If you could amass any five artificial intelligence startup bets right now, which companies would you pick?
My Cerebral Valley co-hosts and I took a stab at answering that question with an artificial intelligence startup draft.
Our startup draft starts at 27:35 after a discussion of some of the biggest themes going into this week’s Cerebral Valley AI Summit.
The draft gave us a chance to dissect some of the most promising startups in artificial intelligence right now.
The goal was to amass five companies with the biggest valuation five years from now. We restricted ourselves to AI startups that had raised more than $100 million.
I encourage you to make your own prediction in the comments.
Give it a listen
Get full access to Newcomer at www.newcomer.co/subscribe
00:00:10
Welcome to the Cerebral Valley Podcast on newcomer I'm Back
00:00:14
with Max Child and James Wilsterman, my Cerebral Valley
00:00:17
Co host and the Volley Co founders.
00:00:20
We are very excited about this episode.
00:00:22
The meat of it will be a draft of the most valuable AI
00:00:29
companies we get. It is like, it's as close as I
00:00:32
think in Silicon Valley you can get to watching sports hopefully
00:00:35
or fan of. Sports.
00:00:37
Sports debate TV show. Yeah, exactly.
00:00:42
Yeah, I I overcomplicated it to make sure I got the.
00:00:47
I won't spoil it anyway, you know we took it very seriously
00:00:51
and even if all our listeners forget about it, I am sure.
00:00:55
James, what was the time window? It's five, is it?
00:00:58
Five years, Yeah. In five years we'll all be
00:01:02
checking to see see how we did. But before that Cerebral Valley,
00:01:07
the conference is on Wednesday. And you know, obviously for the
00:01:13
people going in person, you, you get why you're curious.
00:01:16
It's like OK, But then for everybody else, the videos from
00:01:20
the conference will be posted on YouTube and in the
00:01:23
newsletternewcomer.co. So this conference will be
00:01:27
brought to you digitally. We have not just hyped up a an
00:01:29
event you will not get to consume, you will get to see the
00:01:33
talks. So look forward to that.
00:01:35
I think usually we put some of the best ones in the podcast
00:01:39
feed and we just wanted to sort of take stock going into the
00:01:44
conference sort of big themes that that are going to be on our
00:01:49
mind and then what's changed from the first Cerebral Valley
00:01:55
in March. So yeah, let's get to it.
00:01:58
All right. I said you guys could could go
00:02:00
for themes last, I mean go for themes 1st and I would play
00:02:05
clean up so. James, why don't you go still
00:02:08
thinking about my themes? Yeah, I have noticed a theme
00:02:13
kind of gaining some heat over the last few weeks on Twitter.
00:02:18
Slash X of Kind of this big debate around open source
00:02:23
versus. Close source in terms of safety,
00:02:28
right. I think this ties in with the
00:02:30
executive order put out by the Biden administration and kind of
00:02:36
a raging debate I guess or simmering debate of whether
00:02:40
these larger close source language model companies are in
00:02:45
the business of regulatory capture through their efforts to
00:02:50
speak around model safety and get.
00:02:55
Get the Biden administration involved or whether that comes
00:02:59
from a true concern and how that impacts companies who are
00:03:05
putting out open source models for the community.
00:03:07
OK, I'm going to jump on that. Open source versus closed
00:03:10
source. I I think that's probably the
00:03:11
top in my mind. And in some ways I'd almost
00:03:14
frame it as like I think the first conference ended up having
00:03:18
more of the pro open source people.
00:03:22
You know we ended with Hugging face and Replet.
00:03:25
Both of them are very pro open source.
00:03:29
I mean we have data bricks back they're pretty pro open source.
00:03:33
Whereas this conference we're ending the day with Vinod Khosla
00:03:38
who has been somewhat critical of he he wants more sort of
00:03:43
control and regulation. We'll we'll get the exact nuance
00:03:46
of his open source take and then Mustafa Suleiman at inflection
00:03:51
and the D Mine Co founder who's going to end the day has been
00:03:54
super supportive of regulation and somewhat apprehensive about
00:03:58
open source. So I think this conference in
00:04:00
some ways will be the response to the really pro open source of
00:04:04
the first. Of course I I think open source
00:04:06
is very popular in the not open AI world.
00:04:09
So I wouldn't be surprised still if probably the majority of our
00:04:12
panelists are still pro open source or yeah, what do you guys
00:04:15
think? I guess one thing I wanted to
00:04:18
ask you guys on this topic is. Last time at the conference, a
00:04:23
lot of discussion was about the letter that just came out, with
00:04:26
a lot of signatories around taking a pause on AI
00:04:31
development. A modest facility had signed it.
00:04:34
Totally perplexingly, yeah. Yeah, which, which was?
00:04:38
Confusing to a lot of people. And I guess my question is, at
00:04:43
the time, it felt like there was a lot of heat on companies even
00:04:46
like Open AI and Sam Altman. For being cavalier with their
00:04:52
development of AI and causing, you know, an AI race and putting
00:04:55
out ChatGPT. And you know, before the bigger,
00:04:59
you know, more cautious companies like Google had
00:05:01
released their AIS, right? And I really think in like 6
00:05:05
months that has just totally changed where you know now that
00:05:08
you have this open source community putting out models.
00:05:11
Like Open AI is able to like look like the responsible
00:05:14
stewards of AI going to the White House and you know,
00:05:18
getting regulations in place. Just curious if you guys agree
00:05:22
with that assessment? I do think a lot of the safety
00:05:26
talk has died down. I guess the executive order is
00:05:30
maybe a contrary example of that.
00:05:33
Although I you know, maybe the White House timeline is like six
00:05:35
months behind everyone else's timeline, but.
00:05:39
It does seem like people are a lot less hung up on like is the
00:05:44
AI going to kill everyone and sort of the tech industry and
00:05:47
then I guess simultaneously we got this massive executive order
00:05:50
banning or not banning but regulating models over a certain
00:05:53
size etcetera, etcetera. So it's it is interesting to
00:05:58
think about if there's going to be this sort of waves and and
00:06:02
troughs or peaks and troughs of the the hype cycle around safety
00:06:06
and I feel like. Burn a little bit of a through
00:06:08
inside the industry right now, but six months from now that
00:06:11
could change again. Just got to keep having
00:06:13
conferences in March. GPD 4 was like, you know, a
00:06:16
shockwave and it had come over 3.5 and I think it felt like,
00:06:21
man, if this keeps going like we're going to have generalized
00:06:24
AI tomorrow and then I nobody is really caught up with them,
00:06:27
right? There are open source models and
00:06:30
people can say, oh, this is better than open AI on cost.
00:06:35
But I don't think anyone even in specialized use cases has really
00:06:39
had anything as smart. So then there's just a little
00:06:41
less fear of like, Oh my God, it's becoming like.
00:06:44
So you know, that is just not sort of the pace hasn't been so
00:06:49
out of control. And I think that's part of what
00:06:51
has cooled things down. But I sort of, I mean I think
00:06:54
the binding executive order is like a huge deal and shows that
00:06:57
there's very active sort of government interference and like
00:07:00
Europe has been talking about things.
00:07:03
So I don't know, I think this question of cracking down on
00:07:07
open source is very real. But yeah, but the open source
00:07:10
community has certainly been vibrant and and you know, and
00:07:14
the other piece of that is that firms like Andreessen and others
00:07:17
have been very loud about sort of positioning themselves as
00:07:20
super pro open source. Yeah, interesting.
00:07:23
Next theme. All right, I have a theme for
00:07:25
you guys to think about here. I think one of.
00:07:30
The all time maybe the number one all time Silicon Valley
00:07:33
investing philosophy for all categories of software is you
00:07:38
know we like to invest in companies that sell picks and
00:07:41
shovels, not the companies that dig for gold, right.
00:07:43
We like to, we like to fund the tool makers, not the tool users
00:07:47
in many cases. And I think what's interesting
00:07:51
just in the last six months between the previous conference
00:07:55
and the current conference is. The the sort of what tools were
00:08:00
in vogue 6-7 months ago and then what tools are invoked today.
00:08:04
And then the sort of cherry on top of that is Open AI itself,
00:08:09
you know, very, very recently announced this kind of huge wave
00:08:13
of first party tools improvements whether it's you
00:08:17
know, retrieval improvements, context increases, speed, cost.
00:08:22
Really seeing anyone able to make their own GPT a character
00:08:26
to kind of go after the character AIS of the world.
00:08:28
Like they really came hard on like a bunch of tools categories
00:08:32
that I would say 6-7 months ago were being funded as like pretty
00:08:34
compelling stand alone companies.
00:08:36
And I think it's sort of a A2 way question which is 1.
00:08:42
Is tools investment still good idea in an AI world like or is
00:08:46
open AI going to eat this all alive and then I guess 2 like?
00:08:50
You know who who is, who is still exciting in the tool space
00:08:54
and and who do we think has like a strong foundation for the
00:08:56
future. Yeah, I mean I think that
00:08:59
question of will Open AI just eat everything applies in every
00:09:04
part of the. Yeah, I mean that was a theme of
00:09:07
the first conference definitely, just like how how relevant is
00:09:11
everybody except Open AI and and sort of an ongoing question.
00:09:16
So certainly, yeah, I think now with the developer conference,
00:09:20
this thread on tools and then I think you know, given ChatGPT is
00:09:26
clearly the most popular consumer AI app.
00:09:28
Well, I'd be interested. I mean, you have authentic
00:09:30
conversations with venture capitalists regularly, Eric,
00:09:33
right behind the scenes that don't all get published.
00:09:35
Like, do you ever get some authentic expression of concern
00:09:41
about tools companies that they've backed in the space that
00:09:43
were like, well, maybe we. Jumped on this tools idea too
00:09:47
early and it actually isn't a standalone product or you know
00:09:51
behind the scenes there are definitely a lot of firms that
00:09:54
are shitting on foundation model bets.
00:09:56
Oh, interesting. Oh yeah, that's that's like, I
00:10:00
mean they're expensive you know they're very speculative.
00:10:03
Open AI is in the lead. They feel like even with open AI
00:10:07
open source comes to parity very quickly.
00:10:09
So the, the category I hear the most dunking on is just like
00:10:13
foundation models and you'll you'll see firms that haven't
00:10:16
made a lot of bets there. You know, yeah, the Valley loves
00:10:20
like tools, picks and shovels. So that's not something where I
00:10:25
feel like I've heard a lot of regret.
00:10:26
And if anything I feel like I've been, I mean I've been talking
00:10:29
like vector databases all week and stuff like that.
00:10:32
And literally you're getting VCs talk about chips or getting
00:10:35
companies around chips, which traditionally they've been
00:10:38
scared about. But you have sort of the fact
00:10:41
that NVIDIA is doing so well and then you also have the fact that
00:10:44
it's sort of cool to be like an emerging tech investor at the
00:10:47
moment. So I think those things
00:10:48
combined, you're seeing that. So no, I I think V CS dunking on
00:10:54
foundation models, but bullish about everything else supporting
00:10:57
AI. Yeah.
00:11:00
Are there particular company types you have in mind like do
00:11:02
you think? Well, I mean, I think there's
00:11:04
like obvious companies that are like.
00:11:06
May be less hot than they were in the last conference, right?
00:11:08
I mean like Jasper. Jasper would be an example,
00:11:11
right? Stability would be a huge
00:11:13
example, right? But I think there's company
00:11:14
stability that's just like the unique failings of.
00:11:17
That sure, sure. But there are companies that we
00:11:19
talked to last time, right? You know, Lang Chain, for
00:11:21
example, right. That like it's a big question as
00:11:24
to where Lang Chain fits in the universe, where Open AI is just
00:11:27
going to take care of all this crap, right?
00:11:29
And then? You know, I think we're talking
00:11:31
to some companies this week that, you know, look like they
00:11:33
they could totally be awesome companies.
00:11:35
And I have no earthly clue, but they're providing very
00:11:38
specialized tools like you know, data cleaning or something like
00:11:41
that, right. Like which again I'm excited to
00:11:43
hear the pitch like of of why that's a a great idea for the
00:11:45
company and then probably is. But like, you know, I would be a
00:11:49
little scared that opening. I was just going to like kick
00:11:51
ass in my category if I were a tools company.
00:11:55
And so I just think, and I mean, I think you have to agree that
00:11:57
character has to be pretty upset that Open AI basically announced
00:12:01
their version of character AI tools, right?
00:12:03
Like, so I'm just interested if there's any real feedback of the
00:12:08
fear factor on that front. But it sounds like maybe not.
00:12:10
So yeah. Do you think just jumping diving
00:12:15
in on the character question like there is a theme I've
00:12:18
noticed of people kind of questioning whether?
00:12:22
Open AI for consumers as an end point is is really sufficient.
00:12:27
So there's kind of like the opposite side of that question
00:12:30
from the tooling side, right? Like will they eat all tools?
00:12:32
And then there's from the consumer assistant side, is
00:12:35
there a super assistant that's just ChatGPT or do you end up
00:12:38
using all these characters or mini GBTS that that they are
00:12:43
working to enable developers to build?
00:12:46
How do you guys feel about that debate?
00:12:48
Yeah, I mean, I think that's, that's the open question.
00:12:51
I think it depends. It depends on the quality of
00:12:54
the, you know, the foundation. Models.
00:12:57
I don't know. I mean, I guess like you
00:12:58
selfishly we built games, James, that are built on AI tools,
00:13:01
including in some cases large language models.
00:13:04
So my selfish belief is, is possible to build games on LLMS
00:13:09
without open AI building the exact same thing.
00:13:12
It may depend on the complexity level of the consumer app built
00:13:16
on top of the tool, whereas character is sort of.
00:13:19
Pretty vanilla RE, you know, re prompting or you know tweaked
00:13:23
version of the core, the GPT experience.
00:13:26
But you know, I think it depends on the complexity level and sort
00:13:29
of how much of the stack you control and and how easy it is
00:13:32
to get to your customers. I mean a phrase I've been
00:13:35
hearing over and over again this week, you know, just talking to
00:13:38
VCs has been retrieval augmented generation.
00:13:42
Rags, Rags and I think there are a lot of companies right now
00:13:45
that think we can be really good at giving GPT the sort of
00:13:51
information it needs and then you know Open AI or whoever can
00:13:57
sort of do the thinking based on that.
00:13:59
But we're not, we're not in the business of making the best
00:14:01
thinking company, but we're we're really good at figuring
00:14:04
out like what it needs to consider and and it's possible
00:14:08
that there are a lot of what. Kind of companies do you think
00:14:11
are saying that around like I guess this is kind of.
00:14:15
Pointing to data and unique data sets as kind of a mode of some
00:14:19
kind, right? Like, what kinds of companies do
00:14:22
you think fit that category of potentially having a true data
00:14:26
mode that could be valuable, having a data mode?
00:14:31
Yeah, Well, well, this isn't a data mode, but I mean, I'm one
00:14:38
company as you'll hear in the second.
00:14:40
Isn't it a data? Isn't it a data mode that you
00:14:43
are saying that? These companies are talking to
00:14:45
you saying we have well, like OK, there's this company clean
00:14:49
right, that lets companies run searches, right?
00:14:52
Is it a data mode? They're really like good at
00:14:54
hooking companies up, companies data up to data sources that
00:14:58
then connect, that use rag and then plug into something like
00:15:03
Open AI. So it's not necessarily like a
00:15:06
Bloomberg where it's like we have all the financial data in
00:15:08
the world. It's we're really good at like
00:15:10
getting the relevant data four companies into into this system
00:15:14
is is 1 case. But at least at Open AI Dev Day
00:15:18
this week, they seem to be going after some similar use cases of
00:15:23
allowing companies and enterprise to build their own
00:15:26
internal GPTS or chat GPTS based on internal company data.
00:15:31
So I think that that is still some area that they seem to be
00:15:35
competing in as well. I mean, you know, eventually
00:15:39
there's only so much one company can do unless, unless it's like
00:15:43
the, I feel like James, this is your style of suggestion
00:15:46
throughout this series. Unless there's AGI and like
00:15:49
every way we talk about this is wrong.
00:15:51
But in any sort of traditional business sense where like you
00:15:55
know, Sam Ullman, you know, has joked in the past that like, oh,
00:15:58
we'll just ask open it, we'll ask Chad GBT at some point, you
00:16:02
know, like what our business model should be and we'll do
00:16:04
that. Clearly they didn't need to wait
00:16:06
for that. They figured out some business
00:16:08
models. But you know, I anyway, my my
00:16:11
point is just like in a traditional, normal world that I
00:16:14
expect to continue, You can't do everything And so whatever,
00:16:17
they're going to have to pick their strengths and then
00:16:20
everybody else can do other stuff.
00:16:24
I'm going to, I'm going to give a last, a last theme that's sort
00:16:28
of more in the fun sort of philosophy sort of world.
00:16:32
You know, fitting into that AGI question which is just how
00:16:36
seriously do people take existential risk in a, you know,
00:16:42
is existential risk now is seen as like this is propaganda for
00:16:47
people who want to make it 1 seem really awesome and sexy and
00:16:51
amazing and can do everything. And then two people who want to
00:16:54
be in this world of like shutting it down, Like how much
00:16:57
is existential risk really on people's minds anymore?
00:17:01
Or is it much important that we just make sure it's not racist
00:17:05
and make sure, you know, deal with these like, very practical
00:17:08
concerns that are obviously true now versus imagining this world
00:17:12
where it's. You know, you know, like our AI
00:17:15
Doom episode. You know, figuring out how to
00:17:17
optimize paper clips into blowing up the.
00:17:19
Earth. That's kind of alluding to my
00:17:20
earlier point about I feel like Silicon Valley has stopped
00:17:23
taking safety as seriously, right.
00:17:24
I mean, like, I think that like, yeah, sure, the executive order
00:17:27
is not an example of that, right?
00:17:29
But in Silicon Valley, I think the discussion of, to your
00:17:33
point, the the implied seriousness with which people
00:17:37
take. Existential risk seems pretty
00:17:39
low to me because otherwise people would be like blowing up
00:17:44
GPU factories or whatever. I don't know.
00:17:47
It's like, if you actually believe this, had a 20% chance
00:17:50
of killing all of society, which a lot of people say, like how
00:17:54
could you morally work in this space?
00:17:57
Or at least not like, you know, be actively opposed to any of
00:18:01
the development here, right? And and the only people who seem
00:18:04
to really act as if they truly believe existential risk.
00:18:07
Is real is basically you know, Elisa Yodkowski and the sort of
00:18:10
the AI doomers on Twitter, right, who are really out there
00:18:12
every day saying there was, there was an interview.
00:18:15
Shout out to Logan Bartlett with his podcast.
00:18:19
He interviewed Dario, the Co founder.
00:18:22
Of anthropic, right? And Dario gave like it was like
00:18:25
20% chance, like everything. Goes there was like 10 to 20%
00:18:28
chance that was like really high I think.
00:18:31
Is that like in the next 100 years?
00:18:32
So And again the time frames matter right?
00:18:35
Like. I think it was.
00:18:36
It was reasonably soon. But yeah, I I don't know.
00:18:39
I sort of think you'd have to be insane to be working in an
00:18:43
industry where you truly believe that progress there had a 20%
00:18:47
chance of ending all humanity. I I don't know.
00:18:49
Like, unless you believe in some complicated.
00:18:52
Game theory about deterrence or something.
00:18:53
And back to our earlier episode about, you know, the only thing
00:18:56
that can stop a good guy with an AI, or sorry, a bad guy with an
00:18:59
AI is a good guy with an AII. Said it wrong.
00:19:01
Who coined that line? Did I get that line?
00:19:03
I don't. I can't remember but.
00:19:05
But the only thing that can stop a bad guy with an AI is a good
00:19:08
guy with a better AI, right? Yeah, unless you can explain
00:19:11
some pretty elaborate game theory on how progress in AI by
00:19:16
the good guys is necessary to stop extinction.
00:19:19
It just, you know, it doesn't compute that you believe this
00:19:21
stuff has 20% chance of killing us like you know, and be working
00:19:26
in the industry at the level of Dario, who literally is like the
00:19:29
number two most powerful person in foundation models, right?
00:19:33
I mean it's just, it's nonsense. So I don't know.
00:19:36
I agree it it doesn't seem like people take it that seriously
00:19:39
for real. As we discussed in that episode,
00:19:42
though, there's a lot of, like, very bad things that could
00:19:45
happen that, you know, AI kills some of us, right?
00:19:48
Not. All of us, Yeah.
00:19:49
That was the that was. The that was the take away all.
00:19:52
You're right, right. Maybe it doesn't.
00:19:53
Like, I guess like maybe maybe people have come around to sort
00:19:58
of a middle ground here where it's not existential risk, like
00:20:02
extinction for humanity and more, that there's a lot of bad
00:20:05
things that could happen over the next 5 to 10 years with AI
00:20:08
and I think all of us. Would agree with that, right?
00:20:12
You know there are bad things that could occur if the, you
00:20:16
know, some bad actors got access to some really powerful sure
00:20:20
models or something, right? I mean.
00:20:22
I think you the fact that Cruise is not active in San Francisco,
00:20:27
like even if they sort of bungled their disclosures to the
00:20:30
government, I mean Cruise is a better driver than humans in San
00:20:35
Francisco, right Like. They really they they.
00:20:37
They broke the one rule, which is don't lie to the government,
00:20:41
right? I mean, like, it's just like,
00:20:43
you can be, you can be bad, you can screw up, you know?
00:20:47
But don't lie to the government is like freaking like #1 on the
00:20:51
10 commandments of running a company, right?
00:20:54
I mean, like, you know, that's why SPF travels or whatever, You
00:20:57
know, like, I mean, like like just don't lie to the
00:21:00
government, like, you know? Fake it so you make it can apply
00:21:03
in many scenarios, but not with the government.
00:21:05
There, you can't do that with them.
00:21:08
So anyway can we just go back to this for a second?
00:21:11
So you're you're what are you saying Eric about existential
00:21:17
risk or Max like you you guys are saying companies are not
00:21:20
acting like they believe their own hype or do you agree with
00:21:24
that or? I think, I think people are now
00:21:26
saying calling bullshit in sort of private.
00:21:30
Existential risk. And some of them don't want to
00:21:32
say it publicly because they don't want to be seen as sort of
00:21:35
undermining legitimate concerns. But I think a lot of them are
00:21:39
just like doesn't feel close and then how do you feel that ties
00:21:44
into the executive order? Does that feel to you like over
00:21:50
regulation or you know, pre? Premature.
00:21:53
I'll be sort of waiting until after the conference to stake
00:21:56
out my position partially, just like Get.
00:22:00
Get more information And also, you know I sort of shift between
00:22:06
information gatherer and sometimes opinion haver.
00:22:09
And I don't know. I've been sort of waiting to
00:22:11
come out strong, though I do have a sort of.
00:22:13
Side the last theme I'm actually interested in, and this is very
00:22:17
close to home for James and I in the last couple of months, is.
00:22:21
We are increasingly discovering that the cost of running any
00:22:24
large language model products at consumer scale is unbelievably
00:22:29
high. Like, oh, interesting.
00:22:31
Yeah, is so high that it's it's very hard to come up with a
00:22:36
mental model for how you could build anything on top of a GPT
00:22:40
for quality model right now that would be distributed to
00:22:43
consumers. You could maybe do it on 3/5.
00:22:47
Especially with some recent price cuts with, you know, fine
00:22:50
tuning, yadda yadda yadda. But I would say that it is
00:22:54
shockingly high to run a very basic application that's GPT 4
00:22:58
level quality on LLM. And I think it basically makes
00:23:02
it impossible to build consumer apps on these level of models
00:23:06
today without incinerating money.
00:23:09
And so I think it's an interesting question as to
00:23:13
whether or not a lot of these companies that have raised 100,
00:23:15
two $100 million dollars. I have to imagine their unit
00:23:19
economics are pretty much upside down if they're building any
00:23:21
sort of consumer product. As in every time someone uses
00:23:25
their product they lose more money than they make.
00:23:27
So I wonder if the only outcome of this is that B2B companies
00:23:34
can maybe afford this because they can charge such such
00:23:36
staggering prices. And then?
00:23:39
Open AI is like the only one that can do consumer stuff
00:23:41
because you pretty much have to have no middleman to afford
00:23:44
running these applications. OK.
00:23:46
So that's a great one and I'll broaden it and just say I think
00:23:49
there's an the overall meta theme of that is the rubber
00:23:53
meets the road, right? There's this.
00:23:55
Do people really care about existential risk anymore?
00:23:57
Because it's like, oh, we're just in the practical problems.
00:24:00
Like it's really expensive to run consumer apps and I've heard
00:24:04
people in enterprise ask like are.
00:24:07
Are big companies, customers sticking around or do they try
00:24:11
the demo and say we didn't like it anymore?
00:24:14
And I mean, you know, Chachi, BTI think we're going to see in
00:24:17
a slide during the presentation, you know, activity syncs and
00:24:20
then they have new features and goes up like what's the
00:24:23
resiliency? So just this, yeah, the rubber
00:24:25
mean the road, all right, We've gotten all the hype like how are
00:24:29
people delivering? And now it's sort of the moment
00:24:31
where it's turning into sort of more regular sort of business
00:24:35
style questions. I mean, this stuff is so
00:24:39
expensive to run that it's hard to imagine you would not have to
00:24:42
run into regular business style questions pretty quickly.
00:24:45
Now I was not in the Internet industry in 1993 when you had to
00:24:48
go buy your own servers or on a website, you know, so that this
00:24:54
may be analogous to that time, you know, where you had to be a
00:24:57
garage band server farm to afford to run these kind of
00:25:00
companies. But even that would have been
00:25:03
that would have been fixed cost, right, I mean now we're talking
00:25:06
about. Like we're talking erasing cost.
00:25:08
More customers you get, more expensive.
00:25:09
Yeah, Erasing software margins like.
00:25:13
So like, I'm pretty interested to see if anyone can run a
00:25:17
consumer product without just incinerating money besides Open
00:25:22
AI. So.
00:25:24
Stay tuned or or or use I would say.
00:25:26
I would say the last caveat would just be using smaller
00:25:28
models or. Yeah, using smaller smallers or
00:25:31
fine tuning or something. Or GBT 4 gets 20 times cheaper
00:25:35
in the next year and a half. And then this is a dumb,
00:25:39
irrelevant, you know, retrospective discussion, but it
00:25:41
really has to go down by like a 10X to be interesting.
00:25:44
So they did three XA week ago. We need two more of those to
00:25:49
really be in a good category of cost.
00:25:53
Yeah, we, we should. I mean, that is a good reminder
00:25:55
that we should ask money. Like, how much does this cost?
00:25:59
Like, how? Yeah.
00:26:01
How Where do you make money if like, you know, prices change
00:26:04
one way or the other. Yeah.
00:26:05
Yeah. All right.
00:26:07
Next up, I really think you're going to want to listen to this,
00:26:11
the draft pick. Please give us your feedback.
00:26:14
I want insane like memos about like what the order strategy
00:26:19
should have been, and just feel free to weigh in.
00:26:23
We did, you know, I I think we explained all the particulars in
00:26:26
in James, right. You talked through Yes, yeah,
00:26:28
we're about to jump into the rules and there were, there was
00:26:32
probably more time spent behind the scenes debating the rules
00:26:35
than I preparing for the draft. I was insistent about a
00:26:38
particular rule change that I mean you'll, you'll hear.
00:26:42
So we had a lot of fun. Sent you Eric at newcomer.co.
00:26:48
You can always send me your opinions about who's an idiot I
00:26:52
hopefully. I think I'll try and post the
00:26:54
list. So please forgive us that it was
00:26:57
a set list. There are companies.
00:26:59
I asked if I could pick companies off list as sort of a
00:27:03
insider with Intel. But I was I was not allowed to
00:27:05
pick off list. So I you're restricted to a list
00:27:08
of companies and we didn't pick you, please hire a new marketing
00:27:12
officer. I don't know.
00:27:13
Sorry. Yeah, find find out why your
00:27:17
your valuation was also not, you know, listed in Pitchbug, right
00:27:21
'cause the list was selected from companies who are or who
00:27:26
have have a recent over $100 million rounds in Pitchbug.
00:27:32
There's go. Listen to that.
00:27:37
Welcome. Welcome.
00:27:38
All right. We are gonna move into the draft
00:27:41
we this is probably the thing we're most excited about at
00:27:45
least informed to do taking a page out of.
00:27:50
Sports Podcast, Political Podcast Playbook and we are
00:27:54
going to do a draft pick of the top AI companies.
00:28:00
James, do you want to explain the metrics for who wins this
00:28:05
and how we know and all of that? Here's what we're doing.
00:28:08
Max, Eric and I are drafting teams of AI startups.
00:28:13
The startups must have risen over $100 million.
00:28:18
I wish that wasn't the case, partially because I have Intel
00:28:20
on like companies that are promising, but James isn't
00:28:24
letting me include random. I feel like there's, you know,
00:28:27
it's a risk. They are not even valued at that
00:28:30
yet, but I I, yeah, I think we have, we have 43 companies in
00:28:35
the list and we're only picking five each.
00:28:39
So many of these will not make it on the board.
00:28:42
So I'm excited to see what happens, but this really shows,
00:28:46
I think, the amount of companies who have raised ginormous rounds
00:28:51
in AI. Is this 100 million plus list
00:28:54
you have to have raised raised over 100 million, not valuation
00:28:57
raised. It's probably 100 million
00:28:59
because I I do think also they're probably ones that
00:29:01
aren't. So some other criteria I used to
00:29:05
kind of narrow down this list. You had to be using generative
00:29:09
AI or the company had to be have generated AI or generative AI
00:29:14
infra kind of core to it. I kind of excluded some
00:29:17
industries like bio and healthcare, defense, silicon
00:29:22
chips, robotics, driverless, or a VS and China.
00:29:26
I just wanted to filter down to companies we might have a chance
00:29:30
of knowing something about or being predictive about, rather
00:29:34
than just predicting whether some biotech company is going to
00:29:39
use AI to discover a new drug or something.
00:29:42
And I also wanted companies founded before excluded,
00:29:47
companies founded before 2013. So most of these companies have
00:29:51
been actually founded in the last three years or something,
00:29:54
but there are a few who have been founded maybe earlier.
00:29:58
So that's the game and we are going to determine who goes
00:30:02
first by creating a handicap auction.
00:30:06
So this was, I was adamant, yeah, that we do this.
00:30:09
James and Max were like, oh, you know, that's part of the
00:30:12
randomness. But to me, like a huge part of
00:30:14
this is how how much would you pay to get to pick?
00:30:19
Open AI Because I, I, I honestly think we're all going to pick
00:30:22
Open AI. We should talk about what the,
00:30:24
what the, what the grading criteria is.
00:30:27
Yeah, here. So the way we are drafting these
00:30:31
teams, we are we are trying to create a team with the highest
00:30:35
total valuation on November 1st, 2028.
00:30:40
And final valuations would be determined by pitch book, either
00:30:44
kind of a reported M and a price or a current stock price if if
00:30:48
the company is public or the last evaluation if the company
00:30:52
is still going but is private and if the company is shut down,
00:30:56
I think that would count as a 0 for that team.
00:31:00
If if like 2 year, does it have to have raised at the price
00:31:04
within the last two years or something because?
00:31:09
There's whatever the last fundraise was.
00:31:11
Yeah, just. But like, it might not be worth
00:31:12
that anymore. One question I have is let's
00:31:15
let's say there's a company on this list who has raised, who
00:31:19
has raised a huge round, maybe in the Zerp era but never raises
00:31:23
again, but is a zombie company in five years.
00:31:27
Are we counting that at their? I think the last valuation.
00:31:31
No, it needs to raise another round or do something in the
00:31:35
next five years because otherwise there's a strategy of
00:31:38
just like picking the most highly valued companies.
00:31:41
And just betting that like nothing happens.
00:31:43
That's the efficient market hype.
00:31:44
Out there, I don't think. I think it's theory, current
00:31:47
evaluation, other. Is the best particular of future
00:31:50
value. It needs another either of a, it
00:31:52
needs another fundraiser. Reporting.
00:31:53
Yeah, I think so. Max, are you OK with that or.
00:31:57
I mean, I don't care. I don't care.
00:31:58
Yeah. Yeah, yeah.
00:31:59
So whatever. It needs something, whatever,
00:32:01
right? Are we agreed on that?
00:32:03
Yeah, we're agreed. So if the company does not raise
00:32:07
any round between now and November 1st, 2028, yeah,
00:32:12
according to Pitch Book, a new round, they will count as a zero
00:32:16
as well. Fair.
00:32:18
OK. Yep, brutal.
00:32:21
So how are we doing the bid for 1st and 2nd place?
00:32:28
Well, I think the the idea here is if you you know we're gonna
00:32:34
just go down the line here and name a price that you are
00:32:38
willing to subtract from your total score and then in the end
00:32:42
and then determine you know the order in that in that order of
00:32:47
those 3 handicaps and we will deduct those scores at the.
00:32:51
End. All right, let's just do a live
00:32:52
auction. Like whoever gets the highest
00:32:54
price and then we stop it. OK, so then it's a coin flip.
00:32:57
OK, fine. Coin flip for a second?
00:32:59
Yes. OK, let me pull up my digital
00:33:02
coin flipper. OK, All right.
00:33:06
And then it's a snake draft. Yes.
00:33:08
All right. Let's let's start.
00:33:10
I I bid -45 billion. For open AI.
00:33:15
For first for to go first. For first take, OK, I'll go
00:33:26
I'll, I'll go to 50. Sure, I'll go.
00:33:30
I'll go 55. I'll go. 60 60 billion -60
00:33:37
billion. How is opening eye going to have
00:33:40
an exit? I just am confused by this.
00:33:42
It does they they don't need to have an exit, they just need to
00:33:45
have a fundraise. OK.
00:33:46
I'll go. I'll go to 70. 75 OK, that's
00:33:58
fine. Max, you're done.
00:33:59
I'm done. I'm out.
00:34:00
You got it at 75, all right. 1st pick You have a lunatic open AI.
00:34:05
OK, bad pick. Bad pick now I.
00:34:11
Imagine I did all that to get like anthropic or?
00:34:13
I just, I just I just think the number of companies that are
00:34:16
worth like hundreds of billions of dollars on the secondary
00:34:19
market is just extremely low. So you're you're basically
00:34:21
discounting the net equity value to like 5 or 15 billion now or
00:34:24
whatever it is. So sorry, yeah, added.
00:34:28
Yeah. It's currently valued at 80 or
00:34:29
90 right on secondary markets and.
00:34:31
You took 7075 off that. So you're basically saying I'm,
00:34:35
you know 1010 is where I my starting point and I have to get
00:34:39
another 100 billion or whatever on this valuation?
00:34:41
If it's, is it a trillion dollar company like?
00:34:45
Yeah. Then I crushed it like.
00:34:47
But are they going to be a trillion dollar company as a
00:34:49
subsidiary of Microsoft? I just don't believe that.
00:34:52
But that's fine. This is why it's an exciting,
00:34:54
contentious pick. Yeah, great.
00:34:56
OK, All right. Eric's going to go 1st, and Max
00:35:00
and I are flipping for 2nd place.
00:35:02
Max, name your head or tails here.
00:35:04
I'll take heads. Heads.
00:35:06
You're up second. OK, Eric, we know what you're
00:35:08
picking. Do you want to make it official?
00:35:10
Open AI Congratulations knew what he was doing.
00:35:15
Smart guy. Smart guy.
00:35:17
Anything you want to, Anything you want to add there, or.
00:35:20
Well, I I do also think the fact I think Microsoft be an investor
00:35:24
helps drive up the valuation because they're getting spend
00:35:27
back on their system. There's all this revenue round
00:35:30
tripping going on. I think it'll definitely raise
00:35:32
another round or two. So I feel like it's super safe
00:35:35
bet. I think I'm getting it if we
00:35:37
know it's reported to be 90 billion.
00:35:39
I'm getting it like, yeah, with. With.
00:35:41
But if Microsoft buys, if Microsoft buys an additional 2%
00:35:46
to go from 49% ownership to 51% ownership, that's that's like an
00:35:49
exit in my opinion. Like that's an exit.
00:35:51
I agree. Yeah, yeah.
00:35:52
What do you mean no? They have majority ownership of
00:35:55
Open AI. I mean it's it's a scenario of
00:35:57
Microsoft at that point. No, it's just over then, like.
00:36:00
Yeah, I think. So no, these are if there are
00:36:03
deals on Open AI, they value it as more.
00:36:06
It's worth more, definitely. If other people, if people are
00:36:09
getting shares at a higher price.
00:36:12
There there will be no shares of Open AI, though in this context
00:36:14
I don't know we can we can see what actually plays.
00:36:16
At it, the valuation goes up and there are people who own a lot
00:36:19
of it, then yeah, it it has its own valuation.
00:36:23
There's no way that you're saying, well, well, I I I guess
00:36:27
the shares have to dissolve is what you're saying for it to be
00:36:29
an acquisition that's kind of. Yeah, yeah.
00:36:32
Once they're in Microsoft, I accept that.
00:36:35
Like I don't get to ride up. The Microsoft benefit.
00:36:38
Of. Even though I do think that's a
00:36:40
possible outcome, but. I I just don't think they ever
00:36:42
make it to whatever a trillion dollar valuation is a stand
00:36:46
alone. I don't know if Sam Altman would
00:36:48
sell, oh, you know, away Control to Microsoft.
00:36:51
That's I don't think so either And like there's antitrust and
00:36:53
like I think this is the most Microsoft is probably going to
00:36:56
own. So, but I I'm not willing to
00:36:58
accept like if they own 60%, I'm like no, that's fair, that's
00:37:01
fair. OK.
00:37:02
All right, all right. All right, Mac.
00:37:04
Max. I will take data bricks at #2
00:37:11
Yeah smart, pick smart I. Like I like Data Bricks, I I
00:37:15
like to think that I mean the goal here is obviously to really
00:37:20
get a a couple huge winners or one really huge winner, right.
00:37:24
And I think that Data Bricks is obviously already a very large
00:37:27
company, you know 10s of billions of dollars and I think
00:37:30
there's probably many 10s more or or potentially hundreds in
00:37:33
headroom there and so. Yeah, I don't know.
00:37:36
I think it's a good call option on a really massive company.
00:37:38
So, So I'm grabbing data bricks and also my favorite speaker
00:37:41
from CV1 and Ali goes to the CEO and he's coming back for CV2.
00:37:46
So I'm just a little Homer pick here.
00:37:48
Smart. Yeah, this is giving me now.
00:37:50
My stomach's turning. I do data bricks like I've been
00:37:52
super bullish about them for the long time.
00:37:54
So not to have them in my list is it hurts.
00:37:58
All right, James, pick 3. All right, I've got two that
00:38:01
I'm. You get two in a row, Snake
00:38:04
draft. You get two in a row.
00:38:05
I get two. OK, I just do both of them.
00:38:07
Yes. All right.
00:38:08
I am going to go with hugging face as my first pick.
00:38:13
Wow, that's that's high. That's high.
00:38:15
I think you could have gotten the value bet later in the later
00:38:18
in the draft there. No, I might have.
00:38:20
No, it sounded like Eric was going was eyeing A hugging face
00:38:23
as well. Yeah, I think.
00:38:25
I just, I just really think Hugging Face is going to exit it
00:38:27
under 20 to Microsoft basically. So that's well that's good for
00:38:30
me though that's like a pretty good lock in of 20 million of
00:38:33
you. Need you need to stack those
00:38:35
hundreds to win this draft, James.
00:38:37
We're going up against the trillion dollar open AI here.
00:38:40
So then you should only pick foundation model companies, you.
00:38:43
Should pick whatever you think can go to a trillion Eric, you
00:38:45
know it could be anything hugging face has a lot of room
00:38:48
for I think potentially getting to that 100 billion level if
00:38:53
they kind of maintain with their position in in essential you
00:38:58
know position in the open open source community and they
00:39:02
potentially even build tooling around this that helps other
00:39:06
companies use open source in their in their enterprises.
00:39:10
And yeah, I just think there's a pretty high ceiling there.
00:39:17
GitHub sold for 7.5 billion. Yeah I'm just, I'm just saying
00:39:20
GitHub is is everything hugging faces and more in my opinion and
00:39:23
they decided to exit at the top at 7, 1/2.
00:39:26
So I don't know. I mean, I I love hugging face
00:39:28
for the record, we're, you know, big fans, but I don't know, not
00:39:31
not sure about the pick on the upside.
00:39:33
And there are forum at the end of the day like.
00:39:36
Yeah. All right, James, second pick,
00:39:37
hosting. I I mean, we love them.
00:39:38
But for my second pick, I am going with Anthropic.
00:39:44
Huge second runner in the foundation.
00:39:47
Model Wars, I would argue, just recently struck a pretty major
00:39:52
partnership with Amazon and has a lot of the same people who
00:40:00
were originally working at Open AI.
00:40:02
So I think, yeah, and I've heard really, you know, positive
00:40:06
things from developers about Claude in general.
00:40:09
So yeah, I'm really excited about Anthropic.
00:40:12
OK, this is a tough pick. I'm going to, I'm going to skip
00:40:17
through the really high fundraisers here for my second
00:40:20
pick, go a little more deviant. I'm going to take Pine Cone, the
00:40:25
database company. I think that what I'm hearing
00:40:30
from my little birdies, Little birdies in the, in the AI
00:40:34
communities, Pine Cone is the the database of choice for AI
00:40:40
developers of all stripes. And I think if you look at the
00:40:44
history of Silicon Valley, you see really strong exits from a
00:40:48
lot of different database companies.
00:40:50
So yeah, I mean I I like Pine Cone as I don't know, I mean
00:40:54
this would be a really good pick if I could get you know
00:40:57
discounted for the fact that their valuation is way lower
00:40:59
than most of the companies on this list.
00:41:00
But regardless, I'll take the upside.
00:41:03
I, you know, as a reporter does went went to a VC for advice
00:41:07
before this. I got one because I asked at the
00:41:09
last minute but he said pine cone not there on his list.
00:41:13
He only gave me his five because I think her #3 enterprises are
00:41:18
wasting tons of time building toy prototypes that will
00:41:21
ultimately be replaced. All I all I care about is just
00:41:26
the crazy upside case and I think that's that's what I'm
00:41:29
going for here, so all. Right.
00:41:30
I'm gonna do I get 2 now? I don't know.
00:41:32
Do you get 2 now? Do you get to?
00:41:33
Yeah, you get 2. Now you get to swing.
00:41:34
Back. I'm like, nervous.
00:41:36
I'm gonna take inflection. Please explain.
00:41:46
You got to defend your picks here, Yeah?
00:41:52
With the idea one sort of top talent obviously like DeepMind
00:41:57
Co founder Mustafa Suleiman. I'm not picking him using
00:42:00
speaking but you know I I do think super legit company.
00:42:06
I mean obviously you're betting a lot on the future.
00:42:09
I do think you know if it's go for big swings, huge foundation
00:42:14
model potential is key smart technology I think.
00:42:21
I think also just like give it, I feel like there's a floor and
00:42:24
that it's also like a potential like acquisition to just like
00:42:27
get some big tech company capable worst case.
00:42:33
So yeah, I'm going inflection for #2, OK.
00:42:39
And this is a tough one. All right.
00:42:42
I'm taking character. AII Just feel oh man.
00:42:45
I feel like that bad pick bad. Pick.
00:42:49
You would have taken it, right? Yeah.
00:42:51
Bad. You would have taken it.
00:42:53
Or the pig bad. No, no, no.
00:42:54
Bad pick is just a way I'd joke about every pick being bad.
00:42:57
Sorry, would you have taken character like I never would
00:42:59
have, but it sounds like James would have so.
00:43:01
I think I would have taken it, yes.
00:43:04
I'm going on a foundation model, forward strategy and I I think
00:43:09
character, you know, has. Perhaps a clearer path to a
00:43:13
product than most, I think. I mean they have a very talented
00:43:17
CEOI mean Theresa Horowitz is going big on them.
00:43:21
So what they're going to get lots of follow on funding
00:43:24
rounds. Yeah, we'll see.
00:43:26
I don't know. I you know something I I still
00:43:31
think like the, you know, right now the bot sort of relationship
00:43:35
bot thing is a little overhyped, but.
00:43:38
You know, there's time. What do you think about
00:43:40
character as it is today? What is your?
00:43:44
Do you agree with The Take that it's all sex chat or or?
00:43:47
Not, I don't, I don't know, I'm not like an expert on the
00:43:49
product. But to me there there's sort of
00:43:52
you got a dual bet in that they're super hardcore building
00:43:55
their own foundation model deep on technology and they're
00:43:59
actually have a product that's out there that people are
00:44:02
finding fun ways to use. And like, I mean, you know, Mark
00:44:05
Zuckerberg, that's sort of with hot or not so like anything.
00:44:08
That that they have the sort of instinct to say like let's get
00:44:11
users, that that to me is like a good, good muscle.
00:44:15
And like, even if this first iteration isn't like the one you
00:44:18
want the New York Times to be writing about, I think I think
00:44:21
it's a good sign. I think it's a good sort of out
00:44:25
of the money. You guys would not have.
00:44:26
Picked. Inflection though?
00:44:29
No, I I think. Inflection's a reasonable I like
00:44:31
your picks. I like your picks.
00:44:32
I don't. I honestly don't really know
00:44:34
what these foundation models, which is why I'm going to pick a
00:44:36
foundation model next, baby. Yeah.
00:44:42
I mean, I literally think I could be choosing foundation
00:44:44
models. Pass open AI and anthropic as a
00:44:48
a monkey. Throwing darts at the proverbial
00:44:50
board would do better than me here, but I'll just go with Co
00:44:53
here. I'll just take the market
00:44:54
deciding it's worth a lot. It seems like it's the number 3
00:44:57
#4 best foundation model. Who the F knows is my defense of
00:45:02
this pick, but I got to grab a foundation.
00:45:05
Everybody wants to be in business, so I feel like they're
00:45:08
going to face a ton of competition.
00:45:10
They were early, sure. Yeah, I'm.
00:45:13
I'm not. I'm, I'm.
00:45:14
I'm freely willing to admit this with them.
00:45:15
I'm I'm freely willing to admit this is my lowest conviction
00:45:18
pick here. But I just feel like I got to
00:45:19
have a foundation model, and that to me is the best pick on
00:45:21
the board. So so be it.
00:45:23
Onward and upward. Maybe I'll get lucky.
00:45:25
OK, I'm. I'm ready with my 2 picks.
00:45:27
All right? Please fire away.
00:45:28
All right. 1st pick go go on with another Foundation model of
00:45:33
course. Got to have one.
00:45:35
I will say you can see why VCs are investing big in foundation
00:45:40
models, because if the goal is to go for the fences, you're
00:45:43
like you know, foundation models are are are you know big swings.
00:45:48
And so, like nobody here is really so far trying to
00:45:51
accumulate like a but like I could name some sure things but
00:45:57
then. You know you want like a swing
00:45:59
anyway, James. I'm going with AI 21 Labs, which
00:46:04
is a Israeli foundation model company.
00:46:09
Solid. Just just announced $155
00:46:13
raise in August at a $1.4 billion valuation and.
00:46:21
The. Participants included Google and
00:46:25
NVIDIA. They create foundation models
00:46:29
particularly targeted at writing, and they are known for
00:46:34
their cutting edge Jurassic models.
00:46:37
So yeah, I'm impressed with the founding team and the product.
00:46:41
Yeah. And they they've got some, I
00:46:42
think they have a partnership maybe with Amazon, like I've
00:46:44
seen them come up, but it feels like they're under the the radar
00:46:48
a little bit. I think it's a little bit of an
00:46:49
under the radar pick, but a solid that.
00:46:53
Makes sense? Foundation models, I mean I
00:46:55
would say that is, you know, if if they are Israeli and they're
00:47:01
valued at 1.4, probably in the US they'd be valued at 14
00:47:04
billion. So I think, yeah, exactly like
00:47:09
you know that that seems like a good, good, good cost of living
00:47:13
adjustment right there. All right and with my next pick
00:47:17
I am taking Replet. I'm Jod.
00:47:21
Yeah, another CVCVAI alumni Alumnus if you will.
00:47:28
So replica. You have ugly Face and replica.
00:47:31
They're on the same panel together.
00:47:32
You're just a you're just a big CVAI chapter here.
00:47:36
I love it. I like to meet the founders and
00:47:38
get to know them. Practicing for In the end, it's
00:47:46
really, it's really just a bet on the founders.
00:47:48
It's really about the people. It's about the people.
00:47:49
It's really about the founders. What do you think about this
00:47:52
stuff? Yeah, Founder market fit, if you
00:47:54
will. All right, What?
00:47:55
What's your? I'm definitely as founder.
00:47:57
Twitter are are repla and hogging face like in the like
00:48:01
trillion dollar case. They're like fighting with each
00:48:03
other right? They're both like in the coding.
00:48:05
World, I don't. I don't agree with that.
00:48:06
I think that repla, the trillion dollar case here is that
00:48:10
software engineering. Changes dramatically due to AI
00:48:15
and we enter this world of everyone's an engineer, you
00:48:20
you're using a cloud based IDE, you can use AI really
00:48:25
aggressively as you're coding and you don't even need to
00:48:28
really learn how to code. And I think they're the ones who
00:48:30
are furthest along at kind of building that vision that I, I
00:48:34
do believe is possible with AI that was that was not on my
00:48:37
draft board, all right. Yeah, I don't think I would have
00:48:41
picked it. Yeah, I'm going to go with
00:48:44
another Homer Pig. I will be taking Modular,
00:48:48
another founder. I will be interviewing on stage
00:48:51
at CVA I2 for the for the back story here.
00:48:54
For the back story here. Chris Lattner, who Co founded
00:48:57
Modular, literally invented the Swift programming language that
00:49:01
is on every iPhone, every Mac, every iPad in the world.
00:49:05
Probably one of the top five to 10 most used programming
00:49:08
languages ever invented. So he's obviously a beast.
00:49:12
He has shown tremendous facility in creating really low level
00:49:17
technical products that are used by literally hundreds of
00:49:20
millions of people, if not billions.
00:49:22
Before that he created a tool called Low Level Virtual
00:49:24
Machines that is also basically everywhere.
00:49:26
So I'm just saying Chris Lattner, he gets WS and he is
00:49:31
trying to reinvent the programming language for AI and
00:49:34
replace Python with a new language called the Mojo that
00:49:37
modular is invented. And so, look, I frankly don't
00:49:40
know what the business model is here or where we're going with
00:49:43
this part, but you know, Chris Ladner puts points on the
00:49:46
scoreboard. So that's what I'm hoping he'll
00:49:48
do for my team here. So modular.
00:49:49
AII am breathing like a huge sigh of relief because I took
00:49:54
character instead of this company thinking the character
00:49:56
was going to be a more. Competitive pick and I think you
00:50:01
were right, proved true and this company did not get scooped up.
00:50:05
I am taking what I would have actually probably put as my
00:50:09
third pick, but strategically dropped to 4 clean which is sort
00:50:14
of doing corporate search. Sequoia Back Company,
00:50:17
Lightspeed, Ravi, the key guy at Lightspeed is on it, and Mamoon
00:50:23
Hamid at Kleiner Perkins is an investor in this company.
00:50:26
So I just feel like it's one of these, Like, I don't know.
00:50:29
What do they all know? So you were really like a
00:50:32
venture capital fundamentalist here.
00:50:35
Like, yeah, like, you know, Eric's like, have you ever heard
00:50:40
of that fallacy of appealing to authority?
00:50:42
Well, I go the other way and say it's the way.
00:50:45
Well. Like, you know, like you should
00:50:46
always appeal to authority. I mean what we don't have the
00:50:50
financials, what are you doing? You're looking for a signal
00:50:52
like. I mean what is better signal
00:50:54
than like? Sequoia, Mamoon and Ravi like I,
00:50:57
you know I don't know what it is.
00:50:59
I think they're good VCs and like I I don't have metrics So
00:51:02
what am I going And honestly my VC guru that I went to here's
00:51:06
his case for glean semantic search over business data is the
00:51:10
biggest use case. This opportunity overshadows all
00:51:14
the in house builds people are using link chain etcetera for.
00:51:17
So instead of instead of cobbling it together yourself.
00:51:22
Use glean like I, I, I buy that. Wasn't, wasn't this the Tanium
00:51:26
pitch, which also like you know, is sort of, Oh yeah, laying
00:51:29
around? You know you should.
00:51:30
You need to look into the eyes of the Co founders, right?
00:51:33
Like, are you gonna actually help your investors make money
00:51:35
out of this company anyway? OK, well there you go.
00:51:40
OK. So now I have a last one, This
00:51:43
one I'm I'm debating between like product that's solid,
00:51:47
protect my sort of minimum versus like upside.
00:51:51
Potential man, this is like, I I have empathy for VCs.
00:51:56
It's like, should I go crazy, like hoping to dream, you know,
00:52:02
or should I go sort of like product that people actually use
00:52:07
today. I'd like to think you would do
00:52:09
more than 10 minutes of research, which is what we've
00:52:11
done on this. I don't know.
00:52:13
Did do VCs do that? I don't know though.
00:52:16
They get, they get the financials.
00:52:17
We don't get any. I guess some of these companies
00:52:19
haven't like you. You already did what most VCs
00:52:23
do, which is as their VC. Friends who they should invest
00:52:26
in, just go with that. It works.
00:52:31
Lots of people, a lot of money doing that.
00:52:35
All right, I. I'm going for the the hope and
00:52:40
dream of the entire European continent.
00:52:43
Mistral AI Oh, my gosh. I thought you were going to say
00:52:47
something else, but yeah. Yeah.
00:52:49
Vladimir Zelensky. All right.
00:52:52
Mistral is, you know, super hypey.
00:52:55
I think Lightspeed is an investor European, you know,
00:53:00
they they're very early, super, super hypey, but you know.
00:53:07
Smart, smart founders. I'm really enjoying this draft
00:53:10
because I really would not have picked almost any of the
00:53:12
companies you guys picked. And I feel interesting.
00:53:14
That must apply to me too. So yeah, we can review each
00:53:17
other's teams at the end here. I know who these.
00:53:19
Six would be. So I'll tell you at the end I'm
00:53:21
gonna, I'm gonna I'm I am also going to go hoping to dream here
00:53:25
a little bit more. So I obviously read you know
00:53:30
read and researched all the founders of these companies and
00:53:33
I was very taken by this founder's vision of self
00:53:37
improving foundation models. So the idea of recursive
00:53:42
improvement of a foundation model and so that was the pitch
00:53:46
of imbue our friend Ken who on a hope and a dream and I assume
00:53:53
some really solid demos has raised $234 million.
00:53:57
So so but I think if you're going for true take off trillion
00:54:03
dollar companies as it were the first person to figure out how
00:54:07
to make AI recursively self improving Will has a good shot
00:54:10
at being a trillion dollar company.
00:54:12
So I'll, I'll, I'll put some some chips on Kenjon and see
00:54:15
what happens. Yeah.
00:54:16
I think they're very bullish on like agents, right, I mean,
00:54:18
right and self improving agents as well which I thought was
00:54:21
exciting. So I'll throw, I'll throw $5
00:54:25
into that $234 million pot and help them get there.
00:54:29
All right, James, Final pick, final pick.
00:54:32
All right. With my final pick, I am taking
00:54:35
adept, interesting, very highly valued cerebral values.
00:54:42
CV alum, Yeah. Obviously super high valued and
00:54:46
highly funded. As far as I can tell, they've
00:54:50
raised, yeah, close to half a billion dollars, right?
00:54:53
Woo Hoo, Hoo Hoo. Essentially they are they are
00:54:56
trying to create. A new way to interact, Have AIS
00:55:03
interact with computers. Essentially allow creating
00:55:06
agents that can control computers.
00:55:08
I don't even know I I put this as a copilot I guess, but I
00:55:12
don't even know if this is like this.
00:55:15
Seems like a brand new type of concept.
00:55:19
You know, creating a agents, AI agents that can interact with
00:55:22
computers on your behalf. It's a pretty science fictiony
00:55:26
idea, so you know, could be a. Huge flap.
00:55:31
Or it could change the way you know the world.
00:55:33
Works founders have left in droves.
00:55:35
I feel like lots of internal turmoil.
00:55:39
I feel like. It was hiked earlier other found
00:55:43
you know there there was a founder break up at the company
00:55:47
which which is certainly a negative sign but little scary
00:55:52
but but you never know stop off from from the ashes rises the
00:55:55
phoenix. I mean the.
00:55:58
The CEO is, you know, David Luan, a former I think VP at
00:56:02
Open AI. Certainly is there sort of early
00:56:05
days, super smart guy. But yeah, can't.
00:56:09
Can't keep the original team together.
00:56:12
Interesting, I just think. I just I think working on AI
00:56:15
agents in some capacity that can like to your earlier point about
00:56:20
self self improving AI and you know reinforcement or maybe Max
00:56:26
you were talking about that. I think there's something there
00:56:29
that could unlock a lot of value, so I'm excited that
00:56:32
they're working in that area. Sure, got it.
00:56:35
My zoom out I'm I'm most stressed that I should have just
00:56:41
hoped for a second and gotten data bricks for.
00:56:44
Cheap because I do think next you got a great deal on this
00:56:46
Section 1 there, yeah? And data bricks is like a pretty
00:56:49
sure thing, almost more sure than opening eye.
00:56:52
I feel like the upside. Our producer says he's picking
00:56:58
the script because we're editing the podcast with that and also
00:57:00
that was one I was seriously considering for five working
00:57:04
product, but I didn't know how much technical under like what
00:57:08
their level of. Heck, sophistication was so I
00:57:11
didn't pick it. What?
00:57:11
What are, what are other ones on the board, left on the board,
00:57:14
Anything you guys are? Surprised, notable.
00:57:16
I mean, Jasper obviously is like, everybody's afraid of,
00:57:19
yeah, rapper. I feel like Snorkel has gotten a
00:57:22
lot of hype with like internal data cleaning.
00:57:25
There's companies. That's what I've heard.
00:57:26
Yeah. Yeah, I they're they're always
00:57:28
bouncing around when I'm hearing.
00:57:31
Writer raised a bunch of money at 500, so I'm
00:57:34
interested to see sort of there was sort of a good price.
00:57:38
Lots of money back. Lots of money.
00:57:40
Good price for investors. It's it's interesting, humane.
00:57:43
Which makes that little pin that's supposed to replace your
00:57:45
iPhone that you wear on your jacket or whatever that you talk
00:57:48
to. Nobody put that on the board.
00:57:51
That seems like a a tough, tough sell Hard, but.
00:57:55
Hardware is hard. I just hardware is very hard.
00:57:57
Yeah. Yeah.
00:57:58
They just launched it at Paris Fashion Week, which is odd.
00:58:01
I don't. Know I mean scale you know was
00:58:03
making bank off of. Growing humans and self driving
00:58:06
car problems and now they're trying to make bank at human
00:58:10
reinforcement learning for foundation models.
00:58:13
But it's just hard to believe, you know, it's humans.
00:58:16
Like they just don't seem like that.
00:58:19
You want? We don't need.
00:58:20
We don't need them anymore. We.
00:58:21
Don't need. I know I want a big like don't
00:58:23
need this middle ground thing. I mean stability not being in
00:58:27
the discussion, you know compared to last six months ago
00:58:30
is is quite interesting. Obviously there's been, well,
00:58:33
first of all, there's been a ton of competition and you know, Mid
00:58:36
Journey has gotten super good and all that stuff.
00:58:37
And then secondly, yeah, obviously there was a story
00:58:39
about the founder making a bunch of stuff up, which was
00:58:41
interesting. I think the journey is was left
00:58:45
off the board here because they haven't taken VC funding by the
00:58:48
way, but they could be a super valuable.
00:58:51
Yeah, I know hard to maybe never get a valuation into the cash
00:58:53
flow business, but mid journey man I would definitely have
00:58:56
taken that before Mistral I think interesting the I mean.
00:59:01
Yeah, I think weights and biases really good company.
00:59:04
But I'm just like, what sort of, is it really going to gobble the
00:59:07
world? Runway Love the founder.
00:59:10
Like he's so charming. But, you know, I feel like
00:59:14
maybe, you know, the narrative there is like, does Adobe wish
00:59:18
they could have bought Runway for like way less than it paid
00:59:21
for Figma? You know, when Runway is sort of
00:59:23
the relevant company now? Yeah, I mean, lots of great
00:59:27
companies. I wouldn't write off most of
00:59:30
them. I don't even know what go
00:59:32
student is. I didn't look at him.
00:59:33
That's the most high fund rate. Did anybody research Go Student?
00:59:37
I researched it a bit. So my understanding is they're a
00:59:41
European company who kind of got started in this AI in this
00:59:46
tutoring space and raised a lot of money and now is building LMS
00:59:50
to do tutoring. But I was a little skeptical of,
00:59:53
like, whether that was a, you know, current product or not or
00:59:57
prototype. Yeah.
00:59:59
So that's why I kind of ruled them out.
01:00:00
My last point would just be, I was really struck in doing the
01:00:02
research how unbelievably difficult it is to even tell
01:00:05
what the products of a lot of these companies are or how
01:00:08
they're differentiated. And I mean, you could really
01:00:10
just like throw buzzwords like large model learning like
01:00:15
reinforcement something something internal data,
01:00:18
external data, whatever. It just spit out a new Tech
01:00:21
Crunch article with like a fake company named him every day and
01:00:24
I would find them very similar to a lot of the research I did
01:00:27
for this. So it's.
01:00:29
I mean poolside, you know, like I've written about, it's for
01:00:31
code. I mean partially it's an
01:00:33
earlier, it's like earlier on the foundation model journey.
01:00:36
But yeah, there are a lot where it's like, man, it is.
01:00:38
It is remarkable how many similar ish like either Data or
01:00:43
Foundation model or something about enterprise plus LL Ms.
01:00:47
companies there are that are in this category.
01:00:49
It's it's amazing. To me there's my my zoom out on
01:00:53
my view is that like Max, I mean a lot will fall on data bricks
01:00:59
of course. And that anthropic could be the
01:01:02
people plausibly think anthropic could beat open AI.
01:01:05
In which case, like, James got a great pick discounted open
01:01:10
anthropic. Anthropic is the only one I like
01:01:12
from James. Yeah, really.
01:01:15
Anthropic is when I'm. I want.
01:01:17
I like your, I like your I like your Israeli one.
01:01:20
AI 21 also that's kind of interesting.
01:01:22
Yeah. What?
01:01:22
Which ones do you like from Eric?
01:01:24
I guess the, I mean opening eye obviously is untouchable, but I
01:01:27
think I I think inflection is also quite strong.
01:01:31
The rest, I'm not so sure. Yeah, character.
01:01:33
Man. And whereas your your team's
01:01:36
just rock solid, my my team is Immaculate.
01:01:39
No, I would say other than databrakes, I have no sure
01:01:41
things but I really took I took call options on hopefully giant
01:01:45
upside. Like I could see my team having,
01:01:47
I could see my team having three zeros on it.
01:01:49
But as long as the two are big, I'm, I'm, I'm going for the win
01:01:53
that way, yeah. We we really had to do a
01:01:57
lightning, a speed run of of learning VCs portfolio strategy
01:02:02
here. We we do not have.
01:02:04
The date, you know any we don't get to talk to anybody.
01:02:07
Weird is severe. This is not real VC.
01:02:09
We were playing basically with blindfolds on, but it was very
01:02:12
enjoyable conversation piece. It's kind of funny that we had
01:02:15
to like pick companies that could return the fund.
01:02:17
You know, they all have to. Like, yeah, runs, right.
01:02:21
Yeah. All right.
01:02:22
Very excited to see you in five years.
01:02:24
Hopefully we'll follow up and like some of us will look like
01:02:27
maybe all of us will look like idiots in this space will be
01:02:30
wildly overhyped and maybe my, you know, -75 billion will like
01:02:34
swamp the total value and then we can just.
01:02:38
I don't think that's likely, but as a closer or do you just want
01:02:41
to read off everyone's team? Oh yeah, For people not
01:02:44
watching, the teams are. Team Eric with a negative $75
01:02:49
billion handicap that I paid to go first so I could scoop up
01:02:53
Open AI #1, Inflection #2, Character A, Character AI #3,
01:02:59
Glean #4, and Mistral AI #5. Max you want to read yours?
01:03:05
Yeah. My team is Databricks, Pine
01:03:07
Cone, Cohere Modular, and IMBU. Also known as the best team.
01:03:13
And James. I went with Hugging Face,
01:03:16
Anthropic, AI, 21 Labs, Replet, and Adept.
01:03:22
That's our episode, and that's the Cerebral Valley AI series.
01:03:28
Watch for YouTube videos. I'm sure I'll have Max and James
01:03:31
once I'm like not fatigued of talking back to sort of reflect
01:03:35
on it, but I'm not promising it immediately.
01:03:38
But this has been a great. Run Thank you for sticking with
01:03:42
us. And you missed some of the
01:03:44
earlier episodes. Go back and listen to them.
01:03:48
Thank you so much to Scott Brody, who's been the producer
01:03:51
through all this and help help us figure out the show.
01:03:54
So shout out to him. Thanks to Riley Kinsella, Gabby
01:03:57
Caliendo, both key and sort of getting the conference and
01:04:02
everything together. Thank you for everyone who's
01:04:05
going to speak. And yeah, stick around.
01:04:10
Please, like, comment, subscribe on YouTube, subscribe to
01:04:14
newcomer.co and go play song quiz on your Alexa.
01:04:19
All right? Yeah.
01:04:20
Thanks so much. Thanks guys.
01:04:23
Thank you all. Right.
01:04:24
Bye.