Subscribe
Share
Share
Embed
Fraser & Nabeel explore what it means to build great products in this new world of AI.
Two former founders, now VCs, have an off-the-cuff conversation with friends about the new AI products that are worth trying, emerging patterns, and how founders are navigating a world that’s changing every week.
Fraser is the former Head of Product at OpenAI, where he managed the teams that shipped ChatGPT and DALL-E, and is now an investor at Spark Capital. Nabeel is a former founder and CEO, now an investor at Spark, and has served on the boards of Discord, Postmates, Cruise, Descript, and Adept.
It's like your weekly dinner party on what's happening in artificial intelligence.
Nabeel Hyatt: Do we know what this
thing is gonna be called, Fraser?
Fraser Kelton: No, but we decided that the
world definitely needs another podcast.
That is exactly what,
exactly what's required.
ChatGPT came up with some names.
They were all pretty bad puns playing on
podcasts, AI and startups, none of which
had any taste . I forget what they were.
Nabeel Hyatt: If you can pull them
up at some point, maybe towards the
end of the segment, it'll be great.
So, I am Nabeel Hyatt.
Fraser Kelton: I'm Fraser Kelton.
Nabeel Hyatt: We are doing a thing that
is at least spiritually connected to
a podcast that I used to do with my
old friend and partner, Bijan, when
I basically first got into venture.
We started a thing called Hallway
Chat and the goal was not to be,
I mean, very clearly, not to be
the most well structured, highest
production value place on the block.
The goal was really just to lead with
authenticity about them really crazy time.
At the beginning of the mobile
revolution, where it felt like everything
was changing every single week.
And.
of course, now we have.
a new time.
in AI.
That.
feels an awful lot like that.
And I've got a new partner in Frazier.
Who's uh, now the new VC, who's a former.
Founder and an operator , actually Fraser.
Do you want to talk about yourself at all?
Fraser Kelton: No, I do.
Nobody wants to hear us talk
about ourselves on this.
I don't think.
You say this to founders all the time?
There's nothing worse than VCs bloviating
about themselves and their ideas.
And so we have to keep
the second part of this.
Otherwise we don't have a podcast,
but we can save people from the first.
Nabeel Hyatt: There's these things
called, search engines, although we will
discuss the potential disruption and
those such things where you can look
up our bios and figure out who we are
And if tradition holds and every
week, we're going to focus on our new
AI product that we're actually using
and trying to wrestle with and any of
the insights that come out of that.
Uh, just crying, frankly, stay curious
as we try to navigate all this new space.
And of course.
Frazier we'll navigate the
new venture landscape..
together.
Uh, you can try and keep the
one venture question per week.
Fraser Kelton: Okay.
Yeah.
Yeah.
try my best to limit it to one.
This is a weird world.
So you, you told me the other
week that you use perplexity
for basically everything that
you used to use Google for.
Nabeel Hyatt: Yeah, we did
have this conversation.
I, I don't use Google anymore
really as a search engine.
I, I find it's just not good enough.
That doesn't mean I don't use Google
because it turns out there's a lot of
things Google's done over the years that
are not about search engines per se.
But pretty much all of my searches
have gone to perplexity and then
some percentage of what you would
loosely categorize as search.
Now goes to Claude or ChatGPT.
And yeah, I, I think perplexity
I think there's rumors about IVP
having invested, that we're having
the information this week, some
large round and so on and so forth.
And, on the surface it feels
insane to to be investing and
looking at something that's trying
to compete directly with Google.
But at the same time, how many amazing AI
products have you actually used and stuck
in the last nine months since ChatGPT?
And, and the obvious one that people talk
about is probably CodeComplete, right?
But, but
Fraser Kelton: massive numbers,
user numbers this week.
So I think it's it's GitHub copilot.
It's chat GPT.
There's a lot of interest early on around
RAG like systems for chatting with your
docs chatting with enterprise content.
Nabeel Hyatt: talk to my PDFs.
Fraser Kelton: I think so.
I want my PDFs to talk back to me though.
That's what I want.
I want the synthesized voice of
Sky to come back and tell me that.
That I'm writing this memo wrong.
But, talk a little bit about perplexity
and the product experience and the
user like value that you get from
it versus chat GPT versus Google.
Like it's interesting to hear that
you go there all of the time for what
used to be, quote unquote search.
Nabeel Hyatt: I think it's a great
example of sometimes little things matter.
I think Arvind and that team have
done a lot of subtle things that
make that product really work
that are worth chatting about.
First of all, the, the like six
months ago when agents were suddenly
all the thing for a minute they
launched a little, a little agent
which takes your search engine query.
and then tries to ask follow up
clarifying questions in order to narrow
down and make your search better.
And it's a great example of
does it work all the time?
No.
When it works, it's absolutely magical.
And so, you're searching for something
and then it actually, in dynamic
real time, builds a little UI of,
say, checkboxes or radio buttons that
say, oh, did you mean this or this?
Now, obviously, in the world of agents,
everybody's figuring out exactly, it's
almost like agents are like self driving
cars you can demo them, 80 percent in
a wonderful demo in two months, but
actually getting them to be really
performant and reliable is a, is a
Herculean task, as I think all our builder
friends are, are figuring out right now.
But, but I think it actually works
very well in Perplexity's case.
And not least of which is a good
example is Have you used Google's new
generative engine that they launched?
The conversational
they're basically their
competitor to Perplexity.
Fraser Kelton: yeah.
SGE Search Generative Experience.
I will tell you, it was the
first time that I really had
to reckon with two things.
Number one how painful it is
when you have to ship your org
chart in your business model.
Because, yeah, you can now,
you can now use it to do your
first draft, they call it.
So you go to the search engine that
you've used for 25 years and you
can write write me a poem about...
Tulips.
And it will write the poem about tulips,
and then you have your search results
for, for poems about tulips at the bottom.
Like it just makes
Nabeel Hyatt: Oh, wait, just they still
put in the Google searches, right?
With the Google ads and
things, it makes no sense,
Fraser Kelton: it makes
no sense, no sense at all.
And the, the other thing that I
realized was They're vulnerable and
they realize that they're vulnerable.
If they're willing to ship this type of
janky experience into their core product
they, they don't know what to do because
they have a business model, which is,
maybe the best business model of all,
of all time, that's at real risk here.
And their solution is to allow
you to do your first draft within
the search results below it.
I think maybe the
optimistic story on this is.
They're shipping early, they're going
to learn, they're going to iterate
and they're going to, to figure it
out because it's so economically
important for them to do it.
But I think that this is the first
time in the past 11 months where I
thought, Oh Oh, this is going to be
a really interesting couple of years.
I mean, and that's the optimistic
story for, for Arvind, right?
And perplexity.
Nabeel Hyatt: Yeah, I think it is.
In general with AI products, I've
just tried to take the idea of slowing
down all of my computer use, right?
That's how I've thought about the
next two years, which is just if I'm
gonna go do a search, then one out
of every three times, take the time.
Do the search on three or
four different engines.
Try weird stuff.
And yes, that slows down everything,
but you have to test your defaults.
If you're gonna, if you're gonna be on
the edge, if you're gonna really feel
how all these things are really working,
and you'll take the time to do that, and
have to fit into your day to day life.
Otherwise, it's some random thing
for a half an hour on a Thursday
afternoon, you're like, I'm
supposed to try this product now.
What's my search?
And that just doesn't
Fraser Kelton: Right.
Nabeel Hyatt: I, I, so I've tried,
I've probably done dozens of, of...
of searches on Bing, the new
Google ish thing and Perplexity.
I had a weird one where, I
wear a beanie all the time.
I saw Brad Pitt wearing a beanie.
I was like, where does he get
his, where does he get his beanie?
And if you just think about
the old world of Google, where
would that content come from?
That content would come from
some content farm somewhere.
Someone at BuzzFeed wrote a listicle
Fraser Kelton: Mm hmm.
Mm
Nabeel Hyatt: about, about Brad Pitt's
beanies or, or celebrity beanies,
and then, and then Google scrapes
that, and the economic bundle is
obviously that I hopefully click
on it sometimes, and then, and then
BuzzFeed or whoever else gets a penny.
This whole model , is going to completely
break down because, one, Google, those
searches were way worse on, on their
website it was slow, it gave back only
one answer, not all the answers, and
Perplexity just did a very, very good
job of I don't know what they're doing
under the hood, but very, very good job
of, I think, just doing rag across a deep
corpus set of search results retrieving
out the right things and then actually
presenting them in a good format.
Now, if I have to fast forward a couple
of years, I don't think it'll even
look like a search engine page, right?
It should.
Eventually look almost like that BuzzFeed
page, just dynamically generated,?
It will, it will be nicely laid out.
The photos of the
beanies will be on there.
I'm using a really bad example, but
these are the kinds of ridiculous
things that people search for
randomly on a Friday afternoon.
Fraser Kelton: yeah,
Nabeel Hyatt: And and I also
like, one thing we talked about as
well, I want to come back to is not
all my searches go to Perplexity.
And I think a partner, Yaz, brought
up, and we were talking about the
hallway, is there really going to
be one search engine in the future?
So do you differentiate what
you go to chat GPT for right
now, as an example, versus what
you go to Google for right now?
Fraser Kelton: yeah, when I want to
learn something I go to ChatGPT, right?
So a lot of our job is, is being
curious and trying to understand things.
I no longer go because Google makes you
do both a lot of work to pull out the
information as well as like people have
spent, I don't know, two decades now
trying to game the Google search results.
So that has become, I don't know, this
adversarial thing where you're trying
to put in the work to learn how to.
To get to the answers that you
want, and there's a whole other
side of the equation fighting to
give you the junk on the SEO side.
I used to have some hacks,
everybody does, who's an early
adopter of a pen site reddit.
com, if you want to find out
what the, the best thing is.
And you can now ask ChatGPT
and it gives it to you.
Oh yeah probably the people that I have
the most respect for as early adopters
in terms of taste and product sense and
like discerning input on these types
of things have basically said the same
thing as you about Perplexity , which,
which really caught me off guard.
For me, it's a creature of habit
and then the power of defaults
and distribution where I have
Nabeel Hyatt: Of course.
Fraser Kelton: actively fight that.
And so I'm still finding myself going to.
to Google.
And then when it's an important enough
search, this is, I mean, maybe the
optimistic story, when it's an important
enough search that Google lets me down,
I find myself then going to perplexity.
And how many times am I going to
do that before I just go right to
perplexity or something similar to that?
To your question of when I go to Google,
it's if I want to know how late the
restaurant is open down the street, or
I want to it's, it's all the stuff that
you've mentioned earlier that are, are.
Appendages to the original type of
reasons that we went to Google, right?
Is the one box type of stuff,
Nabeel Hyatt: It is one box.
I think Google as an internet
search engine is terrible, it's
not just because of AI and being
able to ask questions of things.
It's also just the degradation of
content on the internet, which of course,
when there's a bunch of AI generated
content, will only degrade faster.
I think I encapsulate that there's
basically three searches, and maybe
they end up being one product that
rules them all eventually, but it's also
possible sometimes there's bifurcation,
and we see this happen in industries.
And so maybe these end up being
three different companies.
The first one is, I want one box results.
I know it's a one shot answer.
It's probably eight words.
You probably know it.
I know you know it.
Just, just tell me.
And, and Google has done a very good job
of largely ripping a bunch of Yelp and
other people's content off turning it
into beautiful little one box results.
I still go to Google
for that all the time.
Second is inquiry, which
you use the word learning.
When I know that the first answer is
probably wrong, or I will be confused
by it, and I want to ask follow
ups, then I go to GPT or Claude.
And then the third is research, which
is close to learn but different,
where I just want an answer, but
it's a big multivariate answer.
That is, the research questions,
which is obviously the, actually,
Perplexity does an amazing job
on, on the Reddit style questions.
And, frankly, academic
scholarly publications as well,
it does a really good job.
So for me research goes to
Perplexity, OneBox goes to Google
and, and inquiry goes to ChatGPT.
But this is...
This is early in the kind of muck of
everybody messing around with this stuff,
so I'm curious to see how it goes in a
Fraser Kelton: Yeah.
Somebody was telling me that the, the
the pessimistic story for all of this
is that Google and now OpenAI have so
many resources and, and some semblance
of either distribution or brand that
they may not innovate on the right
product experience, but they can just
rip it off once somebody delivers
something that's like clearly resonating.
And we'll see.
I think that that's what makes this
time so fascinating to me is in
2008, 2009, I think you and I would
have had very similar views on how
the mobile landscape was going to
play out between Android and iOS.
Maybe we'd get the relative
share a little bit wrong.
I think in this case, nobody knows
anything and there's a lot of big Big
markets and products that are up for
grabs for the first time in a long time.
Nabeel Hyatt: I mean, this is true,
despite what Sam Lessin says, that I
think there actually will be an AI market.
Fraser Kelton: You told
me you were worked up.
I haven't, I haven't read
anything that it has to say.
There's some semblance of 37 signals
content marketing going on here that
I just can kind of like push a little
bit into the corner, but what's up?
Nabeel Hyatt: no, I like, he's, he's,
okay, yes, of course, he's a little
provocative, but I think these are
long held beliefs not just hot takes.
Sam has a podcast with three other people,
which is actually, I think, the best new
podcast of the last year focused on tech.
Less is more, because specifically, it's
four humans that actually like each
other and actually know each other.
And so there's real authenticity
discussion as they talk about a bunch of
things, which is not most shows, right?
Most shows are canned content marketing.
A anyway So look, there's a first
part of what Sam's rant is and he has
a rant on his show and then he just
released like a 975 page, probably
generated by AI, PowerPoint deck
that kind of like regales how seed is
dead, AI is just a weird pipe dream.
Trying to save VC, it's all going
to crap, and everyone should just
start cottage software businesses.
That's ostensibly the long story short of
Sam's pitch about the evolving ecosystem.
I, there's a part of it that, first,
I agree with, which is that he has a
framing that the kind of industrialization
of venture capital phase is done.
That, that at this point, the factory
farmed B2B SaaS company is over.
And that he sees is that
causing a bunch of late stage.
Crowd into seed spray and pray, they're
not really working with those companies.
Just yeah, you know that
some junior partner has been
told to put money to work.
So the jamming money into random seed
companies and it's messing everything up.
And so if you're a seed or
series a fund oh, wow, you got
a real, you got a real problem..
So, I believe that.
But the second bit where he gets to
AI, I think, he's just anchored in his
priors before AI, where it just felt
like there was not enough innovation.
So, his feeling, I could like, I hope
I don't boil it down the wrong way, is
that fundamentally AI is a sustaining
innovation, not a disruptive innovation,
so it'll change the world, but it'll
just be that Google adds AI, and
Facebook adds AI, and it'll be fine.
If you're Adobe, you win anyway, you
just add AI, and that kind of thing.
And that there's no moat, because
it's all, open source anyway so
what's the moat you're going to
have and there's no network effects.
And that the tech is moving too
fast for, for startups to keep up.
I have a couple of like pretty
basic flawed issues with this.
Fraser Kelton: If the tech is moving so
fast, isn't that the opportune time for
Nabeel Hyatt: There you go!
Fraser Kelton: Arvind is a 25
person team, I think, and he's out
executing like everybody else who has.
tried to push into this space
and, and there's something
remarkable about what he's doing.
He, he's shipping both at the technical
level inference that is like 2.
5 times faster than, than other
inference platforms that are
specifically focused on that.
And then his product decisions are
like tasteful and thoughtful, and
he is not having to fight against
legacy business models and an entire
org chart that has calcified around
an an old way of doing things.
So anyway, you, you teed me up, but
Nabeel Hyatt: Yeah, no, exactly.
I just fundamentally disagree that
big tech is going to dominate.
One of the things I love about what's
happening in AI right now is that
we talk to lots of investors that
are quote unquote active in AI.
And they are activating wildly
different ideas about what's
going to work than we are.
And Sam's going to operate
as part of the ecosystem.
That is ultimately a good
thing for founders, right?
We're going to get a spread
of different capital to try a
bunch of different experiments.
We'll let it play out over time.
But I do think, yeah, there are
two flaws in Sam's argument.
So one is like he overestimates the
big tech's ability to execute, right?
Most large companies.
A struggle to adopt new tech.
And startups can find ways to
sell against Google and Microsoft.
And frankly, one of my little
examples, there was a world where
Microsoft dominated all of GovTech.
Big, entrenched sales.
And Google, when it was still a relatively
early company, actually used cloud and
pushed into that, into that dominance in
a way that I think, frankly, beforehand
you would have said would never work.
And there's lots and lots and
lots of examples of startups
using The disadvantage of a large
company to their benefit, right?
The second is that I think Sam, as
a lot of that entire generation of
founders, grew up in a world where
network effects were the biggest moat.
And so I think there's an anchoring
idea that isn't that the moat?
And, and so, if you grew up in that
world, you look at Facebook, and look,
network effects are an incredible,
incredible barrier to entry, but
at the same time, it's not the
only way startups build durability.
There were many startups
with no network effects.
Frankly, Google Search versus
Microsoft that we were talking about
earlier is an exact example of this.
Vertical integrations like Microsoft,
working from the OSS layer up to the
office layer and launching Microsoft
Office . Apple . Cisco . Cruise
. There's lots of other ways other
than network effects for a startup to
build competitive barriers to entry.
I think he's also over indexing on large
companies that are innovative in AI.
And forgetting that AI is going
to transform every industry
beyond big tech, right?
Manufacturing and healthcare, and
it's going to be a 20 year run
just like it was for the internet.
It's going to go hit every
little part of the economy.
Lots of those parts of the economy
are going to deeply struggle
to come up with a solution.
And there, there's no incumbent advantage.
Yes, some big company could
build a solution for that area,
but also so could a startup.
And so it's all the normal things
that happen with startups, right?
Fraser Kelton: Yeah, yeah, listen, I'm
a Satya Stan and, and that lovely line
that he made Google Dance was, like,
so fun, but come on it wasn't Bing and
Sydney that made Google Dance, I had, it's
just a really bad, crummy product, and
everybody knows it's a bad, crummy product
It is open AI and, and what, what was
shipped there that is making Google dance.
And we just talked earlier about how like
their entire business model is going to
have to be reconsidered for a world where
you can write a poem about tulips and
then have search results put below it.
Like they, they have real
fundamental problems that
they're gonna have to navigate.
Nabeel Hyatt: I think it's very
easy to underestimate how hard
it is for incumbents to change
and trench workflows, right?
It's, it's easier for startups
to start from scratch.
And there's an inherent conflict, and you
called it out, very first thing, between
these two ideas, that, that this is a...
Sustaining innovation, so companies
are just going to add it, and that
it's moving unbelievably fast.
Because it's not just moving unbelievably
fast, and this is a thesis to how
we, how we are navigating the world
right now with founders, which is,
it's not just moving very fast at the
foundational level layer, the model
layer, where there's new capabilities
coming out constantly, it's also moving
really fast at the user expectation
layer, right, at that top layer.
And so the, the, that, the last
revolution in AI 15 years ago, we
actually, as a firm, we didn't go
that deep into it . We had Cruise
and a couple of other investments.
But part of the problem in that age
of small models is that folks with the
data and the distribution win, right?
If you have all the data and you're
Google and you got a distribution,
you're just going to win.
What feels different here is, one,
the data, everybody has access to
the data, with a few big exceptions
. So there's a data advantage.
And then the third D , if I want
to keep this little D thing going,
is is the design is changing, right?
Like the, the kind of concept of a product
like Descript is that you used AI to
completely change the workflow on the
front end of how somebody actually makes
a podcast or edit a video in a way that
was never true for the last 20 years.
When the workflow and design changes.
Then the incumbents, they can't
completely change their workflows
without alienating their current base,
so they have incumbency disadvantage.
And you see this with Adobe, they
didn't, their, their launch of a copycat
product to Descript was Adobe Podcast.
Now, they didn't change a previous
product, they had to start from scratch.
They built a new UI, they launched
a copy, it's pretty good, it's just
getting started, we'll see what happens.
They obviously have money and brand
to go push it out into the market.
They're going to try bundling, all
the things that large companies do.
But that's no different than how large
companies have tried to compete with
startups for the last 30 years and
startups have been
Fraser Kelton: 30 plus years, sure.
Yep, the I think if you look at the
past 12 months, you have seen incumbents
move very quickly into this space.
The whole world has, right?
And I think part of it is, we have a short
term overestimating the impact, and then
long term we underestimate, and there's
going to be a very profound change.
But if you look at what the incumbents
are doing, they're layering in Cheap
ways to leverage this technology, right?
Hey, listen, we can now do some
sort of like really amazing
summarization around unstructured data.
And our app is unstructured data
because it's recording sales calls.
And that is incrementally
more useful for that product.
And it, it's pretty awesome that they
moved so quickly to introduce these
types of features of functionality.
But I think that It's an absence of
imagination to think that there's
not going to be entirely new user
experiences that are delivered off of
this technology, especially because
the technology is strictly only
going to get better from here on out.
And, you said this, I'm stealing your
line, is I don't even think we've seen
from the artists and creatives how
to use this technology appropriately
yet, and I think that that's fair.
Nabeel Hyatt: If you think the
problem is a puzzle, then large
companies have a huge advantage.
Puzzles can get solved by
throwing money at the problem.
But if the problem is a
mystery, Founders have a chance.
to win because agility,
wins and you need judgment.
Something that.
Large companies have problem with.
Fraser Kelton: Yep.
I very much agree.
Like the way that I think about the
world is there's this continuum of
project to product and on the project
end is doing things right and on the
product end is, Doing the right things.
And I feel like the jigsaw metaphor
is like slide it all the way over to
project where you're doing things right.
Nabeel Hyatt: Right, so this is a
puzzles versus mystery situation, and
there will be parts of the market,
absolutely, which feel like puzzles.
Let me just add a chatbot client, let
me get the retrieval properly, it'll
be added to my product, and we're done.
And then the mystery is places
where this technology will lead
to new types of interfaces, and
new types of experiences, that...
Don't sit anywhere comfortably where
an incumbent has previously been.
That's where startups have the advantage.
Fraser Kelton: And you're right.
If you just need to have
a 12 month known roadmap
For shipping AI, then let's coordinate
all your cross functional stakeholders
and make sure that they're bought
in and, and, you check the boxes.
And if it's figuring out the, the right
thing, I think we're so early in, in
understanding the types of experiences
that this technology can provide for us.
And, and we're, we're
just getting started.
Like I think as you said, again,
like the creatives haven't shown
us what's possible just yet
Nabeel Hyatt: Yeah, that's right.
That's right.
I'm, I'm going to throw to
you, actually, on the spot.
I know you're not prepared at all for
this, but ... We just went through
a process where we were talking
to a bunch of limited partners.
So for those who don't know venture
capital, we actually have to raise
money just like founders raise money.
And largely from non profits, endowments
pension funds, places like that.
So when we make money, that's
where the money goes to
those kinds of organizations.
Was there anything about that
process of talking to some LPs over
the course of the AGM in this last
week that caught you by surprise?
Fraser Kelton: It's surprising how much...
Felt very similar to being a founder
raising capital from VCs and then
how some of it felt so different.
You and I sat through a whole host
of different meetings and there
was meetings where the vibe just
was great from the hello, right?
Nabeel Hyatt: Yeah, yeah.
Fraser Kelton: in and it feels.
It just feels great.
The entire experience feels great.
And then there's some where you sit
down and you're like, Well, the vibe
here is not quite the same as the vibe
that just was in the other meeting.
I'm not sure how this
is going to turn out.
The other thing that is very similar is...
You just get into a groove, but you have
to have some feeling of spontaneity in
the discussions, and when, when I was
pitching VCs, you'd get to the point
with the deck where you would have a
pause and you knew, you knew nine times
out of ten they're going to lean in
with this question, and you didn't.
You didn't, you didn't pitch it, so that
you could get ahead of the question.
You pitched it so that they would ask
the question because you knew the answer
to give, but it had to feel spontaneous.
Nabeel Hyatt: There's a
theater to all of it, right?
There's a theater to pitching, and I
do think for an engineering mindset
when they're going out to raise money,
a lot of times the answer is, let me
answer all the questions in the deck.
And let me get it in order, and
somebody asked that question
there, I got a slide for that.
I used to do a thing, it's hard now
in Zoom land, but I used to do a
thing when I was pitching, where I
would take one question that I knew
was just an absolutely plump T up,
and and I would remove the slide,
and then have them ask the question,
and I'd be like, that's interesting.
Do you got a whiteboard marker?
And, and the act of you just, you
get up and you walk over to the
whiteboard and you're sketching
and talking in real time and man it
just, it pulls the room together.
Fraser Kelton: The place where it
was surprising and different, and it
continues to be surprising for me is
that we can spend our day talking to
founders who are trying to create a
future that may or may not take shape
in two years, three years, four years.
I mean, with Cruise, it was eight
years of Kyle grinding, right?
Yeah.
Yeah.
And then we step into a world where we
have to talk about portfolio construction
and it's just very strange to think
that the entire world works because
of portfolio construction, right?
And the risk that you take on so that
these, these outliers can actually
have oxygen and come to life.
Nabeel Hyatt: Well, I mean, I do think
that for the folks who, who came into
venture in the last four or five years,
they're learning a really key lesson
right now, and they have learned a really
key lesson over the last year , there's
a reason why it, it actually works.
I'm a person who like, it's like what
Winston Churchill said about democracy.,
Venture is a very flawed model.
It's just that it's the best model
that we have for sustaining innovation.
I mean, the other alternatives, if
you look at things like corporate R& D
and how many wonderful amazing things
come out of corporate R& D, you're
like, that's not the way to do it.
Okay, so what else do we do?
Do you want the EU to figure
out what we should build next?
That's not the right way to do it.
Like almost all the other models for how
We get an ecosystem to sit on the right
edge of risk and innovate and create
what's going to be next in the world.
I actually am a very big booster
for this particular model.
Like it, it takes builders and it
takes VCs and so on and so forth.
I also hate the industrialization of
venture capital because it, it feels
Fraser Kelton: just different.
Yep.
Nabeel Hyatt: it's not just different,
, I think it ruins parts of that model.
It messes with parts of that model that I
think are becoming more obvious now that
we're out of the, B2B sass go go times,
Fraser Kelton: Yeah.
SASSification of everything.
On a different topic, what, AI
product have you slowed down your
life to explore this past week?
It's not even
Nabeel Hyatt: slowed down.
I have been, so I listened
to a lot of podcasts.
I basically have.
The habit of anytime I have any
downtime, unless it's specifically like
thinking time I I'm doing the dishes.
I'm going outside to
take the dog for a walk.
I have a kind of running
list of podcasts going.
I've been using Marco Armitage
Overcast for many, years.
One of my best go to examples
of everything doesn't
have to be venture scale.
Like sometimes the most amazing product
is a one person effort and it's a
beautiful thing and I love that product
and there's a lot of love in it.
I certainly use it over say I don't
know, like Spotify is trying to
mix in my podcast with my music.
Stupidness or Apple Podcasts.
For the first time, and I don't know how
long it's been, 5, 6, 7, 8 years I, don't
use Overcast anymore, so I switched.
To a product called Snipt, S N I P T.
It is what an amazing difference
it makes . Obviously, it's not
changing how you listen to podcasts.
It is still a podcast player.
It's probably lower on features
than than a product like Overcast,
which has had many years.
But that's what you want with
technological innovation, right?
Is that a startup can start not without
having even hit table stakes for the
market, but there's some feature that's
so good that it drives your behavior.
The thing in particular that I
always want is you're listening to
a podcast, I'm in the car, somebody
says something you want to remember.
And the feature I've always wanted is
like, can you just please just clip?
I want my DVR TiVo feature for my podcast.
Like just clip the last 30 seconds
and do something smart with it so that
I don't lose these ideas as they're
coming, as I'm coming across them.
I think this company has been around
for a little while now, but without
all the AI features that they had.
So I think it used to be a, I
can triple click my earbuds.
And it will grab the last little
30 seconds or something like that.
I think it was the previous
version of this product.
I did, I think I tried
this a year ago or more.
It's fine.
It's just not good enough.
Now what it's doing is a bunch
of incredibly wonderful things.
First of all, as I'm scrolling through.
podcast.
It's using AI to do summarization
of the whole podcast.
So instead of it being the thing that
you or I would write about this podcast,
which may or may not be accurate, it's
actually looking at the transcript
and giving me a real summarization of
the topics that are going on in there.
Very interesting.
It then breaks up the podcast
dynamically into chapter clips.
So if I don't like a topic that we're
talking about right now, like if I
don't want to hear about Snipped.
I can double click on my ear
pods and it will just go to
the next chapter dynamically.
That's awesome.
That is awesome.
Which is just like, it just changes
the experience, because we all know...
Frankly, including ours, all
these podcasts go a little long.
And part of it is, that is the joy
of the podcast, is that it feels like
you're in the room with somebody, you're
listening to them go through a topic,
you can get depth, but sometimes you
just don't want to hear about something.
And so the DoubleClick Next Chapter,
even if they haven't set up, and it's
not, it's chapter rankings on very
finite, like two, three minute scales.
It's not like 20 minute scales,
which is what some podcasts do.
And I just love that.
And then the last feature, AI, Native
feature, which is really wonderful,
is now when you triple click, it is
smart about how much it clips, so
it clips only the topic area, which
could be in the future as well.
It clips the topic area that is related
to the thing that you just triple
clicked on, so somewhere, between 15
seconds and like a minute and a half
is what it's grabbing to try and grab
the concept, which is just brilliant.
And then it transcribes it into text.
So you have a transcription, and
then I have that now plugged into...
It has an API call that then
plugs this into ReadWise , which
is a kind of like note taking for
podcasts and Kindle reading app.
And then I have that, through an
API, plugged into my note taking app.
So that basically, every
single time, I triple click.
And I'm not going to talk about my
note taking app, because I love my note
taking app, but we gotta save that,
you can't blow it all in one week.
But I, triple click.
And basically it now becomes a
pipeline where that becomes a
permanent note for reference later on.
What the heck
Fraser Kelton: do you do with that later?
That's for the note
taking app conversation.
But I'm going back to our
conversation last week that 99
percent of our life is mundane.
I used to, tag things on Delicious
religiously, and I still find myself going
back to that archive every now and then.
And so there is some value there, but
I haven't tagged anything in well over
a decade, so I don't know what that
says about the stickiness and long term
value that I actually derive from it.
Nabeel Hyatt: Anyway.
Well, I like, look, I like when
people say things that are insightful.
Like a good example is I used to
reference of puzzles versus mysteries.
And that has been a guiding
thing for me for a little while.
I came from listening to a
podcast years ago, right?
It's when you, it's when you hear
something insightful and you want to
make sure you don't lose it so you
can internalize it a little bit more.
What I really want now is to feed into
a spaced referential learning product.
But maybe that's just a little
bit too nerdy for most people,
but the pod, but snip is
Fraser Kelton: great.
So yeah I'm a fellow Overcast user, and
I think that Marco's just got great taste.
It's an opinionated product.
There's a bunch of different
flourishes that have really
resonated with me over the years.
What do you miss in, in
Snipped that Overcast gave you?
Nabeel Hyatt: I'm opening up Overcast in
front of me right now to look through.
So he does a really good job of,
having custom clipping at the
beginning and the end of podcasts.
So some of these podcasts start with.
A minute and a half of ads in
the beginning, or a lot of times
I find like the last minute of
the outro is always useless.
Like they just go bloviate for, I lose
a minute and a half of my life every
time I listen to a podcast at the end.
And I actually really like the simple,
clean feature of like how many seconds
of the outro do you want to cut from the
end of every single podcast, which when
you're listening ongoing is helpful.
I also think his shortened silences.
is a very nice feature.
I think those two are probably the
simple ones that kind of like bug me.
There's probably other stuff.
Fraser Kelton: Yeah, isn't it funny?
The I forget what he calls
it, but the shortened silences
feature is so lovingly done.
And when he's spoken about how he built
that in the past, you just realize
here's somebody who loves his craft.
It is so great that when I listen on
other players and I try their speed
up equivalent, it's just jarring.
Nabeel Hyatt: It's not as good.
It's just jarring.
It's not listenable.
Fraser Kelton: That's right.
That's right.
Interesting.
Oh, the last one is
Nabeel Hyatt: Snipped.
Very simple one also is Snipped is very
bad at downloading episodes ahead of time.
So it's fine for streaming, but
like I get on a plane and the
only thing it downloads is things
are already sitting in the queue.
And I want it to just, it's audio, man.
It doesn't take up that much space.
Just, like, download hundreds of
podcasts so I can listen to what I want.
It's still bad at that.
Yeah.
But the whole point is, like, is it
gonna, is it, in a year, is Snipp gonna
be able to download more podcasts?
Like, yeah, probably.
But they, they created a new, for
me at least, they created a reason
we're switching, and now I've got
a new default, and defaults matter.
Yep,
Fraser Kelton: yep.
Not to get too far ahead of things,
but you can imagine that if it knows
the unstructured segments of podcasts
that you actually like, and the ones
that you skip, There can be an entire
discovery piece around, around segments
rather than episodes or podcasts itself.
Like here is some esoteric podcast
talking in, depth around a note
taking app that's going to resonate
with you for these reasons.
Cause
Nabeel Hyatt: That's, your catnip.
Oh yeah, exactly.
That feels like, especially in a
world where it feels like everybody
woke up a year after COVID.
With a great zoom setup and some
mics and decided to start a podcast.
Like we're going to have so much audio.
It's hilarious.
Like I know we're contributing
to the problem, but everybody
with a podcast does create a new
problem, which is content discovery.
And, I don't think for most
podcasts, the answer is.
New things to subscribe to, to
listen to every single episode, but
if Snippet works, you can also see
this cycle where you can listen to
snippets that are more valuable.
It was, I'm sure their original
value proposition, I know it was
because I saw it a while ago.
I don't think it works.
You don't have enough data and
you don't have enough nuance and
unstructured organization before LLMs.
And now with LLMs, I actually think
there's a possibility where a kind
of like okay idea for a product
might actually be a real thing.
We'll see.
Yeah, I
Fraser Kelton: can imagine that you're
not even just getting recommended,
but you're listening to a snipped
podcast that is nothing but segments
that they think you're going to
like on, a topic across a plethora
of different podcasts all stitched
Nabeel Hyatt: together.
Yeah, they have, they're already,
I think they already are in
progress on something similar.
I see.
Well, listen, ship it, guys.
Jeremy was remarking the other night.
I, I do.
Basically record as much of
my life as I can right now.
Partially with the idea that it
will be maybe valuable later on.
I don't know that I'm saying anything
insightful, but maybe something
somebody else will say is insightful.
Or maybe this product can do
something with that audio.
There's got to be a model somewhere
that's going to do something with it.
Fraser Kelton: I think on all of that
stuff, we're going to realize that...
Nothing exciting or interesting or
novel happens in 99 percent of our
lives, and so you'll have terabytes of
content that will remind you of the dull
Nabeel Hyatt: (Laughter).
Fraser Kelton: repetitiveness of
life, by, punctuated by very brief
moments of intense highs and lows.
Nabeel Hyatt: So Fraser, are
you prognosticating, that our AI
summarizing our lives perfectly, will
drive a global nihilism understanding
of the futility of our daily lives?
Fraser Kelton: People don't
even care to that degree, right?
There will be a small pocket of
people who have to confront the
idea that there's nothing going
Nabeel Hyatt: just not that important.
Fraser Kelton: Yeah.
And everybody else will be watching Sunday
night football like they always had been.
Nabeel Hyatt: Emmett Shear, has this
thing about I think he talked about
it at dinner the other night where
he's thinking it's yeah, it turns out.
That the reason Justin.
tv doesn't work is that nobody has an
interesting life on a day to day basis.
And the reason that live reality
television doesn't work all the time
is because even a crazy celebrity
still mostly spends their day doing
mundane and stupid, stupid things.
And so,
Fraser Kelton: Yeah.
And then you get moments like the other
night when a founder is dealing with
immigration hell and you're trying to
figure out how to get this guy who's
just flown into SFO and been held for
23 hours into the country because
he's legally supposed to be there, but
he didn't have the right paperwork.
Nabeel Hyatt: I'm surprised that
tech industry hasn't found some way
to break through with both parties
in Washington on immigration.
You should not have a founder
of a venture backed startup.
Having to flee the country quickly
because of visa problems when they're
just trying to build a company in America.
Fraser Kelton: Yeah.
And, I mean, the visa's
already been approved.
He just presented the
wrong piece of paper.
It's just very strange.
I mean, the idea that we train the
best in the world and then kick
them out some number of months after
doesn't make any sense either, right?
Nabeel Hyatt: Yeah.
That's another thing . This may
be counter to everybody hating on
all the Ivy's and everything else.
Higher education is a magnet that pulls.
Immigrants to us to start with,
because frankly, there's already
a Google office in Mumbai, right?
And so, they're going to come
here for college, and then it's
our job as an ecosystem to make
sure they stay, that the best and
brightest who come here for college.
I actually, I have my issues with higher
education and the way it's structured,
but it is one of the industries
of which we massively dominate.
And so trying to disrupt it and kill it.
Just hurts the whole
ecosystem of startups.
So I fix it But the we should just
tear it all down and colleges don't
matter is ignoring what colleges mean
Fraser Kelton: the idea that you
don't want to take the best electrical
engineer in the world and put them
into a six year PhD in some esoteric
area where they're going to go so
deep that there might just be peer
basic research that comes out of it.
Or it might be some, some other
benefit that gets kicked off.
Like we should just be
investing and supporting that
as much as we possibly can.
Nabeel Hyatt: Yeah, look, Community
colleges are actually very, very
beneficial to local communities.
But right above that, you got a bunch
of schools that aren't doing crap for
anybody except for, creating student debt.
Fine.
But for the top third to 40 percent
it is the beginning of the process of
the engine of what makes America work.
Are there problems after that?
Is the idea of yet yet another
person going 12 years down PhD
physics research only to...
Find out that they're working in some
field so deep that they're likely to
never hit a breakthrough at this point.
And they're throwing their whole life away
as one of the brightest people in America.
I can make that rant, too.
We under indexed how much cross
disciplinary learning there
is when you get to actually
coming through breakthroughs.
.
But I think if you remove it,
you really do a lot of damage
to the whole system, so.
Fraser Kelton: But even that
I'm just looking up Dario
from Anthropic's LinkedIn.
He has a PhD in Biophysics, right?
So he went deep in some sort
of, some sort of weird topic.
I don't even know what
biophysics would be.
And now he's, he's running anthropic.
I think that there's certain types
of minds that, that get attracted to
these deep, programs that also then
learn how to learn and, and grind and
all sorts of different skills that
are quite applicable and valuable
for creating pretty profound things.
Nabeel Hyatt: Yeah.
I think that's fair.
. I think people have a hard time with the
non deterministic nature of it all too.
This idea of who do we let in and what
are they going to study and it's we
as a society seem to want everything
to be rapidly deterministic, right?
I come in to get the biophysics.
How many biophysics jobs
are there in the world?
Do we need more people in biophysics?
It's just a very linear thinking
when, the nature of an economy, the
nature of our, frankly, our human
lives, the nature, certainly the
nature of innovation and startups.
Is that a little bit of randomness and a
little bit of chaos is pretty important.
And that comes to college admissions,
that comes to job admissions, that comes
to what you're going to work on 10 years
from now in your life, all of that.
But we live in an era of data
science, and, and numbers.
So the things that seem to
belie that, that determinist
nature people have trouble with,
Fraser Kelton: So here are the names.
Random Seeds, Latent Chat, Unsupervised
Learning, Weekly Prompts, do you think?
Nabeel Hyatt: So, okay, wait,
wait, let's start over again.
The potential names that, was this,
did we use ChatGPT or Claude for
Fraser Kelton: We use Claude and,
Nabeel Hyatt: this podcast.
and ChatGPT.
Okay, so using both Claude and ChatGPT.
The answers, and this is the, just
so we're clear, this is a cold, this
is like the top 10 percent of the
answers that GPT and Claude gave us
for names for this podcast, right?
Run it down again, Fraser?
Fraser Kelton: Random seeds.
Latent chat.
I'm even editing, so I won't edit it.
Tensor talk.
Tensor talk.
Unsupervised learnings.
And, and the other one that I removed
at the start was gen talk, because
anytime anybody says gen AI, I
cringe and die a little bit inside.
I'm not sure that any of these are great.
Nabeel Hyatt: They do have a certain dad
humor quality to them that I appreciate
Fraser Kelton: Yeah.
Isn't that interesting is
it just reinforces how hard
humor is to capture with
Nabeel Hyatt: well, it's obviously
there's just too much dad humor
in the training data . That's the
Fraser Kelton: well, that's what I mean.
Like the funny people are funny
because that's a scarce resource.
It's not like everybody's
walking around being hilarious.
Nabeel Hyatt: That's right.
I have no strong visceral
response to this at all.
You're right.
Unsupervised learning is the least cringe.
I can at least say that.
But if you are one of the handful of
folks that likely listen and you have
a suggestion as well, let us know.
And that's, let's call it for today.
Fraser Kelton: Let's call it.
Thanks.
Nabeel Hyatt: Awesome.
Talk to you later.