day
and they have.
Speaker 6 (45:53):
A VC fund, we'd smoke them because we'd be like, Okay,
well they're live. I'm going to fly to the founder,
I'm gonna meet with them, and I'm we're gonna win
this deal, and we're gonna be like, yeah, why why
don't you let them put in like two hundred k.
Speaker 4 (46:03):
So the thing to watch out for when the VC
fund is coming, is when you start reducing your live hours,
that'll be the same.
Speaker 1 (46:11):
Yeah.
Speaker 6 (46:11):
I think I think we we didn't get into this
to start a fund, and I think there's a lot
of people that get into content because they see it
as a way to do that and we just love
talking about tech.
Speaker 5 (46:21):
It actually was the key insight was that not enough people.
Yeah it's very fun, but not but very few people
in tech were actually taking media seriously. Yeah, it was
everyone had a fund, everyone that was the high status thing,
and doing media was like lower status or people didn't
think we could have as much of a power all
outcome as it very clearly can. And so this idea
(46:41):
of of just what if you actually took it completely
seriously and just made it the main.
Speaker 6 (46:47):
Thing, and when you were We've had a bunch of
people copy our format, you know, major legacy media companies
all the way through friends of ours. It doesn't really
bother I mean it bothers me more than John. But
at the end of the day, if you want to
spend the tracy, yeah, I totally got it. And so
(47:07):
it's like it doesn't basically know so so John and
I John and I basically hang out for twelve hours
a day, and the entire time, the entire time we're thinking,
we're thinking about the show, we're just talking about the
show doesn't look very different than when we're hanging out offline.
And so I just joke, I'm like, Okay, if you
want to compete with us and you're willing to put
(47:27):
it one hundred hours a week in god by all means,
go for it like this probably your life's work.
Speaker 3 (47:33):
But if it's not good luck, it doesn't matter.
Speaker 2 (47:36):
John Cougan, Jordie Hayes, thank you so much for coming
on Odd Loss that's in good luck and looking forward
to continue watching TVPM.
Speaker 7 (47:45):
We'll be live from Nicey later Odd Lots in the
same day and in the wake of meadow winch mech
me a.
Speaker 3 (47:58):
Thank you for having us. Thank you so much, Tracy.
That was a lot of fun.
Speaker 4 (48:14):
It was a little bit media naval gazing, a little
bit fun.
Speaker 2 (48:17):
A little bit media naval gazing. I mean, there is
this thing that's happening. It's been happening in media for
a while, and of course it happens in Wall Street
and now happening in AI where you just have a
lot of talented people, and they wonder about the degree
to which they need their existing platform there. You know,
(48:38):
like a star banker can take a book of business,
or a star lawyer can take a book of business.
And in the case of a it's not taking a
book of business, right because they don't have like their
individual clients. But it's this knowledge and transport it and
it's instantly worth a lot of money for someone somewhere else.
Speaker 4 (48:54):
The thing I thought was really interesting about that discussion
was the emphasis on how capital intent all of AI
is and so that kind of changes the economics of
why you're paying such a massive money for someone who's
able to eke out even a slight efficiency.
Speaker 3 (49:10):
This is huge.
Speaker 2 (49:13):
This was like this It suddenly is what made it
all made sense to me, right, because we know about
like how costly one training run is, right, and we
know just the insane numbers for data center set up,
et cetera. And I'm sure there's progress being made on
the literal design of like how you string together in video, GPS,
(49:33):
et cetera. And so if you could get some margin,
if you know have that know how to get some
marginal improvement out of it, And this is fundamentally what was.
Speaker 4 (49:40):
One trillion dollars for you?
Speaker 2 (49:42):
Maybe not a trill No, not a trilliont but like
this was not the case when tech was not so
capital intensive in the twenty tens. Where yeah, I'm sure
talented people always made a lot of money and there's
always improvements, but where it's so that link between some
sort of efficiency gain and instant cost savings is so
linear and so straightforward. Yeah.
Speaker 4 (50:04):
And I take their point that, Okay, vcs exist and
they're already investing in some ways and specific talent. But
I do kind of wonder if you're going to get
some sort of like specialized headhunters at the very least,
who are going to like seek out these big AI
talents and try to like graft themselves onto them.
Speaker 2 (50:19):
Yeah, just people whose expertise is in reading through under cited,
under cited AI research papers.
Speaker 3 (50:29):
Yeah, AI research.
Speaker 2 (50:30):
Well, we can't give you Sam Altman, but what if
we could replace Sam Altman in the egg.
Speaker 4 (50:34):
Guy has fifty citations in the following papers? Should we
leave it there?
Speaker 3 (50:38):
Let's leave it there.
Speaker 4 (50:39):
This has been another episode of the All Thoughts podcast.
I'm Tracy Alloway. You can follow me at Tracy Alloway and.
Speaker 2 (50:44):
I'm Jill Wisenthal. You can follow me at the Stalwart,
follow our guest Jordi Hayes He's at Jordi Hayes, and
John Cougan at John Coogan, and check out TBPN at TBPN.
Follow our producers Carmen Rodriguez at Carmen Arman, dash Ol
Bennett at dashbod and kel Brooks at Kale Brooks. And
from our Odlots content good to Bloomberg dot com slash
odd lots were the daily newsletter and all of our episodes,
(51:06):
and you can chat about all of these topics twenty
four to seven in our discord Discord dot gg slash
od loots.
Speaker 4 (51:12):
And if you enjoy odd Lots, if you like it
when we talk about the sportification of AI talent, then
please leave us a positive review on your favorite podcast platform.
And remember, if you are a Bloomberg subscriber, you can
listen to all of our episodes absolutely ad free. All
you need to do is find the Bloomberg channel on
Apple Podcasts and follow the instructions there. Thanks for listening.
(52:04):
H
Bloomberg Audio Studios, Podcasts, radio News.
Speaker 2 (00:18):
Hello and welcome to another episode of The Odd Laws podcast.
Speaker 3 (00:22):
I'm Joe Wisenthal and I'm Tracy Alloway.
Speaker 2 (00:24):
Tracy, I love doing the podcast, but I'm kind of
thinking about becoming a PhD level AI researcher if I
do a career pivot.
Speaker 3 (00:33):
It seems something. Have you seen?
Speaker 2 (00:37):
Have you seen like the salaries supposedly I still don't
actually believe them, But have you seen like the salaries
and comp packages that supposedly Meta and a few others
are paying for it.
Speaker 4 (00:47):
I have seen some of the headlines. I have also
learned a bunch of new words like exploding offer and
acquahireh exploding offer sounds dangerous. What you want is an
explosive offer that you can't turn down.
Speaker 2 (01:01):
I take it the exploding hire is like one that
lasts like five minutes.
Speaker 4 (01:04):
Yeah, so you have to make a decision right away,
and the company that you're currently working for can't counteroffer.
Speaker 2 (01:09):
So I read these headlines about some engineer something getting
paid one hundred million or two hundred and fifty million.
I actually literally don't believe them, Like I actually literally
think that's fake notes. Really, no but I.
Speaker 4 (01:19):
Kind of just think it's probably made up of like
all this weird equity structure.
Speaker 2 (01:23):
But I'm sure it's not all upfront anyway. We sort
of seem to be in a moment and it's actually
not just AI. I would say journalism, finance, et cetera,
where it's like the sportsification of a lot of different industries,
individual talent, the demand for individual superstars, and anything doing
very well.
Speaker 4 (01:44):
I have a bunch of questions when it comes specifically
to AI, such as what makes an AI talent because
I assume if you're Mark Zuckerberg and you're trying to
assemble your AI dream team or something, Okay, sure you
have like a level of technical insight, and maybe you
hear people talk talking about, oh, this one guy is fantastic,
But I assume it's not like you can look at
(02:04):
their like individual levels of code. You can't see what
they're doing on like a daily basis. I don't know.
Speaker 2 (02:11):
I don't know either. I don't know anything about this stuff,
but I'm really excited to say we do have the
perfect guests, A couple of guys who are right in
the middle of all of this and also who have
sort of been ahead of the curve in terms of this,
like I said, the sportification of the industry. We're going
to be speaking with John Coogan and Jordie Hayes. They
are the co hosts of TBPN. It's a live show podcast.
(02:33):
It's become one of my favorite new media properties. It exists,
and I'm really excited we have them both in the
studio here with us today. So John and Jordi, thank
you so much for coming on aut Loots.
Speaker 5 (02:45):
Thanks for having us.
Speaker 3 (02:46):
We're so excited to be here.
Speaker 2 (02:47):
I love by the way you guys like make these
like baseball cards every time, So it's so clever for
those who are not paying attention. Why do you sort
of give the high level overview of like what you
see happening in this like crazy talent for people who
don't know about this, what is actually going on? This
talent war for people who know how to do AI
or train a model or whatever it is.
Speaker 5 (03:09):
Yeah, the League or the mag seven the serious teams
are all worth over a trillion dollars. Tesla's in a
different kind of boat some days, but pretty much there's
multiply seven trillion plus dollar companies now, so there is
an inordinate amount of money to flow around. And if
you think about investing zero point one percent of your
market cap to make your business potentially five percent better,
(03:33):
that's a trade you take all day. The number is staggering,
but now you're seeing companies pay hundreds of millions of dollars.
I think you mentioned that. You know, there's a lot
of debate over whether that was fake news or not.
They seem like they are very real offers and they
are indeed happening, and there's a whole bunch of different
ways that you can kind of underwrite that. But we
could start with going through just a little bit of
(03:55):
the history here of how we got into this place
or what these AI researchers are actually doing. I'm happy
to kind of answer any question.
Speaker 4 (04:03):
Okay, well, why don't we start with the baseball analogy?
Then if I was collecting AI engineer cards, who's the
most valuable? Like who do I actually want to have?
And then secondly, what are the stats that are.
Speaker 6 (04:17):
So taking a little bit of a step back talking
about kind of the sportsification of tech and business, which
has been a huge catalyst for our show. I think
we realized early on people would call TVPN the Sports
Center for Tech or some analogy like that. We were
a little bit it sounded cool. We didn't really know
what that meant. John and I don't watch sports at all,
(04:37):
but for basically, for basically like two decades now, I'm
in my late twenties, John mid thirties, and we've followed
tech in business the way that our college friends follow sports.
So in sports you have players, personalities, coaches, managers, leagues, teams,
and people obsess over all the details. Fully understood that,
(05:01):
I mean, I understood kind of the draw, but it
just was never for me. Whereas John and I would
pick up the newspaper as teenagers and we were tracking
the talent, the CEOs, the companies, the markets, the industries,
and it's just something that we obsessed over. And so
this year has been amazing. As these AI researchers have
(05:21):
been getting these sort of superstar max contracts, which is
what we would call them, we would start joking on
the show and be like, look, this guy just went.
Speaker 3 (05:29):
Over to Meta.
Speaker 6 (05:29):
It's probably you know, four year contracts, you know, one year, Cliff.
And it just got more and more and more real
as the kind of demand for world class AI researchers
completely outstripped the supply, and we actually put out something
a couple of days ago called the Metas List, and
John can go a little bit more into into the
name behind that. It was kind of our take on
(05:51):
the Midas List, which is, we built a list of
top roughly one hundred AI researchers and rank them on
a bunch of different factors. And so when you talk
about what goes into what makes a great AI researcher,
there's a bunch of different factors. One that you can
get right into the numbers with is like citations. So
if somebody is a researcher, they've been publishing, you know, studies, papers,
(06:12):
et cetera for quite a while at this point, and
you can just see quantitatively what their contributions have been
to the industry.
Speaker 3 (06:19):
And so that's like a good starting point to understand.
Speaker 6 (06:21):
And historically Elon even said earlier this week, honestly, it
was like an hour after we posted the Metaalist.
Speaker 3 (06:28):
Probably unrelated, but who knows.
Speaker 6 (06:30):
He's basically saying that AI researchers a and engineers, We're
just going to call them engineers now, and that makes
a lot of sense for someone like Elon to do
as somebody who's always been very engineering focused, less focused
on you know, entirely net new innovation. More so, how
do we make the best possible versions of products that exists.
(06:52):
And it's not to say that he hasn't innovated in
a bunch of different ways, but it is a really
wild moment in time, and it's been a fun moment
because historically you hear about this minas list investor, you know,
making two billion dollars of carry on some deal, or
this founder you know I p O ing. Today we
have you know, the Pigma I p O and uh,
(07:13):
you know there's a bunch of people that are going
to make billions of dollars there, But you don't hear
about the hundredth engineer that was hired, yeah, making you
start a one hundred million dollars signing bonus. And all
of this makes a ton of sense in the context
of what John said, because you look metas up I
checked this morning one hundred and ninety five billion dollars
(07:34):
new market cap and think about how many one hundred
million dollars signing bonuses you can make against that kind
of market cap.
Speaker 3 (07:41):
Should we answer your question, yes, who the who's the bus?
Speaker 5 (07:44):
Who's the whale? To have the who is the light whale?
Speaker 4 (07:51):
Don't say anything about whales or whale products or hunting.
Speaker 2 (07:56):
Large is the way.
Speaker 5 (08:00):
Perhaps he is the Michael Jordan of AI researcher, but
Ilias Sitskiver is really He's at the top of our
list for a variety of reasons. And if you're not
familiar with Ilias Sutzkiver, he is an AI researcher who
was at open AI for a long time, and there's
a few different ways to characterize him. He is both
coming up with new ways to implement AI algorithms the
(08:23):
way you train the model, but he's also very good
at for a long time identifying which the shortest path
in the tech tree. So there are branches of choices
that you need to make as you develop the new
AI models, and he was very early at while he
was at open AI, he identified that the transformer paper
(08:45):
from Google. He didn't invent that it was at Google,
but the transformer technology was extremely important and that it
had the ability to do remarkable things once scaled up massively,
and so he was the driving force between kind of
identifying the transformers as the correct path. Now we look
back on Attention as All you Need, which is the
name of the paper that defines the architecture that is
(09:08):
used in these modern large language models that you use
when you're.
Speaker 3 (09:11):
In chat GPT.
Speaker 5 (09:12):
There were twenty five other potential paths that we could
have gone down. You know, the story goes is that
he really identified that and said, let's go really really
hard on that.
Speaker 4 (09:22):
So he can sort of identify the innovation, what's new,
and then also identify the most efficient path to actually
execute on.
Speaker 5 (09:30):
Yes, and if you're familiar with what the more the
more modern, So the first arc of LMS and these
chatbots scaling up where really this tack the transformer paper,
understand that that's the correct architecture, and then scale it
up really really big. So you need to be able
to not just write the code to implement that that
particular algorithm. You need to marshal the capital to say
(09:52):
we're going really really big and we're going to build
the big data center. We're going to spend a lot
of money, but it's going to be worth it because
we understand the trade offs here. Then the second kind
of innovation or correct call he had was during the
sam Altman, Ouster and return. He was working on a
project that was code named q Star, and there was
a lot of speculation about what q star was. Was
(10:13):
it the secret super intelligence thing?
Speaker 3 (10:15):
What did Ilius see?
Speaker 5 (10:16):
Yes, yeah, because he was going back and forth in
support of Sam Altman and then leaving and then and
then it was back and forth, and there's a lot
of drama there. Always keep them guessing, Always keep them
guessing for sure, and so there was there was a
lot of rumors, but what it wound up being was
just reasoning, which is now what is available in UH
if you use any of the three models in chat GPT,
(10:37):
or you use deep SEKR one and it's it goes
by a number of different names. The project was rebranded
Strawberry and then the O series, And you can think
about this as the test time. Inference is the other buzzword,
but basically the LM, you're using the same foundation model,
but you're running it a ton more to come up
with a bunch of potential answers to questions and then
(10:59):
narrowing that down and essentially applying what's called reinforcement learning
to the transformer architecture and the pre training that's happening,
and so he was very critical in leading that project,
and so that's allowed him to eventually, once he left
open ai, go start a new company called Safe Superintelligence SSI,
and he's gotten that company to what a thirty billion
(11:20):
dollar evaluation, thirty two billion dollar valuation.
Speaker 6 (11:22):
Which ties into the talent war because I think it's
I don't know that it's been officially announced yet, but
people are expecting Daniel Gross to join Meta, who.
Speaker 5 (11:33):
Was the CEO and co founder of that company. Side
and so back to your back to the white dynamic.
Speaker 3 (11:40):
Yeah, yeah, if you guys get together, they start.
Speaker 6 (11:42):
A company within a year, it's worth thirty two billion dollars,
and one of the people on that team decides to
actually leave and join Meta. It sounds insane, probably leaving
billions of dollars on the table, but you can kind
of trace the this talent war actually back to the
open ai founding team. When you think about all the
(12:02):
different players, right, you have Mira Murti who's now with
Thinking Machines, she was a CTO of open Ai. You
have Ilia Sutskiverer with SSI, you have Elon with XAI,
and not to mention, you have these sort of hyperscalers
who are also competing for the same talent. So yeah,
that's really kind of the origin story in the genesis,
And it's been amazing to see open AI's progress despite
(12:24):
you know, the recent reporting is around you know, losing
a bunch of top researchers, which is real, but it
wasn't that long ago that they lost like two very
very very key senior execs in Ilia and Mira.
Speaker 2 (12:53):
So a few stats were recording this. By the way,
July thirty first, Meta came out with earnings last night.
The stock is up like ten percent on stuff like
one hundred and fifty billion more dollars. The other thing is,
and I think this gets ties into the logic behind
the comp so they're going to spend like something like
another seventy billion on capac some stuff like that. So
we know these are incredibly computationally intensive things. They're electricity
(13:18):
intensive things. You mentioned paper citations, but also actually having
the experience of one of these runs, and it does
seem very highly intuitive to me that if you've done
it before, and if you could even cut down the
cost of a training run or set up a big
computer rack for you can pay five very quickly you
(13:39):
have five percent less, right, exactly happened. Instantly pay then
you instantly pay for your salary, I could say, because
and this is different from the B to B SaaS yer, right,
because you can there's these huge kpax costs if you
could just marginally improve the efficiency that that pays that
one hundred million sellar salary.
Speaker 5 (13:57):
And so this this literally just happened right before Mark
Zuckerberg went on the talent acquisition spray that he's been
on for the last few months. Meta's main AI model
is called Lama and they've released a series of versions
of that model, and Lama four was the latest and greatest,
and it didn't go very well. And the rumors about
(14:18):
why it kind of failed to deliver on expectations was
that they kind of went down the wrong path in
the tech tree and they focused a little bit too
much on scale insanely costly. So they've released a part
of Lama four, but they have yet to release Malama
for Behemoth, which is their biggest model, the most the
(14:40):
most aggressive, and they had just little details in the implementation,
little choices of how you chunk the attention in the
transformer model, you know, we're operating in this level of
abstraction up here talking about you know, it's a prediction model,
and then we talk about transformers a little bit. There
are sub algorithms within these systems that you make the
(15:00):
wrong decision and you could get a vastly different.
Speaker 2 (15:03):
Electricity bill goes up by one hundred.
Speaker 5 (15:04):
Many exactly, so all of that money that was spent
on that electricity and then build out of course that
it can absorb that. But when you think about the
cost of getting correct.
Speaker 6 (15:14):
One thing is is a lot of these projects will
be energy constrained too, So efficiency is going to matter
a lot. Oh, it hasn't been the core focus today.
If you talk to anybody at the big labs, they
are focused on maxing out intelligence at the cost of
efficiency because they know they can. You know, you want
to be on the bleeding edge. You want to have
the smartest model.
Speaker 3 (15:34):
You want to be.
Speaker 6 (15:35):
You know, there's debates around which benchmarks actually matter, how
important AI benchmarks really are. But energy is going to
you know, a lot of people are saying Zuck won't
even be able to spend as much as he wants
to spend on data center development purely because of the
energy constraint.
Speaker 4 (15:52):
Can you talk a little bit about what happened with Windsurf,
because this is this is the one that seems to
have like captured everyone's attention, in part because there was
the drama. It almost seemed like it came out of
like a Silicon Valley the show script or something where
all the employees were gathering expecting to hear that they
were going to be bought by Open Ai, and then
they find out that actually their CTO has just been
(16:14):
bought by someone else completely different, and they're sort of
left in the lart ceo.
Speaker 3 (16:18):
Oh is top CEO in top fifty?
Speaker 5 (16:21):
Well, how about I give some prehistory on acquires and
you can take take us through that actual weekend. So
there has been a trend of sort of these zombie
aqua hires. Everyone has different names for this, but effectively,
when a very large company, usually a hyperscaler, wants to
go and acquire a company that has AI talent, they
used to just buy the whole company. And this was
(16:42):
part of the Silicon Valley social contract that even if
I am in operations or sales or finance or HR,
if I join a hot startup and it gets acquired
I'm coming along for the ride to exit and I
get the job at Google at least for a little bit.
And then yeah, if I underperform Google or Meta Amazon,
they might lay me off. But even if I'm somewhat redundant,
(17:03):
I'm coming along for the ride and my cashing out
my shares, and these the reason that you go and
take the risk and take the little bit of the
rougher ride that is a startup. You don't get as
many amenities, but you get the lottery tickets in the
form of stock options that hopefully pay out. But there's
a big question about how much of this is the
acquirer just not wanting to deal with the d duplication
(17:23):
of the back office. There's also the question of the FTC.
A lot of the FTC rulings have made it much
harder to get these these big acquisitions across the finish line.
And even if they even if the FDC does approve,
they can often hold it up for six months. And
we're in a race where if you deliver the best
model today, you're going to make more money. You're going
to be the hot company, you're going to acquire. It's
(17:45):
all this big snowball so companies. This started with the
very few, but character Ai was the big one with
Noam Shazir.
Speaker 6 (17:51):
Who was in a very very interesting situation where so
Nome wanted to go back to Google, which made sense.
Google wanted him there as well researches of a yes,
and so having him go back there made a lot
of sense.
Speaker 3 (18:05):
I think he had built this platform.
Speaker 6 (18:06):
He built like the first at scale AI companionship platform.
If you look at character AI's site traffic today, I
mean it's it's one of the largest sites on the
internet still, which is wild. But I think he realized,
I want to work on AI research. I don't want
to work on AI girlfriends, boyfriends, that that kind of thing.
Speaker 3 (18:24):
And as part of that.
Speaker 6 (18:26):
Deal, character Ai effectively became entirely employee owned and they
had a really strong balance sheet. They had a crazy
amount of users. They weren't monetizing maybe that well yet.
But I think a lot of the employees in that
situation where like this is pretty cool. We're basically running
a co op where we all own a lot of
this company.
Speaker 3 (18:43):
We don't have.
Speaker 6 (18:43):
Investors, we don't have the same kind of pressure to perform,
and that space is competitive, right, even Elon is competing
there now Chat GPT gets used as a companion, but.
Speaker 3 (18:55):
Not competitive like Cogen.
Speaker 6 (18:56):
And so that was the dynamic here that was insane
because Google clearly cares about code generation. They see anthropic adding.
They went from one to four billion dollars in run
rate this year. They're pacing to be somewhere around ten
at the end of the year, and so that's a
Google size market, right. Google is going to ultimately care
about Cogen, So it made sense for them to say, hey,
(19:18):
let's get you know, fifty more hyper talented engineers. This
is winter we're talking about. The issue though, is if
you didn't get brought over to the Google ship, you
were on what I what I was calling a ghost ship.
Speaker 3 (19:31):
John.
Speaker 5 (19:31):
Everyone who was pro the strategy, they would call it
the remain co I was.
Speaker 6 (19:38):
But imagine you're at a company and the dynamic with
Windsurf was fascinating because the company, if you had joined
in August of last year Windsurf, you could have the
product hadn't launched it, so you could have worked and
worked up to the product launch launched it. Seeing this
meteoric growth, the last reported number they had with some
something like eighty million of air R. You have a
term sheet to get acquired from open Ai for three
(20:00):
billion dollars and that falls through, and then all these
employees are looking around and they're like, wait, I didn't
even hit my I haven't even hit my one year cliff.
Yet I don't actually have a right to. I mean,
they might technically own shares or options depending on how
it was structured, but they as I'm sure everyone here
knows you, oftentimes, like sometimes founders would would try to
(20:25):
accelerate their employees, get them compensated as part of a
transaction like that, but there's not necessarily a contractual right
to do that, and it's part of a negotiation. And
so in this case, you have this team that's effectively
split up. And we were covering the whole thing live
because we were hearing that. You know, employees that Friday
were like crying, and there was a ton of confusion
(20:45):
and chaos and everybody was learning facts kind of over
that forty eight hour period. Luckily, our friend Scott Wu
at Cognition, the company that ultimately bought the remain co,
flew to meet the Windsurf team, the new CEO Windsurf
and Basic. They spent the weekend doing this insane deal
and it ended up being a great outcome, I think
for everybody.
Speaker 4 (21:05):
Didn't involve Windsurf have their like marketing department in the
room or something. When they made the announcement, the initial announcement,
and everyone was like because they were expecting to be
bought by open Ai, right, They're like, we're going to
film this for posterity, and then it turns out to
be something.
Speaker 5 (21:21):
Yeah. Yeah, I believe the timeline is that, you know,
Windsurf launches a year ago, is in a knockout, drag
out fight with Cursor. Cursor's doing very well. They're also
going to eighty million or so opening.
Speaker 3 (21:34):
Cursors in the hundreds of millions of revenue.
Speaker 6 (21:37):
Yeah, Windsurf was very clearly the number two player in
the AI id E mark.
Speaker 5 (21:41):
Yeah, and so Cursor is staying independent, but open Ai
wants to continue to get a foothold in this space,
so they make an offer that falls through. The rumor
was because Microsoft would have had ip look through exposure
via the more complex open ai structure.
Speaker 3 (22:00):
Which is ongoing, which is continuous.
Speaker 5 (22:02):
Be ongoing, and then when when Google came through, they
said that they wanted just to buy the team leave
the remain co because it's potentially cleaner from an FTC perspective.
But there's a whole bunch of different reasons and no
one really comments on exactly what happens. But when these
deals happen. This just happened with SCALEAI and Meta. The CEO,
even though there's some sort of amorphous FTC risk, the
(22:26):
CEO if they're going through one of these zombie acquisitions,
does have the ability to kind of set the team
up with certain expectations and say, hey, we're going to
take care of you. Trust me. You're going to get
a check that you would expect based on your ownership.
So if you own point zero one percent of the company,
you would expect that you get x of the headline number,
and yes, it's coming to you. The weird thing about
(22:49):
the Windsurf deal was that that was not messaged and
so there was like this z the zombie ship was
more zombie five.
Speaker 6 (22:57):
So like yeah, and and imagine, imagine you're a you're
an hyper competitive market where you're competing with Google and
Anthropic and Cursor and all these different players, and then
you lose your CEO and your top fifty engineers, and
they're like, and you guys are gonna do against don't worry.
You guys have a strong balance sheet. You guys are
going to do great. And we're also gonna be We're
(23:17):
also gonna be competing with you at at at at Google,
But you guys are gonna be fod luck guys and
so and.
Speaker 2 (23:23):
Thanks, thanks for the help. But this break but like, okay,
cognition came in. It sounds like all the employees will
get something. It's something in the world because there was
cash on the balance but it didn't have to be
that outcome. And so there's two things here that strike me.
One is, again, this does not seem like the B
to B SaaS era, where it's like, if you created
(23:44):
the hottest building product for dentist offices and that was
great gathering traction, no one would like just buy the
talent that built that. So that's really different, right, because
so the the ability to take value out via the
talent channel rather than buying the product itself. But then
also and I'm you know, the knock on effects. Okay, fine,
the Windsurf employees did fine, but going forward, there's no
(24:05):
guarantee that they could have. And so I'm wondering, from
either a VC perspective or a future employees perspective, how
this is going to change the sort of calculation that
anyone makes when participating in a new AI startup, the
fact that the enterprise may not be where the value
actually is.
Speaker 5 (24:20):
So the first question is, like, this kind of is
the example you gave of, like buying the team that
built the dentist B to B SaaS because Windsurf does
not train foundation models and Google is exceptional in training
foundation models. With Gemini and deep Mind, the deep Mind
team is extremely well staffed on AI researchers and continually
(24:42):
seems to push the frontier both the qualitative and quantitative metrics.
And if they were weak anywhere, it was maybe products exactly.
And so this is the like, I don't want to
be like they're just product people. Sure, Windsurf has some
amazing AI researchers, some amazing AI engineers, but what Windsurf
really did. They did not train a frontier model that
was about to disrupt Gemini go deep minds foundation model
(25:07):
they were going after. You know, would you use a
Google product, No, you would use Windsurf on top of
anthropy person or another foundation model.
Speaker 3 (25:15):
And ultimately don't.
Speaker 6 (25:17):
I don't think it's a systemic risk to the social
contract of our industry. And the reason for that is
that when you see a deal like the Google Windsurf
deal get done, it was very concerning. A lot of
people were extremely angry. It ended up being a good outcome.
A lot of employees were looking around, probably concerned about
(25:37):
you know, is the million dollars of stock that I've
been working for for years worth anything at all? I
think that was a good question to ask. But I
think there's two things. One is is AI is a category,
and of course it's touching kind of every category of
venture in the private markets. But we're not seeing You're
not seeing a defense tech founding team or engineering group
(25:58):
get There's no real.
Speaker 3 (25:59):
Act hire value there.
Speaker 6 (26:01):
The acquihires that are getting done in hard tech are hey,
you clearly have good engineering capabilities. We're happy to have
you join the team, but there's no real premium being
placed on that kind of talent. Might be able to
get a great comp packages, but they're certainly not getting
these sort of multi hundred million dollar premiums. The other
factor here is, like we have a set like right now,
(26:23):
if you are one of the top one hundred AI researchers,
you could probably get one hundred million dollar comp package
like within a week if you really wanted it, right,
And there's even people that are at Thinking you know,
the reporting from this week was that and it was
kind of hotly debated, but that people at Thinking Machines,
Miramrati's company, former CTO of Open AI, we're turning down
these sort of one hundred million, multi hundred million dollar
(26:45):
There was a rumor that somebody had turned down a
billion dollar five year contract. And I don't believe that
those deals will be getting done in three years now.
I might be wrong, and there's going to be like
probably some exceptions, but the idea that the thirtieth AI
researcher on your team is going to make more than
a mags Heaven CEO, like that doesn't feel hyper.
Speaker 3 (27:09):
Something has to change either.
Speaker 5 (27:10):
Tim Cook has to make more money.
Speaker 3 (27:13):
Tim Cook is looking.
Speaker 2 (27:15):
For assistances on Wall Street were like a top trader
to consistently make more than or will often make more
than the CEO, absolutely because.
Speaker 4 (27:22):
They're the ones that got the money in the door. Yeah,
can you talk a little bit more about I guess
the fungibility of the IP here in both the practical
(27:46):
and legal sense. So if I'm a VC or some
sort of investor, I invest in an AI company because
I get to own that technology, that intellectual property. But
then let's say the main guy walks out the door,
gets hired by Google or whoever, Like, how much of
what he learned at his original firm or developed is
(28:06):
immediately going to be replicated at Google?
Speaker 3 (28:08):
I bite almost all of it.
Speaker 5 (28:10):
Yeah, I guess this is like, well, that's exactly why
they're paying that much money.
Speaker 4 (28:13):
This is a lot more way of saying how many
lawsuits are we going to get after all these aquhiers?
Speaker 6 (28:19):
So I think one way you can look at some
of Zuck's deals are like unauthorized aquhiers. It's like the
other CEO is not even participating in the deal. But
you know, Zuck is able to come in and be like,
would I pay if this, if these ten people were
working on a company independently, would I pay a billion
dollars to bring them onto my team? One hundred percent? Okay,
let's see the deal. It doesn't matter that you're kind
(28:39):
of piecing it all up from a. I think going
back to like the kind of Wall Street example of
like a top trader who maybe is like putting up
consistently incredible returns with some sort of like differentiated approach,
that person can go and as long as they have
capital at the new company, they're going to be able
to imagine to continue to just follow that same type
of strategy. I do think that AI and large language
(29:02):
models are somewhat different in that you have to look
at what is the AI product that these companies are
going to sell. How much of an impact is that
individual engineer researcher are going to have. And you know,
it's fairly obvious with open Ai their consumer tech company,
right they sell subscriptions to chat GBT, they have some
other use cases. It's obvious that companies like Anthropic, where
(29:23):
they're in the cogeneration business, they don't really have a
consumer business. And at Meta it's a little bit less
clear right now because the sort of ongoing product strategy
is not entirely clear yet. The thing that is obvious
is that if you can make a twenty billion dollars
training around more efficient than you pay for yourself, pretty quickly,
we should.
Speaker 4 (29:42):
Just point out the Wall Street analogy is not perfect, right,
because if you're a high flyer at a Wall Street firm,
either client facing or if you're trading, if you join
someone else, you usually have a non compete agreement in
one form or another. You can't take all your existing
clients with you and then be you're often on an
enforced garden for a really long time. Yeah, I imagine
(30:02):
in the massive race that is AI at the moment,
there's no gardening.
Speaker 5 (30:06):
Also, the nature of the intellectual property I believe is
a little bit different still. I think of this. I
think it's a twenty twelve example from Citadel in Chicago
where a hygh ferquency trader stole some code and they
sent and he had it on his hard drive. You
know this story, Yes, it was really.
Speaker 4 (30:24):
Famous when it happened. He like sent it to himself
for yeah.
Speaker 5 (30:27):
Yeah, So he expltraded some code, some very you know,
definitive code of how to make money in the market
and get an edge, and they he figured out that
they were on his trail, and he threw his hard
drive into the river and they sent scuba divers into
the river and got it back.
Speaker 3 (30:42):
Here.
Speaker 6 (30:43):
The other high profile SV example was the guy going
from Google self driving to yes.
Speaker 5 (30:50):
Yeah and so and so taking specific code.
Speaker 3 (30:53):
That's not happening.
Speaker 5 (30:53):
I mean, there's probably going to be like one example
of this that pops up, but mostly it's just you
get someone who says, I understand that at my previous job,
we scaled up the transformer a little too far and
we didn't focus on reasoning enough, and so we need
to shift this spend to test time inference. And it's
almost like you're leaving you're leaving a company and you're
(31:14):
and you're just you're you're leaving with like a mindset
of like the right balance of capex toopics or something.
Speaker 6 (31:19):
What was the recent Jane Street example that came out
in that lawsuit around it?
Speaker 3 (31:23):
Was it the strategy?
Speaker 4 (31:25):
Yeah?
Speaker 3 (31:26):
It only we we only really all.
Speaker 6 (31:27):
Learned what was happening because there was this lawsuit over
over basically the team bringing over some type of IP
or methodology.
Speaker 5 (31:37):
Yeah, can we talk.
Speaker 2 (31:38):
About later the broader sort of economic world right now,
because there's a bunch of other sort of things going
on and things that you talk about.
Speaker 6 (31:46):
Touch on them last thing, because I think it's pretty interesting.
So we put out the metaslist I think it was Monday,
and we immediately had a ton of inbound from a
lot of these researchers basically like critiquing like the ranking right,
and the ranking was like we got we got people that.
Speaker 3 (32:02):
Work in AI to kind of like rank them. We
the beauty of ranks? Do you get so much the beauty?
Speaker 6 (32:11):
And there's one uh, there was one example that was
interesting where somebody that was ranked fairly high that I
know has gotten one of these nine figure comp packages
and somebody that worked with him basically said this person
was kicked off of every team they worked on and
just like effectively like over multi year period, like consistently.
Speaker 3 (32:31):
Demoted over and over and over.
Speaker 6 (32:34):
And now now like got you know this incredible.
Speaker 4 (32:37):
And soft skills in ranking?
Speaker 3 (32:40):
Well yeah, and it was it was more of like
a technical ability thing.
Speaker 6 (32:44):
It was not even because I think skills are getting
a little bit uh you know, they don't they don't.
You don't care that Lebron James like might get a
little angry at somebody if they underperform, right, No.
Speaker 2 (32:56):
No, no, no, no, there's so this phenomenon of like superstars,
like it's not just it's not just in tech and
like we see it. I mentioned in journalism, where you know,
the sort of like median reporter job in a newsroom.
A lot of that's hollowing out, but you have some
people who become insanely well compensated, either because they're really
smart and they write a great newsletter, or they're really
(33:17):
good looking and they can do front facing video, et cetera,
or you know, like the four of us, you know,
like can do something on video or something like that.
You see it in the range of areas. And then
like the amount of betting that's going on, which of
course adds to this, and so the fact that all
of these events there's some market out there that you
can bet on that. And I'm curious, like from your coast,
(33:39):
you know, you're here in New York, but from your coast,
what does the world look like in terms of just
like the state of this economy.
Speaker 5 (33:45):
Yeah, there's something interesting going on in tech where the
idea of a company going from one hundred billion to
a trillion seemed unfathomable. Yeah, and there's this take that
in fact, that was the easiest ten X of all,
where the hardest ten X was going from zero to one.
From going to zero to a million dollars and getting
these systems, inventing PageRank, and then once you had Google
(34:07):
humming at one hundred billion dollar market cap getting to
a trillion, not to you knock of the work that
they did to get that, but these the numbers have
just kept growing and growing at Internet scale, at Internet speed,
and so the leverage that you're getting from an AI
researcher is ever more increasing, and it just adds ten
x every couple of years. And so ye seeing the
(34:29):
compackages and increase that way, have.
Speaker 6 (34:31):
A couple of things like the Internet is the greatest
distribution engine for information and digital products and apps and
services ever and so at least in venture, that just
means that everything is faster now.
Speaker 3 (34:43):
Even in media too.
Speaker 6 (34:44):
If you're if you are somebody's working at a at
a legacy media company and they set up a sub stack,
they can get a million dollars of ARR on the
first day that they launch. If somebody is working at
a media company and really good at a certain type
of YouTube video, they can launch and immediately the YouTube
algorithm will serve that video to all of their fans
(35:06):
within the first week. And that what that does is
it just changes. You know, the power dynamic between media
companies you see is.
Speaker 5 (35:13):
Leverage come in through financial markets. If you can lever
up or just marshal more capital one idea can generate
billions of dollars in value. Same thing in technology, but
we aren't seeing it everywhere. I don't think we're seeing
it in hard skills would working like unless you get
into true like art territory. But there is this interesting
question about like what's up next, and we were kind
(35:34):
of noodling on like there might be a law firm
in the future that's extremely high leverage in the same
sense that you might have a lawyer who's so good
at what they do and they are so good at
resolving these and the and the connections and everything that
they're doing. They're making billions, but they have a very
very lean team. And so you see a much a
(35:55):
much steeper power law in law. And there's a there's
a number of other professional services that are going through
like a transformation with technology bringing increased leverage to the
profession and that could drive more of that power li
outcome in terms of earning.
Speaker 6 (36:11):
Yeah, the thing you know in the private markets on
the West Coast, which is the dynamic right now that's
fascinating is like, I feel like a lot of people
have like a little bit of PTSD from the twenty
twenty to twenty twenty two era, where many of the
things that in hindsight were incredible top signals are like
all pop popping up again right now, and like we like,
(36:31):
we like trap, we track them just for fun. We
are not in the business of like calling the top
or anything like that. And in many ways it feels like,
you know, there's so many positive indicators. But the interesting
dynamic and in venture is like, as an angel investor,
I've invested in probably sixty five or so different companies
at the pre seed to Series A stage, and there
(36:53):
was a bunch of deals that I did in twenty
twenty one twenty twenty two that ultimately like I paid
too much or just like the team wasn't good enough
to execute against the vision they had. But there was
also just like a handful of deals that I did
that that have been you know, performed so well that
it doesn't matter that I did a bunch of silly deals.
And so venture right now is this interesting kind of
(37:13):
dynamic where everybody knows that it's crazy, Right, you have
hundreds of billions of dollars of value created in the
private markets that there's a lot of real revenue growth,
but there's also hundreds of billions of dollars of value
tied to zero revenue, right. And I think that there's
something that we've been tracking is like this underlying kind
of feeling from people that are maybe under thirty of
(37:34):
there's like this meme that's become very prevalent, and I
think it explains a lot of economic activity today, which
is young people feel that they have two years to
accumulate capital to escape the permanent underclass. And so I
think that drives a lot of investing activity today in
that you know, you'll see young people on the timeline
(37:55):
saying that like you'd be stupid not to use leverage,
you know, and they're and they're just like, you know,
they're not a professional.
Speaker 3 (38:01):
Investor in any capacity. Don't do it.
Speaker 6 (38:05):
And you know, any anytime you have people in venture
making public market stock predictions, we get a little concerned.
Speaker 4 (38:13):
Are we going to get AI talent agents? Not AI agents,
We have those already, but AI talent agents because like.
Speaker 5 (38:19):
We do so we do they're called venture capitalists.
Speaker 4 (38:22):
Oh well, but yes, this was going to be a sentence.
So if the money is no longer in, you know,
I invest in a startup and eventually they get the exit,
they get acquired, or they list or whatever. If the
money is in, eventually the guy that started the startup
gets bought by someone like Meta or whoever. Wouldn't I
invest my money in that person versus in the startup
(38:45):
inventure service?
Speaker 5 (38:46):
Yes, yeah, you can't quite do that, but there are
a ton of roles that are indexed to these high
performance packages. And yes, there are people right now in
Silicon Valley who are effectively talent agents who get a
cut of those big packages and they help me get
siate just.
Speaker 3 (39:01):
Called just talent, called the venture capitalists. No, no, so so
you can.
Speaker 5 (39:05):
You can you find a great researcher and you think that, okay,
maybe they will build a business, but I'm sure that
they are going to be worth hundreds of millions of
dollars a year, and you can just invest in their company.
You will almost certainly see a good return on that investment,
even if the product never gets to scale, because when
they get an ACQUA higher, you will get a.
Speaker 2 (39:25):
Payout, not going there unlike the.
Speaker 6 (39:33):
Are like, oh absolutely no, it's now it's now so
normalized to invest in college dropouts that you actually see
people like investing in high school.
Speaker 5 (39:43):
I'm not kidding about this. So the Iomo Gold Medal
is the math Olympiad, and there are venture capitalists who
will give calls to every single student that performs well
on them on the math Olympiad. And these are high
school students.
Speaker 4 (39:54):
So do they what are these investments? Exactly like, are
you're funding their tuition or no?
Speaker 5 (39:59):
No, no, it's it's it's I am taking ten or twenty
percent of a company of a Delaware c core. Most
likely that you will build something in and who knows
where it goes. Maybe it turns into a great business,
maybe it turns into aquhire. But the downside is extremely
limited because there's always this at least right now, there's
this aqui hire, there's this AQUHIRER on the table where
(40:22):
it used to be.
Speaker 6 (40:23):
That to be clear that it is only an AI
yes only for the people that you would consider to
be the top five hundred yes in there industry.
Speaker 5 (40:32):
So to go back to look at Cursor windsurf thing
like Instagram was the power law winner got the billion
dollar acquisition from Meta Hipstomatic was the second largest photo
filtering app, did not get a billion dollar aquhier from Google. Right,
But now we're in the market where if there's a
leading product with a bunch of AI researchers over here,
and they get a multi billion dollar acquisition for the product,
(40:54):
and the product's working and it's growing and it is
a great business, and you buy it for the value
of the business, then the second best team might get
acquired just for talent, which is a completely different downside protection.
Speaker 6 (41:04):
Yeah, and there's a lot of talent acquisitions traditional acquires
where it's only the talent that benefit, right in the
sense that they get basically a job offer at the
new company, and vcs get some capital back or in
some cases a small amount of money.
Speaker 2 (41:19):
Can I ask a TVPN question This should be like
the ultimate compliment, which is that I got a d
M from some random person I've never seen them the
other day and he said I can build tvpns for X,
So like that's that's deck. That's sort of which means
it's sort of like it's become like Kleenex or one
(41:39):
of these things where it's like a category and where
are you going with it? What are your playing?
Speaker 6 (41:44):
It's so funny that people are calling it TVPN for
X because our show, while it's unique in a variety
of ways, looks very much like traditional is interesting, and so.
Speaker 3 (41:56):
We can't we can't take credit for we held it.
Speaker 5 (42:00):
We invented media, we invented live streaming.
Speaker 2 (42:03):
Of course, course normally it doesn't really I don't know,
why would I watch cable news on.
Speaker 6 (42:09):
I think thing that's exciting is that in the private
markets and venture and tech, if you had a podcast,
there was one format which was a once a week
interview show, and it was a great strategy to do
that ten years ago, around the time that you guys.
Speaker 3 (42:26):
It locked established.
Speaker 5 (42:30):
But if we if we tried to clone this, yeah,
everyone would be like, why is that a knockoff? There's
nothing new about this, nothing fresh about this, Like why
do I want to go on your knockoff? I just
go on the real thing.
Speaker 3 (42:40):
Yeah.
Speaker 6 (42:40):
And so our edge in launching the show at the
beginning of the year is that media is not zero sum.
Speaker 3 (42:48):
Content is not zero s.
Speaker 6 (42:49):
We have a friend that jokes that he's so competitive
he wishes he wishes media zero sum it was truly
zero s. But our edge early was that we just
took it ten times more seriously than anyone else. So
everybody that was creating content for the private markets was
doing it as a part time gig, and we were
happy to compete with a bunch of people that were
part time.
Speaker 2 (43:10):
You know what, Actually, I was wondering in twenty twenty
one or twenty twenty two that craziness. Someone once reached
out to me and they're like, you know what, you
should like, no direct, it's probably the no, but this
was this was the interesting thing. He said, you should
go independent and what you should do is attach a
(43:31):
VC armed to it, and because of the quality of
the guests that you could get, you would get really
good access to deal flow. But it seems like basically
it's very uncomfortable to me because I'm not going to
highlight startup founders and say, oh, you're the perfect guest.
And it's like because like I have like five percent.
But I'm curious about because you mentioned doing Angel the
(43:51):
sort of link of the TVPN at some point, like, yeah,
so an investing.
Speaker 6 (43:57):
Platform Angel investing. I look at it as a hobby.
It's not a great financial activity, right, locking up, I've
heard locking out. I mean, like you can generate great returns,
but it's not the most logical way to invest your time.
But it's really fun. Like we just enjoy supporting founders
early when it's an idea and a team and it's
just it is genuinely addicting. I always joke about people
(44:20):
in San Francisco with Angel. You know, people across the country,
sports betting addictions san Francisco, you know, Angel investing addictions
are I think real. But for us, for us, that was,
you know, as we started having some success a lot
of people, they would probably get a message today, when's
the fun coming, when's the fund coming? And that's been
(44:40):
a way to monetize an audience within tech. If you
have an audience, go raise one hundred million dollar fund,
you get the fee streams do that, you get a
bunch of upside. And we joke because we have the shows.
Every single day we go live at eleven. We have
to eat, we have to work out, we have to
prep the show, we have to talk with partners, we
(45:02):
have to manage our team. There's all these different things
and so somebody might immediately think, Okay, these guys talk
to six founders and investors a day. They have this
like media property that like, I think everybody in ventured
now is going to see our content like once a
week in some form or another if they're if they're online.
But I joke with them, I'm like, so we're live
(45:25):
for three hours every single day during the middle of
the day. And so if you think that we with
a show would would effectively compete at VC is about
winning allocation, right. You can be cool and connected and
have an audience that might get you one hundred K allocation,
but if you have one hundred million dollar fund, you
need to be putting size into deals. And so I
(45:47):
know that if John and Jordi have a VC fund
and we're competing with these other guys that have a.
Speaker 3 (45:51):
Live show that they're live for three hours a day
and they have.
Speaker 6 (45:53):
A VC fund, we'd smoke them because we'd be like, Okay,
well they're live. I'm going to fly to the founder,
I'm gonna meet with them, and I'm we're gonna win
this deal, and we're gonna be like, yeah, why why
don't you let them put in like two hundred k.
Speaker 4 (46:03):
So the thing to watch out for when the VC
fund is coming, is when you start reducing your live hours,
that'll be the same.
Speaker 1 (46:11):
Yeah.
Speaker 6 (46:11):
I think I think we we didn't get into this
to start a fund, and I think there's a lot
of people that get into content because they see it
as a way to do that and we just love
talking about tech.
Speaker 5 (46:21):
It actually was the key insight was that not enough people.
Yeah it's very fun, but not but very few people
in tech were actually taking media seriously. Yeah, it was
everyone had a fund, everyone that was the high status thing,
and doing media was like lower status or people didn't
think we could have as much of a power all
outcome as it very clearly can. And so this idea
(46:41):
of of just what if you actually took it completely
seriously and just made it the main.
Speaker 6 (46:47):
Thing, and when you were We've had a bunch of
people copy our format, you know, major legacy media companies
all the way through friends of ours. It doesn't really
bother I mean it bothers me more than John. But
at the end of the day, if you want to
spend the tracy, yeah, I totally got it. And so
(47:07):
it's like it doesn't basically know so so John and
I John and I basically hang out for twelve hours
a day, and the entire time, the entire time we're thinking,
we're thinking about the show, we're just talking about the
show doesn't look very different than when we're hanging out offline.
And so I just joke, I'm like, Okay, if you
want to compete with us and you're willing to put
(47:27):
it one hundred hours a week in god by all means,
go for it like this probably your life's work.
Speaker 3 (47:33):
But if it's not good luck, it doesn't matter.
Speaker 2 (47:36):
John Cougan, Jordie Hayes, thank you so much for coming
on Odd Loss that's in good luck and looking forward
to continue watching TVPM.
Speaker 7 (47:45):
We'll be live from Nicey later Odd Lots in the
same day and in the wake of meadow winch mech
me a.
Speaker 3 (47:58):
Thank you for having us. Thank you so much, Tracy.
That was a lot of fun.
Speaker 4 (48:14):
It was a little bit media naval gazing, a little
bit fun.
Speaker 2 (48:17):
A little bit media naval gazing. I mean, there is
this thing that's happening. It's been happening in media for
a while, and of course it happens in Wall Street
and now happening in AI where you just have a
lot of talented people, and they wonder about the degree
to which they need their existing platform there. You know,
(48:38):
like a star banker can take a book of business,
or a star lawyer can take a book of business.
And in the case of a it's not taking a
book of business, right because they don't have like their
individual clients. But it's this knowledge and transport it and
it's instantly worth a lot of money for someone somewhere else.
Speaker 4 (48:54):
The thing I thought was really interesting about that discussion
was the emphasis on how capital intent all of AI
is and so that kind of changes the economics of
why you're paying such a massive money for someone who's
able to eke out even a slight efficiency.
Speaker 3 (49:10):
This is huge.
Speaker 2 (49:13):
This was like this It suddenly is what made it
all made sense to me, right, because we know about
like how costly one training run is, right, and we
know just the insane numbers for data center set up,
et cetera. And I'm sure there's progress being made on
the literal design of like how you string together in video, GPS,
(49:33):
et cetera. And so if you could get some margin,
if you know have that know how to get some
marginal improvement out of it, And this is fundamentally what was.
Speaker 4 (49:40):
One trillion dollars for you?
Speaker 2 (49:42):
Maybe not a trill No, not a trilliont but like
this was not the case when tech was not so
capital intensive in the twenty tens. Where yeah, I'm sure
talented people always made a lot of money and there's
always improvements, but where it's so that link between some
sort of efficiency gain and instant cost savings is so
linear and so straightforward. Yeah.
Speaker 4 (50:04):
And I take their point that, Okay, vcs exist and
they're already investing in some ways and specific talent. But
I do kind of wonder if you're going to get
some sort of like specialized headhunters at the very least,
who are going to like seek out these big AI
talents and try to like graft themselves onto them.
Speaker 2 (50:19):
Yeah, just people whose expertise is in reading through under cited,
under cited AI research papers.
Speaker 3 (50:29):
Yeah, AI research.
Speaker 2 (50:30):
Well, we can't give you Sam Altman, but what if
we could replace Sam Altman in the egg.
Speaker 4 (50:34):
Guy has fifty citations in the following papers? Should we
leave it there?
Speaker 3 (50:38):
Let's leave it there.
Speaker 4 (50:39):
This has been another episode of the All Thoughts podcast.
I'm Tracy Alloway. You can follow me at Tracy Alloway and.
Speaker 2 (50:44):
I'm Jill Wisenthal. You can follow me at the Stalwart,
follow our guest Jordi Hayes He's at Jordi Hayes, and
John Cougan at John Coogan, and check out TBPN at TBPN.
Follow our producers Carmen Rodriguez at Carmen Arman, dash Ol
Bennett at dashbod and kel Brooks at Kale Brooks. And
from our Odlots content good to Bloomberg dot com slash
odd lots were the daily newsletter and all of our episodes,
(51:06):
and you can chat about all of these topics twenty
four to seven in our discord Discord dot gg slash
od loots.
Speaker 4 (51:12):
And if you enjoy odd Lots, if you like it
when we talk about the sportification of AI talent, then
please leave us a positive review on your favorite podcast platform.
And remember, if you are a Bloomberg subscriber, you can
listen to all of our episodes absolutely ad free. All
you need to do is find the Bloomberg channel on
Apple Podcasts and follow the instructions there. Thanks for listening.
(52:04):
H
Hosts And Creators
Joe Weisenthal
Tracy Alloway
Popular Podcasts
On Purpose with Jay Shetty
I’m Jay Shetty host of On Purpose the worlds #1 Mental Health podcast and I’m so grateful you found us. I started this podcast 5 years ago to invite you into conversations and workshops that are designed to help make you happier, healthier and more healed. I believe that when you (yes you) feel seen, heard and understood you’re able to deal with relationship struggles, work challenges and life’s ups and downs with more ease and grace. I interview experts, celebrities, thought leaders and athletes so that we can grow our mindset, build better habits and uncover a side of them we’ve never seen before. New episodes every Monday and Friday. Your support means the world to me and I don’t take it for granted — click the follow button and leave a review to help us spread the love with On Purpose. I can’t wait for you to listen to your first or 500th episode!
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Special Summer Offer: Exclusively on Apple Podcasts, try our Dateline Premium subscription completely free for one month! With Dateline Premium, you get every episode ad-free plus exclusive bonus content.
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.