AI Eats the World: Benedict Evans on the Next Platform Shift

0:00:03 Chad GPT has got eight or nine hundred million
0:00:03 weekly out of users.
0:00:06 And if you’re the kind of person who is using this
0:00:10 for hours every day, ask yourself why five times more people
0:00:12 look at it, get it, know what it is, have an account,
0:00:15 know how to use it, and can’t think of anything to do with it
0:00:17 this week or next week.
0:00:19 The term AI is a little bit like the term technology.
0:00:22 When something’s been around for a while, it’s not AI anymore.
0:00:24 Is machine learning still AI? I don’t know.
0:00:27 In actual general usage, AI seems to mean new stuff.
0:00:30 And AGI seems new scary stuff.
0:00:32 AGI seems to be a little bit like this.
0:00:35 Like either it’s already here and it’s just more software
0:00:38 or it’s five years away and will always be five years away.
0:00:41 We don’t know the physical limits of this technology
0:00:43 and so we don’t know how much better it can get.
0:00:46 You’ve got Sam Altman saying, we’ve got PhD level researchers right now.
0:00:48 And Demis Asibis says, no we don’t, shut up.
0:00:51 Very new, very, very big, very, very exciting,
0:00:53 world-changing things tend to lead to bubbles.
0:00:55 So yeah, if we’re not in a bubble now, we will be.
0:00:59 Is AI just another platform shift
0:01:01 or the biggest transformation since electricity?
0:01:05 Benedict Evans, technology analyst and former A16Z partner,
0:01:08 has spent years studying waves like PCs, the internet, and cell phones
0:01:11 to understand what actually changed and who captured the value.
0:01:14 Now he’s turned that same lens on AI
0:01:17 and the picture is far more complex than benchmarks or hype cycles suggest.
0:01:20 Some industries may be rewritten from the ground up.
0:01:22 Others may barely notice.
0:01:25 Tech giants like Google, Meta, Amazon, and Apple
0:01:27 are racing to reinvent themselves before someone else does.
0:01:29 Yet for all the excitement,
0:01:33 most people still struggle to find something they truly need AI for every single day.
0:01:36 A disconnect Benedict thinks is an important signal
0:01:37 about where we really are in the curve.
0:01:40 In today’s episode, we get into where bottlenecks emerge,
0:01:42 why adoption looks the way it does,
0:01:44 what kinds of products still haven’t shown up,
0:01:46 and how history can actually guide us here.
0:01:49 And finally, what would have to happen over the next few years
0:01:51 is for us to look back and say,
0:01:52 AI wasn’t just another wave.
0:01:54 It was bigger than the internet.
0:01:59 Benedict, welcome back to the A16Z podcast.
0:02:00 Good to be back.
0:02:02 We’re here to discuss your latest presentation,
0:02:03 AI Eats the World.
0:02:05 So for those who haven’t read it yet,
0:02:07 maybe we can share the high-level thesis
0:02:11 and maybe contextualize it in light of recent AI presentations.
0:02:13 I’m curious how your thinking has evolved.
0:02:16 Yeah, it’s funny, one of the slides in the debt reference
0:02:18 is a conversation where I had with a big company CMO
0:02:21 who said, we’ve all had lots of AI presentations now.
0:02:22 The Google one.
0:02:24 We’ve had the Google one and the Microsoft one.
0:02:26 We’ve had the Bain one and the BCG one.
0:02:29 We’ve had the one from Accenture and the one from our ad agency.
0:02:30 So now what?
0:02:33 So there’s sort of 90-odd slides.
0:02:35 So there’s a bunch of different things I’m trying to get at.
0:02:37 One of them is, I think, just to say,
0:02:40 well, if this is a platform shift or more than a platform shift,
0:02:42 how do platform shifts tend to work?
0:02:43 What are the things that we tend to see in it?
0:02:47 And how many of those patterns can we see being repeated now?
0:02:51 And, of course, some of the patterns that come out of that
0:02:52 are things like bubbles,
0:02:56 but others are that lots of stuff changes inside the tech industry.
0:02:58 And there are winners and losers
0:03:00 and people who were dominant end up becoming irrelevant.
0:03:03 And then there were new billion, trillion-dollar companies created.
0:03:07 But then there’s also, what does this mean outside the tech industry?
0:03:11 Because if we think back over the last waves of platform shifts,
0:03:13 there were some industries where this changed everything
0:03:15 and created and uncreated industries.
0:03:18 And there are others where this was just kind of a useful tool.
0:03:20 So, you know, if you’re in the newspaper business,
0:03:23 the last 30 years looked very different to if you were in the cement business,
0:03:25 where the internet was just kind of useful,
0:03:27 but didn’t really change the nature of your industry very much.
0:03:33 And so what I tried to do is give people a sense of,
0:03:36 well, what is it that’s going on in tech?
0:03:37 How much money are we spending?
0:03:38 What are we trying to do?
0:03:39 What are the unanswered questions?
0:03:43 What might or might not happen within the tech industry?
0:03:46 But then outside technology,
0:03:48 how does this tend to play out?
0:03:50 What seems to be happening at the moment?
0:03:54 How is this manifesting into tools and deployment
0:03:57 and new use cases and new behaviors?
0:04:01 And then as we kind of step back from all of this,
0:04:03 again, how many times have we gone through all of this before?
0:04:05 It’s funny, I went on a podcast this summer
0:04:07 and I sort of opening line,
0:04:07 I said something like,
0:04:09 well, I’m a centrist.
0:04:11 I think this is as big a deal as the internet or smartphones,
0:04:13 but only as big a deal as the internet or smartphones.
0:04:16 And there’s like 200 YouTube commentators under these
0:04:18 saying this more and he doesn’t understand how big this is.
0:04:21 And I think, well, the internet was kind of a big deal.
0:04:23 It was kind of a big deal.
0:04:25 And, you know, I sort of finished the day
0:04:26 by looking at elevators
0:04:28 because I live in an apartment building in Manhattan
0:04:30 and we have an attended elevator,
0:04:32 which means there’s no buttons,
0:04:33 there’s an accelerator and a brake.
0:04:35 And the doorman gets in and drives you to your floor,
0:04:35 this streetcar.
0:04:40 And in the 50s, Otis deployed automatic elevators
0:04:42 and then you get in and you press a button.
0:04:44 And they marketed it by saying,
0:04:46 it’s got electronic politeness,
0:04:48 which means the infrared beam.
0:04:52 And today when you get into an elevator,
0:04:52 you don’t say,
0:04:56 ah, I’m using an electronic elevator.
0:04:57 It’s automatic.
0:04:58 It’s just a lift,
0:05:01 which is what happened with databases
0:05:04 and with the web and with smartphones.
0:05:06 And I kind of think now,
0:05:06 this is funny,
0:05:08 I’ve done a couple of polls on this in LinkedIn and threads
0:05:10 of is machine learning still AI?
0:05:13 The term AI is a little bit like the term technology
0:05:15 or automation.
0:05:17 It only kind of applies when something’s new.
0:05:19 When something’s been around for a while,
0:05:20 it’s not AI anymore.
0:05:21 So like databases certainly aren’t AI.
0:05:23 Is machine learning still AI?
0:05:24 I don’t know.
0:05:25 And there’s obviously,
0:05:26 there’s like an academic definition
0:05:28 where people say this guy’s an idiot.
0:05:28 No, of course,
0:05:30 I’m going to explain the definition of AI.
0:05:32 But then in actual general usage,
0:05:33 AI seems to mean new stuff.
0:05:38 AI seems like new, scary stuff.
0:05:40 Yeah, it’s funny.
0:05:41 I was thinking about this.
0:05:42 There’s an old theologian’s joke
0:05:43 that the problem for Jews
0:05:44 is that you wait and wait and wait
0:05:46 for the Messiah and he never comes.
0:05:47 And the problem for Christians
0:05:48 is that he came and nothing happened.
0:05:50 You know, the world didn’t change.
0:05:51 There was still sin.
0:05:52 All practical purposes, nothing happened.
0:05:55 And AI seems to be a little bit like this.
0:05:57 Like either it’s already here.
0:05:58 And so you’ve got Sam Altman saying,
0:06:00 we’ve got PhD level researchers right now.
0:06:02 And Demis Asibis says,
0:06:02 what?
0:06:03 No, we don’t.
0:06:03 Shut up.
0:06:06 And so either it’s already here
0:06:08 and it’s just more software
0:06:10 or it’s five years away
0:06:12 and will always be five years away.
0:06:13 Yeah, yeah.
0:06:16 Let’s compare back to previous platform chefs
0:06:17 because some people look at,
0:06:18 you know,
0:06:19 some of the internet and say,
0:06:20 hey, there were net new
0:06:22 trillion dollar companies,
0:06:23 Facebook and Google
0:06:24 that were created from it
0:06:26 and just sort of all sorts of
0:06:28 new emerging winners.
0:06:29 Whereas they look at something like mobile
0:06:30 and say, hey,
0:06:30 you know,
0:06:31 there were big companies
0:06:32 like Uber and Snap
0:06:33 and Instagram and WhatsApp,
0:06:35 but these were billion dollar outcomes
0:06:37 or tens of billion dollar outcomes.
0:06:38 But really the big winners were,
0:06:39 were in fact,
0:06:40 Facebook and Google.
0:06:41 And so in some sense,
0:06:44 mobile perhaps was sustaining.
0:06:45 You feel free to quibble
0:06:46 with the definition
0:06:47 of sustaining disruptive,
0:06:48 but sustaining in the sense
0:06:49 that maybe more of the value
0:06:51 went to incumbents,
0:06:51 companies that existed
0:06:53 prior to the shift.
0:06:53 I’m curious
0:06:55 how you think about AI
0:06:56 in light of that
0:06:57 in terms of
0:06:58 is more of the gains
0:06:59 going to come to net new companies
0:07:01 like OpenAI and Anthropic
0:07:02 and others that follow
0:07:04 or are more of the gains
0:07:04 going to be captured
0:07:05 by Microsoft
0:07:06 and Google
0:07:07 and Meta
0:07:07 and companies
0:07:08 that existed prior?
0:07:11 So I think there’s several answers
0:07:11 to this.
0:07:11 One of them is like
0:07:13 you kind of have to be careful
0:07:14 about like framings
0:07:15 and structures and things
0:07:15 because you end up
0:07:16 arguing about the framing
0:07:17 and the definition
0:07:18 rather than arguing
0:07:19 about what’s going to happen.
0:07:21 And they’re all useful,
0:07:22 but they’ve all got holes in them.
0:07:24 And, you know,
0:07:26 what mobile did was,
0:07:26 you know,
0:07:27 there’s a bunch of things
0:07:28 that it changed fundamentally.
0:07:29 It shifted us from the web
0:07:30 to apps, for example.
0:07:32 And it gave everybody
0:07:32 in the world
0:07:33 a pocket computer.
0:07:34 So even today,
0:07:35 there’s less than a billion
0:07:36 consumer PCs on Earth
0:07:36 than there’s something
0:07:38 between five and six billion
0:07:38 smartphones.
0:07:41 And it made possible things
0:07:42 that would not have been
0:07:43 possible without it,
0:07:44 whether that’s TikTok
0:07:45 or arguably,
0:07:46 I think,
0:07:47 things like online dating.
0:07:49 And you can map those
0:07:50 against dollar value.
0:07:51 You can also map those
0:07:53 against kind of structural change
0:07:54 in consumer behavior
0:07:55 and access to information
0:07:56 and things.
0:07:58 And I think you could
0:07:59 certainly argue that Meta
0:08:00 would be a much smaller company
0:08:01 if it wasn’t for mobile,
0:08:02 for example.
0:08:04 So you can kind of argue
0:08:05 the puts and calls
0:08:06 on this stuff a lot.
0:08:07 There’s certainly
0:08:08 not all platform shifts
0:08:09 are the same.
0:08:10 And, you know,
0:08:11 you can do the sort of
0:08:12 standard sort of teleology
0:08:12 of say, well,
0:08:13 there were mainframes
0:08:14 and then PCs
0:08:14 and then the web
0:08:15 and then smartphones.
0:08:16 But you kind of want
0:08:17 to put SAS in there somewhere
0:08:18 and you kind of want
0:08:19 to put open source in there
0:08:19 and maybe you want
0:08:20 to put databases.
0:08:22 And so these are kind
0:08:22 of useful framings,
0:08:24 but they’re not predictive.
0:08:24 They don’t tell you
0:08:25 what’s going to happen.
0:08:27 They just kind of give you
0:08:28 one way of understanding
0:08:30 what seems some of the
0:08:31 patterns that we have here.
0:08:32 And of course,
0:08:33 the big debate around
0:08:34 generative AI
0:08:34 is just another
0:08:35 platform shift
0:08:36 or is it something
0:08:36 more than that?
0:08:37 And of course,
0:08:38 the problem is we don’t
0:08:38 know and we don’t have
0:08:39 any way of knowing
0:08:40 other than waiting to see.
0:08:41 So this may be as big
0:08:43 as PCs or the web
0:08:44 or SAS or open source
0:08:44 or something,
0:08:45 or it may be as big
0:08:46 as computing
0:08:46 and then you’ve got
0:08:47 the very overexcited
0:08:48 people living in group
0:08:49 houses in Berkeley
0:08:50 who think this is
0:08:50 as big as fire
0:08:51 or something.
0:08:52 Well, great.
0:08:53 But does this
0:08:53 bring new companies?
0:08:54 I mean,
0:08:54 you go back to the mobile,
0:08:55 there was a time
0:08:56 when people thought
0:08:57 that blogs were going
0:08:58 to be indifferent to the web,
0:08:59 which seems weird now.
0:09:00 Like Google needed
0:09:02 like a separate blog search.
0:09:02 This was seriously,
0:09:03 this was a thing.
0:09:04 There was a time
0:09:06 when it was really not clear
0:09:06 and I think you kind of
0:09:07 generalize this point.
0:09:09 You go back to the internet
0:09:11 in the mid-90s,
0:09:12 we kind of knew
0:09:13 this was going to be
0:09:13 a big thing.
0:09:15 We didn’t really know
0:09:16 it was going to be the web.
0:09:17 So before that,
0:09:17 we didn’t know
0:09:18 it was going to be the internet.
0:09:19 We knew there were
0:09:20 going to be networks.
0:09:20 It wasn’t clear
0:09:21 it was going to be the internet.
0:09:21 Then it wasn’t clear
0:09:22 it was going to be the web.
0:09:23 Then it wasn’t really clear
0:09:24 how the web was going to work.
0:09:27 And when Netscape launched,
0:09:28 like Mark Zuckerberg
0:09:29 was in junior high
0:09:30 or elementary school
0:09:30 or something
0:09:32 and Larry and Sergei
0:09:32 were students
0:09:34 and like Amazon
0:09:34 with a bookstore.
0:09:37 So you can know it
0:09:38 but not know it.
0:09:39 And you could make
0:09:39 the same point
0:09:40 about smartphones.
0:09:41 Like it was,
0:09:42 we knew everyone
0:09:42 was going to have
0:09:43 an internet connected
0:09:44 thing in their pocket,
0:09:45 but it was not clear
0:09:46 it was basically
0:09:47 going to be a PC
0:09:48 from this has-been
0:09:49 PC company
0:09:50 from the 80s
0:09:52 and a search engine company.
0:09:52 It was not clear
0:09:53 it wasn’t going to be
0:09:54 Nokia or Microsoft.
0:09:54 See, I think you have
0:09:55 to be super careful
0:09:56 in making kind of
0:09:57 deterministic predictions
0:09:58 about this.
0:09:59 What you can do
0:10:00 is say,
0:10:00 well,
0:10:02 when this stuff happens
0:10:03 everything changes
0:10:04 and that’s happened
0:10:06 five or ten times before.
0:10:07 I’m curious how you got
0:10:08 conviction in this idea
0:10:09 or what you got
0:10:09 as a prediction
0:10:10 that hey,
0:10:11 AI is going to be
0:10:12 as big as the internet
0:10:12 which of course
0:10:13 is pretty big
0:10:15 but I’m not yet,
0:10:16 I’m not yet at the conviction
0:10:17 that it’s going to be
0:10:17 any bigger.
0:10:18 I’m curious
0:10:19 what sort of
0:10:21 inspires that
0:10:21 sort of
0:10:24 statement
0:10:24 and then also
0:10:26 what might change
0:10:27 your mind either way
0:10:28 that it might not
0:10:28 be as big as the internet
0:10:29 because of course
0:10:29 the internet
0:10:30 was obviously very big
0:10:31 but also that
0:10:31 hey,
0:10:33 perhaps it might be bigger.
0:10:34 Well,
0:10:35 so I think
0:10:36 I don’t want to
0:10:37 I made a diagram
0:10:38 of kind of S-curves
0:10:39 kind of going up
0:10:39 and someone said
0:10:40 well,
0:10:40 what’s the axis
0:10:41 on this diagram?
0:10:42 I don’t want to
0:10:43 kind of get into
0:10:45 is this 5% bigger
0:10:46 than internet
0:10:47 or is it 20% bigger?
0:10:48 I think the question
0:10:48 is more like
0:10:49 is it another
0:10:50 of these industry cycles
0:10:51 or is it a much
0:10:52 more fundamental
0:10:52 change
0:10:53 in what technology
0:10:54 can be?
0:10:54 Is it more like
0:10:55 computing
0:10:56 or electricity
0:10:57 as a sort of
0:10:58 structural change
0:11:00 rather than
0:11:01 here’s a whole bunch
0:11:01 more stuff we can do
0:11:02 with computers?
0:11:03 I think that’s sort of
0:11:04 the question
0:11:05 and there’s a funny
0:11:06 sort of disconnect
0:11:07 I think in looking
0:11:07 at debates about this
0:11:08 within tech
0:11:09 because I watched
0:11:10 this is one of
0:11:12 the open AI
0:11:12 live streams
0:11:13 a couple of
0:11:13 weeks ago
0:11:16 and they spend
0:11:17 the first 20 minutes
0:11:17 talking about
0:11:18 how they’re going
0:11:18 to have like
0:11:18 human level
0:11:19 PhD level
0:11:20 AI researchers
0:11:21 like next year
0:11:22 and then the
0:11:23 second half
0:11:23 of the stream
0:11:23 is
0:11:24 oh and here’s
0:11:25 our API stack
0:11:26 that’s going to
0:11:26 enable hundreds
0:11:27 and thousands
0:11:27 of new software
0:11:28 developers
0:11:28 just like Windows
0:11:29 and in fact
0:11:30 literally quote
0:11:30 Bill Gates
0:11:31 and you think
0:11:31 well those can’t
0:11:32 kind of both
0:11:33 be true
0:11:34 like either
0:11:35 I’ve got a thing
0:11:35 which is a
0:11:36 PhD level
0:11:37 AI researcher
0:11:38 which by implication
0:11:39 is like a PhD
0:11:40 level CPA
0:11:42 or I’ve got
0:11:43 a new piece
0:11:43 of software
0:11:44 that does my
0:11:44 taxes for me
0:11:45 and well
0:11:45 which is it
0:11:47 either this thing
0:11:47 is going to be
0:11:48 like human level
0:11:49 and some
0:11:50 and that’s a very
0:11:51 very challenging
0:11:52 problematic
0:11:53 complicated statement
0:11:55 or this is
0:11:56 going to let us
0:11:57 make more software
0:11:57 that can do
0:11:58 more things
0:11:58 that software
0:11:59 couldn’t be
0:12:00 and I think
0:12:01 there’s a real
0:12:02 like schizophrenia
0:12:04 in conversations
0:12:04 around this
0:12:05 because like
0:12:06 scaling laws
0:12:06 and it’s going
0:12:07 to scale all
0:12:07 the way
0:12:07 and meanwhile
0:12:08 I’m going
0:12:08 to hear
0:12:09 look how good
0:12:09 it is at
0:12:10 writing code
0:12:10 and again
0:12:10 like well
0:12:10 is it
0:12:11 writing code
0:12:12 or do we
0:12:12 not need
0:12:13 software anymore
0:12:13 because in
0:12:14 principle
0:12:15 if the models
0:12:15 keep scaling
0:12:16 nobody’s going
0:12:16 to write code
0:12:17 anymore
0:12:17 you’ll just
0:12:18 say to the
0:12:18 model like
0:12:18 hey can you
0:12:19 do this
0:12:19 thing for me
0:12:21 is it a little
0:12:21 bit of a hedge
0:12:22 or like a
0:12:23 sequencing thing
0:12:23 or
0:12:24 well
0:12:25 some of it’s
0:12:26 a sequencing
0:12:26 thing
0:12:27 but you know
0:12:27 in principle
0:12:28 if you think
0:12:28 this stuff is
0:12:29 going to keep
0:12:29 scaling
0:12:30 like why are
0:12:30 you investing
0:12:31 in a software
0:12:31 company
0:12:33 because you know
0:12:34 people just have
0:12:34 this like
0:12:37 and I think
0:12:38 this is
0:12:39 kind of
0:12:40 the funny
0:12:40 kind of
0:12:40 challenge
0:12:41 and this is
0:12:41 I think
0:12:42 the fundamental
0:12:43 way that this
0:12:43 is different
0:12:43 from previous
0:12:44 platform shifts
0:12:45 is that with
0:12:46 the internet
0:12:47 or with mobile
0:12:48 or being
0:12:48 deemed with
0:12:48 mobile
0:12:49 you didn’t
0:12:50 know what
0:12:50 was going to
0:12:50 happen in the
0:12:51 next couple
0:12:51 of years
0:12:52 you didn’t
0:12:52 know what
0:12:53 Amazon would
0:12:53 become
0:12:53 and you didn’t
0:12:54 know how
0:12:54 Netscape was
0:12:55 going to work
0:12:55 out
0:12:56 and you didn’t
0:12:56 know what
0:12:56 next year’s
0:12:57 iPhone was going
0:12:57 to be
0:12:58 and 10 years
0:12:58 ago when we
0:12:59 cared about
0:12:59 that
0:12:59 you kind of
0:13:00 knew the
0:13:00 physical limits
0:13:00 it’s like
0:13:01 you knew
0:13:02 in 1995
0:13:02 you knew
0:13:03 that telcos
0:13:03 were not
0:13:03 going to
0:13:04 give everybody
0:13:05 gigabit fiber
0:13:05 next year
0:13:07 and you knew
0:13:07 that the iPhone
0:13:08 wasn’t going
0:13:08 to like have
0:13:09 a year’s
0:13:09 battery life
0:13:10 and unroll
0:13:10 and have a
0:13:11 projector
0:13:11 and fly
0:13:12 or whatever
0:13:13 but we don’t
0:13:14 know the
0:13:15 physical limits
0:13:15 of this
0:13:15 technology
0:13:16 because we
0:13:16 don’t really
0:13:17 have a good
0:13:17 theoretical
0:13:18 understanding
0:13:18 of why
0:13:18 it works
0:13:19 so well
0:13:20 nor indeed
0:13:20 do we have
0:13:20 a good
0:13:21 theoretical
0:13:21 understanding
0:13:21 of what
0:13:21 human
0:13:22 intelligence
0:13:22 is
0:13:23 and so we
0:13:24 don’t know
0:13:24 how much
0:13:25 better it can
0:13:25 get
0:13:27 so you can
0:13:28 do a chart
0:13:28 and you could
0:13:29 say well
0:13:31 for modems
0:13:31 and this is
0:13:32 the roadmap
0:13:32 for DSL
0:13:33 and this is
0:13:33 how fast
0:13:34 DSL will be
0:13:34 and then you
0:13:34 can make
0:13:35 some guesses
0:13:36 about how
0:13:37 quickly telcos
0:13:37 will deploy
0:13:38 DSL
0:13:38 and then you
0:13:39 can say well
0:13:39 clearly we’re
0:13:39 not going to
0:13:40 be able to
0:13:41 replace broadcast
0:13:41 TV
0:13:42 we’re streaming
0:13:43 in 1998
0:13:45 but we don’t
0:13:46 have an
0:13:47 equivalent way
0:13:47 of modelling
0:13:48 this stuff
0:13:50 to know
0:13:51 what is the
0:13:51 fundamental
0:13:52 capability
0:13:52 of it
0:13:53 going to
0:13:53 look like
0:13:54 in three
0:13:54 years
0:13:55 which gets
0:13:56 you to
0:13:56 these kind
0:13:56 of slightly
0:13:57 vibes based
0:13:58 forecasting
0:13:59 where no one
0:14:00 really knows
0:14:00 so you
0:14:01 know
0:14:01 Jeff Hinton
0:14:02 says well
0:14:02 I feel
0:14:03 like
0:14:03 and Demis
0:14:03 Sassabas
0:14:04 says well
0:14:04 I feel
0:14:05 like
0:14:05 but no one
0:14:06 knows
0:14:06 and then
0:14:07 Karpathy
0:14:07 goes into
0:14:08 Rakesh’s
0:14:08 podcast
0:14:09 and says
0:14:09 I feel
0:14:09 like
0:14:10 you know
0:14:10 it’s
0:14:10 a decade
0:14:11 out
0:14:11 yeah I
0:14:12 know
0:14:12 when I saw
0:14:13 this meme
0:14:14 of what’s his
0:14:14 name
0:14:14 Ilya
0:14:15 Suskeva
0:14:15 but he
0:14:16 says like
0:14:16 the answer
0:14:17 will reveal
0:14:17 itself
0:14:18 and somebody
0:14:18 like
0:14:20 memed
0:14:20 I’m going
0:14:20 to say
0:14:21 photoshopped
0:14:21 but of course
0:14:22 it wouldn’t
0:14:22 have been
0:14:22 photoshopped
0:14:23 turned him
0:14:23 into a
0:14:23 Buddhist
0:14:23 monk
0:14:24 wearing
0:14:24 like an
0:14:24 orange
0:14:25 like an
0:14:25 orange
0:14:25 outfit
0:14:26 the future
0:14:27 will reveal
0:14:27 itself
0:14:29 but this is
0:14:30 the problem
0:14:30 we don’t
0:14:30 know
0:14:31 we don’t
0:14:31 have a way
0:14:32 of modeling
0:14:32 this
0:14:32 yeah
0:14:33 and so
0:14:35 let’s connect
0:14:36 this to sort
0:14:36 of the
0:14:37 you know
0:14:37 the upfront
0:14:37 investment
0:14:38 that some
0:14:38 of these
0:14:38 companies
0:14:39 are making
0:14:40 because we
0:14:40 don’t know
0:14:40 you know
0:14:41 is there
0:14:41 a risk
0:14:42 of over
0:14:43 investment
0:14:43 leading to
0:14:44 some
0:14:44 you know
0:14:45 potential
0:14:46 you know
0:14:46 bubble like
0:14:46 mechanics
0:14:47 or how
0:14:48 do you
0:14:48 think about
0:14:48 that question
0:14:50 well
0:14:51 deterministically
0:14:52 very new
0:14:53 very very
0:14:53 big
0:14:54 very very
0:14:54 exciting
0:14:55 world changing
0:14:55 things tend
0:14:56 to lead
0:14:56 to bubbles
0:14:58 and you
0:14:58 don’t think
0:14:59 anybody would
0:14:59 dispute that
0:15:00 you can see
0:15:00 some bubbly
0:15:01 behavior now
0:15:02 and you
0:15:02 know
0:15:02 you can
0:15:03 argue about
0:15:03 what kind
0:15:04 of bubble
0:15:04 but again
0:15:05 like that
0:15:05 doesn’t have
0:15:06 very much
0:15:06 predictive power
0:15:08 and you
0:15:09 know one
0:15:09 of the
0:15:10 features of
0:15:11 bubbles is
0:15:11 that when
0:15:11 everything’s
0:15:12 going
0:15:13 everything goes
0:15:14 up all at
0:15:14 once and
0:15:15 everyone looks
0:15:15 like a genius
0:15:16 and everyone
0:15:17 leverages and
0:15:18 cross leverages
0:15:18 and does
0:15:19 circular revenue
0:15:19 and that’s
0:15:20 great until
0:15:20 it’s not
0:15:22 and then
0:15:22 you get
0:15:23 kind of a
0:15:23 ratchet effect
0:15:24 as it goes
0:15:24 back down
0:15:24 again
0:15:26 so yeah
0:15:26 if we’re not
0:15:26 in a bubble
0:15:27 now we will
0:15:28 be I remember
0:15:28 Mark
0:15:29 Andreessen
0:15:29 saying you
0:15:29 know
0:15:30 1997 was not
0:15:31 a bubble
0:15:31 98 was not
0:15:32 a bubble
0:15:32 99 was a
0:15:33 bubble
0:15:34 are we in
0:15:35 97 now
0:15:36 or 98 or
0:15:36 99
0:15:37 you know
0:15:38 if we could
0:15:39 predict that
0:15:39 you know we’d
0:15:40 live in a
0:15:40 parallel universe
0:15:42 I think
0:15:43 you know
0:15:43 to the
0:15:44 there’s I
0:15:45 suppose maybe
0:15:46 kind of two
0:15:46 more specific
0:15:47 more tangible
0:15:48 answers to this
0:15:49 the first
0:15:49 the first
0:15:49 the first
0:15:49 of them
0:15:50 is we
0:15:50 don’t
0:15:51 really
0:15:51 know
0:15:53 what
0:15:53 the
0:15:54 compute
0:15:54 requirements
0:15:55 of this
0:15:55 stuff
0:15:55 are going
0:15:56 to be
0:15:57 and
0:15:58 forecasting
0:15:58 that
0:15:59 except like
0:16:00 more
0:16:01 and forecasting
0:16:02 that feels a lot
0:16:03 like trying to
0:16:03 forecast like
0:16:04 bandwidth use
0:16:05 in the late
0:16:06 90s
0:16:07 imagine if you were
0:16:08 trying to do the
0:16:09 algebra on that
0:16:09 you say well
0:16:10 this many users
0:16:11 you know how much
0:16:12 bandwidth does a
0:16:12 web page use
0:16:13 how will that
0:16:14 change how will
0:16:14 that change
0:16:15 as bandwidth
0:16:15 gets faster
0:16:16 what happens
0:16:16 with video
0:16:17 what kind
0:16:17 of video
0:16:18 what bandwidth
0:16:19 what bit rate
0:16:19 of video
0:16:20 how long
0:16:20 do people
0:16:21 watch a video
0:16:21 how much
0:16:22 video
0:16:23 and then
0:16:24 you’d like
0:16:25 you could
0:16:25 build the
0:16:26 spreadsheet
0:16:26 and it would
0:16:27 tell you
0:16:27 what bit rate
0:16:28 what global
0:16:29 bandwidth consumption
0:16:30 would be in
0:16:30 10 years
0:16:31 and then you
0:16:31 could try and
0:16:31 use that to
0:16:32 back calculate
0:16:32 how many
0:16:33 routers is this
0:16:34 going to sell
0:16:35 and you could
0:16:36 get a number
0:16:37 but it wouldn’t
0:16:37 be the number
0:16:38 you know
0:16:38 there’d be a
0:16:39 hundred fold
0:16:40 range of possible
0:16:41 outcomes from that
0:16:42 and you could
0:16:42 you know
0:16:42 you could make
0:16:43 the same point
0:16:44 about algebra
0:16:45 of consumption
0:16:45 now
0:16:47 so you know
0:16:48 right now
0:16:49 we have a bunch
0:16:50 of rational
0:16:50 actors saying
0:16:51 well this stuff
0:16:52 is transformative
0:16:53 and a huge threat
0:16:55 and we can’t
0:16:56 keep up with
0:16:56 demand for it
0:16:57 now
0:16:58 and as far
0:16:58 as we know
0:16:59 the demand
0:16:59 is going
0:16:59 to keep
0:17:00 going up
0:17:02 and you know
0:17:02 we’ve had a
0:17:03 variety of quotes
0:17:03 from all of the
0:17:04 hyperscalers
0:17:05 basically saying
0:17:05 the downside
0:17:06 of not investing
0:17:06 is bigger
0:17:07 than the downside
0:17:08 of over investing
0:17:13 that kind of
0:17:14 thing always
0:17:14 works well
0:17:15 until it doesn’t
0:17:16 yeah
0:17:17 and I saw
0:17:18 slightly strange
0:17:18 quote from
0:17:19 Mark Zuckerberg
0:17:20 saying well
0:17:21 if it turns out
0:17:21 that we’ve
0:17:21 over invested
0:17:22 we can just
0:17:23 resell the
0:17:23 capacity
0:17:24 and I thought
0:17:24 let me just
0:17:26 like stop you
0:17:27 there Mark
0:17:28 because if it
0:17:28 turns out
0:17:29 you can’t
0:17:29 use your
0:17:30 capacity
0:17:30 everybody else
0:17:31 is going to
0:17:31 have loads
0:17:31 of spare
0:17:32 capacity
0:17:32 as well
0:17:33 yeah
0:17:33 all these
0:17:34 people
0:17:34 now who are
0:17:35 desperate for
0:17:35 more capacity
0:17:36 if it turns
0:17:37 out we can
0:17:37 get the same
0:17:38 results for
0:17:38 hundreds of
0:17:39 the compute
0:17:41 that will be
0:17:41 true for
0:17:42 everyone else
0:17:42 two not
0:17:43 just you
0:17:46 so yeah
0:17:47 you know
0:17:48 in an investment
0:17:49 cycle like this
0:17:49 you tend to
0:17:50 get over
0:17:50 investment
0:17:52 but then
0:17:53 after that
0:17:54 there’s very
0:17:55 limited predictions
0:17:55 you can make
0:17:56 about what’s
0:17:56 going to happen
0:17:57 I think the
0:17:58 more useful
0:18:00 kind of way
0:18:01 to look at
0:18:02 this is to
0:18:03 think well
0:18:05 you’ve got
0:18:06 these kind
0:18:07 of transformative
0:18:08 capabilities
0:18:10 that are
0:18:11 already
0:18:12 increasing
0:18:13 the value
0:18:13 of your
0:18:13 existing
0:18:14 products
0:18:14 if you’re
0:18:15 Google or
0:18:15 Meta or
0:18:16 Amazon
0:18:17 and you’re
0:18:17 going to be
0:18:18 able to use
0:18:19 them to build
0:18:19 a bunch
0:18:20 more stuff
0:18:22 and
0:18:24 why would
0:18:24 you want
0:18:25 to let
0:18:25 somebody else
0:18:26 do that
0:18:27 rather than
0:18:27 you doing
0:18:28 it
0:18:28 as long
0:18:28 as you’re
0:18:29 able to
0:18:29 keep
0:18:30 funding
0:18:31 and selling
0:18:31 what you’re
0:18:32 building
0:18:35 and it
0:18:35 will turn
0:18:36 out that
0:18:37 we have
0:18:37 an evolution
0:18:38 of models
0:18:38 in the next
0:18:39 year
0:18:39 that means
0:18:39 you can
0:18:40 get the
0:18:40 same result
0:18:41 for
0:18:42 hundreds
0:18:43 of the
0:18:43 compute
0:18:43 that you’re
0:18:44 using
0:18:44 today
0:18:45 bearing in
0:18:45 mind that
0:18:46 it’s already
0:18:47 going down
0:18:48 pick your
0:18:49 numbers
0:18:49 20,
0:18:49 30,
0:18:50 40 times
0:18:50 a year
0:18:53 but then
0:18:53 the usage
0:18:54 is going
0:18:54 up
0:18:54 so you’re
0:18:54 in this
0:18:55 very
0:18:55 as I said
0:18:55 it’s like
0:18:56 trying to
0:18:56 predict
0:18:56 bandwidth
0:18:57 consumption
0:18:57 in the
0:18:57 late
0:18:57 90s
0:18:57 early
0:18:58 2000s
0:18:59 you can
0:19:00 throw
0:19:00 all the
0:19:00 parameters
0:19:00 out
0:19:01 but it
0:19:01 doesn’t
0:19:01 get you
0:19:01 something
0:19:01 useful
0:19:02 you just
0:19:02 kind of
0:19:02 need to
0:19:03 step back
0:19:03 and say
0:19:04 yeah but
0:19:04 is this
0:19:04 internet
0:19:05 thing
0:19:05 any good
0:19:07 well
0:19:08 yeah because
0:19:08 I’m curious
0:19:08 I’m curious
0:19:09 if the
0:19:09 bottlenecks
0:19:09 are
0:19:10 if you
0:19:10 see them
0:19:10 as more
0:19:10 on the
0:19:10 supply
0:19:11 side
0:19:11 or the
0:19:11 demand
0:19:12 side
0:19:12 you know
0:19:13 more
0:19:14 technical
0:19:14 constraints
0:19:15 or is
0:19:15 just
0:19:16 is AI
0:19:16 any good
0:19:17 are there
0:19:17 enough
0:19:18 use cases
0:19:19 to justify
0:19:22 what are you
0:19:22 seeing and
0:19:23 what are you
0:19:23 predicting
0:19:24 so
0:19:26 maybe two
0:19:26 answers to
0:19:27 this question
0:19:27 the first
0:19:27 of them
0:19:28 is I
0:19:28 think we’ve
0:19:28 had the
0:19:29 sort of
0:19:30 a bifurcation
0:19:30 of what
0:19:31 all the
0:19:31 questions
0:19:31 are
0:19:32 so there
0:19:32 are now
0:19:33 very very
0:19:34 detailed
0:19:34 conversations
0:19:35 about chips
0:19:35 and then
0:19:36 very very
0:19:36 detailed
0:19:37 conversations
0:19:38 about data
0:19:38 centers
0:19:38 and about
0:19:39 funding
0:19:39 for data
0:19:40 centers
0:19:40 and then
0:19:41 about
0:19:42 what is
0:19:43 a new
0:19:44 enterprise
0:19:44 SaaS
0:19:44 company
0:19:45 built on
0:19:45 AI
0:19:46 what margins
0:19:47 will it
0:19:47 have
0:19:48 and how
0:19:49 much money
0:19:49 does it
0:19:49 need to
0:19:50 raise
0:19:50 and so
0:19:50 there are
0:19:50 venture
0:19:51 capital
0:19:51 conversations
0:19:52 and so
0:19:52 there are
0:19:52 many
0:19:53 different
0:19:53 conversations
0:19:54 within which
0:19:55 like I don’t
0:19:55 know anything
0:19:56 about chips
0:19:56 you know I
0:19:56 can spell
0:19:57 ultraviolet but
0:19:58 like I don’t
0:19:58 know what
0:19:59 like an
0:19:59 ultraviolet
0:20:00 process is
0:20:01 it’s like
0:20:02 it’s more
0:20:02 violets
0:20:03 I don’t
0:20:03 know
0:20:06 and so
0:20:07 you’ve got
0:20:07 this you know
0:20:08 it’s like the
0:20:08 Milton Friedman
0:20:09 line
0:20:09 no one knows
0:20:10 how to build
0:20:10 a pencil
0:20:10 you’ve got
0:20:11 you know
0:20:11 we’ve got
0:20:12 this you know
0:20:12 it’s turned
0:20:12 into deployment
0:20:13 I think
0:20:14 a second
0:20:15 answer might
0:20:16 be I think
0:20:17 there’s two
0:20:17 kinds of
0:20:18 AI deployment
0:20:19 generative AI
0:20:19 deployment
0:20:20 one of them
0:20:21 is there
0:20:22 are places
0:20:23 where it’s
0:20:24 very easy
0:20:25 and obvious
0:20:25 right now
0:20:26 to see what
0:20:26 you would
0:20:27 do with
0:20:27 this
0:20:28 which is
0:20:29 basically
0:20:29 software
0:20:30 development
0:20:31 marketing
0:20:33 point
0:20:33 solutions
0:20:35 for many
0:20:36 very boring
0:20:37 very specific
0:20:38 enterprise
0:20:38 use cases
0:20:40 and also
0:20:40 basically
0:20:41 people like
0:20:41 us
0:20:42 which are
0:20:43 people who
0:20:43 have kind
0:20:43 of very
0:20:44 open
0:20:45 very free
0:20:45 form
0:20:46 very flexible
0:20:46 jobs
0:20:47 with many
0:20:47 different
0:20:48 things
0:20:49 and people
0:20:50 who are always
0:20:50 looking for ways
0:20:51 to optimize
0:20:51 that
0:20:52 and so
0:20:53 you get
0:20:53 people in
0:20:53 Silicon Valley
0:20:54 who are like
0:20:54 you know
0:20:54 I spend
0:20:55 all my
0:20:55 time in
0:20:56 chat GPT
0:20:56 I don’t
0:20:56 use
0:20:57 Google
0:20:57 anymore
0:20:58 you know
0:20:58 I’ve
0:20:59 replaced
0:20:59 my
0:20:59 CRM
0:21:00 with
0:21:00 this
0:21:02 and
0:21:02 you
0:21:03 kind of
0:21:03 and then
0:21:04 obviously
0:21:04 people who
0:21:04 write
0:21:05 if you’re
0:21:05 writing
0:21:05 codes
0:21:05 this works
0:21:06 really well
0:21:06 if you’re
0:21:07 in marketing
0:21:07 you know
0:21:07 all these
0:21:08 stories of
0:21:08 big companies
0:21:08 where you know
0:21:09 they’re making
0:21:10 300 assets
0:21:10 where they would
0:21:10 have made
0:21:11 30
0:21:13 and then
0:21:13 Accenture
0:21:14 and Bain
0:21:14 and McKinsey
0:21:15 and Infosys
0:21:16 and so on
0:21:16 sitting and
0:21:17 solving very
0:21:18 specific problems
0:21:18 inside big
0:21:18 companies
0:21:20 then there’s a
0:21:20 whole bunch
0:21:21 of other
0:21:22 people who
0:21:22 look at it
0:21:23 and they’re
0:21:23 like
0:21:25 it’s
0:21:25 okay
0:21:28 and
0:21:28 you go
0:21:28 and look
0:21:29 at the
0:21:30 usage data
0:21:31 and you
0:21:31 see
0:21:32 okay
0:21:33 chat GPT
0:21:33 has got
0:21:34 800 or
0:21:34 900 million
0:21:35 weekly
0:21:35 active users
0:21:37 5% of
0:21:37 people are
0:21:38 paying
0:21:40 and
0:21:40 then you
0:21:41 go and look
0:21:42 at all the
0:21:42 survey data
0:21:43 and you know
0:21:43 it’s very
0:21:44 fragmented
0:21:44 and inconsistent
0:21:45 but it all
0:21:45 sort of
0:21:46 points to
0:21:46 like
0:21:47 something like
0:21:48 10 or 15%
0:21:49 of people
0:21:49 into the
0:21:50 developed world
0:21:50 are using
0:21:51 this every
0:21:51 day
0:21:53 another 20 or
0:21:53 30% of people
0:21:54 are using it
0:21:55 every week
0:21:56 and if you’re
0:21:56 the kind of
0:21:56 person who
0:21:57 is using
0:21:57 this for
0:21:58 hours every
0:21:58 day
0:21:59 ask yourself
0:21:59 why
0:22:01 five times
0:22:02 more people
0:22:02 look at it
0:22:03 get it
0:22:03 know what
0:22:04 it is
0:22:04 have an
0:22:04 account
0:22:05 know how
0:22:05 to use it
0:22:06 and can’t
0:22:06 think of
0:22:07 anything to do
0:22:07 with it
0:22:08 this week
0:22:08 or next
0:22:09 week
0:22:11 why is
0:22:11 that
0:22:13 is it
0:22:13 because
0:22:13 it’s
0:22:13 early
0:22:14 and it’s
0:22:14 not like
0:22:15 a young
0:22:15 people
0:22:15 thing
0:22:16 either
0:22:17 incidentally
0:22:17 so is
0:22:18 that just
0:22:18 because
0:22:18 it’s
0:22:19 early
0:22:20 is it
0:22:20 because
0:22:21 of the
0:22:22 error
0:22:22 rates
0:22:23 is it
0:22:23 because
0:22:24 you have
0:22:25 to map
0:22:25 it against
0:22:26 what you
0:22:26 do
0:22:26 every
0:22:26 day
0:22:27 and
0:22:29 one of
0:22:29 the
0:22:30 analogy
0:22:30 I was
0:22:30 used
0:22:30 to
0:22:30 use
0:22:30 which
0:22:31 isn’t
0:22:31 in the
0:22:31 current
0:22:31 presentation
0:22:32 I’ve
0:22:32 used
0:22:32 in
0:22:32 previous
0:22:33 presentations
0:22:33 is
0:22:34 imagine
0:22:34 you’re
0:22:34 an
0:22:35 accountant
0:22:35 and you
0:22:35 see
0:22:35 software
0:22:36 spreadsheets
0:22:36 for the
0:22:36 first
0:22:36 time
0:22:37 this
0:22:37 thing
0:22:38 can
0:22:38 do
0:22:39 a month
0:22:39 of work
0:22:39 in 10
0:22:40 minutes
0:22:40 almost
0:22:41 centrally
0:22:43 you want
0:22:43 to change
0:22:44 you want
0:22:45 to recalculate
0:22:45 that DCF
0:22:46 that 10 year
0:22:47 DCF
0:22:47 with a different
0:22:48 discount rate
0:22:48 I’ve done it
0:22:49 before you
0:22:50 finished asking
0:22:50 me to
0:22:50 and that
0:22:51 would have
0:22:51 been like
0:22:51 a day
0:22:52 or two
0:22:52 days
0:22:52 or three
0:22:52 days
0:22:53 of work
0:22:53 to recalculate
0:22:53 all these
0:22:54 numbers
0:22:55 great
0:22:55 now imagine
0:22:56 you’re
0:22:56 a lawyer
0:22:57 and you
0:22:57 see it
0:22:58 and you
0:22:58 think
0:22:59 well that’s
0:23:00 great
0:23:01 my accountant
0:23:01 should see
0:23:02 it
0:23:02 maybe I’ll
0:23:03 use it
0:23:03 next week
0:23:04 when I’m
0:23:04 making a
0:23:05 table of
0:23:05 my billable
0:23:06 hours
0:23:06 but that’s
0:23:07 not what
0:23:07 I do
0:23:07 all day
0:23:09 and Excel
0:23:10 doesn’t
0:23:11 use
0:23:11 do things
0:23:12 that a
0:23:12 lawyer can
0:23:12 do every
0:23:13 day
0:23:14 and I
0:23:14 think
0:23:15 there’s
0:23:15 this other
0:23:16 class
0:23:16 of person
0:23:17 that’s
0:23:17 like
0:23:18 I’m not
0:23:19 sure what
0:23:19 to do
0:23:19 with this
0:23:20 and some
0:23:20 of that
0:23:21 is habit
0:23:21 some
0:23:22 of that
0:23:22 is like
0:23:23 realizing
0:23:23 no
0:23:24 instead
0:23:24 of doing
0:23:24 it
0:23:25 that way
0:23:25 I could
0:23:25 do it
0:23:26 this way
0:23:27 but that’s
0:23:28 also what
0:23:29 products are
0:23:30 like every
0:23:30 entrepreneur
0:23:31 who comes
0:23:32 into A16Z
0:23:33 when I was
0:23:33 there from
0:23:34 2014 to
0:23:35 2019
0:23:35 and I’m
0:23:36 sure now
0:23:37 like you
0:23:37 could look
0:23:38 at any
0:23:38 company that
0:23:39 comes in
0:23:39 and say
0:23:40 that’s
0:23:40 basically
0:23:41 a database
0:23:42 that’s
0:23:43 basically
0:23:44 a CRM
0:23:45 that’s
0:23:46 basically
0:23:47 Oracle
0:23:47 or Google
0:23:48 Docs
0:23:49 except
0:23:50 that they
0:23:51 realize
0:23:51 there’s
0:23:51 this
0:23:52 problem
0:23:52 or this
0:23:53 workflow
0:23:53 inside
0:23:54 this
0:23:54 industry
0:23:55 and
0:23:55 worked
0:23:55 out
0:23:56 how
0:23:56 to
0:23:56 use
0:23:56 a
0:23:57 database
0:23:58 or CRM
0:23:59 or basically
0:23:59 concepts
0:24:00 from 5,
0:24:00 10,
0:24:01 20 years
0:24:01 ago
0:24:02 and solve
0:24:03 that problem
0:24:03 for people
0:24:04 in that
0:24:04 industry
0:24:05 and go
0:24:05 in and
0:24:06 sell it
0:24:06 to them
0:24:06 and work
0:24:07 out how
0:24:07 they can
0:24:07 get it
0:24:07 to use
0:24:08 it
0:24:09 and
0:24:10 so this
0:24:10 is why
0:24:11 you look
0:24:11 at data
0:24:12 on this
0:24:12 that depending
0:24:13 on how
0:24:13 you count it
0:24:13 the typical
0:24:14 big company
0:24:14 today has
0:24:15 400 to
0:24:15 500
0:24:15 SaaS
0:24:15 apps
0:24:16 in the
0:24:17 US
0:24:18 400 to
0:24:18 500
0:24:18 SaaS
0:24:19 applications
0:24:20 and they’re
0:24:20 all
0:24:21 basically
0:24:21 doing
0:24:22 something
0:24:22 you could
0:24:22 do in
0:24:22 Oracle
0:24:23 or Excel
0:24:24 or email
0:24:28 and that’s
0:24:29 the other
0:24:29 side
0:24:30 I’m monologuing
0:24:31 I’m afraid
0:24:31 but this is
0:24:32 the other
0:24:32 side of
0:24:33 what do
0:24:33 you do
0:24:34 with these
0:24:34 things
0:24:34 do you
0:24:35 just go
0:24:35 to the
0:24:35 bot
0:24:36 and ask
0:24:36 it to
0:24:36 do a
0:24:36 thing
0:24:37 for you
0:24:38 or does
0:24:39 an enterprise
0:24:40 salesperson
0:24:40 come to
0:24:41 your boss
0:24:41 and sell
0:24:42 you a
0:24:42 thing
0:24:43 that means
0:24:44 now you
0:24:44 press a
0:24:44 button
0:24:45 and it
0:24:45 analyzes
0:24:46 this process
0:24:47 that you
0:24:47 needed
0:24:48 that you
0:24:49 never realized
0:24:49 you were
0:24:50 even doing
0:24:52 and I
0:24:53 feel like
0:24:53 that’s
0:24:54 I mean
0:24:55 that’s
0:24:56 why there
0:24:56 are AI
0:24:57 software
0:24:57 companies
0:24:59 really
0:25:00 isn’t that
0:25:00 what they’re
0:25:00 doing
0:25:01 they’re
0:25:01 unbundling
0:25:01 chat
0:25:02 GPT
0:25:02 just as
0:25:03 the
0:25:03 enterprise
0:25:03 software
0:25:04 company
0:25:04 of 10
0:25:04 years ago
0:25:04 was
0:25:05 unbundling
0:25:05 Oracle
0:25:05 or Google
0:25:06 or Excel
0:25:07 do you
0:25:07 have the
0:25:07 view
0:25:07 that
0:25:08 what
0:25:08 Excel
0:25:09 did
0:25:09 for
0:25:10 accountants
0:25:14 AI
0:25:14 is now
0:25:14 doing
0:25:15 for
0:25:15 coders
0:25:16 and
0:25:17 developers
0:25:17 but hasn’t
0:25:18 quite
0:25:18 figured out
0:25:19 that daily
0:25:20 critical
0:25:20 workflow
0:25:20 for
0:25:21 other
0:25:21 job
0:25:21 positions
0:25:22 and so
0:25:22 it’s
0:25:22 unclear
0:25:23 for
0:25:23 people
0:25:23 who
0:25:23 aren’t
0:25:24 developers
0:25:25 why I
0:25:25 should
0:25:25 be
0:25:25 using
0:25:25 this
0:25:26 for
0:25:26 many
0:25:26 hours
0:25:27 a
0:25:27 day
0:25:28 I
0:25:29 think
0:25:29 there’s
0:25:29 a lot
0:25:29 of
0:25:30 people
0:25:30 who
0:25:31 don’t
0:25:31 have
0:25:32 tasks
0:25:32 that
0:25:33 work
0:25:33 very
0:25:33 well
0:25:33 with
0:25:34 this
0:25:35 and
0:25:35 then
0:25:35 there’s
0:25:36 a lot
0:25:36 of
0:25:36 people
0:25:37 who
0:25:37 need
0:25:37 it
0:25:37 to
0:25:37 be
0:25:38 wrapped
0:25:38 in
0:25:38 a
0:25:39 product
0:25:39 and
0:25:39 a
0:25:39 workflow
0:25:40 and
0:25:40 tooling
0:25:40 and
0:25:41 UX
0:25:41 and
0:25:41 someone
0:25:41 to
0:25:42 come
0:25:42 and say
0:25:43 hey
0:25:43 have you
0:25:43 realized
0:25:43 you
0:25:43 could
0:25:44 do
0:25:44 it
0:25:44 with
0:25:44 this
0:25:46 I
0:25:46 had
0:25:46 this
0:25:46 conversation
0:25:47 in the
0:25:47 summer
0:25:48 with
0:25:48 Balaji
0:25:49 who’s
0:25:49 another
0:25:50 former
0:25:50 A16Z
0:25:51 person
0:25:52 and
0:25:53 he was
0:25:53 making
0:25:53 this
0:25:53 point
0:25:54 about
0:25:54 validation
0:25:55 that
0:25:55 can
0:25:55 you
0:25:56 because
0:25:56 these
0:25:56 things
0:25:57 still get
0:25:57 stuff
0:25:57 wrong
0:25:57 and
0:25:58 people
0:25:58 in
0:25:58 the
0:25:58 valley
0:25:58 often
0:25:58 kind
0:25:59 of
0:25:59 hand
0:25:59 wave
0:25:59 this
0:26:00 away
0:26:01 but
0:26:01 you
0:26:01 know
0:26:01 there
0:26:01 are
0:26:02 questions
0:26:02 that
0:26:03 have
0:26:04 specific
0:26:04 answers
0:26:05 where it
0:26:05 needs
0:26:05 to be
0:26:06 the right
0:26:06 answer
0:26:06 or one
0:26:07 of a
0:26:07 limited
0:26:08 set
0:26:08 of
0:26:08 right
0:26:08 answers
0:26:09 can
0:26:09 you
0:26:10 validate
0:26:10 that
0:26:10 mechanistically
0:26:12 if
0:26:13 not
0:26:13 is it
0:26:13 efficient
0:26:14 to
0:26:14 validate
0:26:14 it
0:26:14 with
0:26:15 people
0:26:16 so
0:26:16 you
0:26:16 know
0:26:16 the
0:26:17 marketing
0:26:17 use
0:26:17 case
0:26:17 it’s
0:26:18 a lot
0:26:18 more
0:26:18 efficient
0:26:18 to
0:26:18 get
0:26:19 a
0:26:19 machine
0:26:19 to
0:26:19 make
0:26:19 you
0:26:20 200
0:26:20 pictures
0:26:20 and
0:26:21 have
0:26:21 a
0:26:21 person
0:26:21 look
0:26:21 at
0:26:21 them
0:26:22 and
0:26:22 pick
0:26:22 10
0:26:22 that
0:26:22 are
0:26:22 good
0:26:23 than
0:26:23 to
0:26:23 have
0:26:24 people
0:26:25 make
0:26:25 10
0:26:25 good
0:26:26 images
0:26:26 or
0:26:26 100
0:26:26 even
0:26:27 more
0:26:27 if
0:26:27 you’re
0:26:27 going
0:26:27 to
0:26:27 make
0:26:28 500
0:26:28 images
0:26:28 and
0:26:28 pick
0:26:28 100
0:26:29 that
0:26:29 are
0:26:29 good
0:26:29 that’s
0:26:30 a lot
0:26:30 more
0:26:30 efficient
0:26:31 than
0:26:31 having
0:26:31 a
0:26:31 person
0:26:31 make
0:26:32 100
0:26:32 images
0:26:33 but
0:26:33 on
0:26:33 the
0:26:34 other
0:26:34 hand
0:26:34 if
0:26:34 you’re
0:26:34 doing
0:26:34 something
0:26:34 like
0:26:35 data
0:26:35 entry
0:26:35 and
0:26:35 I
0:26:36 wrote
0:26:36 something
0:26:36 about
0:26:36 this
0:26:37 about
0:26:38 OpenAI
0:26:38 launch
0:26:39 deep
0:26:39 research
0:26:40 OpenAI
0:26:40 launch
0:26:40 deep
0:26:41 research
0:26:41 their
0:26:41 whole
0:26:42 marketing
0:26:42 case
0:26:42 is
0:26:42 it
0:26:42 goes
0:26:43 off
0:26:43 and
0:26:43 collects
0:26:44 data
0:26:44 about
0:26:44 the
0:26:44 mobile
0:26:45 market
0:26:45 I
0:26:45 used
0:26:45 to be
0:26:45 a
0:26:45 mobile
0:26:46 analyst
0:26:46 the
0:26:46 numbers
0:26:46 are
0:26:46 wrong
0:26:48 their
0:26:49 use
0:26:49 case
0:26:49 of
0:26:50 look
0:26:50 how
0:26:50 useful
0:26:50 this
0:26:51 is
0:26:51 their
0:26:51 numbers
0:26:51 are
0:26:52 wrong
0:26:53 and
0:26:53 in
0:26:53 some
0:26:54 cases
0:26:54 they’re
0:26:54 wrong
0:26:55 because
0:26:55 they’ve
0:26:56 literally
0:26:56 transcribed
0:26:57 the
0:26:57 number
0:26:58 incorrectly
0:26:58 from
0:26:58 the
0:26:58 source
0:26:59 in
0:26:59 other
0:27:00 cases
0:27:00 it’s
0:27:00 wrong
0:27:00 because
0:27:01 they’ve
0:27:01 used
0:27:01 a
0:27:01 source
0:27:02 they
0:27:02 shouldn’t
0:27:02 have
0:27:03 used
0:27:03 but
0:27:04 like
0:27:04 if
0:27:04 I
0:27:04 asked
0:27:04 an
0:27:05 intern
0:27:05 to
0:27:05 do
0:27:05 it
0:27:05 for
0:27:05 me
0:27:06 then
0:27:06 an
0:27:06 intern
0:27:06 would
0:27:06 probably
0:27:06 have
0:27:07 picked
0:27:07 that
0:27:08 and
0:27:08 to
0:27:09 the
0:27:09 point
0:27:09 about
0:27:10 verification
0:27:11 if you’re
0:27:12 going to do
0:27:12 data
0:27:12 entry
0:27:13 if I’m
0:27:13 going to
0:27:14 ask
0:27:14 a
0:27:14 machine
0:27:14 to
0:27:15 copy
0:27:15 200
0:27:15 numbers
0:27:16 out
0:27:16 of
0:27:16 200
0:27:17 PDFs
0:27:17 and
0:27:17 then
0:27:17 I’m
0:27:17 going to
0:27:17 have
0:27:18 to
0:27:18 check
0:27:18 all
0:27:18 200
0:27:18 of
0:27:19 those
0:27:19 numbers
0:27:19 I’m
0:27:19 going to
0:27:19 as well
0:27:20 just
0:27:20 do
0:27:20 it
0:27:20 myself
0:27:22 so
0:27:22 you’ve
0:27:23 got
0:27:23 like
0:27:23 a
0:27:24 whole
0:27:24 swirling
0:27:25 matrix
0:27:26 of
0:27:27 how
0:27:27 do
0:27:27 you
0:27:28 map
0:27:28 this
0:27:29 against
0:27:30 existing
0:27:30 problems
0:27:31 but
0:27:31 the
0:27:31 other
0:27:32 side
0:27:32 of
0:27:32 it
0:27:32 is
0:27:33 how
0:27:33 do
0:27:33 you
0:27:34 map
0:27:34 this
0:27:34 against
0:27:35 new
0:27:35 things
0:27:35 that
0:27:35 you
0:27:36 couldn’t
0:27:36 have
0:27:36 done
0:27:36 before
0:27:38 and
0:27:39 this
0:27:39 comes
0:27:39 back
0:27:39 to
0:27:40 my
0:27:40 point
0:27:40 about
0:27:40 platform
0:27:41 ships
0:27:41 because
0:27:41 you know
0:27:42 I see
0:27:42 people
0:27:42 looking
0:27:43 at
0:27:44 genitive
0:27:44 AR
0:27:44 and
0:27:45 saying
0:27:45 well
0:27:45 this
0:27:45 is
0:27:45 useless
0:27:46 because
0:27:46 it
0:27:46 makes
0:27:46 mistakes
0:27:47 and
0:27:47 I
0:27:47 think
0:27:48 that’s
0:27:48 kind
0:27:48 of
0:27:48 like
0:27:49 looking
0:27:49 at
0:27:49 an
0:27:50 Apple
0:27:50 2
0:27:50 in
0:27:50 the
0:27:50 late
0:27:51 70s
0:27:51 and
0:27:51 saying
0:27:51 could
0:27:52 you
0:27:52 use
0:27:53 these
0:27:53 to
0:27:53 run
0:27:53 banks
0:27:54 to
0:27:54 which
0:27:54 the
0:27:54 answer
0:27:54 is
0:27:55 no
0:27:56 but
0:27:56 that’s
0:27:56 kind
0:27:57 of
0:27:57 the
0:27:57 wrong
0:27:58 question
0:27:59 could
0:27:59 you
0:27:59 build
0:28:00 professional
0:28:01 video
0:28:01 editing
0:28:02 inside
0:28:02 Netscape
0:28:03 no
0:28:04 but
0:28:04 that’s
0:28:05 the
0:28:05 wrong
0:28:06 question
0:28:06 and
0:28:07 later
0:28:07 20
0:28:08 years
0:28:08 later
0:28:08 you
0:28:08 can
0:28:08 but
0:28:09 meanwhile
0:28:09 it
0:28:09 does
0:28:09 a
0:28:10 whole
0:28:10 bunch
0:28:10 of
0:28:10 other
0:28:10 stuff
0:28:10 the
0:28:11 same
0:28:11 with
0:28:11 mobile
0:28:12 can
0:28:12 you
0:28:12 use
0:28:13 mobile
0:28:13 to
0:28:13 replace
0:28:14 your
0:28:15 five
0:28:15 screen
0:28:16 professional
0:28:16 programming
0:28:16 rig
0:28:17 no
0:28:17 therefore
0:28:17 it
0:28:17 can’t
0:28:18 replace
0:28:18 PCs
0:28:18 well
0:28:20 guess
0:28:20 what
0:28:20 5 billion
0:28:21 people
0:28:21 have got
0:28:21 a
0:28:21 smartphone
0:28:21 and
0:28:22 700 or 800
0:28:22 million
0:28:22 people
0:28:22 have got
0:28:23 a
0:28:23 consumer
0:28:23 PC
0:28:23 so
0:28:24 it
0:28:24 kind
0:28:24 of
0:28:24 did
0:28:25 but
0:28:25 did
0:28:25 a
0:28:25 different
0:28:25 thing
0:28:26 and
0:28:27 the
0:28:27 new
0:28:27 thing
0:28:27 is
0:28:30 generally
0:28:30 not
0:28:31 very
0:28:31 good
0:28:31 or
0:28:32 terrible
0:28:32 at
0:28:32 the
0:28:32 stuff
0:28:32 that
0:28:32 was
0:28:33 important
0:28:33 to
0:28:33 the
0:28:33 old
0:28:33 thing
0:28:34 but
0:28:34 it
0:28:34 does
0:28:35 something
0:28:35 else
0:28:36 right
0:28:37 and
0:28:38 a lot
0:28:38 of
0:28:38 the
0:28:39 question
0:28:39 is
0:28:44 old
0:28:44 tasks
0:28:44 that
0:28:45 generative
0:28:45 AI
0:28:45 is
0:28:45 good
0:28:46 at
0:28:46 there’s
0:28:46 also
0:28:47 many
0:28:47 more
0:28:47 old
0:28:48 tasks
0:28:48 that
0:28:48 generative
0:28:49 AI
0:28:49 is
0:28:49 maybe
0:28:49 not
0:28:49 very
0:28:50 good
0:28:50 at
0:28:51 but
0:28:51 then
0:28:51 there’s
0:28:51 a whole
0:28:51 bunch
0:28:52 of
0:28:52 other
0:28:52 things
0:28:52 that
0:28:53 you
0:28:53 would
0:28:53 never
0:28:53 have
0:28:54 done
0:28:54 before
0:28:55 that
0:28:55 generative
0:28:56 AI
0:28:56 is
0:28:56 really
0:28:57 really
0:28:57 good
0:28:57 at
0:28:59 and
0:28:59 then
0:28:59 how
0:28:59 do
0:28:59 you
0:29:00 find
0:29:00 those
0:29:00 or
0:29:01 think
0:29:01 of
0:29:01 those
0:29:01 and
0:29:01 how
0:29:01 much
0:29:02 of
0:29:02 that
0:29:02 is
0:29:02 the
0:29:02 user
0:29:03 thinking
0:29:03 of it
0:29:03 faced
0:29:03 with
0:29:03 the
0:29:04 general
0:29:14 product
0:29:15 with
0:29:15 a
0:29:15 button
0:29:15 that
0:29:16 will
0:29:16 do
0:29:16 it
0:29:16 for
0:29:16 you
0:29:18 and
0:29:18 that’s
0:29:19 why
0:29:19 there
0:29:19 are
0:29:19 software
0:29:20 companies
0:29:20 right
0:29:21 and
0:29:21 on
0:29:22 mobile
0:29:22 some
0:29:22 of
0:29:22 the
0:29:23 new
0:29:23 use
0:29:23 cases
0:29:24 getting
0:29:24 in
0:29:25 strangers
0:29:25 cars
0:29:26 we
0:29:26 mentioned
0:29:26 left
0:29:26 an
0:29:26 Uber
0:29:27 or
0:29:27 sort
0:29:27 of
0:29:28 you
0:29:28 know
0:29:28 lending
0:29:28 people
0:29:28 you
0:29:28 met
0:29:28 via
0:29:29 an
0:29:29 app
0:29:29 or
0:29:31 sort
0:29:31 of
0:29:32 you
0:29:32 know
0:29:32 lending
0:29:32 your
0:29:33 spare
0:29:33 bedroom
0:29:34 out
0:29:35 etc
0:29:36 and
0:29:36 those
0:29:37 were
0:29:37 net
0:29:37 new
0:29:37 companies
0:29:38 that
0:29:39 were
0:29:39 built
0:29:39 around
0:29:39 those
0:29:40 behaviors
0:29:40 and
0:29:40 I
0:29:40 think
0:29:40 for
0:29:41 AI
0:29:41 there’s
0:29:41 still
0:29:41 questions
0:29:41 of
0:29:42 what
0:29:42 are
0:29:43 those
0:29:43 net
0:29:43 new
0:29:43 behaviors
0:29:44 we’re
0:29:44 starting
0:29:44 to see
0:29:45 some
0:29:45 in
0:29:45 terms
0:29:46 of
0:29:46 people
0:29:47 engaging
0:29:47 and
0:29:48 talking
0:29:48 with
0:29:49 chat
0:29:49 bots
0:29:49 instead
0:29:50 of
0:29:50 humans
0:29:50 or
0:29:51 in
0:29:52 addition
0:29:53 and
0:29:53 then
0:29:53 there’s
0:29:53 a
0:29:53 question
0:29:53 of
0:29:54 are
0:29:54 these
0:29:54 done
0:29:55 by
0:29:55 the
0:29:56 model
0:29:56 providers
0:29:57 that
0:29:57 currently
0:29:57 exist
0:29:58 or
0:29:58 are
0:29:58 these
0:29:58 done
0:29:58 by
0:29:59 net
0:30:03 is
0:30:03 how
0:30:04 far
0:30:04 up
0:30:04 the
0:30:04 stack
0:30:04 does
0:30:04 the
0:30:04 new
0:30:05 thing
0:30:05 go
0:30:07 and
0:30:08 I
0:30:08 was
0:30:08 talking
0:30:08 about
0:30:08 this
0:30:09 with
0:30:09 another
0:30:09 former
0:30:10 A16Z
0:30:10 person
0:30:10 who
0:30:11 pointed
0:30:11 out
0:30:11 that
0:30:11 in
0:30:12 the
0:30:12 mid
0:30:13 90s
0:30:14 people
0:30:14 argued
0:30:15 that
0:30:15 the
0:30:15 operating
0:30:16 system
0:30:16 does
0:30:16 all
0:30:16 of it
0:30:17 and
0:30:18 Windows
0:30:18 apps
0:30:18 are
0:30:18 basically
0:30:19 just
0:30:19 thin
0:30:20 Win32
0:30:20 wrappers
0:30:21 and
0:30:22 Office
0:30:22 is
0:30:23 basically
0:30:23 just
0:30:23 a
0:30:23 thin
0:30:24 Win32
0:30:24 wrapper
0:30:24 like
0:30:25 all the
0:30:25 important
0:30:25 stuff
0:30:25 is
0:30:26 being
0:30:26 done
0:30:26 by
0:30:26 the
0:30:26 OS
0:30:27 whether
0:30:27 it’s
0:30:27 the
0:30:28 document
0:30:28 management
0:30:29 and
0:30:29 printing
0:30:29 and
0:30:29 storage
0:30:30 and
0:30:30 display
0:30:31 which
0:30:31 all
0:30:31 stuff
0:30:31 that
0:30:31 used
0:30:31 to be
0:30:31 done
0:30:32 by
0:30:32 apps
0:30:32 like
0:30:33 on
0:30:33 DOS
0:30:33 the
0:30:33 apps
0:30:33 had
0:30:33 to
0:30:34 do
0:30:34 printing
0:30:34 the
0:30:34 apps
0:30:34 had
0:30:34 to
0:30:35 manage
0:30:35 the
0:30:35 display
0:30:36 we
0:30:36 moved to
0:30:36 Windows
0:30:37 90%
0:30:37 of the
0:30:37 stuff
0:30:37 that
0:30:37 the
0:30:37 app
0:30:38 used
0:30:38 to do
0:30:38 is
0:30:38 now
0:30:38 being
0:30:38 done
0:30:39 by
0:30:39 Windows
0:30:40 and
0:30:41 so
0:30:41 Office
0:30:42 is
0:30:42 just
0:30:42 like
0:30:42 a
0:30:42 thin
0:30:43 Win32
0:30:43 wrapper
0:30:44 and
0:30:44 all the
0:30:44 stuff
0:30:44 has
0:30:44 been
0:30:45 done
0:30:45 by
0:30:45 the
0:30:46 OS
0:30:46 and
0:30:46 it
0:30:46 turned
0:30:46 out
0:30:46 well
0:30:46 that
0:30:47 was
0:30:47 again
0:30:48 frameworks
0:30:48 are
0:30:48 useful
0:30:48 but
0:30:48 that’s
0:30:49 not
0:30:49 maybe
0:30:49 not
0:30:49 a
0:30:50 useful
0:30:50 way
0:30:50 of
0:30:50 thinking
0:30:50 about
0:30:50 what’s
0:30:50 going
0:30:51 on
0:30:51 and
0:30:52 the
0:30:52 same
0:30:52 thing
0:30:53 now
0:30:53 like
0:30:54 how
0:30:54 much
0:30:54 does
0:30:55 this
0:30:55 need
0:30:56 single
0:30:57 dedicated
0:30:58 understanding
0:30:59 of how
0:30:59 that
0:30:59 market
0:31:00 it
0:31:00 works
0:31:00 or
0:31:00 what
0:31:01 that
0:31:01 market
0:31:01 is
0:31:01 and
0:31:01 what
0:31:02 you
0:31:02 would
0:31:02 do
0:31:02 with
0:31:02 that
0:31:03 I
0:31:04 remember
0:31:04 when
0:31:04 we
0:31:04 were
0:31:04 at
0:31:05 A16Z
0:31:05 there
0:31:05 was
0:31:05 an
0:31:06 investment
0:31:06 in
0:31:06 a
0:31:06 company
0:31:07 called
0:31:07 Everlaw
0:31:08 which
0:31:08 is
0:31:10 legal
0:31:10 discovery
0:31:11 in
0:31:11 the
0:31:11 cloud
0:31:12 and
0:31:13 so
0:31:13 machine
0:31:13 learning
0:31:14 happens
0:31:14 and
0:31:14 so
0:31:14 now
0:31:15 they
0:31:15 can
0:31:15 do
0:31:15 translation
0:31:16 are
0:31:16 they
0:31:17 worried
0:31:17 that
0:31:17 lawyers
0:31:17 are
0:31:18 going
0:31:18 to
0:31:18 say
0:31:18 we
0:31:18 don’t
0:31:18 need
0:31:18 you
0:31:19 guys
0:31:19 anymore
0:31:19 we’re
0:31:19 just
0:31:19 going
0:31:19 to
0:31:20 get
0:31:20 a
0:31:20 translate
0:31:21 app
0:31:21 and
0:31:21 a
0:31:21 sentiment
0:31:22 analysis
0:31:22 app
0:31:22 from
0:31:22 AWS
0:31:26 law firms
0:31:26 want to
0:31:26 buy a
0:31:27 thing
0:31:28 legal
0:31:28 discovery
0:31:29 software
0:31:29 management
0:31:29 they
0:31:30 don’t
0:31:30 want to
0:31:31 write
0:31:31 their
0:31:31 own
0:31:31 do
0:31:32 API
0:31:32 calls
0:31:32 I
0:31:32 mean
0:31:32 very
0:31:33 big
0:31:33 law firms
0:31:33 might
0:31:34 but
0:31:34 typical
0:31:35 law firm
0:31:35 isn’t
0:31:35 going
0:31:35 to
0:31:35 do
0:31:36 that
0:31:36 people
0:31:36 buy
0:31:37 solutions
0:31:37 they
0:31:37 don’t
0:31:37 buy
0:31:37 technologies
0:31:39 and
0:31:39 the
0:31:39 same
0:31:39 thing
0:31:40 here
0:31:40 like
0:31:40 how
0:31:41 far
0:31:41 up
0:31:41 the
0:31:41 stack
0:31:42 do
0:31:42 these
0:31:43 models
0:31:43 go
0:31:45 how
0:31:46 much
0:31:46 can
0:31:46 you
0:31:46 turn
0:31:47 things
0:31:47 into
0:31:49 a
0:31:50 widget
0:31:51 how
0:31:51 much
0:31:51 can
0:31:51 you
0:31:51 turn
0:31:52 things
0:31:52 into
0:31:53 an
0:31:53 LLM
0:31:54 request
0:31:55 and
0:31:55 how
0:31:55 much
0:31:56 know
0:31:56 does
0:31:56 it
0:31:56 turn
0:31:57 out
0:31:57 that
0:31:57 you
0:31:57 need
0:31:58 that
0:31:58 dedicated
0:31:59 UI
0:31:59 the fun
0:31:59 thing
0:31:59 is
0:32:00 you can
0:32:00 see
0:32:00 this
0:32:00 around
0:32:00 Google
0:32:00 because
0:32:01 Google
0:32:01 had
0:32:01 this
0:32:01 whole
0:32:01 idea
0:32:02 that
0:32:02 everything
0:32:02 would
0:32:02 just
0:32:02 be
0:32:02 a
0:32:02 Google
0:32:03 query
0:32:03 and
0:32:03 Google
0:32:03 would
0:32:04 work
0:32:04 out
0:32:04 what
0:32:04 the
0:32:04 query
0:32:04 was
0:32:05 and guess
0:32:05 what
0:32:06 now you
0:32:06 want
0:32:07 Google
0:32:08 flights
0:32:08 is not
0:32:08 a
0:32:08 Google
0:32:09 query
0:32:09 you know
0:32:10 they use
0:32:10 certain
0:32:11 point
0:32:12 and one
0:32:12 of the
0:32:13 interesting
0:32:13 things
0:32:13 about
0:32:13 this
0:32:13 and
0:32:14 it’s
0:32:14 interesting
0:32:15 to think
0:32:15 about
0:32:16 what a GUI
0:32:16 is doing
0:32:17 that
0:32:17 some
0:32:18 of what a GUI
0:32:19 is doing
0:32:19 and the obvious
0:32:20 thing that a GUI
0:32:20 is doing
0:32:22 is that it
0:32:22 enables
0:32:23 Office to have
0:32:24 500
0:32:25 features
0:32:25 and you
0:32:26 can find
0:32:26 them all
0:32:27 or at least
0:32:27 you don’t
0:32:28 have to
0:32:28 memorize
0:32:28 keyboard
0:32:29 commands
0:32:30 you can
0:32:30 now have
0:32:31 effectively
0:32:31 infinite
0:32:32 features
0:32:32 and you
0:32:32 can just
0:32:33 keep
0:32:33 adding
0:32:33 menus
0:32:33 and
0:32:34 dialogue
0:32:34 boxes
0:32:35 and eventually
0:32:35 you run
0:32:36 out of
0:32:36 screen space
0:32:36 for dialogue
0:32:37 boxes
0:32:38 but you
0:32:38 can have
0:32:38 hundreds
0:32:39 of features
0:32:40 without people
0:32:40 needing to
0:32:40 memorize
0:32:41 keyboard
0:32:41 commands
0:32:42 but the
0:32:42 other side
0:32:43 of it
0:32:43 is
0:32:44 you’re
0:32:45 in that
0:32:45 dialogue
0:32:45 box
0:32:46 or you’re
0:32:46 in that
0:32:47 screen
0:32:47 in that
0:32:47 workflow
0:32:48 in
0:32:49 Workday
0:32:49 or Salesforce
0:32:50 or whatever
0:32:51 the enterprise
0:32:51 software is
0:32:53 or the
0:32:54 airline
0:32:54 website
0:32:55 or Airbnb
0:32:55 or whatever
0:32:55 it is
0:32:56 and there
0:32:56 aren’t 600
0:32:57 buttons on the
0:32:57 screen
0:32:58 there’s seven
0:32:58 buttons on the
0:32:58 screen
0:32:59 because a bunch
0:32:59 of people
0:33:00 at that
0:33:00 company
0:33:01 have sat
0:33:01 down
0:33:01 and thought
0:33:02 what is it
0:33:03 that the
0:33:03 users should
0:33:04 be asked
0:33:04 here
0:33:05 what questions
0:33:05 should we
0:33:06 give them
0:33:07 what choices
0:33:08 should there
0:33:08 be at this
0:33:09 point in the
0:33:09 flow
0:33:11 and that
0:33:12 reflects a
0:33:13 lot of
0:33:13 institutional
0:33:14 knowledge
0:33:14 and a lot
0:33:15 of learning
0:33:15 and a lot
0:33:16 of testing
0:33:17 and a lot
0:33:17 of really
0:33:18 careful thought
0:33:18 about how
0:33:19 this should
0:33:19 work
0:33:20 and then
0:33:20 you give
0:33:21 somebody
0:33:21 a raw
0:33:22 prompt
0:33:22 and you
0:33:23 just say
0:33:24 okay
0:33:24 you just
0:33:25 tell the
0:33:25 thing
0:33:26 how to
0:33:26 do
0:33:26 the
0:33:26 thing
0:33:26 and
0:33:28 you’ve
0:33:28 kind of
0:33:29 got to
0:33:29 shut your
0:33:29 eyes
0:33:30 screw your
0:33:30 eyes up
0:33:31 and think
0:33:31 from first
0:33:32 principles
0:33:32 how does
0:33:32 this all
0:33:33 of this
0:33:33 work
0:33:34 it’s kind
0:33:34 of like
0:33:34 I always
0:33:35 used to talk
0:33:35 about machine
0:33:36 learning as
0:33:36 giving you
0:33:36 infinite
0:33:37 interns
0:33:38 you know
0:33:38 imagine
0:33:39 imagine you’ve
0:33:39 got a
0:33:39 task
0:33:40 and you’ve
0:33:40 got an
0:33:41 intern
0:33:42 and the
0:33:42 intern
0:33:43 doesn’t
0:33:43 know what
0:33:44 venture
0:33:44 capital
0:33:45 is
0:33:47 how
0:33:47 are they
0:33:47 going to be
0:33:48 and like
0:33:50 they and
0:33:50 they don’t
0:33:51 know that
0:33:53 companies publish
0:33:54 quarterly reports
0:33:55 and that
0:33:56 we’ve got a
0:33:57 Bloomberg account
0:33:58 that lets us
0:33:58 look up
0:33:59 multiples
0:34:00 and that
0:34:01 then you
0:34:01 should probably
0:34:02 use
0:34:04 pitch book
0:34:04 for this
0:34:05 data
0:34:06 and rather
0:34:06 than using
0:34:07 Google
0:34:07 this is my
0:34:08 point about
0:34:08 deep research
0:34:09 like no you
0:34:09 should use
0:34:10 this source
0:34:10 and not
0:34:11 that source
0:34:14 do you want
0:34:14 to have to
0:34:15 work that out
0:34:15 from scratch
0:34:16 or do you
0:34:16 want a bunch
0:34:17 of people
0:34:17 who know a
0:34:18 lot about
0:34:18 this stuff
0:34:19 to have spent
0:34:19 five years
0:34:20 working out
0:34:21 what the
0:34:21 choices should
0:34:22 be on the
0:34:22 screen for
0:34:23 you to click
0:34:23 on it
0:34:24 I mean it’s
0:34:24 the old
0:34:25 user interface
0:34:25 saying the
0:34:26 computer should
0:34:26 never ask
0:34:26 you a question
0:34:27 that you
0:34:27 should have
0:34:27 to work out
0:34:28 that it
0:34:28 should know
0:34:29 by itself
0:34:30 you go to
0:34:30 a blank
0:34:32 raw chatbot
0:34:32 screen
0:34:34 it’s asking
0:34:34 you literally
0:34:35 everything
0:34:36 it’s not just
0:34:37 asking you
0:34:37 one question
0:34:38 it’s asking
0:34:39 you absolutely
0:34:39 everything
0:34:40 about what
0:34:41 it is that
0:34:41 you want
0:34:42 and how
0:34:42 you’re going
0:34:42 to work
0:34:43 out how
0:34:43 to do
0:34:43 it
0:34:45 and so
0:34:46 you’re
0:34:46 mentioning
0:34:47 chat
0:34:49 isn’t
0:34:50 sort of
0:34:50 a product
0:34:51 as much
0:34:51 as a chatbot
0:34:52 disguised
0:34:53 as a product
0:34:54 I’m curious
0:34:56 when we
0:34:56 sort of
0:34:57 look back
0:34:58 at this
0:34:58 sort of
0:35:00 platform
0:35:00 shift
0:35:01 do you
0:35:01 think
0:35:02 that
0:35:03 there will
0:35:03 be
0:35:03 another
0:35:04 sort
0:35:04 of
0:35:04 iPhone
0:35:05 sort
0:35:06 of
0:35:06 Excel
0:35:07 product
0:35:07 that
0:35:07 kind
0:35:08 of
0:35:08 defines
0:35:09 the
0:35:10 platform
0:35:11 shift
0:35:11 in a way
0:35:11 that
0:35:12 chat
0:35:12 GPT
0:35:12 won’t
0:35:13 or
0:35:14 is it
0:35:14 sort
0:35:14 of
0:35:14 that
0:35:15 the world
0:35:15 has to
0:35:15 catch up
0:35:16 to
0:35:16 how
0:35:17 to use
0:35:17 chat
0:35:18 GPT
0:35:18 or
0:35:19 something
0:35:19 like
0:35:19 so
0:35:19 both
0:35:20 of these
0:35:21 can be
0:35:21 true
0:35:21 because
0:35:21 there
0:35:22 was
0:35:22 a lot
0:35:22 of
0:35:22 like
0:35:23 it
0:35:23 took
0:35:23 time
0:35:23 to
0:35:24 realize
0:35:24 how
0:35:24 you
0:35:24 would
0:35:24 use
0:35:24 Google
0:35:25 Maps
0:35:25 and
0:35:25 what
0:35:25 you
0:35:25 could
0:35:26 do
0:35:26 with
0:35:26 Google
0:35:26 and how
0:35:27 you
0:35:27 could
0:35:27 use
0:35:27 Instagram
0:35:28 and
0:35:28 all
0:35:28 of
0:35:28 these
0:35:28 products
0:35:29 have
0:35:29 evolved
0:35:29 a huge
0:35:30 amount
0:35:30 over
0:35:30 time
0:35:31 so
0:35:31 some
0:35:31 of
0:35:31 it
0:35:32 is
0:35:32 like
0:35:32 you
0:35:32 grow
0:35:33 towards
0:35:33 realizing
0:35:33 what
0:35:33 you
0:35:34 could
0:35:34 do
0:35:34 with
0:35:34 this
0:35:34 like
0:35:34 you
0:35:35 realize
0:35:35 that’s
0:35:35 just
0:35:35 a Google
0:35:36 query
0:35:36 now
0:35:36 you
0:35:37 realize
0:35:37 that you
0:35:37 could
0:35:37 just
0:35:37 do it
0:35:38 like
0:35:38 that
0:35:57 10,000
0:35:57 10,000
0:35:58 really
0:35:58 clever
0:35:58 people
0:35:58 sitting
0:35:58 and
0:35:59 trying
0:35:59 to
0:35:59 work
0:35:59 out
0:35:59 what
0:36:00 those
0:36:00 things
0:36:00 are
0:36:00 and
0:36:00 then
0:36:00 showing
0:36:00 it
0:36:01 to
0:36:01 you
0:36:01 as
0:36:01 a
0:36:02 product
0:36:03 I
0:36:03 think
0:36:03 another
0:36:03 side
0:36:04 of
0:36:04 this
0:36:04 is
0:36:04 like
0:36:05 you
0:36:06 know
0:36:06 there
0:36:06 were
0:36:06 always
0:36:06 these
0:36:07 precursors
0:36:08 so
0:36:09 like
0:36:09 there
0:36:09 were
0:36:09 lots
0:36:10 of
0:36:10 other
0:36:10 things
0:36:11 before
0:36:11 Instagram
0:36:13 YouTube
0:36:14 didn’t
0:36:14 start
0:36:15 as
0:36:15 YouTube
0:36:16 it
0:36:16 started
0:36:16 as
0:36:16 video
0:36:17 dating
0:36:17 I
0:36:17 think
0:36:18 there
0:36:18 were
0:36:19 lots
0:36:19 of
0:36:19 attempts
0:36:20 to
0:36:20 do
0:36:20 online
0:36:21 dating
0:36:21 that
0:36:21 all
0:36:22 kind
0:36:22 of
0:36:22 worked
0:36:23 until
0:36:23 Tinder
0:36:24 kind
0:36:24 of
0:36:24 pulled
0:36:24 the
0:36:24 whole
0:36:25 thing
0:36:25 inside
0:36:25 out
0:36:26 and
0:36:26 so
0:36:27 there
0:36:27 were
0:36:27 always
0:36:28 lots
0:36:28 of
0:36:28 things
0:36:29 what’s
0:36:29 the
0:36:29 phrase
0:36:29 local
0:36:30 maxima
0:36:30 in
0:36:31 fact
0:36:31 this
0:36:31 is
0:36:31 where
0:36:31 we
0:36:31 were
0:36:32 particularly
0:36:32 with
0:36:32 the
0:36:33 iPhone
0:36:35 before
0:36:35 because
0:36:35 I
0:36:36 was
0:36:36 working
0:36:36 in
0:36:36 mobile
0:36:37 for
0:36:37 the
0:36:37 previous
0:36:38 decade
0:36:39 it
0:36:40 didn’t
0:36:40 feel
0:36:40 like
0:36:40 we
0:36:41 were
0:36:41 waiting
0:36:41 for
0:36:41 a
0:36:42 thing
0:36:42 it
0:36:42 felt
0:36:42 like
0:36:43 it
0:36:43 was
0:36:43 kind
0:36:43 of
0:36:43 working
0:36:44 like
0:36:44 every
0:36:44 year
0:36:44 the
0:36:45 network
0:36:45 got
0:36:45 faster
0:36:45 and
0:36:46 the
0:36:46 phone
0:36:46 got
0:36:46 better
0:36:47 and
0:36:48 we
0:36:48 had
0:36:49 apps
0:36:49 and
0:36:49 we
0:36:49 had
0:36:49 app
0:36:50 stores
0:36:50 and
0:36:50 we
0:36:50 had
0:36:51 cameras
0:36:52 and
0:36:53 stuff
0:36:53 seemed
0:36:53 to be
0:36:54 every
0:36:54 year
0:36:54 it was
0:36:54 a bit
0:36:55 better
0:36:55 and
0:36:55 then
0:36:55 the
0:36:55 iPhone
0:36:56 arrives
0:36:56 and
0:36:56 it
0:36:56 just
0:36:59 blowed
0:36:59 the
0:37:00 chart
0:37:00 kind
0:37:00 of
0:37:01 you’ve
0:37:01 got
0:37:01 this
0:37:01 line
0:37:01 doing
0:37:02 this
0:37:02 and
0:37:02 then
0:37:02 there’s
0:37:03 a
0:37:03 line
0:37:03 that
0:37:03 does
0:37:03 that
0:37:04 although
0:37:04 remember
0:37:05 also
0:37:05 the
0:37:05 iPhone
0:37:05 took
0:37:06 two
0:37:06 years
0:37:06 before
0:37:06 it
0:37:06 worked
0:37:07 because
0:37:07 the
0:37:07 price
0:37:07 was
0:37:08 wrong
0:37:08 and
0:37:08 the
0:37:08 feature
0:37:08 set
0:37:09 was
0:37:09 wrong
0:37:09 and
0:37:09 the
0:37:10 distribution
0:37:10 model
0:37:10 didn’t
0:37:10 quite
0:37:10 work
0:37:12 and
0:37:13 so
0:37:13 yeah
0:37:14 you can
0:37:14 think
0:37:16 everything
0:37:17 is
0:37:17 going
0:37:17 well
0:37:17 and
0:37:17 then
0:37:18 something
0:37:18 comes
0:37:18 along
0:37:19 and
0:37:19 you
0:37:19 realize
0:37:19 no
0:37:19 no
0:37:20 no
0:37:20 no
0:37:20 no
0:37:20 no
0:37:20 that
0:37:21 which
0:37:21 is
0:37:21 the
0:37:21 same
0:37:21 for
0:37:22 Google
0:37:23 search
0:37:23 was
0:37:23 a
0:37:24 thing
0:37:24 before
0:37:24 Google
0:37:24 it
0:37:24 just
0:37:25 wasn’t
0:37:25 very
0:37:25 good
0:37:26 so
0:37:26 there
0:37:27 was
0:37:27 lots
0:37:27 of
0:37:28 social
0:37:28 stuff
0:37:28 before
0:37:28 Facebook
0:37:29 and
0:37:29 that
0:37:30 was
0:37:30 the
0:37:30 thing
0:37:30 that
0:37:31 catalyzed
0:37:31 it
0:37:31 so
0:37:32 I
0:37:32 just
0:37:32 think
0:37:33 deterministically
0:37:33 this
0:37:33 whole
0:37:33 thing
0:37:33 is
0:37:33 so
0:37:34 early
0:37:34 that
0:37:34 it
0:37:35 feels
0:37:35 like
0:37:35 of
0:37:35 course
0:37:35 there
0:37:35 are
0:37:35 going
0:37:36 to
0:37:36 be
0:37:36 you
0:37:36 know
0:37:37 dozens
0:37:37 hundreds
0:37:38 of
0:37:38 new
0:37:38 things
0:37:38 otherwise
0:37:39 HGC and Z
0:37:40 would just
0:37:40 kind of
0:37:40 shut down
0:37:40 and give
0:37:41 the money
0:37:41 back
0:37:41 to the
0:37:41 LPs
0:37:42 because
0:37:43 the
0:37:44 models
0:37:44 will
0:37:44 just
0:37:45 do
0:37:45 the
0:37:45 whole
0:37:45 thing
0:37:45 and
0:37:45 I
0:37:46 don’t
0:37:46 think
0:37:46 you’re
0:37:46 going
0:37:46 to
0:37:46 do
0:37:46 that
0:37:47 at
0:37:47 least
0:37:47 I
0:37:47 hope
0:37:47 not
0:37:47 no
0:37:48 if
0:37:48 we
0:37:48 have
0:38:01 sub
0:38:01 sector
0:38:01 that
0:38:01 there
0:38:01 would
0:38:02 be
0:38:03 net
0:38:04 new
0:38:04 companies
0:38:04 created
0:38:05 that
0:38:05 would
0:38:05 be
0:38:06 better
0:38:06 than
0:38:06 the
0:38:07 model
0:38:08 providers
0:38:08 that
0:38:08 there
0:38:08 would
0:38:08 be
0:38:09 even
0:38:09 multiple
0:38:10 model
0:38:10 providers
0:38:10 that
0:38:11 in
0:38:11 every
0:38:12 category
0:38:13 one
0:38:14 thing
0:38:14 we’ve
0:38:14 always
0:38:15 in the
0:38:15 web
0:38:15 two
0:38:15 era
0:38:15 we’ve
0:38:15 always
0:38:15 been
0:38:16 on
0:38:16 the
0:38:16 category
0:38:16 winner
0:38:19 but
0:38:20 these
0:38:21 markets
0:38:21 are
0:38:21 so
0:38:21 big
0:38:22 and
0:38:23 there’s
0:38:23 so
0:38:24 much
0:38:24 expertise
0:38:25 and
0:38:25 specialization
0:38:26 that
0:38:28 one
0:38:28 there
0:38:28 can
0:38:28 be
0:38:28 winners
0:38:29 in
0:38:29 every
0:38:29 category
0:38:29 it’s
0:38:29 not
0:38:30 just
0:38:30 the
0:38:31 model
0:38:31 providers
0:38:31 taking
0:38:31 everything
0:38:32 but
0:38:32 that
0:38:32 even
0:38:32 in
0:38:33 every
0:38:33 category
0:38:33 including
0:38:34 the
0:38:34 model
0:38:34 providers
0:38:35 there
0:38:35 can
0:38:35 be
0:38:35 multiple
0:38:35 winners
0:38:36 and
0:38:36 increasing
0:38:38 specialization
0:38:38 and
0:38:38 the
0:38:39 markets
0:38:39 are
0:38:39 just
0:38:40 big
0:38:40 enough
0:38:40 to
0:38:41 contain
0:38:42 multiple
0:38:42 winners
0:38:43 I think
0:38:43 that’s
0:38:44 right
0:38:45 and I
0:38:45 think
0:38:46 the
0:38:46 categories
0:38:47 themselves
0:38:48 aren’t
0:38:48 clear
0:38:49 and
0:38:50 many
0:38:51 things
0:38:51 you think
0:38:51 this is
0:38:51 a
0:38:52 category
0:38:52 and
0:38:52 it
0:38:52 turns
0:38:52 out
0:38:53 no
0:38:53 it
0:38:53 was
0:38:53 actually
0:38:53 that
0:38:53 whole
0:38:54 other
0:38:54 thing
0:38:54 and
0:38:54 the
0:38:55 categories
0:38:55 kind
0:38:55 of get
0:38:56 unbundled
0:38:56 and
0:38:57 recombined
0:38:57 in
0:38:57 different
0:38:57 ways
0:38:57 I
0:38:58 remember
0:38:58 I
0:38:58 was
0:38:58 a
0:38:58 student
0:38:59 in
0:39:00 1995
0:39:01 and
0:39:03 I
0:39:03 think
0:39:03 I
0:39:03 had
0:39:03 like
0:39:03 four
0:39:03 or
0:39:04 five
0:39:04 different
0:39:04 web
0:39:04 browsers
0:39:05 on my
0:39:05 PC
0:39:08 because
0:39:08 Tim
0:39:08 Berners-Lee’s
0:39:09 original
0:39:09 web
0:39:09 browser
0:39:10 had a
0:39:10 web
0:39:10 editor
0:39:10 in
0:39:11 it
0:39:11 because
0:39:11 he
0:39:11 thought
0:39:11 this
0:39:11 was
0:39:12 kind
0:39:12 of
0:39:12 like
0:39:12 a
0:39:12 network
0:39:12 drive
0:39:13 and
0:39:13 it
0:39:13 was
0:39:13 a
0:39:14 sharing
0:39:14 system
0:39:15 and
0:39:15 not
0:39:16 really
0:39:16 a
0:39:16 publishing
0:39:17 system
0:39:17 so
0:39:17 you
0:39:17 would
0:39:18 have
0:39:18 your
0:39:18 web
0:39:18 pages
0:39:18 on
0:39:19 your
0:39:19 PC
0:39:21 turned
0:39:21 on
0:39:21 and
0:39:21 that
0:39:21 would
0:39:21 be
0:39:22 how
0:39:22 your
0:39:22 colleagues
0:39:22 would
0:39:22 look
0:39:23 at
0:39:23 your
0:39:23 Word
0:39:23 documents
0:39:25 or
0:39:25 your
0:39:25 web
0:39:25 pages
0:39:26 and
0:39:26 so
0:39:26 again
0:39:27 we
0:39:27 just
0:39:27 don’t
0:39:28 know
0:39:29 how
0:39:34 picking
0:39:35 up
0:39:35 on
0:39:35 a
0:39:36 strand
0:39:36 within
0:39:36 what
0:39:36 you
0:39:36 just
0:39:36 said
0:39:37 though
0:39:37 the
0:39:37 interesting
0:39:38 one of
0:39:38 the
0:39:38 things
0:39:38 I’m
0:39:38 sort
0:39:39 of
0:39:39 thinking
0:39:39 about
0:39:39 a lot
0:39:39 is
0:39:40 looking
0:39:40 at
0:39:40 open
0:39:40 AI
0:39:42 because
0:39:44 I’m
0:39:44 sort
0:39:44 of
0:39:44 fascinated
0:39:45 by
0:39:45 disconnections
0:39:46 and
0:39:46 we’ve
0:39:46 got
0:39:47 this
0:39:47 interesting
0:39:47 disconnect
0:39:48 now
0:39:48 which
0:39:48 is
0:39:49 you
0:39:49 know
0:39:49 if
0:39:49 you
0:39:49 look
0:39:49 at
0:39:49 the
0:39:49 benchmark
0:39:50 scores
0:39:51 so
0:39:51 you’ve
0:39:51 got
0:39:51 these
0:39:51 general
0:39:52 purpose
0:39:52 benchmarks
0:39:52 where
0:39:52 the
0:39:52 models
0:39:53 are
0:39:53 basically
0:39:53 all
0:39:53 the
0:39:54 same
0:39:54 and
0:39:54 if
0:39:55 you’re
0:39:55 spending
0:39:56 hours
0:39:56 a day
0:39:56 and
0:39:57 you’ve
0:39:57 got
0:39:57 this
0:39:57 opinion
0:39:57 about
0:39:58 I like
0:40:00 GPT
0:40:00 and I
0:40:00 like
0:40:01 GPT
0:40:01 5.1
0:40:01 more than
0:40:02 GPT
0:40:03 4.9
0:40:03 or whatever
0:40:03 the hell
0:40:03 it’s
0:40:04 called
0:40:05 if
0:40:05 you’re
0:40:05 using
0:40:05 this
0:40:06 once
0:40:06 a
0:40:06 week
0:40:06 you
0:40:06 really
0:40:06 don’t
0:40:07 notice
0:40:07 this
0:40:07 stuff
0:40:08 and
0:40:09 the
0:40:09 benchmark
0:40:09 scores
0:40:09 are all
0:40:10 roughly
0:40:10 the
0:40:10 same
0:40:11 but the
0:40:12 usage
0:40:12 isn’t
0:40:12 it’s
0:40:13 basically
0:40:13 the
0:40:13 only
0:40:14 the
0:40:14 Claude
0:40:14 has
0:40:15 basically
0:40:15 no
0:40:15 consumer
0:40:16 usage
0:40:16 even
0:40:17 though
0:40:17 on
0:40:17 the
0:40:17 benchmark
0:40:17 score
0:40:18 it’s
0:40:18 the
0:40:18 same
0:40:19 and
0:40:19 then
0:40:19 it’s
0:40:20 chat
0:40:20 GPT
0:40:20 and
0:40:21 then
0:40:21 halfway
0:40:22 down
0:40:22 the
0:40:22 chart
0:40:23 it’s
0:40:23 Meta
0:40:24 and
0:40:25 Google
0:40:25 and
0:40:25 the
0:40:26 funny
0:40:26 thing
0:40:26 is
0:40:26 you
0:40:27 read
0:40:27 all
0:40:27 the
0:40:28 AI
0:40:28 newsletters
0:40:28 then
0:40:29 Meta
0:40:29 is
0:40:30 lost
0:40:30 they’re
0:40:30 out
0:40:30 of
0:40:30 the
0:40:30 game
0:40:30 they’re
0:40:31 dead
0:40:32 Mark Zuckerberg
0:40:32 is
0:40:32 spending
0:40:32 a billion
0:40:33 dollars
0:40:33 of
0:40:33 researcher
0:40:34 to get
0:40:34 back
0:40:34 in
0:40:34 the
0:40:34 game
0:40:35 but
0:40:35 from
0:40:35 the
0:40:36 consumer
0:40:36 side
0:40:36 well
0:40:36 it’s
0:40:37 distribution
0:40:39 and
0:40:39 the
0:40:40 interesting
0:40:40 thing
0:40:41 here
0:40:41 is
0:40:42 that
0:40:42 you’ve
0:40:43 got
0:40:43 what
0:40:43 I’m
0:40:43 kind
0:40:43 of
0:40:44 circling
0:40:44 around
0:40:44 is
0:40:45 if
0:40:45 the
0:40:46 model
0:40:46 for
0:40:47 a
0:40:47 casual
0:40:48 consumer
0:40:48 user
0:40:48 certainly
0:40:49 is a
0:40:50 commodity
0:40:51 and
0:40:51 there’s
0:40:51 no
0:40:52 network
0:40:52 effects
0:40:52 or
0:40:52 winner
0:40:53 takes
0:40:53 all
0:40:53 effects
0:40:54 yet
0:40:54 those
0:40:54 may
0:40:55 emerge
0:40:55 but
0:40:55 we
0:40:55 don’t
0:40:55 have
0:40:55 them
0:40:56 yet
0:40:56 and
0:40:57 things
0:40:57 like
0:40:57 memory
0:40:58 aren’t
0:40:58 network
0:40:58 effects
0:40:58 so
0:40:59 stickiness
0:40:59 but
0:40:59 they
0:40:59 can
0:40:59 be
0:41:00 copied
0:41:04 how
0:41:04 is
0:41:05 it
0:41:05 that
0:41:05 you
0:41:06 compete
0:41:06 do
0:41:06 you
0:41:07 just
0:41:07 compete
0:41:07 on
0:41:08 being
0:41:08 the
0:41:08 recognized
0:41:09 brand
0:41:09 and
0:41:09 adding
0:41:10 more
0:41:10 features
0:41:10 and
0:41:11 services
0:41:11 and
0:41:11 capabilities
0:41:12 and
0:41:12 people
0:41:12 just
0:41:13 don’t
0:41:13 switch
0:41:13 away
0:41:14 which
0:41:14 is
0:41:14 kind
0:41:14 of
0:41:14 what
0:41:15 happened
0:41:15 with
0:41:15 Chrome
0:41:16 for
0:41:16 example
0:41:17 there’s
0:41:17 not a
0:41:18 network
0:41:18 effect
0:41:18 for
0:41:19 Chrome
0:41:19 and
0:41:20 it’s
0:41:20 not
0:41:20 actually
0:41:20 any
0:41:21 better
0:41:21 than
0:41:23 Safari
0:41:23 but
0:41:23 you
0:41:24 use
0:41:24 Chrome
0:41:24 because
0:41:24 you
0:41:24 use
0:41:25 Chrome
0:41:26 or
0:41:28 is
0:41:28 it
0:41:28 that
0:41:29 you
0:41:29 get
0:41:30 left
0:41:30 behind
0:41:31 on
0:41:32 distribution
0:41:34 or
0:41:34 network
0:41:35 effects
0:41:36 that
0:41:36 emerge
0:41:37 somewhere
0:41:37 else
0:41:39 and
0:41:39 meanwhile
0:41:40 you
0:41:40 don’t
0:41:40 have
0:41:40 your
0:41:40 own
0:41:41 infrastructure
0:41:42 so
0:41:42 I
0:41:42 suppose
0:41:42 what
0:41:42 I’m
0:41:43 getting
0:41:43 at
0:41:43 is
0:41:43 you’ve
0:41:43 got
0:41:44 these
0:41:44 800
0:41:44 or
0:41:44 900
0:41:45 million
0:41:45 weekly
0:41:45 active
0:41:46 users
0:41:46 but
0:41:47 that
0:41:48 feels
0:41:48 very
0:41:49 fragile
0:41:49 because
0:41:49 all
0:41:50 you’ve
0:41:50 really
0:41:50 got
0:41:50 is
0:41:51 the
0:41:51 power
0:41:51 of
0:41:51 the
0:41:51 default
0:41:52 and
0:41:52 the
0:41:52 brand
0:41:52 you
0:41:53 don’t
0:41:53 have
0:41:53 a
0:41:53 network
0:41:53 effect
0:41:54 you
0:41:54 don’t
0:41:54 really
0:41:54 have
0:41:54 feature
0:41:55 lock
0:41:55 in
0:41:56 you
0:41:56 don’t
0:42:00 infrastructure
0:42:00 so
0:42:00 you
0:42:00 don’t
0:42:01 control
0:42:01 your
0:42:01 cost
0:42:01 base
0:42:02 you
0:42:02 don’t
0:42:02 have
0:42:02 a
0:42:02 cost
0:42:03 advantage
0:42:04 you
0:42:05 get
0:42:05 a bill
0:42:05 every
0:42:05 month
0:42:06 from
0:42:06 Satya
0:42:09 so
0:42:09 you
0:42:09 kind
0:42:10 of
0:42:10 got
0:42:10 to
0:42:10 scramble
0:42:11 as
0:42:11 fast
0:42:11 as
0:42:11 you
0:42:12 can
0:42:12 in
0:42:13 both
0:42:13 of
0:42:13 those
0:42:14 directions
0:42:15 to
0:42:15 on the
0:42:15 one
0:42:16 side
0:42:16 build
0:42:17 product
0:42:17 and
0:42:18 build
0:42:18 stuff
0:42:18 that
0:42:19 on top
0:42:19 of
0:42:19 the
0:42:20 model
0:42:20 which is
0:42:20 our
0:42:20 earlier
0:42:21 conversation
0:42:21 is it
0:42:22 just
0:42:22 the
0:42:22 model
0:42:23 you’ve
0:42:23 got
0:42:23 to
0:42:23 build
0:42:24 stuff
0:42:24 on
0:42:24 top
0:42:24 of
0:42:25 the
0:42:25 model
0:42:25 in
0:42:25 every
0:42:26 direction
0:42:27 it’s
0:42:27 a
0:42:27 browser
0:42:28 it’s
0:42:28 a
0:42:28 social
0:42:29 video
0:42:29 app
0:42:30 it’s
0:42:30 an
0:42:30 app
0:42:31 platform
0:42:31 it’s
0:42:31 this
0:42:31 it’s
0:42:32 that
0:42:32 it’s
0:42:32 like
0:42:32 you
0:42:32 know
0:42:32 the
0:42:32 meme
0:42:33 of
0:42:33 the
0:42:33 guy
0:42:33 with
0:42:33 the
0:42:33 map
0:42:34 with
0:42:34 all
0:42:34 the
0:42:34 strings
0:42:34 on
0:42:34 it
0:42:36 it’s
0:42:36 all
0:42:36 of
0:42:36 these
0:42:36 things
0:42:37 we’re
0:42:37 going to
0:42:37 build
0:42:37 all
0:42:37 of
0:42:37 them
0:42:38 yesterday
0:42:38 and
0:42:38 then
0:42:39 in
0:42:39 parallel
0:42:40 it’s
0:42:40 infrastructure
0:42:43 we’ve
0:42:43 got to
0:42:44 deal
0:42:44 with
0:42:46 NVIDIA
0:42:47 with
0:42:48 Broadcom
0:42:48 with
0:42:48 AMD
0:42:49 with
0:42:49 NVIDIA
0:42:49 with
0:42:50 Oracle
0:42:51 and
0:42:51 with
0:42:52 Petrodollars
0:42:53 because
0:42:53 you’re
0:42:53 kind of
0:42:54 scrambling
0:42:55 to get
0:42:55 from
0:42:57 this
0:42:57 amazing
0:42:58 technical
0:42:58 breakthrough
0:42:59 and
0:43:00 these
0:43:01 800
0:43:01 900
0:43:01 million
0:43:02 wows
0:43:03 to
0:43:03 something
0:43:04 that
0:43:04 has
0:43:05 like
0:43:05 really
0:43:06 sticky
0:43:07 defensible
0:43:07 sustainable
0:43:08 business
0:43:08 value
0:43:08 and
0:43:08 product
0:43:09 value
0:43:10 yeah
0:43:11 and
0:43:11 so
0:43:12 as
0:43:13 you’re
0:43:13 evaluating
0:43:14 the
0:43:14 competitive
0:43:15 landscape
0:43:16 among
0:43:16 the
0:43:17 hyperscalers
0:43:18 what
0:43:18 are
0:43:18 the
0:43:19 questions
0:43:20 that
0:43:20 you’re
0:43:20 asking
0:43:21 that
0:43:21 you
0:43:21 think
0:43:21 are
0:43:21 going
0:43:21 to
0:43:21 be
0:43:22 most
0:43:23 important
0:43:23 in
0:43:23 determining
0:43:25 who’s
0:43:25 going
0:43:25 to
0:43:26 gain
0:43:27 durable
0:43:27 competitive
0:43:27 advantages
0:43:28 or
0:43:28 how
0:43:28 this
0:43:29 competition
0:43:29 is going
0:43:30 to play
0:43:31 out
0:43:32 well
0:43:32 this
0:43:32 kind of
0:43:33 comes
0:43:33 back
0:43:33 to
0:43:33 your
0:43:33 point
0:43:33 about
0:43:34 sustaining
0:43:35 advantage
0:43:35 and
0:43:35 we
0:43:36 talked
0:43:36 about
0:43:36 Google
0:43:36 like
0:43:37 if we
0:43:37 think
0:43:37 about
0:43:38 the
0:43:38 shift
0:43:38 to
0:43:39 mobile
0:43:40 for
0:43:40 meta
0:43:41 this
0:43:41 turned out
0:43:41 to be
0:43:42 transformative
0:43:43 like it
0:43:43 made the
0:43:43 products
0:43:44 way more
0:43:44 useful
0:43:46 for Google
0:43:47 it turned out
0:43:48 mobile search
0:43:48 is just
0:43:48 search
0:43:52 and
0:43:53 maps
0:43:53 changed
0:43:53 probably
0:43:54 and
0:43:54 YouTube
0:43:54 changed
0:43:55 a bit
0:43:55 but
0:43:55 basically
0:43:56 for
0:43:56 Google
0:43:56 search
0:43:56 Google
0:43:57 search
0:43:57 is search
0:43:57 and the
0:43:57 web
0:43:58 search
0:43:59 means more
0:43:59 people doing
0:44:00 more search
0:44:00 more
0:44:00 more of the
0:44:01 time
0:44:03 and
0:44:05 the
0:44:06 default
0:44:06 view
0:44:07 now
0:44:07 would
0:44:07 seem
0:44:08 to be
0:44:08 well
0:44:09 Gemini
0:44:09 is as
0:44:09 good
0:44:09 as
0:44:09 anybody
0:44:10 else
0:44:10 next week
0:44:10 like the
0:44:11 new
0:44:11 model
0:44:11 I haven’t
0:44:12 looked at
0:44:12 the benchmarks
0:44:13 for GPT 5.1
0:44:14 which is out
0:44:14 today
0:44:15 is it better
0:44:15 than Gemini
0:44:15 probably
0:44:16 will it still
0:44:16 be better
0:44:17 next month
0:44:17 no
0:44:20 so
0:44:21 so
0:44:21 that’s a
0:44:22 given
0:44:22 like you’ve
0:44:23 got a
0:44:23 frontier
0:44:23 model
0:44:24 fine
0:44:25 what does
0:44:25 that cost
0:44:26 it costs
0:44:26 you
0:44:27 pick a
0:44:27 number
0:44:28 250
0:44:28 billion
0:44:28 dollars
0:44:28 a year
0:44:29 100
0:44:29 billion
0:44:29 dollars
0:44:30 a year
0:44:31 what’s
0:44:32 our earlier
0:44:32 conversation
0:44:32 about
0:44:33 CapEx
0:44:33 okay
0:44:33 so
0:44:34 Google
0:44:35 can pay
0:44:35 that
0:44:36 because
0:44:36 they’ve
0:44:36 got the
0:44:36 money
0:44:36 they’ve
0:44:37 got the
0:44:37 cash
0:44:37 away
0:44:37 from
0:44:38 everything
0:44:38 else
0:44:39 and so
0:44:39 you do
0:44:39 that
0:44:40 and your
0:44:40 existing
0:44:40 products
0:44:41 you optimize
0:44:42 search
0:44:42 you optimize
0:44:42 your ad
0:44:43 business
0:44:43 you build
0:44:45 new experiences
0:44:46 maybe you
0:44:46 invent the
0:44:47 new iPhone
0:44:48 of AI
0:44:48 maybe there
0:44:48 is no
0:44:49 iPhone
0:44:49 of AI
0:44:50 maybe someone
0:44:50 else does it
0:44:51 and you do
0:44:51 an Android
0:44:51 and just
0:44:52 copy it
0:44:55 so fine
0:44:56 it’s a new
0:44:56 mobile
0:44:58 we’ll just
0:44:58 carry on
0:44:58 search as
0:44:59 such
0:45:00 AI is AI
0:45:01 we’ll do
0:45:01 the new
0:45:01 thing
0:45:02 we’ll make
0:45:02 it a
0:45:02 feature
0:45:02 we’ll just
0:45:03 carry on
0:45:03 doing it
0:45:05 for meta
0:45:06 it feels
0:45:06 like there
0:45:06 are bigger
0:45:07 questions
0:45:07 on what
0:45:08 this means
0:45:08 for search
0:45:10 or what
0:45:10 it means
0:45:10 for content
0:45:11 and social
0:45:12 and experience
0:45:12 and recommendation
0:45:13 which makes
0:45:13 it all that
0:45:14 more imperative
0:45:15 that they have
0:45:15 their own
0:45:15 models
0:45:15 just as
0:45:16 it is
0:45:16 for Google
0:45:18 for
0:45:19 Amazon
0:45:20 okay
0:45:21 well
0:45:21 on the
0:45:22 one side
0:45:22 it’s
0:45:22 commodity
0:45:22 infra
0:45:23 and we’ll
0:45:23 sell it
0:45:24 as commodity
0:45:24 infra
0:45:25 and on the
0:45:26 other side
0:45:27 maybe
0:45:29 stepping back
0:45:30 if you’re
0:45:30 not a
0:45:31 hyperscaler
0:45:31 if you’re
0:45:32 a web
0:45:32 publisher
0:45:33 a marketer
0:45:34 a brand
0:45:35 an advertiser
0:45:35 a media
0:45:36 company
0:45:38 you could
0:45:38 make a list
0:45:39 of questions
0:45:39 but you don’t
0:45:39 even know
0:45:40 what the
0:45:40 questions are
0:45:41 right now
0:45:43 what is
0:45:43 this
0:45:44 what happens
0:45:45 if I ask
0:45:46 a chatbot
0:45:46 a thing
0:45:47 instead of
0:45:47 asking
0:45:47 Google
0:45:48 even if
0:45:48 it’s
0:45:48 Google
0:45:49 from
0:45:49 Google’s
0:45:50 point of view
0:45:50 what I’ll ask
0:45:51 Google’s
0:45:51 chatbot
0:45:51 it’s fine
0:45:52 but as a
0:45:53 marketer
0:45:53 what does that
0:45:53 mean
0:45:54 what happens
0:45:55 if I ask
0:45:56 for a recipe
0:45:57 and the LLM
0:45:57 just gives me
0:45:58 the answer
0:45:58 what does that
0:45:59 mean if my
0:45:59 business is
0:46:00 having recipes
0:46:02 do you have
0:46:02 do you have
0:46:02 do you have
0:46:02 do you have
0:46:02 a kind
0:46:03 of split
0:46:04 between
0:46:04 and this
0:46:04 is also
0:46:05 an Amazon
0:46:05 question
0:46:06 how does
0:46:07 the purchasing
0:46:07 decision
0:46:08 happen
0:46:08 how does
0:46:09 the decision
0:46:09 to buy
0:46:10 a thing
0:46:10 that I didn’t
0:46:11 know existed
0:46:12 before happen
0:46:13 what happens
0:46:13 if I wave
0:46:14 my phone
0:46:14 at my
0:46:14 living room
0:46:15 and say
0:46:15 what should
0:46:16 I buy
0:46:17 where does
0:46:17 that take
0:46:17 me
0:46:18 in ways
0:46:18 that it
0:46:19 wouldn’t
0:46:19 have taken
0:46:19 me
0:46:19 in the
0:46:20 past
0:46:21 so there’s
0:46:21 a lot
0:46:22 of questions
0:46:22 further
0:46:23 downstream
0:46:23 and that
0:46:24 goes
0:46:24 upstream
0:46:25 to meta
0:46:25 and to
0:46:26 stomach
0:46:26 stent for
0:46:26 Google
0:46:27 it’s a much
0:46:28 bigger question
0:46:29 in the long
0:46:29 term for
0:46:30 Amazon
0:46:31 do LLMs
0:46:32 mean that
0:46:32 Amazon can
0:46:33 finally do
0:46:34 really good
0:46:34 at scale
0:46:35 recommendation
0:46:36 and discovery
0:46:36 and suggestion
0:46:37 in ways that
0:46:37 it couldn’t
0:46:38 really do
0:46:38 in the past
0:46:40 because of
0:46:40 this kind
0:46:40 of pure
0:46:41 commodity
0:46:41 retailing
0:46:41 model
0:46:42 that it
0:46:42 has
0:46:44 Apple
0:46:46 sort of
0:46:46 off on
0:46:47 one side
0:46:48 you know
0:46:48 interestingly
0:46:49 they produced
0:46:50 this incredibly
0:46:50 compelling
0:46:51 vision of
0:46:51 what Siri
0:46:52 should be
0:46:52 two years
0:46:52 ago
0:46:53 it just
0:46:54 turned out
0:46:54 that they
0:46:54 couldn’t
0:46:55 make it
0:46:56 interestingly
0:46:56 nobody
0:46:57 else
0:46:57 could
0:46:57 have
0:46:57 made
0:46:57 it
0:46:57 either
0:46:58 you
0:46:58 go back
0:46:58 and watch
0:46:59 the Siri
0:46:59 demo
0:47:00 that they
0:47:00 gave
0:47:00 and you
0:47:00 think
0:47:01 okay
0:47:02 so we’ve
0:47:02 got
0:47:03 multimodal
0:47:03 instantaneous
0:47:04 on device
0:47:05 tool using
0:47:06 agentic
0:47:07 multi-platform
0:47:08 e-commerce
0:47:09 in real time
0:47:10 with no prompt
0:47:10 injection
0:47:11 problems
0:47:11 and zero
0:47:11 error
0:47:12 rates
0:47:12 well that
0:47:12 sounds
0:47:13 good
0:47:14 I mean
0:47:15 has anyone
0:47:16 got that
0:47:16 working
0:47:16 like no
0:47:17 open
0:47:18 eye
0:47:18 open
0:47:18 google
0:47:19 and open
0:47:19 AI
0:47:19 I don’t
0:47:19 have that
0:47:19 working
0:47:20 I don’t
0:47:20 think
0:47:20 Google
0:47:21 or
0:47:21 open
0:47:21 AI
0:47:21 could
0:47:22 deliver
0:47:22 the
0:47:22 Siri
0:47:22 demo
0:47:23 that Apple
0:47:23 gave
0:47:23 two years
0:47:23 ago
0:47:25 I mean
0:47:25 they could
0:47:25 probably
0:47:25 do
0:47:25 the demo
0:47:26 but they
0:47:26 couldn’t
0:47:26 like
0:47:27 consistently
0:47:28 reliably
0:47:28 make it
0:47:28 work
0:47:29 I mean
0:47:29 that
0:47:30 demo
0:47:30 that product
0:47:30 isn’t in
0:47:31 Android
0:47:31 today
0:47:33 and
0:47:33 Apple
0:47:34 I mean
0:47:35 Apple to me
0:47:35 has the most
0:47:35 kind of
0:47:36 intellectually
0:47:36 interesting
0:47:36 question
0:47:37 which is
0:47:41 so I saw
0:47:42 Craig Federighi
0:47:42 make this
0:47:42 point
0:47:43 which is
0:47:43 like
0:47:43 we don’t
0:47:44 have our
0:47:44 own
0:47:44 chatbot
0:47:44 fine
0:47:45 we also
0:47:45 don’t
0:47:45 have
0:47:45 YouTube
0:47:46 or
0:47:46 Uber
0:47:50 explain
0:47:51 why
0:47:51 that is
0:47:51 different
0:47:52 which is
0:47:52 a harder
0:47:52 question
0:47:53 to answer
0:47:53 than it
0:47:53 sounds
0:47:54 like
0:47:55 and of
0:47:55 course
0:47:55 the answer
0:47:56 is
0:47:56 if this
0:47:56 actually
0:47:57 fundamentally
0:47:57 change
0:47:58 the nature
0:47:58 of computing
0:47:59 then it’s
0:47:59 a problem
0:48:00 if it’s
0:48:00 just a
0:48:01 service
0:48:01 that you
0:48:01 use
0:48:02 like
0:48:02 Google
0:48:03 then that’s
0:48:03 not a
0:48:03 problem
0:48:04 which is
0:48:05 kind of
0:48:05 the point
0:48:05 about
0:48:06 you know
0:48:07 where
0:48:07 does
0:48:07 Siri
0:48:07 go
0:48:08 but the
0:48:08 interesting
0:48:09 candle
0:48:09 example
0:48:09 here
0:48:09 would be
0:48:10 to think
0:48:10 about
0:48:10 what
0:48:10 happened
0:48:10 to
0:48:11 Microsoft
0:48:11 in the
0:48:12 2000s
0:48:12 which is
0:48:13 the entire
0:48:13 dev environment
0:48:14 gets away
0:48:14 from them
0:48:15 and no one
0:48:15 builds Windows
0:48:16 apps after
0:48:16 like 2001
0:48:17 or something
0:48:18 but you need
0:48:19 to use the
0:48:19 internet
0:48:19 to use the
0:48:19 internet
0:48:20 you need
0:48:20 a PC
0:48:21 and what
0:48:21 PC
0:48:21 are you
0:48:21 going to
0:48:22 buy
0:48:22 well
0:48:23 like
0:48:23 Apple’s
0:48:24 not really
0:48:24 a player
0:48:25 at that
0:48:26 time
0:48:26 and
0:48:27 just
0:48:27 getting
0:48:27 back
0:48:27 into
0:48:28 the
0:48:28 game
0:48:28 Linux
0:48:28 is
0:48:29 obviously
0:48:29 not
0:48:29 an option
0:48:29 for any
0:48:35 magnitude
0:48:36 more
0:48:36 PCs
0:48:40 as a result
0:48:41 of this
0:48:41 thing
0:48:41 that
0:48:42 Microsoft
0:48:42 lost
0:48:43 and then
0:48:44 it takes
0:48:44 until
0:48:44 mobile
0:48:45 that
0:48:46 they lose
0:48:46 the device
0:48:47 as well
0:48:47 as the
0:48:47 development
0:48:48 environment
0:48:49 so here’s
0:48:49 this kind
0:48:49 of
0:48:49 question
0:48:50 is
0:48:50 if all
0:48:50 the new
0:48:50 stuff
0:48:51 is built
0:48:51 on AI
0:48:52 and I’m
0:48:52 accessing it
0:48:53 in an app
0:48:53 that I download
0:48:54 from the App Store
0:48:56 to what extent
0:48:56 is this a problem
0:48:57 for Apple
0:49:01 and what would
0:49:01 have to
0:49:02 you would need
0:49:03 a much more
0:49:04 fundamental shift
0:49:04 in what it was
0:49:05 that was happening
0:49:06 for that to be
0:49:06 a problem
0:49:06 for Apple
0:49:07 and even if
0:49:08 you take
0:49:08 like the
0:49:08 you know
0:49:09 not the
0:49:09 like the full
0:49:10 like the
0:49:11 rapture arrives
0:49:11 and we all
0:49:12 just kind of
0:49:12 go and live
0:49:13 sleep in pods
0:49:13 like the guys
0:49:14 in Up
0:49:15 not Up
0:49:16 yes
0:49:16 what is it
0:49:17 the one with
0:49:17 the robot
0:49:18 that’s capturing
0:49:18 the trash
0:49:19 which one
0:49:19 is that
0:49:19 WALL-E
0:49:20 WALL-E
0:49:21 WALL-E
0:49:21 yeah
0:49:21 you know the
0:49:22 guys in the
0:49:22 pods in that
0:49:23 movie
0:49:23 maybe we’ll
0:49:24 be like that
0:49:25 in which case
0:49:25 fine
0:49:27 but like
0:49:27 there’s a
0:49:27 sort of
0:49:27 a mid case
0:49:28 which is
0:49:28 like the
0:49:28 whole nature
0:49:29 of software
0:49:29 changes
0:49:29 and there
0:49:30 are no
0:49:30 apps
0:49:30 anymore
0:49:30 and you
0:49:31 just go
0:49:31 and ask
0:49:31 the LLM
0:49:31 a thing
0:49:32 fine
0:49:32 what is
0:49:33 the device
0:49:33 on which
0:49:33 you ask
0:49:34 the LLM
0:49:34 a thing
0:49:35 well it’s
0:49:36 probably
0:49:36 going to
0:49:36 have a nice
0:49:36 big color
0:49:37 screen
0:49:38 and it’s
0:49:38 probably
0:49:38 going to
0:49:39 have like
0:49:39 a one
0:49:39 day battery
0:49:40 life
0:49:40 probably
0:49:40 needs a
0:49:41 microphone
0:49:42 probably a
0:49:42 good camera
0:49:48 kind of
0:49:48 sounds like
0:49:49 an iPhone
0:49:52 am I going
0:49:52 to buy the
0:49:53 one that’s
0:49:53 a tenth of
0:49:54 the price
0:49:54 and just
0:49:54 use the
0:49:55 LLM
0:49:55 on it
0:49:58 no
0:49:59 because I’ll
0:49:59 still want the
0:50:00 good camera
0:50:00 and the
0:50:01 good screen
0:50:01 and the
0:50:02 good battery
0:50:02 life
0:50:02 so
0:50:03 it’s
0:50:04 not
0:50:05 there’s a
0:50:05 bunch of
0:50:06 kind of
0:50:06 interesting
0:50:07 strategic
0:50:07 questions
0:50:08 when you
0:50:08 start poking
0:50:09 away
0:50:09 well what
0:50:09 does this
0:50:10 mean for
0:50:10 Amazon
0:50:11 those are
0:50:12 completely
0:50:12 different
0:50:12 questions
0:50:13 to what
0:50:13 does it
0:50:14 mean for
0:50:14 Google
0:50:15 or what
0:50:15 does it
0:50:15 mean for
0:50:15 Apple
0:50:16 what does
0:50:16 it mean
0:50:16 to Facebook
0:50:17 or what
0:50:17 does it
0:50:17 mean to
0:50:17 Salesforce
0:50:18 what does
0:50:19 it mean
0:50:20 to Uber
0:50:21 and then
0:50:22 right back
0:50:22 to what we
0:50:22 were saying
0:50:23 at the
0:50:23 beginning
0:50:23 of this
0:50:24 conversation
0:50:24 what does
0:50:24 this
0:50:25 mean
0:50:25 for Uber
0:50:27 well
0:50:27 their
0:50:28 operations
0:50:29 get X
0:50:29 percent
0:50:29 more
0:50:30 efficient
0:50:30 and now
0:50:31 the fraud
0:50:31 detection
0:50:31 works
0:50:33 and
0:50:33 you know
0:50:34 okay maybe
0:50:34 they’re
0:50:34 autonomous
0:50:35 cars
0:50:35 different
0:50:35 conversation
0:50:36 but
0:50:36 presume
0:50:37 no
0:50:37 autonomous
0:50:37 cars
0:50:37 that’s
0:50:37 a whole
0:50:38 other
0:50:38 conversation
0:50:39 otherwise
0:50:39 as Uber
0:50:39 what does
0:50:40 this
0:50:40 change
0:50:40 well
0:50:42 not a
0:50:42 huge
0:50:43 amount
0:50:44 I want to
0:50:44 sort of zoom
0:50:45 out a little
0:50:45 bit
0:50:46 this whole
0:50:46 framing
0:50:48 so you’ve
0:50:48 been doing
0:50:48 these
0:50:49 presentations
0:50:49 for a while
0:50:49 now
0:50:50 you’ve
0:50:50 bumped
0:50:50 them up
0:50:51 two times
0:50:52 because
0:50:52 so much
0:50:53 is changing
0:50:54 and one
0:50:55 of the
0:50:55 things you do
0:50:55 in each
0:50:56 presentation
0:50:56 is you’re
0:50:57 famous for
0:50:57 asking
0:50:58 really great
0:50:58 questions
0:50:59 and chronicling
0:50:59 what are
0:50:59 the important
0:51:00 questions to
0:51:01 be asking
0:51:02 I’m curious
0:51:03 as you reflect
0:51:04 you know
0:51:05 maybe post
0:51:05 you know
0:51:06 chat GPT in
0:51:06 2022
0:51:08 or GPT 3
0:51:08 rather
0:51:10 the questions
0:51:10 you were asking
0:51:11 then
0:51:12 and you reflect
0:51:12 on to now
0:51:14 to what extent
0:51:15 do we have
0:51:15 some direction
0:51:16 on some of
0:51:16 those questions
0:51:17 or to what
0:51:17 extent
0:51:18 are they
0:51:18 the same
0:51:19 questions
0:51:20 or new
0:51:21 and different
0:51:22 questions
0:51:23 or what
0:51:23 what is
0:51:23 sort
0:51:23 of your
0:51:24 you know
0:51:24 if I
0:51:25 woke up
0:51:25 in a coma
0:51:27 after reading
0:51:27 your you know
0:51:29 your original
0:51:29 presentation
0:51:30 let’s say
0:51:30 the one
0:51:32 after GPT 3
0:51:32 launch
0:51:33 came out
0:51:34 and then
0:51:35 seeing this
0:51:35 one now
0:51:36 what were
0:51:36 the sort
0:51:36 of most
0:51:37 surprising
0:51:37 things
0:51:38 or things
0:51:38 that we
0:51:39 learned
0:51:40 that updated
0:51:40 those questions
0:51:42 so I
0:51:42 think we
0:51:43 have
0:51:44 a lot
0:51:45 of new
0:51:45 questions
0:51:46 this year
0:51:48 so I
0:51:49 feel like
0:51:49 you know
0:51:49 you could
0:51:50 make a list
0:51:50 of as it
0:51:51 might be
0:51:51 half a dozen
0:51:52 questions
0:51:53 in spring
0:51:53 of 23
0:51:55 like
0:51:56 open source
0:51:57 China
0:51:58 Nvidia
0:51:59 does scaling
0:52:00 continue
0:52:02 what happens
0:52:03 to images
0:52:05 how long
0:52:05 does open
0:52:06 AI’s lead
0:52:06 remain
0:52:07 and those
0:52:07 questions
0:52:08 didn’t really
0:52:08 change in
0:52:09 23 and
0:52:09 24
0:52:11 and most
0:52:11 of those
0:52:11 questions
0:52:11 are kind
0:52:12 of still
0:52:12 there
0:52:12 like the
0:52:12 Nvidia
0:52:13 question
0:52:13 hasn’t really
0:52:13 changed
0:52:15 the answer
0:52:17 on how many
0:52:18 models will
0:52:18 there be
0:52:18 the answer
0:52:19 is okay
0:52:19 there’s going
0:52:19 to be
0:52:20 anybody
0:52:21 who can
0:52:21 spend a couple
0:52:23 of billion dollars
0:52:24 can have a
0:52:25 frontier model
0:52:25 and that was
0:52:27 pretty obvious
0:52:28 in early 23
0:52:29 it took a while
0:52:29 for everyone
0:52:30 to understand
0:52:30 that
0:52:31 and big
0:52:31 models
0:52:31 and small
0:52:31 models
0:52:32 will we have
0:52:32 small models
0:52:33 running on
0:52:33 devices
0:52:34 no because
0:52:34 the small
0:52:34 models
0:52:36 the capabilities
0:52:36 keep moving
0:52:37 too fast
0:52:38 for the small
0:52:39 models to shrink
0:52:39 the small
0:52:39 model onto
0:52:40 the device
0:52:40 but those
0:52:41 questions
0:52:41 kind of
0:52:41 didn’t
0:52:41 change
0:52:42 for two
0:52:42 and a half
0:52:42 years
0:52:43 I think
0:52:43 we now
0:52:44 have
0:52:44 I think
0:52:45 a bunch
0:52:46 of more
0:52:47 product
0:52:47 strategy
0:52:48 questions
0:52:49 as you see
0:52:50 real consumer
0:52:50 adoption
0:52:51 and open
0:52:51 AI and
0:52:52 Google
0:52:53 building stuff
0:52:53 in different
0:52:54 directions
0:52:54 Amazon
0:52:55 going in
0:52:55 different
0:52:56 directions
0:52:57 Apple
0:52:57 trying and
0:52:58 obviously
0:52:58 failing and
0:52:59 then trying
0:53:00 again to do
0:53:00 stuff
0:53:01 there’s some
0:53:02 sense of
0:53:03 like there
0:53:03 is something
0:53:04 more going
0:53:04 on in the
0:53:05 industry
0:53:05 than just
0:53:06 well let’s
0:53:06 just build
0:53:07 another model
0:53:07 and spend
0:53:08 more money
0:53:09 there’s
0:53:09 more
0:53:10 questions
0:53:10 and more
0:53:10 decisions
0:53:11 now
0:53:12 there’s
0:53:12 also
0:53:13 more
0:53:13 questions
0:53:14 outside
0:53:15 of
0:53:15 tech
0:53:16 in
0:53:16 certainly
0:53:17 on like
0:53:17 the
0:53:17 retail
0:53:18 media
0:53:18 side
0:53:19 of
0:53:21 how do you
0:53:21 start
0:53:22 thinking
0:53:23 about
0:53:24 what you
0:53:25 would do
0:53:25 with this
0:53:26 and again
0:53:27 you know
0:53:27 classic
0:53:27 framing
0:53:28 in my
0:53:28 deck
0:53:28 is
0:53:28 like
0:53:28 step
0:53:29 one
0:53:29 is
0:53:29 you
0:53:29 make
0:53:29 it
0:53:29 a feature
0:53:30 and you
0:53:30 absorb
0:53:30 it
0:53:30 and you
0:53:31 do the
0:53:31 obvious
0:53:31 stuff
0:53:31 step
0:53:31 two
0:53:32 is
0:53:32 you do
0:53:32 new
0:53:32 stuff
0:53:33 step
0:53:33 three
0:53:33 is
0:53:34 maybe
0:53:34 someone
0:53:34 will
0:53:34 come
0:53:35 and pull
0:53:35 the whole
0:53:35 industry
0:53:36 inside out
0:53:36 and completely
0:53:37 redefine
0:53:37 the question
0:53:38 and so
0:53:38 you could
0:53:39 kind of
0:53:39 do
0:53:39 like
0:53:39 an
0:53:39 imagine
0:53:40 if
0:53:40 here
0:53:41 of
0:53:41 like
0:53:42 step
0:53:42 one
0:53:43 is
0:53:44 you know
0:53:45 you’re
0:53:45 a manager
0:53:45 at a
0:53:46 Walmart
0:53:46 in
0:53:48 the
0:53:48 Bay Area
0:53:49 or DC
0:53:50 or whatever
0:53:50 it is
0:53:50 step
0:53:51 one
0:53:51 is
0:53:51 find me
0:53:51 that
0:53:52 metric
0:53:52 step
0:53:53 two
0:53:53 is
0:53:53 build
0:53:53 me
0:53:53 a
0:53:54 dashboard
0:53:55 step
0:53:55 three
0:53:56 is
0:53:56 it’s
0:53:57 Black
0:53:57 Friday
0:53:59 and I’m
0:53:59 running
0:53:59 managing
0:54:00 a
0:54:00 Walmart
0:54:01 outside
0:54:01 of DC
0:54:03 what should
0:54:03 I be
0:54:03 worried
0:54:04 about
0:54:05 like
0:54:06 and that
0:54:06 might be
0:54:07 the wrong
0:54:07 one
0:54:08 but it’s
0:54:08 like
0:54:08 you know
0:54:09 step
0:54:09 one
0:54:09 for
0:54:10 Amazon
0:54:10 is
0:54:10 you bought
0:54:11 light bulbs
0:54:11 so here’s
0:54:12 so you bought
0:54:12 bubble wrap
0:54:13 so here’s
0:54:13 some packing
0:54:14 tape
0:54:15 but what
0:54:15 Amazon
0:54:16 should actually
0:54:16 be doing
0:54:17 is saying
0:54:17 hmm
0:54:18 it’s like
0:54:18 this person
0:54:19 is moving
0:54:19 home
0:54:19 we’ll show
0:54:20 them a
0:54:20 home
0:54:20 insurance
0:54:21 ad
0:54:21 which is
0:54:22 something
0:54:22 that
0:54:22 Amazon’s
0:54:23 correlation
0:54:23 systems
0:54:24 wouldn’t
0:54:24 get
0:54:24 because
0:54:24 they
0:54:24 wouldn’t
0:54:25 have
0:54:25 that
0:54:25 in
0:54:25 their
0:54:25 purchasing
0:54:26 data
0:54:27 and
0:54:27 we’re
0:54:28 still
0:54:28 very much
0:54:29 at the
0:54:29 like
0:54:31 we’re
0:54:31 still
0:54:32 starting
0:54:32 we’re
0:54:33 still on
0:54:33 the
0:54:33 step
0:54:33 one
0:54:34 of
0:54:34 that
0:54:34 but
0:54:34 thinking
0:54:35 much
0:54:35 more
0:54:35 what
0:54:35 would
0:54:36 the
0:54:36 step
0:54:36 two
0:54:36 step
0:54:37 three
0:54:37 be
0:54:38 what
0:54:38 would
0:54:38 new
0:54:39 revenue
0:54:39 be
0:55:03 what
0:55:04 kinds
0:55:04 of
0:55:04 content
0:55:05 were
0:55:05 predicated
0:55:05 on
0:55:06 Google
0:55:06 rooting
0:55:07 that
0:55:07 question
0:55:07 to
0:55:08 you
0:55:08 and
0:55:08 what
0:55:09 kind
0:55:09 of
0:55:09 content
0:55:11 isn’t
0:55:11 really
0:55:12 that
0:55:12 question
0:55:13 like
0:55:13 do I
0:55:14 want
0:55:15 a
0:55:16 bolognese
0:55:16 recipe
0:55:17 or do I
0:55:17 want
0:55:18 to hear
0:55:19 Stanley Tucci
0:55:20 talking about
0:55:21 cooking in
0:55:21 Italy
0:55:22 like
0:55:23 do I
0:55:23 just
0:55:23 want
0:55:24 the
0:55:24 do I
0:55:25 want
0:55:25 that
0:55:25 SKU
0:55:26 or do I
0:55:27 want
0:55:27 to work
0:55:27 out
0:55:28 which
0:55:28 product
0:55:28 I
0:55:29 should
0:55:29 buy
0:55:29 which
0:55:29 is
0:55:30 Amazon
0:55:30 is
0:55:30 great
0:55:30 at getting
0:55:31 you
0:55:31 the SKU
0:55:31 terrible
0:55:32 at telling
0:55:32 you what
0:55:32 SKU
0:55:35 do I
0:55:35 just
0:55:35 want
0:55:36 the
0:55:36 slide
0:55:36 deck
0:55:37 or do
0:55:37 I
0:55:37 want
0:55:37 to
0:55:38 spend
0:55:38 a
0:55:38 week
0:55:39 talking
0:55:39 to
0:55:39 a
0:55:39 bunch
0:55:39 of
0:55:40 partners
0:55:40 from
0:55:40 Bain
0:55:41 about
0:55:41 how
0:55:41 I
0:55:41 could
0:55:41 think
0:55:42 about
0:55:42 doing
0:55:42 this
0:55:43 do I
0:55:43 just
0:55:44 want
0:55:44 money
0:55:45 or
0:55:45 do I
0:55:46 want
0:55:46 to
0:55:46 work
0:55:46 with
0:55:48 A16Z’s
0:55:49 operating
0:55:50 groups
0:55:51 what is
0:55:51 it that
0:55:52 I’m
0:55:52 doing
0:55:53 here
0:55:54 and I
0:55:54 think
0:55:54 the
0:55:55 LLM
0:55:57 thing
0:55:57 is
0:55:57 starting
0:55:57 to
0:55:58 crystallize
0:55:58 that
0:55:59 question
0:55:59 in
0:55:59 lots
0:56:00 of
0:56:00 different
0:56:00 ways
0:56:02 what am
0:56:02 I
0:56:02 actually
0:56:03 trying
0:56:03 to
0:56:03 do
0:56:03 here
0:56:03 do I
0:56:04 just
0:56:04 want
0:56:04 a
0:56:04 thing
0:56:04 that
0:56:04 a
0:56:05 computer
0:56:05 can
0:56:05 now
0:56:06 answer
0:56:06 for
0:56:06 me
0:56:06 or
0:56:07 do I
0:56:07 want
0:56:07 something
0:56:08 else
0:56:08 that
0:56:09 isn’t
0:56:09 because
0:56:09 the
0:56:09 LLMs
0:56:10 can do
0:56:10 a bunch
0:56:10 of
0:56:10 stuff
0:56:11 that
0:56:11 computers
0:56:11 couldn’t
0:56:11 do
0:56:12 before
0:56:12 right
0:56:13 is
0:56:14 that
0:56:14 thing
0:56:14 that
0:56:15 the
0:56:15 computer
0:56:15 couldn’t
0:56:15 do
0:56:16 before
0:56:16 my
0:56:16 business
0:56:18 or
0:56:18 or
0:56:18 am
0:56:18 I
0:56:18 actually
0:56:19 doing
0:56:19 something
0:56:20 else
0:56:20 we’re
0:56:20 about
0:56:21 to
0:56:21 figure
0:56:21 out
0:56:21 what
0:56:21 is
0:56:22 the
0:56:22 in
0:56:22 a
0:56:22 much
0:56:22 more
0:56:23 granular
0:56:23 way
0:56:23 what
0:56:24 is
0:56:24 the
0:56:24 true
0:56:24 job
0:56:24 to be
0:56:25 done
0:56:25 for
0:56:26 many
0:56:26 of
0:56:26 these
0:56:28 yeah
0:56:29 and
0:56:29 you know
0:56:29 going back
0:56:29 to the
0:56:30 internet
0:56:30 there was
0:56:30 you know
0:56:30 the
0:56:31 observation
0:56:32 about
0:56:32 newspapers
0:56:32 is that
0:56:33 newspapers
0:56:33 looked on
0:56:33 the internet
0:56:34 and they
0:56:34 talked about
0:56:34 you know
0:56:35 expertise
0:56:36 and curation
0:56:36 and journalism
0:56:37 and everything
0:56:37 else
0:56:38 and didn’t
0:56:38 really say
0:56:39 well we’re
0:56:39 a light
0:56:40 manufacturing
0:56:40 company
0:56:41 and a
0:56:41 local
0:56:41 distribution
0:56:41 and
0:56:42 trucking
0:56:42 company
0:56:43 and that
0:56:44 was the
0:56:44 bit
0:56:44 that was
0:56:44 the
0:56:45 problem
0:56:46 and until
0:56:46 the internet
0:56:46 arrived
0:56:47 like that
0:56:48 wasn’t a
0:56:48 conversation
0:56:49 you thought
0:56:49 about
0:56:50 and then
0:56:51 the internet
0:56:51 suddenly
0:56:52 makes that
0:56:52 clear
0:56:52 and suddenly
0:56:53 creates an
0:56:54 unbundling
0:56:54 that didn’t
0:56:55 exist before
0:56:56 and so
0:56:56 there will
0:56:57 be those
0:56:58 kinds of
0:56:58 like you
0:56:59 didn’t
0:56:59 realize you
0:57:00 were that
0:57:00 before
0:57:01 until an
0:57:01 LLM
0:57:02 comes along
0:57:03 and points
0:57:03 to
0:57:04 someone
0:57:04 comes along
0:57:05 with an
0:57:05 LLM
0:57:05 and says
0:57:06 I can use
0:57:07 this to do
0:57:07 this thing
0:57:07 that you
0:57:08 didn’t really
0:57:09 realize was
0:57:09 the basis
0:57:10 of your
0:57:11 defensibility
0:57:11 or the
0:57:11 basis
0:57:11 of your
0:57:12 profitability
0:57:13 I mean
0:57:13 it’s like
0:57:15 the joke
0:57:15 about
0:57:16 US health
0:57:16 insurance
0:57:17 the basis
0:57:18 of US
0:57:18 health
0:57:18 insurance
0:57:19 profitability
0:57:19 is making
0:57:20 it really
0:57:20 boring
0:57:21 and difficult
0:57:21 and time
0:57:21 consuming
0:57:22 that’s where
0:57:23 the profits
0:57:23 come from
0:57:24 maybe it isn’t
0:57:24 I don’t know
0:57:25 for the sake
0:57:26 of argument
0:57:27 say that’s
0:57:27 your
0:57:28 defensibility
0:57:28 well an
0:57:28 LLM
0:57:29 removes
0:57:30 boring
0:57:31 time consuming
0:57:32 mind numbing
0:57:32 tasks
0:57:35 so what
0:57:35 industries
0:57:36 are protected
0:57:36 by having
0:57:37 that
0:57:37 and they
0:57:37 didn’t
0:57:38 realize
0:57:38 that
0:57:39 and these
0:57:39 you know
0:57:40 it’s like
0:57:41 you could
0:57:41 have asked
0:57:42 these
0:57:42 questions
0:57:42 about the
0:57:43 internet
0:57:43 in the
0:57:44 mid-90s
0:57:44 or about
0:57:44 mobile
0:57:45 a decade
0:57:46 later
0:57:47 and generally
0:57:47 you’d have
0:57:48 half of the
0:57:48 questions you’d
0:57:49 ever asked
0:57:49 would have been
0:57:49 the wrong
0:57:50 questions in
0:57:51 hindsight
0:57:51 I remember
0:57:52 as a baby
0:57:53 analyzed in
0:57:53 2000
0:57:54 everyone kept
0:57:55 saying what’s
0:57:55 the killer
0:57:55 use case for
0:57:56 3G
0:57:58 what’s a good
0:57:58 use case
0:57:59 for 3G
0:57:59 and it
0:57:59 turned out
0:57:59 that having
0:58:00 the internet
0:58:00 in your
0:58:00 pocket
0:58:01 everywhere
0:58:01 was the
0:58:02 use case
0:58:02 for 3G
0:58:03 but that
0:58:04 wasn’t the
0:58:04 question that
0:58:05 people were
0:58:05 asking
0:58:06 and I’m
0:58:06 sure that
0:58:07 will be
0:58:07 the thing
0:58:08 now
0:58:10 is
0:58:11 there’s so
0:58:11 much that
0:58:12 will happen
0:58:13 and get
0:58:13 built
0:58:14 where you
0:58:15 go and
0:58:15 you realize
0:58:16 oh
0:58:18 that’s
0:58:18 how you
0:58:19 would do
0:58:19 this
0:58:19 you can
0:58:20 turn it
0:58:20 into
0:58:21 that
0:58:23 and I’m
0:58:23 sure you’ve
0:58:23 had this
0:58:24 experience
0:58:24 seeing
0:58:24 entrepreneurs
0:58:25 you get
0:58:26 every now
0:58:26 and then
0:58:27 they come
0:58:28 in and they
0:58:28 pitch the
0:58:28 thing
0:58:28 and you’re
0:58:29 like
0:58:30 oh
0:58:31 okay
0:58:33 you can
0:58:33 turn it
0:58:34 into that
0:58:35 I didn’t
0:58:35 realize it
0:58:36 was that
0:58:37 100%
0:58:39 my last
0:58:39 question
0:58:41 if we’re
0:58:42 talking
0:58:42 two or
0:58:42 three years
0:58:43 from now
0:58:43 you’re
0:58:44 doing a
0:58:44 presentation
0:58:44 you say
0:58:45 oh
0:58:46 this is
0:58:46 actually
0:58:47 bigger than
0:58:47 the internet
0:58:47 or maybe
0:58:48 this is
0:58:49 like
0:58:50 computing
0:58:51 what would
0:58:52 need to
0:58:52 be true
0:58:53 what would
0:58:53 need to
0:58:53 happen
0:58:54 what would
0:58:55 evolve our
0:58:56 thinking
0:58:57 I mean I
0:58:58 kind of
0:58:59 you know
0:58:59 sort of
0:59:00 come back to
0:59:01 my point
0:59:01 about you
0:59:01 know
0:59:03 Jews and
0:59:03 Christians
0:59:04 and the
0:59:04 Messiah came
0:59:05 nothing happened
0:59:10 we forget
0:59:11 I mean it’s
0:59:11 maybe two
0:59:12 very brief ways
0:59:13 to think about
0:59:13 this
0:59:14 one of them
0:59:14 is I think
0:59:15 we forget
0:59:16 how enormous
0:59:17 the iPhone
0:59:17 was
0:59:17 and how
0:59:18 enormous
0:59:18 the internet
0:59:19 was
0:59:19 and you
0:59:20 can still
0:59:21 find people
0:59:22 in tech
0:59:22 who claim
0:59:22 that
0:59:22 smartphones
0:59:23 aren’t a
0:59:23 big deal
0:59:25 and this
0:59:26 was the
0:59:26 basis of
0:59:26 people
0:59:27 complaining
0:59:27 about me
0:59:28 like this
0:59:28 idiot
0:59:28 he thinks
0:59:29 like
0:59:29 generative
0:59:29 AI
0:59:30 as big
0:59:30 as those
0:59:30 silly
0:59:31 phone
0:59:31 things
0:59:33 come on
0:59:35 I think
0:59:36 another answer
0:59:36 would be
0:59:37 like
0:59:39 I don’t want
0:59:40 to get into
0:59:40 the argument
0:59:41 about you
0:59:41 know what
0:59:41 is the
0:59:42 grace rating
0:59:43 capability
0:59:43 and benchmarks
0:59:44 and all
0:59:44 you know
0:59:45 you can see
0:59:45 lots of
0:59:46 five hour
0:59:46 long podcasts
0:59:47 of people
0:59:47 talking about
0:59:48 this stuff
0:59:50 but the
0:59:50 stuff we
0:59:51 have now
0:59:52 is not a
0:59:53 replacement
0:59:55 for an
0:59:55 actual
0:59:56 person
0:59:58 outside of
0:59:59 some very
1:00:00 narrow and
1:00:00 very tightly
1:00:01 constrained
1:00:02 guardrails
1:00:02 which is why
1:00:02 you know
1:00:03 Demis’s point
1:00:04 that it’s
1:00:05 absurd to say
1:00:05 that we have
1:00:06 PhD level
1:00:07 capabilities
1:00:07 now
1:00:13 what we
1:00:13 would have
1:00:14 to be
1:00:14 seeing
1:00:15 something
1:00:15 that would
1:00:16 really
1:00:16 shift
1:00:17 our
1:00:18 perception
1:00:19 of the
1:00:19 capability
1:00:20 of this
1:00:20 stuff
1:00:22 so that
1:00:22 it’s
1:00:23 actually
1:00:24 a person
1:00:26 as opposed
1:00:26 to
1:00:27 it can
1:00:27 kind of
1:00:28 do these
1:00:29 people like
1:00:29 things
1:00:30 really well
1:00:31 sometimes
1:00:31 but not
1:00:32 other times
1:00:33 and it’s
1:00:34 a very
1:00:35 tough
1:00:35 conceptual
1:00:38 thing to
1:00:38 think about
1:00:39 because
1:00:40 I’m
1:00:41 conscious
1:00:41 I’m not
1:00:41 giving you
1:00:42 a
1:00:42 falsifiable
1:00:42 answer
1:00:43 but I’m
1:00:44 not sure
1:00:44 what a
1:00:44 falsifiable
1:00:45 answer
1:00:45 would be
1:00:46 to that
1:00:47 when would
1:00:47 you know
1:00:48 whether this
1:00:48 was AGI
1:00:49 it’s the
1:00:50 Larry Tesla
1:00:50 line
1:00:50 AI is
1:00:51 whatever
1:00:51 doesn’t
1:00:51 work
1:00:51 yet
1:00:52 as soon
1:00:52 as people
1:00:53 say it
1:00:53 works
1:00:54 people say
1:00:54 well that’s
1:00:54 just not
1:00:54 AI
1:00:55 that’s
1:00:55 just
1:00:55 software
1:00:57 it’s
1:00:57 a
1:00:59 you know
1:00:59 it’s
1:00:59 an
1:01:00 and it
1:01:01 becomes
1:01:01 like a
1:01:02 kind
1:01:02 of
1:01:03 a
1:01:03 slightly
1:01:03 drunk
1:01:04 philosophy
1:01:05 grad
1:01:05 student
1:01:06 kind
1:01:06 of
1:01:06 conversation
1:01:07 as much
1:01:08 as it
1:01:08 is a
1:01:08 technology
1:01:09 conversation
1:01:09 like what
1:01:10 would it
1:01:10 have you
1:01:10 ever
1:01:11 considered
1:01:11 Eric
1:01:12 that
1:01:12 maybe
1:01:13 we’re
1:01:13 not
1:01:14 either
1:01:20 all I
1:01:21 can say
1:01:21 to give
1:01:22 a tangible
1:01:22 answer to
1:01:22 this
1:01:23 question
1:01:23 is
1:01:23 what
1:01:23 we
1:01:24 have
1:01:24 right
1:01:24 now
1:01:26 isn’t
1:01:26 that
1:01:27 will
1:01:27 it
1:01:28 grow
1:01:28 to
1:01:28 that
1:01:29 we
1:01:30 don’t
1:01:30 know
1:01:31 you
1:01:31 may
1:01:32 believe
1:01:32 it
1:01:32 will
1:01:32 I
1:01:33 can’t
1:01:33 tell you
1:01:33 that
1:01:33 you’re
1:01:34 wrong
1:01:34 we’ll
1:01:35 just
1:01:35 have
1:01:35 to
1:01:35 find
1:01:35 out
1:01:36 I
1:01:36 think
1:01:36 that’s
1:01:37 a good
1:01:37 place
1:01:37 to
1:01:38 wrap
1:01:38 the
1:01:39 presentation
1:01:39 is
1:01:39 AI
1:01:39 used
1:01:39 the
1:01:40 world
1:01:40 we’ll
1:01:40 link
1:01:40 to
1:01:40 it
1:01:41 it’s
1:01:41 fantastic
1:01:42 Benedict
1:01:42 thanks
1:01:42 so much
1:01:42 for
1:01:42 coming
1:01:42 on the
1:01:43 podcast
1:01:43 to
1:01:43 discuss
1:01:43 it
1:01:44 sure
1:01:44 thanks
1:01:45 a lot
1:01:47 thanks
1:01:48 for
1:01:48 listening
1:01:48 to
1:01:48 this
1:01:49 episode
1:01:49 of the
1:01:50 a16z
1:01:50 podcast
1:01:51 if you
1:01:51 liked this
1:01:52 episode
1:01:52 be sure
1:01:53 to like
1:01:53 comment
1:01:54 subscribe
1:01:54 leave us
1:01:55 a rating
1:01:55 or review
1:01:56 and share
1:01:56 it with
1:01:57 your friends
1:01:57 and family
1:01:58 for more
1:01:59 episodes
1:01:59 go to
1:02:00 youtube
1:02:00 apple
1:02:01 podcast
1:02:01 and spotify
1:02:02 follow us
1:02:03 on x
1:02:04 a16z
1:02:05 and subscribe
1:02:05 to our
1:02:06 substack
1:02:06 at
1:02:07 a16z
1:02:07 substack
1:02:08 dot com
1:02:08 thanks
1:02:09 again
1:02:09 for
1:02:09 listening
1:02:09 and I’ll
1:02:10 see you
1:02:10 in the
1:02:10 next
1:02:11 episode
1:02:12 as a
1:02:13 reminder
1:02:13 the content
1:02:13 here is
1:02:14 for
1:02:14 informational
1:02:14 purposes
1:02:15 only
1:02:15 should
1:02:16 not be
1:02:16 taken
1:02:16 as
1:02:16 legal
1:02:17 business
1:02:17 tax
1:02:17 or
1:02:18 investment
1:02:18 advice
1:02:19 or be
1:02:19 used
1:02:19 to
1:02:19 evaluate
1:02:20 any
1:02:20 investment
1:02:20 or
1:02:21 security
1:02:21 and
1:02:21 is
1:02:21 not
1:02:22 directed
1:02:22 at
1:02:22 any
1:02:22 investors
1:02:23 or
1:02:23 potential
1:02:24 investors
1:02:24 in
1:02:24 any
1:02:25 a16z
1:02:25 fund
1:02:26 please
1:02:26 note
1:02:26 that
1:02:27 a16z
1:02:27 and
1:02:27 its
1:02:28 affiliates
1:02:28 may
1:02:28 also
1:02:28 maintain
1:02:29 investments
1:02:29 in
1:02:29 the
1:02:29 companies
1:02:30 discussed
1:02:30 in
1:02:30 this
1:02:31 podcast
1:02:31 for
1:02:31 more
1:02:32 details
1:02:32 including
1:02:33 a link
1:02:33 to
1:02:33 our
1:02:33 investments
1:02:34 please
1:02:34 see
1:02:35 a16z
1:02:36 dot com
1:02:36 forward
1:02:37 slash
1:02:38 disclosures