00:00at the end of 2022 our team at a16z
00:02asked dozens of Partners across the firm
00:04to Spotlight one big idea that startups
00:07in their fields could tackle in 2023
00:09emerging from this exercise came
00:11literally 40 plus builder-worthy
00:13Pursuits for the year and that ranged
00:15from entertainment franchise games to
00:17Precision delivery of medicines small
00:19modular reactors and of course loads of
00:23and in this two-part series we'll be
00:25covering 12 of those big ideas in
00:28today's part one we'll cover ideas
00:29spanning consumer AI games and
00:31Enterprise featuring the voices and big
00:33ideas of Connie Chan and Lee skates Jack
00:36soslow Doug McCracken Sarah Wang and
00:39Sumit Singh in these conversations
00:40you'll hear several important themes
00:42including how physical transfers to
00:44digital what the next wave of AI might
00:46look like and specifically what role
00:48fault tolerance might play in this
00:50equation what we can learn about social
00:51commerce from the East and what role the
00:54growing number of bots in our lives
00:55might play as a reminder the content
00:59here is for informational purposes only
01:00should not be taken as legal business
01:02tax or investment advice or be used to
01:05evaluate any investment or security and
01:07is not directed at any investors or
01:09potential investors in any ACC fund for
01:12more details please see acc.com
01:25hi I'm Connie Chan a general partner on
01:28the consumer team and this is my big
01:32the big idea is around social commerce
01:34so social commerce Discovery Commerce
01:36video Commerce these Trends are
01:39inevitable in 2023 will be the year it
01:42all becomes obvious social platforms
01:44will become a natural place for product
01:46Discovery and platforms will seek to
01:48ease the friction between inspiration to
01:50purchase intent and finally the
01:52completed purchase transaction whether
01:54it's short form long form or live stream
01:57video is going to be a fantastic way to
01:59sell things and teach consumers about a
02:01product's value proposition this will
02:04birth an entirely new ecosystem where
02:06anyone can become a seller and new
02:08companies can help everyday creators
02:10curate and sell products and services
02:12all right Connie so I like your
02:13prediction but I think some people could
02:15argue that some of these aspects of
02:17social video and Discovery Commerce that
02:20you're describing are already in certain
02:21applications I'm just going to use
02:23Instagram as an example here so what
02:25gaps do you see in terms of how social
02:28implements Discovery or changes its
02:30involvement with e-commerce and how
02:32perhaps does this maybe compare to other
02:34parts of the world that have a different
02:37so I think you're right in the sense
02:40that on Instagram we already all shop
02:42from Instagram most of the time we get
02:44lots of product inspiration from the ads
02:46but whether or not we actually complete
02:48the transaction right there and then
02:49that's still a TBD and I think it's
02:52because a lot of the platforms today the
02:54Western platforms still have way too
02:55much friction between inspiration to
02:59discovering more and learning about the
03:00product to finally doing the actual
03:02checkout the actual payment completion
03:04today you might find inspiration on
03:07Instagram you might find a video of an
03:09influencer talking about a product but
03:11you might not click and complete the
03:13purchase right there and then you might
03:14instead open a browser and go to
03:16sephora.com and go search for that item
03:18or go to nordstrom.com and go search for
03:20that outfit the com the transaction
03:23right now is still kind of being broken
03:24up and that's because there's a lot of
03:26friction in the actual implementation of
03:27it's not great but I think the other
03:29thing about video that's still very
03:31underdeveloped is the idea of something
03:32like live video and live video is a
03:35fantastic way to sell things and it's
03:38something that I'd say we haven't really
03:39explored out in the west yet outside of
03:42these other platforms like whatnot that
03:43does live shopping for Collectibles and
03:47and the reason why live is so powerful
03:49is a bunch of things one there's
03:50Community the people who are all buying
03:52at the same time are talking to each
03:54other through comments in the chat
03:56they're talking to the seller you can
03:58ask the seller a question directly say
04:00you are buying a used luxury purse you
04:03can say hey turn it inside out can you
04:05flip it over I see a little stain can
04:07you put it up closer to the screen I
04:09want to look at it more closely so
04:11basically as a consumer you get more
04:14to buy and shopping is all about giving
04:17consumers confidence that this is the
04:19thing that's going to change their life
04:20and make their lives better right so
04:22live video just gives you more ways to
04:24answer questions and get that
04:25information to the consumer but also
04:28live video allows you to do these kind
04:31of limited time discounts or these
04:34window duration Promos in the same sense
04:36that you have Black Friday sales why not
04:39have a sale for a few hours where the
04:42price is just different and then drive
04:44people to do volume purchasing to do
04:46group purchasing there's just a lot more
04:47stuff that gets unlocked when it's live
04:50so whether it's again short video
04:52pre-recorded video long video I still
04:55think there's more to explore in the
04:56same sense that you and I when we grew
04:58up we had these like as seen on TV
05:00infomercials and they worked they work
05:03for so many of us and for a lot of it it
05:05was because it had entertainment value
05:07right now the video that we see on
05:09Instagram today most of them are still
05:11ads or maybe an influencer kind of
05:13campaign sharing a product with you is
05:16sharing their endorsement but they're
05:17not entertainment yet and there's still
05:19much ways to go before video shopping in
05:21the the West looks like entertainment
05:24yeah so I think you're painting a pretty
05:25compelling picture but I still am trying
05:28to understand why we haven't seen some
05:30of these buying behaviors I mean to your
05:31point the infomercial has existed in our
05:34Western culture for a long time and you
05:36said it worked and so what is stopping
05:39us from having these platforms or having
05:41the habits really of a consumer to go
05:43seek out that kind of shopping
05:45experience does it just really not exist
05:47yet has the right company not sprung up
05:49or why do you think we maybe see some of
05:52these behaviors in other parts of the
05:54world but not really in the West
05:57we see them more exemplified in a place
06:00like China where live shopping is very
06:01big I think it's because the platforms
06:05quite frankly years ago and there's also
06:08a huge ecosystem built out where you
06:10have people who are just curating
06:11products handling Logistics customer
06:13service and so forth but there is no
06:15reason that that user behavior that
06:18customer behavior is only going to work
06:20in Asia in fact in the US with whatnot
06:22you already see lots of people loving
06:25the community of buyers and sellers
06:27and finding things that they're buying
06:29that they didn't think to go by there's
06:32already an element of discovery that
06:33happens with love but you start that
06:36question asking about infomercials and
06:37and let me ask you when you saw those
06:40infomercials as a kid
06:42it wasn't just that the person was
06:44demonstrating the product it was because
06:46if you buy it now and that like 10
06:49minute window you had all these freebies
06:52or if you bought it right then in that
06:5610 minute window it was discounted but
06:58once that infomercial ended it might not
07:02that's why infomercials worked and that
07:05kind of time element is very possible
07:08with something like live video but not
07:10what we're seeing today with kind of
07:12these pre-recorded Instagram videos with
07:15celebrities because quite frankly the
07:16platforms haven't built out the tools to
07:18allow them to do something like that yet
07:21I love that you brought that up because
07:22I think through digital scarcity and how
07:25in an infomercial you're like if I don't
07:26buy this now it's not I'm not going to
07:29have access to this discount I won't
07:31know where to get it either yeah exactly
07:33and I feel like there's almost like the
07:36opposite Dynamic that has
07:38become a habit at least for me on
07:40Instagram using that example again where
07:42I will literally see an ad for something
07:43and if I don't click it right then and
07:45there I'm like I know I'm going to be
07:47retargeted I know I'm gonna get this ad
07:49again and so I don't feel this urgency
07:51this scarcity but another aspect of kind
07:54of my shopping experience from when I
07:56was younger that I don't see in the
07:58digital experience is I used to think
08:00shopping was fun it was a social
08:01activity I'd go with my friends we'd
08:03talk about you know what we wanted to
08:05get I'd look at what they're buying and
08:06I think what you're pointing at is also
08:11other parts of the world and in the East
08:13are it's a social experience It's
08:16Entertainment to actually shop which I
08:18don't think really exists in I mean for
08:21years now in in China when you were
08:23shopping you could put four or five
08:24dresses in a shopping cart send that
08:27link to several friends and they could
08:28comment on every single one and then you
08:31could see under every dress the comments
08:33from each of your friends
08:34and that feature's been there for years
08:36right so the idea of unlocking social
08:40commerce or product Discovery excites me
08:42a lot like when I walk into Target or
08:45I might have like one thing I need to
08:47get and I always end up with a lot more
08:49stuff in my shopping cart yeah totally
08:52inevitable right a cart from Costco less
08:55than 200 even when I went there for one
08:59but does that happen when you shop on
09:01Amazon it doesn't it doesn't because the
09:04West shopping is still search based and
09:08we are not focused on pushing kind of
09:10this discovery based
09:12full 100 recommendation engine shopping
09:15experience to the end user I say Shan is
09:19temu the new player in e-commerce pushed
09:22by pinged they're they're pushing that
09:24too but we have still yet to see
09:27more Commerce companies take on this
09:29product Discovery Focus as opposed to
09:32search and filter and let me just help
09:36that person navigate all the skus I have
09:38the reality is consumers when you went
09:40shopping with your friends you might
09:42have picked out coats or dresses that
09:43you wouldn't have gone to search for if
09:45you were just searching by you know neck
09:48type or or sleeve length or bestseller
09:51right it's not going to surface the same
09:53things to you as if you were just
09:54walking in a store that's a great
09:56analogy something else I want to ask you
09:58about is I've heard you talk about how
10:00social apps in particular which you
10:02could say are becoming Commerce focused
10:04apps if they aren't already a lot of
10:06them have kind of devolved into copying
10:09one another maybe the most prominent
10:11example is when Instagram copied snap
10:13Stories We also saw a version of this
10:16last year when Clubhouse was copied by
10:18just about every social app out there in
10:21terms of having live audio and so why
10:25does it at least seem like social
10:27platforms today are just copying the
10:29same features from one another and not
10:32testing some of these more creative
10:34features especially when we've seen some
10:36of these creative features work in other
10:39parts of the world yeah I think there's
10:42there's both logic to it if you think
10:44about it but also opportunity right to
10:46change the status quo the logical part
10:48is that hey this is already a proven
10:50idea I already have Distribution on my
10:53app why don't I just copy this feature
10:54it's going to work on my app too and
10:57that's actually a very I call it super
10:59app the phrase super app almost a super
11:01app mentality which is that I have
11:04my users will want the same thing
11:07I should stick that same feature or that
11:09same product in my app too and it's
11:11so there's Logic for why they're doing
11:13it but to your point there is missed
11:15opportunity because there are all these
11:17other business models that they're not
11:19exploring yet it's almost like they have
11:20to wait till some startup gets traction
11:22with it or some other company gets
11:24traction with it before they try and
11:26implement it and if they don't do it
11:28correctly it doesn't always work but you
11:30know some of these Eastern to Western
11:33translations they take time this has
11:35been part of my investment thesis for
11:37the last 10 years I take ideas that work
11:39in Asia and other parts of the world I
11:41try and figure out what can be copied in
11:43the US and then I try and teach it to as
11:45many people as possible so we can learn
11:48from companies outside of just the US
11:51my learning is that there's usually a
11:53time lag of at least a year and
11:54generally three to five years before
11:56some of those Concepts that are already
11:58mainstream in Asia become very popular
12:01and is that just a reality that there's
12:04just friction there's inertia in the way
12:06we do things which sometimes is
12:08different to other parts of the world or
12:10is there some sort of misnomers or some
12:12sort of misunderstanding where people
12:13think oh there's like a different
12:15culture and therefore it won't apply
12:17here I'm just trying to understand why
12:19it takes so long for us to absorb these
12:22learnings it requires a very open mind
12:25because you have to go at it or think
12:30with the assumption that you know just
12:32like Ratatouille anyone can be a cook
12:34like anyone can come up with a great
12:36product idea and if you come up with
12:38that mentality become much more
12:40open-minded and you actually want to see
12:41as many examples of success or failure
12:43as possible and I think another big
12:46reason though why some of this stuff
12:47takes a while to implement is twofold
12:51it requires huge organizational changes
12:53this is really a top-down decision it's
12:55very difficult to change this type of
12:56mentality Bottoms Up and the reality is
12:59if you are changing your business model
13:01of say like an Instagram news feed and
13:03you start pushing all these
13:05live shoppable videos that's taking up
13:08space that otherwise would have gone to
13:10an ad that generates money right so your
13:13p l takes a near-term hit you are using
13:16a pixel space that could have gone to
13:17something else and so this is a very
13:19high level decision or high level
13:21framework that needs to really come from
13:23the top and have really strong buy-in
13:25from the top because there are near term
13:28concept not consequences there there are
13:30near-term impacts when you try and make
13:32these big shifts right
13:34the second reason I'd say is as much as
13:37we use our phones and are addicted to
13:38our phones in the US I think we are
13:41still mobile but also PC oriented we are
13:45not mobile only maybe we're mobile first
13:48maybe younger Generations but we're not
13:50right whereas I say in China in a lot of
13:53developing countries they are mobile
13:55only there are people who literally do
13:58not have a computer and everything is
14:00done on their phone and therefore
14:01they're experimenting much much faster
14:03on mobile product experiences and mobile
14:07business models and in the US because a
14:10lot of platforms try and be backwards
14:11compatible they have to make sure that
14:13whatever they have on their app is also
14:15supported on the computer
14:17it limits their creativity or it limits
14:19the rate at which they can innovate I'll
14:22give you an example to kind of solidify
14:25only China is versus the US in the U.S
14:28if I'm buying a plane ticket
14:31to another country to another state
14:33whatever I might research that plane
14:35ticket on my phone but I might still buy
14:39and that's the data that actually comes
14:41back from the travel companies
14:43that those transactions the high ticket
14:45transactions are still happening on your
14:47computer and that's even phone in right
14:52some people might even phone in to have
14:54that Assurance right and whereas like
14:57something like that it would all be on
14:59the phone 100 of that kind of
15:01transaction would all happen on the
15:02phone in these developing countries and
15:04so this kind of Disconnect between are
15:06you mobile first are you mobile only is
15:09another big reason why the US moves
15:12I like that frame and I saw the same
15:14kind of leap frogging from living in
15:16Indonesia for a while where truly as
15:17you're saying it's it's not mobile first
15:19it's mobile only I like that frame a lot
15:21and something that you articulated in
15:24your idea for the year is that these
15:26Trends are in your own words inevitable
15:29so what do you see specifically changing
15:31in 2023 this upcoming year as we look to
15:35some of these perhaps long-term shifts
15:37but what do you see happening very soon
15:40I see platforms taking a
15:43I don't know if more risk is the right
15:44word but platforms are taking more
15:46action they are drawing they are going
15:50to experiment more specifically with
15:52things that related to Commerce or
15:53things related to creators there's going
15:55to be more experimentation happening on
15:57that front and because I'm so bullish on
15:59on the efficacy of it because I know I I
16:03have like an ab roller an OxiClean and
16:05like all this random stuff that I bought
16:06from infomercials they just work right
16:10um and I think it's going to work on the
16:11phone too there's no reason why it
16:13and because platforms are waking up to
16:16this and also brands are waking up to
16:18this I think that combination is the
16:21important Catalyst now that we're still
16:23missing ingredients we still need
16:24products to sell we still need creators
16:26who are really good at selling now the
16:29creators who are big on Instagram might
16:31not be the best sellers you know just go
16:35back to your infomercial days like there
16:37are some people really good at selling
16:39they're not necessarily going to win
16:41Academy Awards because they're not
16:42necessarily great actresses but they're
16:45really good at selling
16:46and in the same sense I also think even
16:49if we see an increase in Commerce for
16:52video you might also see a new category
16:55of creators emerge next up we have Anne
16:57hi I'm Anne Lee skates and I'm a
17:00consumer partner at a16z my big idea for
17:032023 is unlocking the third space
17:07Ray Oldenburg coined the term third
17:09space places that host regular voluntary
17:12informal and happily anticipated
17:14Gatherings outside of home or work
17:18these used to be in Real Life Community
17:20Gathering spots such as bars coffee
17:23shops churches gyms and clubs
17:26but remote and hybrid work as well as
17:29this generation's natively digital
17:31habits have given way to an online first
17:35so many interactions and first
17:37encounters occur online whether it's
17:39over Zoom or meeting like-minded friend
17:41with similar interests over Twitter or
17:43Tick Tock or in a Discord group or a
17:47what are the next generation of tools
17:49and platforms that are built to serve
17:50consumers Community Builders and
17:53creators in the post-covet hybrid world
17:55and how can new technologies such as
17:58generative AI create user experiences
18:00that facilitate deeper discussions and
18:02relationship building between real
18:04humans and not Bots all right and I am
18:06so excited to talk about this topic
18:08because we've been seeing this theme
18:09across several of our big Ideas this
18:11transition from physical to digital and
18:14before we get into that I just want to
18:15hear from you I mean we heard Ray
18:17oldenburg's definition but how would you
18:19define this idea of a third space I
18:22think the third spaces is physical
18:23locations that are not your home and not
18:26your work uh where you can explore
18:28different but important parts of your
18:30identity to me those are two Dimensions
18:32one is obviously the actual physical
18:35location but the other is a person's
18:38mental and psychological State third
18:40spaces are places that enable freedom
18:42and safety outside of our workplaces and
18:45our homes I think in terms of the
18:47physical location Dimension There Are
18:49Places you congregate socially and
18:51you're able to explore different parts
18:53of your uh identity freely and safely
18:56psychologically to me these are spaces
18:58where you don't have a defined role
19:00responsibility for instance at home you
19:02might be a parent right or you might be
19:03a breadwinner or a head of household or
19:06both but at work you have a specific job
19:09Duty uh and Associated responsibility
19:11neither of these places uh offer true
19:14Freedom colleges to me are really good
19:17earlier life examples of good third
19:19spaces so there's lots of communities
19:22and spaces where you're free to explore
19:25and free to interact with whoever you
19:28want for adults this could be religion
19:31in churches uh it could be games and
19:34gaming bars cafes arcades for Fitness
19:38enthusiasts this could be gyms and for
19:41art lovers you know this could be a
19:43museum or a particular Museum or even
19:45space within a museum but there's been
19:47definitely a dwindling of these types of
19:48public spaces and especially with covid
19:51actually there's an interesting stat
19:52from the Atlanta that in 2019 two thirds
19:55of of Americans said they had a favorite
19:57local place they went to regularly but
20:00that two-thirds has actually dropped to
20:02just a little more than half that's
20:04really interesting and I like the frame
20:06of having a place in society or feeling
20:08like you can almost shape or mold how
20:11you show up because as you said there's
20:12kind of like predefined Notions in the
20:15way that you show up to work or at home
20:16but these third spaces are really
20:18important you mentioned College like
20:19that was such a formative time of so
20:21many people's lives because they could
20:22kind of reshape themselves and say hey I
20:24actually want to show up differently
20:26here we're also starting to see in
20:28addition to the dwindling of these
20:29physical spaces maybe some of these
20:31digital spaces arise which is really
20:33interesting and something that I talk
20:35about a lot on the podcast is how we
20:38tend to copy and paste what exists in
20:40the physical world straight to the
20:41digital so you gave examples of like
20:43bars or churches during the pandemic we
20:46saw the equivalent of those again kind
20:48of being copied and pasted like the same
20:49routines the same ways of interacting
20:52just you know on a web page or in a zoom
20:56um but I think when things get really
20:57interesting it's when we start to
20:59rethink these spaces with the tools that
21:01we have online so have you seen any
21:03examples of where people have kind of as
21:06people say from first principles thought
21:07about how do I use this technology to
21:10actually facilitate some of these human
21:12connections that we previously formed in
21:14physical third spaces before I go into
21:17an example I think there's two
21:19particularly interesting things here
21:21about what digital native unlocks
21:25um for for these third spaces so the
21:28first is you can connect to more
21:30specialized people and have a much wider
21:32selection of people because literally
21:34now online you have a global poll to
21:37choose from that's like everybody at
21:39billions and billions of people in the
21:40world so like let's say you're an anime
21:43fan instead of connecting with 800k
21:45people in San Francisco because you live
21:47in San Francisco who may or may not
21:49share your interest in anime you can now
21:51actually uh be have a selection set of
21:54100 million people globally who are true
21:57fans of anime and you know more
21:59concretely like taking whatnot which is
22:01one of our portfolio companies that's a
22:03live shopping platform focused on
22:05Collectibles and communities as an
22:06example if you think about what not it
22:09might either look like just a
22:11Marketplace or it might just look like a
22:14digital games toy collectible store but
22:17it's actually not a copy and paste of a
22:19physical game store there's gamification
22:21built into the product via just a live
22:24auction mechanisms it's kind of like
22:25you're you know in this real-time
22:27auction uh with all these people and
22:30their digital goods and services that
22:32are transacted live and you can jump
22:34from one stream to another in real time
22:36and actually have tens of thousands of
22:38choices of different rooms or even
22:40different streams different events that
22:42you want to be a part of that you can
22:44literally immediately go to without
22:46physically changing your location
22:48this relates to my second thought which
22:50is in a physical event your experience
22:53is really limited by presence you can
22:56physically only be present in one place
22:59at one time uh your physical belongings
23:02and items can also only be in one place
23:04at one time but in a digital world this
23:07concept of presence is completely
23:09different you can actually reimagine
23:10what this means you can be in a digital
23:13environment but also know what's
23:15happening or be somewhat immersed in a
23:18completely different environment without
23:20having to physically go there and you
23:23can immediately change the experience of
23:24Presence by changing the environment in
23:26a digital world we can you know change
23:27this phase change and move change the
23:30furniture change the background what
23:32you're wearing anything and everything
23:33games is an area it's probably the most
23:35advanced in digitally native and I think
23:37in general it's always a good place to
23:39look for the future of consumer
23:40technology Dungeons and Dragons is a
23:43good example so it's obvious it's
23:45original OG RPG game you know there was
23:48sort of like a physical tabletop version
23:50you had to physically gather around with
23:53other players and have enough people to
23:54fill in various roles that are
23:56stipulated by the games but you're
23:58limited by the specific people you pick
24:00their real life imagination and actually
24:02having these people available but today
24:05in RPGs you're able to explore
24:08completely different worlds play with AI
24:10or humans but these humans can be from
24:13all over the world and you can immerse
24:16in any uh you know Visual and audio
24:18environment of your choosing that's
24:20literally not possible when you're
24:22playing in real life
24:23it's exciting I think about Chess is
24:26something I played when I was younger
24:27and I only had access to the chess
24:29coaches in my immediate area where the
24:31chess clubs in my immediate area and the
24:33people within those clubs and the skill
24:34sets that they were at and so yes it
24:36does completely change who you have
24:37access to but I also like this idea of
24:40different modalities of fun or
24:43engagement that you mentioned and
24:45something that comes to mind there is
24:46I've been talking a lot to my friends
24:49recently about how all we ever do is go
24:51and eat or drink and that's what you do
24:54you just meet up with friends and you
24:55talk but I remember when I was younger
24:58you know college comes to mind we would
24:59go play tennis we would go I don't know
25:02go join a dance so many more fun
25:04activities right things to do replic
25:08comes to mind here where digitally we've
25:09turned coding into a game you also
25:12mentioned you know the gaming space
25:13overall fortnite is something that we
25:15see many kids and older folks playing
25:19together and so I think that's that's
25:21encouraging to me I think a lot of
25:23people looked again aiming in a negative
25:24way but to me we're seeing again these
25:27new modalities of how people are
25:29engaging with one another instead of
25:31just talking to one another and so are
25:33there any other examples that come to
25:35mind there in terms of like how the
25:37internet or how the digital tools we
25:39have really change our ability to engage
25:44there's quite a few that come to mind
25:45and I definitely think gaming is the
25:47best example but actually for the
25:49purposes of this question since we're
25:50looking at all sorts of activities I'm
25:53hopeful that there will be a lot more
25:54activity digital activities that you can
25:56do with your friends
25:58um outside of gaming I think if you look
26:00at new relationships you can build
26:01through Tech what not Tick Tock Reddit
26:04discords these are all really great at
26:07facilitating intro space and sometimes
26:09even a stranger first relationships and
26:12actually deepening those and giving
26:14people a place to outlet and nerd out
26:17about very specific Niche hobbies and
26:20Niche interests for existing
26:22relationships you know there can be
26:24different forms of gaming ranging from
26:26really casual types of games like
26:28raising a virtual pets all the way
26:30through to more serious games like
26:31fortnite and I also think there's new
26:33forms of media creation would be real
26:36Tick Tock being the obvious most recent
26:38example one interesting one is lenza AI
26:42which is you know the sort of
26:44GPT generative AI created a visual of
26:48yourself based on a few photos of the
26:51person they actually have this creative
26:53thing where you can now make gen AI art
26:55for your friends and family and even for
26:57your pets so not just for yourself and
27:00it's an interesting new way uh to use
27:02technology to engage yeah I love how as
27:06you're as you're speaking about people
27:08are creating together I I can't think of
27:10other examples like this in the physical
27:12world other than going to maybe a random
27:14pottery class once a year with friends
27:16like I think it's really exciting that
27:18you actually see friends and family from
27:20far away connecting over the creation of
27:23these digital products I'm curious
27:25though I could see some pushback from
27:27people saying you know this is great
27:29like you can foster your online pet but
27:31we have to exist in the in the physical
27:33world and of course there's Merit to
27:35that we eat in the physical world we
27:37sleep in the physical world
27:38um and connection in that world is
27:40important too but I'm curious to know if
27:41you've seen any Trends in terms of some
27:43of habits that are inevitably being
27:45formed by the time the sheer time we're
27:48spending online if those habits or those
27:50new modalities that we've talked about
27:52are they reshaping any ways of
27:55interacting in the physical world have
27:56you seen any of the transfer in reverse
27:58because so far we've talked about how
28:00physical transfers to digital
28:03yeah yeah and I I also think it's
28:05interesting to see uh just the interplay
28:08one interesting example I I think from a
28:10technological lens is VR
28:13um it is obviously by nature in a
28:15completely immersive experience and so
28:18for instance we have this portfolio
28:21education portfolio company called
28:22prisms VR that's teaching kids math uh
28:25in a spatial world and what we see is
28:29actually physically in classrooms kids
28:31are you know using these headsets and
28:32learning math and VR uh the more
28:35advanced kids are actually more eager to
28:37help out kids who are struggling with a
28:39certain concept after being in this new
28:43novel digital environment even though
28:46they were solving problems alone in
28:48these digital environments so I find
28:50that really fascinating I would say
28:52underexport area here is thanks to zoom
28:56covet and remote work we're now really
28:58comfortable interacting with people
29:00online whether it's strangers or College
29:03or even you know your loved ones but I
29:06think there's an opportunity in digital
29:07products you know for digital products
29:09and features that prompt or create
29:11in-person interactions so Pokemon go to
29:14me was last such Mass Market product and
29:16now we're seeing some Innovations around
29:19the Creator economy of fandoms and
29:21creators and celebrities Gathering
29:24um and translating those to the physical
29:26space imagine like you showed up for a
29:28physical concert and you're prompted to
29:30meet other fans that you've already
29:32interacted with digitally before who
29:34maybe share the same favorite songs or
29:36share a same favorite experience with a
29:38celebrity or collect the same digital
29:40good I've been thinking a lot about how
29:42we have in many cases so much data from
29:46the digital world but then we don't as
29:48you're saying translate that back to in
29:50their physical Gatherings like I see so
29:51many podcasts or communities thrive in
29:54the digital world and then they organize
29:56these in-person meetups and they're just
29:58like hey everyone stand together like
29:59figure it out meet people and they have
30:02so much info information about the types
30:04of topics people like talking about even
30:06potentially demographic data where they
30:08can connect people who have the same
30:09problems or who are at the same phase of
30:11life I think you can facilitate such
30:13richer discussions and richer
30:16interactions once you use all those
30:19pieces of Digital Data that you're
30:20talking about exactly I love this idea
30:23of blending the two because there's
30:24positives to both the physical and
30:25digital and again I think when we go
30:27physical to digital we often just copy
30:29and paste in strange ways and then when
30:31we go the reverse way we almost like
30:33forget all of the benefit that we got
30:35from the digital world that could be
30:37implemented in the physical next up we
30:39have Jack hi I'm Jax Oslo I'm a partner
30:41on the games team focused on arvr games
30:44infrastructure and Ai and this is my big
30:46idea games as a never-ending Turing test
30:50it's not hard for a computer to deceive
30:54from Eliza to chat gbt computers have
30:57successfully masqueraded their biology
30:58this phenomena also occurs in games
31:01through the mass prevalence of bots
31:04Bots have historically been scripted
31:06procedures but are increasingly becoming
31:08true neural network based AIS
31:11the Bots are all around you you just
31:14don't realize look to the rumored
31:16concurrency bots in Words With Friends
31:18the close win bots in Call of Duty
31:20Mobile's onboarding the prevalence of
31:22cheaters thoughts and chess or the
31:23never-ending war of bots and Platforms
31:26in games like RuneScape
31:28the next generation of these thoughts
31:29will take human-like to a whole new
31:31level startups like in World ai conv ai
31:34or charisma.ai are making in-game agents
31:36that understand game State and have
31:38conversations emotions real objectives
31:42imagine walking through the Wilderness
31:43not knowing whether your Clan's Mage is
31:45a bot building a town in a Township tale
31:48with strangers without knowing the
31:50humanity of the local farmer playing a
31:53game of diplomacy but not knowing if
31:54Turkey is an AI gunning for European
31:58in the Airhead you may not know who's
32:00who anymore and you won't mind
32:02games are good alone but better together
32:04or so you think as you've said it's not
32:08that hard for a computer to trick a
32:10human it turns out the Turing test was
32:12probably too low of a bar and we're
32:14seeing this play out everywhere I think
32:15maybe a recent example of this is chat
32:17gbt which is throwing out a lot of
32:19correct answers but also a lot of
32:20incorrect answers and humans are just
32:22kind of taking it at face value so maybe
32:24we don't even need human-like Bots to
32:27trick humans it seems like less evolved
32:30Bots are already doing that so what do
32:32you think about that have we already
32:34kind of crossed that chasm
32:38um and I think the question here is like
32:39at what point do you cross the chasm in
32:43whatever domain to deceive humans so for
32:46example it's like you used to read the
32:48words like don't believe everything you
32:50read on the internet now I think this
32:52applies to AI like don't believe
32:53everything you read from AI
32:56um the Turing test as you say is like a
32:58flawed metric and the reason for this is
33:00in part because it is too low of a bar
33:03it is a binary bar so it's either
33:04intelligent or not intelligent and a
33:06test for deception not real
33:08understanding what these AIDS do is they
33:11try to understand the underlying rules
33:13that generate the data set that they're
33:15trained on and then generate an
33:17approximation of what they believe is
33:19the next token but it doesn't mean that
33:21it's right like in the case of Chachi BT
33:23it communicates with a lot of confidence
33:25a great example of this might be say
33:26asking chat GPT to multiply 153 times
33:31257. like it might get an answer that's
33:34close to correct and it communicates it
33:36like it is correct but it's not correct
33:38and this is a deterministic domain where
33:40you know what the right answer is there
33:42are a lot of domains where there isn't
33:43actually a right answer the the right
33:46answer is much more amorphous say like
33:48and you ask about China's what's your
33:50China's AI strategy be like this is a
33:53very complex question and it is dictated
33:55by say like your bias underlying what
33:58data you've seen in the past what you
34:01believe about China at large and then
34:03also like what you believe the
34:04importance of AI should be at large too
34:06this is a complicated question with a
34:08lot of different answers open Ai and
34:10chat jbt might have one answer and then
34:12say maybe a Chinese AI might have
34:14another because these things are
34:16information sources how much can you
34:19really trust them in the case of like
34:21Google you ascribe some trust to the
34:23outputs but you don't really know like
34:26the underlying model that contributed to
34:28generating the sources that you see you
34:31might trust your mother but what went
34:32into her like reasoning process you
34:35probably don't trust the onion but I'm
34:37sure some people's grandmothers do and
34:39the question of like should you trust
34:42the answer is like probably for most
34:45things but it's like not always that
34:48clear we don't know exactly what went
34:50into the training data which is the
34:52underlying rule set that they're
34:53generating their model upon and then
34:55additionally there's like a second layer
34:57of filtering so like what do they
34:59consider sensitive and then not give you
35:01a gender generation for with these two
35:03layers of filtering it's hard to
35:04understand exactly what is or what is
35:07isn't going on in the model but for the
35:09most part like when people have been
35:11testing the outputs they've been fairly
35:13trustworthy so in my opinion my faith is
35:16like I trust the AIS for the most part
35:18and then like will validate their
35:20results on like Google or other pieces
35:22of information online that I I've like
35:25have been more properly validated and
35:27what makes me positive and optimistic
35:31that trusting AIS is going to work out
35:34well is because there's been a large
35:36transition more towards open source and
35:38openness so you understand the data
35:40that's going into the models then you
35:42also under understand like the weights
35:43and biases so you know exactly what
35:45you're getting and then additionally
35:46these models get better and better over
35:48time so where 153 times 257 may have
35:51been incorrect in the past it might be
35:53right in the future with gpt4 and then
35:56because it's been trained on the
35:58entirety of the internet it actually
35:59does understand certain things much
36:02better than a human would so like I do
36:04trust it as an information source for
36:07the most part I like that you mentioned
36:09trust is really a sliding scale and
36:10that's not just with Bots but that's
36:11with humans as well and maybe one
36:13differentiating aspect of bots or in the
36:16case of CHA gbt as an example is that
36:18when there are falsehoods they can be
36:21corrected at scale as in like to use
36:23your example if an answer is incorrect
36:25and you can retrain the AI to no longer
36:27do that it will no longer do that not
36:29just in the single instance but globally
36:31so that's interesting but I think
36:33another interesting aspect of your big
36:34idea for the year is that we
36:36are really experiencing this exponential
36:39age as it relates to software and that
36:41includes Bots includes Ai and by nature
36:44of that exponential technology I think
36:46we can expect to see a lot more Bots and
36:49these Bots becoming a lot more
36:50convincing which is exciting maybe scary
36:53there's all sorts of implications of
36:55this where we're not just talking math
36:57true or false it's it's like do I know
37:00if this person's real or is it really
37:01their video right is it a deep fake and
37:04so I think there's going to be all sorts
37:06of implications but focusing it back on
37:08game specifically how do you think this
37:11is going to impact the gaming world is
37:13this going to make certain games not
37:14even worth playing because you're just
37:17really encountering a bunch of bots
37:20a game is a structured form of play and
37:22the key word there is structure like
37:24There Are Rules involved in a game so
37:26like when I go and play a game I enter
37:28to say like Call of Duty under the
37:30auspices that everybody is going to be a
37:32human and they're playing by the rules
37:33when there's a bot involved say someone
37:35has Aimbot or something like that then I
37:38feel like they I've been treated
37:39unjustly and then the game becomes Less
37:41the question then becomes like should
37:44you design rules for Bots if you're
37:46going through like the RuneScape
37:48never-ending war of like Banning Bots
37:50then the Bots come up with a different
37:52strategy to get around the band and then
37:53you have to ban those and your strategy
37:54Bend those but it's like is that worth
37:57the effort for just maybe designing the
37:59rules for bots in the first place maybe
38:01you lower the resource constraints or
38:03make it particularly hard for Bots to
38:05configure in this world or make it so
38:07that the humans have to interact with
38:08another human in order to get the Rune
38:11or to Smith whatever sword that you want
38:13these are this these are the types of
38:15questions that you have to ask so if
38:17it's a structured form of play perhaps
38:18the structure should be built around
38:20bots in the first place if you can't
38:21really defend against them then
38:23additionally it's like there is some
38:26times like in these games like you're
38:28interacting with Bots and you just don't
38:29know it especially when they're like
38:31non-complexed games like it may be a
38:33good analogy to think about this is
38:35imagine a bot versus a human who played
38:37tic-tac-toe like tic-tac-toe is a solved
38:40game it would be pretty much impossible
38:42to distinguish between the two now think
38:44about Chess now think about chatbots
38:45there's a sliding scale it's the
38:46complexity of tasks and like what you
38:48can distinguish to be a bug and not in
38:51an exponential age we're looking at say
38:53chat Bots that pass the Turing test but
38:55what does this mean for say like in
38:57Skyrim say like the local guard like in
39:00that case they might eventually get to
39:01the point where they are purely
39:03human-like maybe the the types of
39:05interactions that you have with them
39:06aren't like scripted chatbot lines but
39:09really the types of interactions that
39:11you might have with somebody that you
39:12build a relationship with perhaps the
39:14first time you enter a town the the
39:16guard is pretty guarded you know they
39:17give you a little sass and they ask for
39:19money but after you beat the werewolf
39:21that has been haranguing the town for
39:23years and you put them back in their
39:25cave the guard treats you with the love
39:27and respect that you deserve and you go
39:28to the local bar and people celebrate
39:30you these are the types of things that
39:32like over time will get more and more
39:34believable worlds which with more and
39:35more believable entities the reason why
39:38games like Minecraft have like added
39:40pets or villagers and things like that
39:42is because these single player games
39:44like survival Minecraft is a pretty
39:45lonely experience you want to have like
39:48real companionship or at least the
39:50feeling of pseudo companionship
39:52throughout your journey so if you have
39:54like really believable villagers in
39:56Minecraft that would be a nice ad or say
39:58for example if Navi in like link or the
40:01Ocarina of Time like there's literally
40:03someone who is with you the entirety of
40:05the game who kind of tutorializes the
40:07experience and gives you the sense of
40:08companionship throughout like what
40:10happens when that's an AI that can
40:11understand all the different mistakes
40:13that you might be making ahead of time
40:14or is actually building a relationship
40:17with you that's personalized for your
40:18experience there are a million different
40:20ways where this can kind of Bubble Up
40:22the ones that I'm most excited about are
40:24with like single player games and then
40:25also with multiplayer games where the
40:27multiplayer games become personalized
40:29where you actually don't play with other
40:31humans anymore you're playing with
40:33perfectly catered AI eyes for the
40:37experience that you want a good example
40:38of this is Call of Duty mobile where
40:40when you onboard you're playing against
40:41really bad thoughts and then they give
40:44you an initial win that helps you learn
40:46the actual skills that go into being
40:48good at this game and then slowly but
40:49surely you introduce more and more
40:51humans into the equation to level up the
40:53skill cap this is the type of experience
40:55that could be done at scale with games
40:57like fortnite say for example groups of
41:00different AIS that give you a close
41:02shootout these types of close wins are
41:05really good factors for predicting the
41:07likelihood that you come back in the
41:09future and there's this chart that game
41:11makers try to bring you on which is this
41:13kind of like Flow State which is like
41:14over time like the dopamine spikes hit
41:17and then they go down and then they go
41:18up a little bit more and then they go
41:19down a little bit more and then so on so
41:21forth all the way through and they try
41:22to cater the game experience for this
41:25exact feeling but in the case of
41:27multiplayer games you can't actually
41:28control what other people do but if you
41:30had AI Bots that were believable
41:32um you could so those are the kinds of
41:34things that I'm I'm looking forward to
41:36yeah I'm hearing a couple important
41:38things there one is it does seem
41:39unrealistic to try to void games of all
41:43Bots it doesn't seem like that's
41:45possible and so again recurating the way
41:47that we think about how games are
41:49designed is important and the second
41:51part is how do we make these Bots
41:52additive instead of reductive right
41:54there are the kinds of bots where
41:55actually you're describing they can make
41:57the game experience better and then we
41:59all know of bots where it just like
42:01completely destroys the game it makes it
42:02seem like it's not even worth playing
42:04because I can never achieve the skill
42:06level of a particular bot as an example
42:09the important distinction there is like
42:10a bot is something that is made by the
42:13game creator and then a hack is made by
42:16a game player to bypass the rules so
42:19like a bot is supposed to improve the
42:21game and then a hack is detracting from
42:22the game from everyone else right but at
42:25the same time we can't prevent the hacks
42:27most likely either right that's going to
42:29have to be part of the game design as
42:31well yes yes well I mean to that point
42:34it my limited game experience is in the
42:38realm of Chess and it reminds me of some
42:39of the stuff we've been seeing in chess
42:41recently where we've actually known that
42:43Bots are better than humans at chess for
42:45decades now since 1997 the famous deep
42:48blue versus Kasparov game but for
42:50decades since then people have still
42:52willingly and enjoyingly played chess
42:54because there was still a sentiment of I
42:56can improve at chess relative to other
42:58humans I'm still not facing bots in this
43:01sense more recently there's obviously
43:03been some controversy because there is
43:05now this gray area of can we control
43:07Bots kind of coming into the human realm
43:10of Chess but even then that's kind of
43:13skyrocketed chess into the Zeitgeist and
43:15it's become more interesting or
43:16important than ever and then you almost
43:18see this new spin-off it's actually
43:20existed for a while but it seemed to
43:21gain relevance which is the Fischer
43:23random test us which basically has the
43:26same fundamentals as chess but the
43:29pieces are swapped around and so it's
43:31almost like humans are trying to
43:33extend the game so that's still relevant
43:36so that they can still kind of test
43:37their humanness their human skill while
43:41avoiding the influence of bots so I'm
43:43interested to see how that might play
43:44out in other types of games as you said
43:46kind of designing games so that they are
43:49not so much bot proof because the Bots
43:51will play a role but that they are still
43:53interesting and exciting with the
43:55influence of bots 100 and what I really
43:58appreciate about games like Fisher
44:00random chess is that it's about
44:01understanding chess versus say like
44:05memorizing which a computer can do
44:06really well and what's essentially what
44:08like deep blue was doing at like a
44:10greater extent it was just like a really
44:11glorified search function I'm excited
44:13for other games that kind of consider
44:15Bots as first class citizens within them
44:18and then additionally new game modes
44:20that are built specifically for humans
44:22where instead of saying hey
44:25um no Bots you can play on bots on these
44:28servers but also these servers these are
44:30for humans and oh we'll let like
44:31Darwinism decide what is played more
44:33often that's it that's a good way to put
44:35it we'll let the consumers decide let's
44:37close this out by talking about 2023 in
44:39particular because we're going to see
44:41games and the influence of bots change
44:42over the coming years the coming decades
44:45that's going to be really hard to
44:46predict but what specifically in 2023
44:48are you excited about in this space I'm
44:50excited for foundation models to come
44:52out more domains so we saw image pretty
44:54much be solved last year and then texts
44:56continually get better and better this
44:58year I'm excited about 3D Foundation
45:00models audio Foundation models video
45:02Foundation models those sorts of things
45:04and then additionally I'm excited about
45:06how they all come together Under One
45:08Roof so how do the text models inform
45:10the 3D models inform the Audio models in
45:13the case of games like every game asset
45:15is not really just one 3D asset it is an
45:19asset with sound attached to it with
45:21code and scripts and each of these
45:24informs the other aspects say for
45:26example changing an old wooden door it
45:28might have a rickety slow moving like
45:30script as well as a sound attached to it
45:32but if I were to change change the door
45:34to be a metal door how does that
45:36actually change the script and then also
45:37the sound and then if I were to change
45:38the code how does that actually change
45:40the 3D asset and so on so forth so all
45:43of these things are connected and right
45:44now they've been primarily Point
45:45Solutions and then additionally there's
45:48like this kind of AI Unity that is born
45:50that really helps creators and game
45:52creators kind of abstract away a lot of
45:54the things that are mostly wrote or at
45:56least have been solved by experts in
45:58their respective fields
46:00yeah I think you co-wrote an article
46:02with James quartzman from our team and I
46:05thought the way you put it was really
46:07really strong in in the sense that if we
46:09look at how easy it is to publish online
46:11for so many creators to like tweet or
46:13write up a blog post that has happened
46:16in the last couple decades and if we
46:18look back to like the 1990s early 2000s
46:20that was not the case it was not that
46:22easy to publish online and I'm excited
46:24for kind of the same Renaissance within
46:26games it's kind of crazy actually to
46:27imagine that like anyone like myself who
46:31has very little gaming experience could
46:33actually spin up a game potentially in
46:34the coming years I'm sure you've had
46:36games in your head that you've wanted to
46:38create but you've got it because it's
46:39too hard in the same way that maybe like
46:41you wanted to make an image in the past
46:44if I don't know what I mean and it
46:46wasn't really possible until you saw mid
46:48Journey or Dolly like this this type of
46:49capability is right around the corner
46:51and it's a bet on that derivative AI is
46:54um it's uh you have to like kind of ride
46:56the wave and see where it's going to go
46:57so excited about all this I think games
47:00is coming for New Renaissance there's
47:02going to be tons and tons of games that
47:04are made but the real skill is in
47:06understanding what a good game looks
47:08like next up we have Doug hey I'm Doug
47:10McCracken from the games team here's my
47:12big idea for the year the metaverse goes
47:16Gamers know that character skins in
47:18games like League of Legends and
47:19fortnite are an important form of
47:22self-expression as they become part of a
47:23player's identity that's why character
47:26skins are big business despite having no
47:28gameplay benefits digital natives gen Z
47:31and gen Alpha demand that Brands enable
47:33self-expression in the metaverse of that
47:36group two and five already believe that
47:38self-expression via fashion is more
47:40important in the digital world than the
47:42physical in three and four say they will
47:45spend money on digital fashion
47:47brands that lead in like Gucci will be
47:50rewarded by the hearts and wallets both
47:51physical and digital of consumers and as
47:54physical Brands go digital more digital
47:56brands will go physical creating even
47:58stronger competition and broader
48:00adoption brands that don't go all in
48:02will be left behind consumers will
48:05demand interoperability across
48:07experiences in the metaverse so over
48:09time Brands will favor platforms that
48:11enable them to wear their Nike shoes in
48:13different games and Virtual Worlds gen Z
48:16and gen Alpha move seamlessly between
48:17the physical and digital Worlds the
48:19fashion brands that Embrace this will
48:22all right Doug so I feel like maybe as
48:26someone who's not a super active gamer
48:28fortnite comes to mind as a great
48:30example of where game players are
48:31spending money on digital assets that as
48:34you said don't offer specific gameplay
48:36benefits can you give some other
48:38examples for people who may not be
48:40Familiar of this same dynamic
48:42so first of all for games 90 of gen Z
48:46play games right and gen outfits it's
48:49even higher I think it's 94 for Gen
48:51Alpha so the idea the definition of
48:53games is very broad at this point and
48:55games are becoming increasingly social
48:57I'm not sure if you've read the book
48:58status and culture but it's uh it's a
49:01really interesting look into what drives
49:03identity and culture and it really a lot
49:06of it is about status so you know games
49:08that have a strong sense of identity
49:10tend to be social multiplayer games in
49:13particular have very strong sense of
49:15identity so so fortnite has really built
49:18an amazing game that essentially built
49:20for Partnerships like what other game
49:21can have DC Superheroes Ariana Grande
49:25and Rick and Morty y'all in the same
49:27world right so they've just done an
49:29incredible job but another example is
49:30League of Legends the Skins really
49:32started as kind of fully imagined
49:34alternate versions of the characters in
49:36alternate universes but they've also
49:38done Partnerships they partner with
49:39Louis Vuitton for example for their
49:41World Finals when it was a in Paris and
49:43they launched a set of skins and there
49:46are other examples as well so Minecraft
49:48GTA online both have really strong skin
49:51ecosystems especially with ugc
49:54components right so so people that are
49:56players actually creating skins for
49:59those game experiences themselves and
50:01then two other examples are Roblox and
50:03Animal Crossing that focus more on on
50:06got it and I love that you went through
50:08so many examples but I do feel like we
50:10have a limited number of examples just
50:12by the nature of this being relatively
50:14nascent based on those examples are
50:16there any Trends any playbooks that
50:18you've seen work particularly well is it
50:20just about partnering with these large
50:21Brands like Louis Vuitton or artists
50:23like Ariana Grande is it about
50:25introducing some level of scarcity is it
50:27the ugc component have you seen any kind
50:30of indicators of what people have been
50:33starting to figure out in the space as
50:35as it comes to what really works here
50:37yeah I mean stay more focused on fashion
50:39for the moment I think we're in this
50:41experimental phase where you know the
50:45the brands seem to be thinking about
50:46their Partnerships with the digital
50:48worlds more as marketing and over time I
50:51think that will build into a bigger
50:52business for them and back on that book
50:55status and culture one of the other
50:56hypotheses is that that you know High
50:59status individuals and I will apply this
51:01to Brands too tend to deviate from the
51:03norm so it's interesting to see the
51:05luxury Brands seem to have embraced this
51:07trend first with the virtual worlds and
51:09the games and and really digging into
51:11these Partnerships I'll give you an
51:12example so Kim Kardashian were a
51:15Balenciaga dress at New York Fashion
51:17Week and then fortnite players noticed
51:20that there was actually a skin that
51:23looked very similar to that dress that
51:24was locked they couldn't see the full
51:26skin but they started conversation and
51:28then of course fortnite comes out and
51:29reveals that they've done a partnership
51:31with Balenciaga and created skins in the
51:33game related to this New York Fashion
51:35Week event so I think that's just an
51:37interesting way that the brands have
51:39been really building awareness with the
51:41player bases of these games and you know
51:44excitement with those player bases and
51:46that's one example but creating scarcity
51:48is another example right and I think
51:50over time as we look at web3 and other
51:54types of technologies that enable
51:55ownership and resale that actually
51:57amplifies this idea of of scarcity and
52:00the value of scarcity you know another
52:02example is feature differentiation so
52:04the digital world enables a lot more
52:07feature differentiation than the
52:08physical world I'll give an example of
52:10League of Legends there are different
52:12tiers of skins and this is kind of OG
52:14for all the League of Legends players
52:16out there but DJ Sona was the third
52:18ultimate skin which is the highest tier
52:20to be released and you know they charged
52:23I think it was roughly 40 50 at the time
52:25for that skin and it was a complete
52:28rework of the character she had three
52:30different versions of herself she was a
52:32DJ so there was music that was custom
52:35and you could share that music with your
52:37team and then all of the effects were
52:39redone and so there are many different
52:41opportunities to really try new things
52:44and add really cool benefits into the
52:47experience with fashion or with skins or
52:50cosmetics and so I think that there's a
52:51lot of experimentation happening proven
52:54the business kind of in the core games
52:55like League of Legends and fortnite and
52:57what's going to happen for the brands
52:58how are they going to participate in
53:00that business over the long term that
53:01remains to be seen I love that you
53:03mentioned that because I think it's very
53:05natural for us as humans to just kind of
53:07replicate what we have in the physical
53:08world in the digital world so to give an
53:11example that's parallel to your Kim
53:12Kardashian example people have been
53:14doing this all the time when they see a
53:15celebrity wear a particular outfit
53:17they'll go and figure out what designer
53:19created that outfit but then they'll
53:21also find a bunch of different brands
53:22who have created similar clothing and
53:24they'll say hey if you want to be like
53:26Meghan Markle you know wear these jeans
53:28or wear this dress and so it looks like
53:30we're again kind of copying and pasting
53:32some of the behaviors from the physical
53:34world into digital but as you said
53:37there's this opportunity to rethink it
53:39it's like what does fashion look like
53:41natively within games as you're saying
53:43like there's all types of signaling from
53:45music to different tiering of the
53:49specific assets that that people are
53:51selling and it just reminds me of Snow
53:54Crash from 30 years ago where I'm
53:57reminded of when people went to the
53:59metaverse in that book they could show
54:02up as a better version of themselves as
54:03in like what they physically look like
54:05in real life or they could show up as
54:06like a duck or something that doesn't
54:08even exist in the quote unquote real
54:10world so I think that's really exciting
54:12but I want to ask you about something
54:14that I've been noticing which is that we
54:17spend so much time in this Digital World
54:19many of us I can speak for myself I've
54:21probably spend more time realistically
54:24in the online world than the offline
54:25world but my share of spend you could
54:28call it does not represent that so
54:31people are still much more likely today
54:33to buy a suit in the physical world than
54:35you know buy maybe some sort of
54:36equivalent in the digital world are you
54:39seeing any indicators any data any
54:41trends that maybe show that this might
54:44be changing yeah so it's interesting I
54:47the global apparel and Footwear business
54:49I've seen stats that say it's roughly
54:52two trillion dollar business right and
54:54so that's massive but I think it's it's
54:55hard to compare Revenue the cost
54:57structure is completely different so I
54:59don't think I don't think digital will
55:01fully disrupt physical we still need to
55:03wear outfits and you know dress for for
55:05daily life but I do see that for maybe a
55:08specific example is fortnite earned over
55:10two billion dollars uh for multiple
55:12years right more than half of their
55:14Avenue comes from cosmetics and I also
55:16think the ugc component for digital is
55:20very different because if you're able to
55:22immediately download something that a
55:24Creator created maybe for free maybe for
55:27a small amount of money how do Brands
55:28compete in that environment and so the
55:31pricing structure may be different and I
55:33do see that the features that we were
55:35talking about earlier could play a big
55:36role in that in improving the value in
55:38the digital world I would have loved
55:40when I was a kid to have those those
55:42Nike shoes that Marty McFly put on in
55:45Back to the Future too and you know they
55:47they're self-lacing that's a basic
55:49feature in the metaverse right like what
55:51out what other cool features can we add
55:53but maybe just one last last point is
55:55that with so much content
55:58people will likely change their outfits
56:00more depending on their experiences and
56:02so the consumption of different types of
56:05of content could be much much broader
56:07than in the physical world
56:09I like that you also mentioned the
56:10pricing structure because I it reminds
56:13me of apps in the early days for
56:15whatever reason people felt like it was
56:18too much to ask for people to spend 99
56:20cents on an app even though often it
56:22provided much more value than maybe it's
56:24parallel in the physical world and so it
56:26will be interesting to see all of these
56:28experiments come to be and see where the
56:30pricing really ends up because this is
56:33as we've been talking about in nascent
56:34space and perhaps it is also a
56:36generational thing whereas more
56:38digitally native people start
56:41with these games maybe they'll have less
56:43friction as you know people who maybe
56:45are older who you know who question this
56:48idea of spending on this digital asset
56:51within a game without the game benefits
56:53but something I want to get to that you
56:55spoke about in your big idea is the need
56:57for interoperability so you said in your
57:00big idea that consumers will demand the
57:02ability to interoperate their assets
57:04between platforms but I want to push you
57:06on this because we don't see that many
57:09examples of this today most games do
57:11kind of operate in their silos and I
57:13want to ask you is this due to maybe the
57:15technology we've had up until now or is
57:18it due to Consumers truly just enjoying
57:21the quote unquote best games and by
57:24nature of that these are the games where
57:26the creators of these games have decided
57:28to allocate engineering resources to
57:30just their games and not necessarily
57:32interoperating with other games do you
57:34have any thoughts on that and how we
57:35might progress there I do I have two
57:37main points here I think that the first
57:39is really related to there are kind of
57:41standards and Technical hurdles for
57:44implementing interoperability of course
57:46and then you know certain games have
57:48created their own modes and a lot of the
57:50most popular games now have existed for
57:5210 plus years right and they Continue to
57:54update and innovate but what I'm
57:56interested in is what are kind of the
57:58next generation of companies doing and
58:00there's a company for example Ready
58:02Player me who is a cross-game avatar
58:06company for the metaverse and they have
58:096 000 developers that have signed up to
58:11leverage their Avatar technology right
58:14and so what happens when one of those
58:17developers becomes the Next Generation
58:20Roblox or League of Legends you know so
58:24I think that it's the Builders of the
58:26future who are leveraging this Tech it
58:28actually solves an engineering problem
58:29for these startups because they don't
58:32have to develop their own avatar Tech
58:33right so it actually helps them solve
58:36that and then adds interoperability and
58:38so I think there is Tech that's being
58:40developed to solve this problem and
58:42there is adoption happening pretty
58:43rapidly and I think the second Point
58:45really is on the consumer side you know
58:47if you think about back to the Gen Z gen
58:49Alpha demographics kids are growing up
58:52today playing games like Roblox where
58:55there's an expectation that you can
58:57customize your avatar and you can go
59:00into different experiences with that
59:01Avatar you come out of the experience
59:03you customize it again and you go in a
59:05different experience right and so I
59:07think that that is really creating this
59:09environment where kids are going to
59:10expect that with the next generation of
59:12games in Virtual Worlds as well so you
59:14know we'll see how this all plays out
59:15and I don't think it's it's black or
59:17white like there's an interoperable
59:18future or there's not an interoperable
59:21future I think both can coexist but I
59:23think interoperability will grow over
59:25time yeah yeah and we see parallels of
59:28that on the web it's not binary where
59:30everything is interoperable or not but
59:32there are layers of interoperability
59:34that do exist so excited to see how that
59:36progresses but a lot of the trends we've
59:38spoken to so far are maybe macro trends
59:41trends that at least will play out over
59:42the next several years if not decades
59:45What specifically are you excited to see
59:47in this space in 2023 with the user
59:51generated content environment I'm very
59:53excited to see more digital focused or
59:56digital only creators and businesses
59:59forming around digital fashion we have
01:00:01examples like artifact which was sold to
01:00:03Nike we have a company that's focused on
01:00:05building fashion I'll dress X you know
01:00:07there are designers that are creating
01:00:09amazing content I want to see this this
01:00:11grow even more I'm also interested in
01:00:16content and Trends will translate more
01:00:19and more back into the physical world as
01:00:21well I think that that's going to
01:00:22continue to happen like I mentioned kids
01:00:24move seamlessly Between Worlds they're
01:00:26both the real world to them right and so
01:00:28how will these influence each other I
01:00:30think it's gonna be really interesting
01:00:31to track that and then finally just
01:00:33continued growth of Avatar
01:00:35interoperability of cross-platforms next
01:00:37up we have Sarah hi everyone I'm Sarah
01:00:39Wang and I'm a general partner on the
01:00:41a16z growth fund focused on Enterprise
01:00:44software and this is my big idea for
01:00:49generative AI advances Beyond text to
01:00:51image to complex workflows
01:00:53we're starting to see the next momentous
01:00:55platform shift in technology AI is
01:00:59in 2022 a lot of this took the form of
01:01:01AI generated images of dogs flying in
01:01:04outer space or AI avatars reflecting the
01:01:06best version of our selves selfies but
01:01:09in 2023 and Beyond the Enterprise
01:01:11productivity implications will really
01:01:15AI will drive 10 to 100 times
01:01:17performance improvements showing
01:01:19companies that there is a new way to
01:01:20work advancing from text to image to
01:01:23more complex workflows such as
01:01:24text-to-sql queries or eventually text
01:01:27to excel modeling and more
01:01:29we're tremendously excited for the next
01:01:31generation of AI native infrastructure
01:01:32and application companies to emerge
01:01:34along with the enduring existing
01:01:36companies that incorporate AI into their
01:01:38products as AI becomes increasingly
01:01:40democratized and underlying models
01:01:42potentially commoditized applications
01:01:44will need to differentiate on the basis
01:01:46of winning Mission critical workloads as
01:01:48they did in the last great platform
01:01:50shift and moving from on-prem to Cloud
01:01:52sir I'm so glad your prediction touched
01:01:55on this idea of practical applications
01:01:57of some of the generative AI tools that
01:01:59we've seen arise in the last year I feel
01:02:01like 2020 was so exciting we got to you
01:02:04know create Avatar versions of ourselves
01:02:06write up these like half-baked
01:02:07screenplays it was really fun but I'm
01:02:10excited to see The Next Step there in
01:02:12terms of how this is applied maybe on a
01:02:15more fundamental level in applications
01:02:16that we use every day and I think one of
01:02:19the reasons that we saw some of these
01:02:21more creative spheres take on AI first
01:02:23is because there's not really a correct
01:02:25answer right if you get an avatar sure I
01:02:27want it to look like me but there's no
01:02:29perfect Avatar that I'm looking for in
01:02:31fact I love the fact that I don't know
01:02:33what I'm gonna get back but in some
01:02:34other cases there is perhaps a more
01:02:38correct solution to a problem and
01:02:40there's a little less what you might
01:02:42call fault tolerance in the system and
01:02:44so as we think about maybe workflow
01:02:46applications or Enterprise applications
01:02:48are there any that come to mind with
01:02:50similar potential maybe a little more
01:02:51tolerance in terms of getting wrong
01:02:53answers or they're not being Maybe
01:02:55a specifically correct solution
01:02:59yeah first of all I love how you put
01:03:01that fault tolerant
01:03:03um you know I think that's notably
01:03:04different from other areas where the bar
01:03:06is incredibly High autonomous vehicles
01:03:09would be a good example of that the
01:03:12you know the highest right people die if
01:03:15the product isn't right
01:03:16um and and as a result the process of
01:03:19getting there may take a bit longer on
01:03:21the other hand I believe that the
01:03:22potential workflow applications of AI
01:03:24are enormous and Incredibly near term
01:03:26and a lot of that is due to this again
01:03:29more tolerance for for fault and right I
01:03:32think a lot of these workflows are
01:03:33actually oriented around AI assist so in
01:03:36other words where AI is making the human
01:03:38more productive but humans still stay in
01:03:40the loop and as a result they can be
01:03:43deployed immediately
01:03:45um I think coding has been pretty
01:03:46Salient recently I've even heard
01:03:48developers say that they can't work
01:03:50without it you know that's not to say
01:03:51that AI is you know developers are think
01:03:54hey you know create this app Soup To
01:03:56Nuts right that that may be in the
01:03:58future in the very near future but um
01:04:00today what developers are doing are um
01:04:03using it to code complete right and and
01:04:05figuring out how they can make their
01:04:06process especially the the repetitive
01:04:08parts of their job just way more
01:04:10efficient and faster and um you know we
01:04:13always talk about finding new solutions
01:04:14that are 10x plus better and that's
01:04:17that's arrived in that Arena there's so
01:04:19many use cases we could touch on um but
01:04:21ones that we've seen lately that are
01:04:22that have been mind-blowingly good video
01:04:24editing with companies like Runway
01:04:26really come to mind you know this has
01:04:27the potential to replace software that's
01:04:29been around for for 30 years
01:04:32um Pitch Pitch editing and Pitch
01:04:34creation the list really goes on but I
01:04:37think the key theme here to your point
01:04:39on immediate value uh really centers
01:04:41around how AI is assisting the human and
01:04:44making them more productive and
01:04:45efficient and then we'll slowly get to
01:04:47the idea that hey we can actually make a
01:04:50perfect pitch deck that reflects exactly
01:04:51what we want we can code it up that you
01:04:53know is exactly what we want and we can
01:04:56edit a video Soup To Nuts you know
01:04:58without humans in the loop but until
01:04:59then I think the the intermediate step
01:05:01has already been pretty groundbreaking
01:05:03yeah I also like the idea of and AI
01:05:07pushing us to be more creative so one
01:05:09concept that Mark spoke about in an
01:05:11interview he did with Vijay recently is
01:05:14Mad Men was a show that was known for
01:05:16always having these really really
01:05:17creative plots and one of the reasons
01:05:20behind that is they would sit in a room
01:05:21and they would talk about what are all
01:05:23the different plot lines that people are
01:05:25going to expect and then let's avoid all
01:05:27of those let's make sure we don't give
01:05:29people what they think is naturally
01:05:31coming next and I think you know they
01:05:33kind of pontificated this idea of you
01:05:35could do that same thing with AI you
01:05:37could ask the AI to come up with you
01:05:39know 10 different plot lines and then
01:05:41say the same thing and so it's again
01:05:43it's pushing you to think more
01:05:44creatively and you know in this case
01:05:46again have the human in the loop but
01:05:48another interesting part of AI um that I
01:05:50think is is maybe underrated is its
01:05:54ability to transition from not just text
01:05:55to text but we've talked about text to
01:05:57image you talked about text to code and
01:06:00I think something you talked about in
01:06:01your big idea was us getting to those
01:06:03more complex models like text to sequel
01:06:07as we do that how do you see maybe the
01:06:10foundational layers changing if at all
01:06:12because right now we are relying on a
01:06:14few models like gpt3 which influence
01:06:17copilot or some of these other layers do
01:06:20you see that changing where we then have
01:06:22many many more kind of
01:06:25fine-tuned models to the use cases that
01:06:28people need or do you see it all kind of
01:06:30layering back to the few foundational
01:06:32models that we have today like gbt
01:06:36yeah this is a really fascinating
01:06:37question and one that we've been
01:06:39discussing internally a lot especially
01:06:41as to your point a lot of the focus more
01:06:43recently is on these horizontal large
01:06:45language models and the sheer
01:06:47capabilities of these models has has
01:06:49really been astounding
01:06:50but to your other point on fine-tuning
01:06:53or you know the process of using
01:06:54additional data to further train a
01:06:56pre-trained language model has a really
01:06:59big impact on how accurate and useful
01:07:01these models are especially if you think
01:07:03about the B2B use cases where a lot of
01:07:06the truly useful data is actually a
01:07:08proprietary Enterprise data and not
01:07:10available widely on the internet to
01:07:12train on so if you put this together
01:07:14what does it actually look like in the
01:07:17um honestly the the real answer is
01:07:19likely that it's use case dependent
01:07:21um you know I think one of the exciting
01:07:22aspects of blms is that they've really
01:07:25democratized access to large and
01:07:27expensive models via their API you don't
01:07:29need tens or even hundreds of millions
01:07:31of dollars to deploy them in your
01:07:33product and if companies are then
01:07:35further able to fine-tune these models
01:07:37to their particular industry there's an
01:07:39argument that specific use cases don't
01:07:41need expensive models and we're seeing
01:07:44some companies start to make the
01:07:45decision to own the model which affords
01:07:47them more control but also maybe more
01:07:50cost effective in the long run and so
01:07:51it's it's still an open question in my
01:07:54mind and again it will be likely use
01:07:56case dependent given the capital demands
01:07:59of developing large language models uh
01:08:02there is a likelihood that it points to
01:08:05a potential you know more oligopolistic
01:08:07structure the way that you have with the
01:08:10big clouds and not everyone will go down
01:08:13the route of attempting to build a model
01:08:16of that size but the real question is
01:08:17which use cases need what and how much
01:08:20will it cost and how do you trade that
01:08:22off between control having in-house
01:08:24talent and not relying on an independent
01:08:26vendor versus just being able to move
01:08:29in addition to that I also think about
01:08:31competition right if we think about gbt3
01:08:34as an example there have been several
01:08:36products built off of that some that
01:08:38come to mind Jasper Lex these writing
01:08:40tools that again are all built on the
01:08:43same foundation and the question has
01:08:44come up repeatedly and from my
01:08:47experience of how are they going to
01:08:49differentiate and of course there's
01:08:50differentiation through distribution
01:08:52through product through ux but do you
01:08:55have any thoughts there on If there
01:08:57really are only a few models that so
01:08:59many products are built off of how do
01:09:02you really differentiate in that
01:09:03environment you know it goes back to I
01:09:06touched on this a little bit in the big
01:09:07idea but it really goes back to owning
01:09:09more of the workflows and
01:09:12um you know if we work in the idea of
01:09:14fine-tuning your model right how do you
01:09:16collect the data that enables you to do
01:09:18so and makes your product better I think
01:09:20it goes back to product Improvement
01:09:21Cycles if you will this isn't a perfect
01:09:23parallel but if you look at the last
01:09:25great platform shift which is the shift
01:09:27to cloud and who the Big Apple
01:09:29application winners from that shift
01:09:31where you could look at a Salesforce or
01:09:33a workday a service now and the
01:09:36competitors that they faced right and
01:09:37all of their competitor direct
01:09:38competitors not sort of last gen had
01:09:41also moved to Cloud right and so the the
01:09:43benefit of this platform shift that they
01:09:45were touting was not at all unique to
01:09:48um but came down the ability for them to
01:09:51build out workflows for their end
01:09:52customer that were useful and sticky and
01:09:56has really enabled them to dominate
01:09:57their continued uh mind share in front
01:10:00of customers in the decade plus since
01:10:01then and so I think in this next gen of
01:10:04applications built on open Ai and open
01:10:06ai's competitors the way to really
01:10:08differentiate yourself is to further
01:10:10build and develop workflows that are not
01:10:12just sort of a one-click button of add
01:10:14AI to XYZ uh which you know is really
01:10:18neat over time it has that novelty wow
01:10:20factor but is frankly not sticky and not
01:10:23something that captures the attention
01:10:25and differentiation in front of your
01:10:27customers long term but to really say
01:10:29hey uh you know we're actually going to
01:10:32develop the entire video editing
01:10:33workflow or whatever workflow you've
01:10:35decided to tackle and to that point
01:10:37Steph I actually think there's an
01:10:38interesting opportunity for companies
01:10:41that already exist and own these
01:10:43workflows to incorporate AI into their
01:10:46current products a lot of our own
01:10:47portfolio companies are doing this and
01:10:49in fact I would say this is so important
01:10:51to venture that if your company doesn't
01:10:53have an AI strategy it should really be
01:10:54thinking about one yesterday because the
01:10:58fact that owning the workflow and maybe
01:11:00the data associated with it the user is
01:11:03already living your product all day long
01:11:04to further enhance that with AI features
01:11:06just makes a very powerful Loop that you
01:11:09know I think contributes to the point
01:11:10that you mentioned on hey how do you
01:11:11really differentiate if you're built on
01:11:12the same infrastructure absolutely I've
01:11:15been thinking about that a lot this idea
01:11:16where again 2022 I feel like was Maybe
01:11:19This sounds too negative but a little
01:11:21bit of a gimmicky period as it related
01:11:23to AI where it's like look what we can
01:11:25do and it's in your face and when a
01:11:27company is using it they need everyone
01:11:28to know but upon reflection there have
01:11:31been so many companies that have
01:11:32Incorporated AI or machine learning into
01:11:35their products for years and it hasn't
01:11:38been front and center and one company
01:11:39that comes to mind there is the script
01:11:40which we use for editing and they're
01:11:43over feature is using machine learning
01:11:46to train on your voice and that has
01:11:49existed for quite some time it's been
01:11:51called overdub it's not like AI assisted
01:11:53voice Learning System right and so it is
01:11:56it is going to be interesting to see how
01:11:58companies differentiate in the
01:11:59technology they have but also maybe the
01:12:01marketing The Branding the positioning
01:12:03of it whether they choose to put it out
01:12:05in front or whether it's just as you
01:12:07said like existing companies almost like
01:12:10improving solidifying their existing
01:12:14features but to that end it feels like
01:12:17almost like every white collared
01:12:20is going to be reshaped in some way by
01:12:22AI to differing degrees but I'm curious
01:12:26in 2023 specifically because that's what
01:12:29we're talking about with these big ideas
01:12:30are there any kind of short-term
01:12:32near-term companies products industries
01:12:36that you think opportunity lies the most
01:12:39yeah I think there's a lot of talk or
01:12:41attention on some of the things that we
01:12:43mentioned right copy editing coding a
01:12:46lot of the consumer use cases so maybe
01:12:48I'll throw out one that's a little bit
01:12:50less than news but that I'm really
01:12:51excited for and I think it will be a
01:12:532023 uh change um and that's just the
01:12:56potential for AI and the data workflow
01:12:58so I'm no data scientist but as an Excel
01:13:01junkie you know I know there's a lot of
01:13:02repetitive workflow involved in getting
01:13:04to the Insight or the answer that AI can
01:13:07help you Wade through and from the demos
01:13:10that we've seen that have frankly I
01:13:14think the 10x potential that you see in
01:13:16a lot of these other Arenas can be
01:13:18applied to data workflows and that's
01:13:20incredibly exciting because data-driven
01:13:22decision making has become the norm it's
01:13:24uh the backbone of our data
01:13:25infrastructure investing thesis but for
01:13:28AI to tackle that as an arena next I
01:13:31think is incredibly exciting in a very
01:13:33very large End Market and again you know
01:13:35I think a lot of the companies that own
01:13:36the workflows already will come out with
01:13:38some of these breakthrough screws and AI
01:13:40features they really do change the the
01:13:42use case from novelty to something
01:13:45that's actually useful that will save
01:13:47you hours of your day that will take you
01:13:49to an outcome that is greater than you
01:13:51would have gotten to had you not used
01:13:53the AI and I think that's what that's
01:13:55what I'm most excited for next up we
01:13:57have some meat hey everyone uh this is
01:13:59Sumit Singh I'm on the a16z fintech team
01:14:02and this is my big idea for 2023.
01:14:05in 2023 fintech companies will need to
01:14:08strike a delicate balance between
01:14:09pushing the envelope by building with
01:14:11new technology rails such as large
01:14:12language models or llms but also
01:14:15maintaining customer trust while
01:14:17potential use cases with infintech are
01:14:19still emerging llms like gpt3 and the
01:14:22upcoming gpt4 may help businesses train
01:14:24data sets much more quickly and cheaply
01:14:27in addition they may finally be able to
01:14:29automate data heavy and manual tasks
01:14:31such as insurance claims processing or
01:14:33loan origination that have really only
01:14:35been semi-automated in the past
01:14:37but while llms can address some
01:14:39low-hanging fruit more complex use cases
01:14:42will require reserves of user trust when
01:14:44dealing with say fully automated
01:14:46investment decisions or automatic
01:14:48automatic financial reporting for
01:14:49businesses with complex money flows
01:14:52companies will need to balance these new
01:14:54services and experiences with potential
01:14:56skepticism from consumers all right
01:14:58Sumit so something I want to call out
01:15:00here is it seems like robustness to
01:15:01error really matters here so to give an
01:15:04example AI today being applied to
01:15:06potential drug candidates sounds like
01:15:09maybe a better option than actually
01:15:11prescribing a drug to a patient
01:15:13definitively and I say for now today the
01:15:16the models that we have because this may
01:15:18change but what applications stand out
01:15:20to you again today as near-term
01:15:22opportunities that maybe have some of
01:15:24this fault tolerance I think what we're
01:15:26going to see is you know the initial
01:15:29tools that have this sort of level of
01:15:31fault tolerance being around customer
01:15:33engagement tools for example in fintech
01:15:34where both Banks and fintechs you know
01:15:37can use these sort of models to
01:15:40basically automate what might be a CX
01:15:43agent in the past but today you see a
01:15:45sort of virtual assistance when you log
01:15:48and I think there's a level of fault
01:15:49tolerance in there in that you know
01:15:51while it's obviously not great if the
01:15:53customer doesn't get what they want
01:15:54right away but there still is going to
01:15:55be CX agents in the back right I don't
01:15:57think these people are going away
01:15:58anytime soon and so you know they might
01:16:00be able to be plugged in should there be
01:16:01an issue versus something that you know
01:16:05is just more sort of higher value right
01:16:08like a automatic underwriting or
01:16:10automated accounting where things need
01:16:12to be 100 correct so for automated
01:16:14accounting for example if things are
01:16:15incorrect there could be huge potential
01:16:17implications uh like IRS fines I like
01:16:20the example of customer service because
01:16:22something I wanted to ask you about is
01:16:24the transparency that you think is
01:16:26required here or how Founders should
01:16:28maybe be thinking about that and just to
01:16:30give a couple examples here there are
01:16:32several applications today that use
01:16:36machine learning within their feature
01:16:38set and some of them choose to be very
01:16:40very transparent about it and some of
01:16:41them don't an example that I can think
01:16:44of is descript has had this awesome
01:16:45feature within their application for
01:16:47years I think and they call it overdub
01:16:50and they're basically using machine
01:16:51learning to train your voice so that you
01:16:53can replicate it as you need that's
01:16:56obviously using machine learning but
01:16:57they're not branding it as such but I
01:17:00could also see how different
01:17:01circumstances would be the opposite
01:17:02where you'd want to vouch and say hey
01:17:04look look at this awesome new technology
01:17:06that we're using and maybe a silly
01:17:09example is just when you use chat GPT
01:17:10you know I'm using an llm you know that
01:17:13front and center and so how do you think
01:17:15about that how transparent should
01:17:18Founders be when they're thinking about
01:17:19implementing this technology
01:17:21I think it's you know a bit of
01:17:23pontification at this point because it's
01:17:25so early but I think in some cases there
01:17:27will be interfaces for Consumer you know
01:17:29consumers to for example do their taxes
01:17:32via chat TPT like interface or for an
01:17:35earnings call to be automated for where
01:17:37an equity research analysts can
01:17:38basically use a chat GPT like interface
01:17:40to you know search based on specifically
01:17:42tuned Enterprise data for that company
01:17:44but I think on sort of like more sort of
01:17:48consumer fintech products or other
01:17:49products that we haven't even thought of
01:17:51these might just be sort of back-end
01:17:52products or back-end platforms where the
01:17:55consumer is never really seeing the
01:17:57interface if you will they're just
01:17:59beneficiaries of that technology shift
01:18:01right and if they're beneficiaries I'm
01:18:03also curious to know if it's in the back
01:18:05end and maybe they're aware maybe
01:18:07they're not do you think that there will
01:18:11for example to disclose that something
01:18:13is being run through an llm or how do
01:18:16you see regulation maybe coming into
01:18:18play in this Arena
01:18:20yeah no I think it's a bit too early to
01:18:22tell today but I'm sure there will be
01:18:25you know let's say in a world where loan
01:18:27decisioning or tenant things like tenant
01:18:29screening are being fully automated by
01:18:31llms and say there are consumer issues
01:18:34with that or you know it turns out that
01:18:36the models that were being trained upon
01:18:39were using sort of discriminatory data
01:18:42there could be huge implications from
01:18:43you know folks like the cfpb and other
01:18:45Regulators who crack down on these types
01:18:47of companies but again I think it's just
01:18:49something the founders are going to have
01:18:50to keep in mind given it's no one is
01:18:53doing that just yet right so it's it's
01:18:54kind of warning if you are going to use
01:18:57these things to be aware of the
01:18:59potential regulatory implications
01:19:04lesso and fintech
01:19:06knowing that something is being run
01:19:08through a computer instead of a human
01:19:10just changes your relationship with it I
01:19:12think I saw a tweet recently where
01:19:14someone was using llms for therapy and
01:19:17it kind of as you'd imagine got a bunch
01:19:19of blowback because just the sentiment
01:19:21of a human not being there in the
01:19:24process people were not a fan of even
01:19:26though in that example I think actually
01:19:27humans were part of that process but
01:19:30what else do you think Founders should
01:19:31be thinking about here we've kind of
01:19:33touched on regulation we've talked
01:19:34touched on the transparency that might
01:19:36be required is there anything else that
01:19:38you think Founders should be really
01:19:39paying attention to as they're thinking
01:19:40about integrating this technology
01:19:42yeah for sure I think you know on the
01:19:44product side I think these product leaps
01:19:46will likely come from to an existing
01:19:48large models with four specific sort of
01:19:50commercial use cases with specific data
01:19:53however I think the one thing that
01:19:55Founders can't forget is that no matter
01:19:57the sort of platform shift whether it's
01:19:58been internet Cloud mobile distribution
01:20:02is always key and so even with this new
01:20:05sort of product leap where maybe you
01:20:06know you have consumers flocking to the
01:20:08product because you are advertising say
01:20:10hey this automated tax product is
01:20:13powered by llms and you're gaining a
01:20:15bunch of customers through that
01:20:16in the event that this sort of product
01:20:18become or technology becomes table
01:20:20Stakes well then we're sort of back at
01:20:22square one which is distribution is
01:20:23still key we often say that it's almost
01:20:26more important than product but
01:20:28hopefully that's not the case given sort
01:20:30of a lot of the last generation of
01:20:31financial services fintech products that
01:20:33were more financial engineering than
01:20:35software engineering and so it'll be
01:20:37interesting to see if these new product
01:20:38leaps can account for distribution modes
01:20:41that folks have had in the past yeah so
01:20:43what I'm hearing is one it'll be
01:20:45important for people to train their own
01:20:46data sets to a degree or or kind of
01:20:48fine-tune these models because if they
01:20:50don't they're using essentially the same
01:20:52models that everyone else has but then
01:20:54even if that's the case as things
01:20:55Advance there will be kind of this
01:20:57Baseline layer that most applications
01:20:59are using and some and so from there as
01:21:02you're saying distribution really
01:21:03matters how can you reach the customers
01:21:05that matter the most right and that's a
01:21:07bit of pattern matching from previous
01:21:09sort of Technology shifts right and so
01:21:11you know there's a chance this time is
01:21:12different however you know people more
01:21:14wise than me have always said it's not
01:21:16it's never ever different just you know
01:21:18just maybe just a different flavor but
01:21:20yeah I mean we always think it's
01:21:21different but no I
01:21:23know it's something that I've been
01:21:26initially if you're using some
01:21:28mine llms like if everyone is using gpd4
01:21:31when it comes out then it's all being
01:21:33built on the same foundation so how do
01:21:35you stand out how do you build a real
01:21:36moat around that it'll be interesting to
01:21:38see you mentioned insurance claims
01:21:40processing and loan origination as two
01:21:43potential Avenues of this being applied
01:21:45are there other developments that you
01:21:46see specifically coming in 2023 or then
01:21:51I think in the near term it probably
01:21:53even won't be you know loan origination
01:21:56for example I think where we'll see the
01:21:57low hanging fruit is likely around kind
01:21:59of back office automation of like
01:22:02workflows if you will and so you know
01:22:04we've been very excited on the fintech
01:22:06team around vertical SAS businesses that
01:22:08help for example smps across Industries
01:22:11manage their back office finances and
01:22:14right now this software or the software
01:22:16products that we've seen is really
01:22:17augmenting people and augmenting for the
01:22:20fact that there are basically less back
01:22:22office people to do this work and so
01:22:23software can help make each employee
01:22:25even more powerful I think what we might
01:22:28start seeing is this ability to fully
01:22:31replace that person and the reality is
01:22:33in a lot of cases you're not even
01:22:35replacing anyone it's just these people
01:22:36have left the workforce this sort of
01:22:38like typical back office employee at an
01:22:40SMB and so you know perhaps we'll see
01:22:42models being used by companies of All
01:22:46Sorts uh to automate workflows fully
01:22:49and I think something that we've you
01:22:50know said in the past on our team is
01:22:52it's better to automate 100 of five
01:22:55percent rather than trying to automate a
01:22:57little bit of 100 of a workflow
01:22:59and I think the big question is now like
01:23:01okay we have we may have the potential
01:23:03to automate everything and what are the
01:23:05implications of that I think that's
01:23:06going to be really exciting to see yeah
01:23:08I think that'll be really interesting to
01:23:09watch I mean I think the canonical
01:23:10example is the ATM and how that replaced
01:23:13one aspect of what bank tellers did
01:23:15in the Arc of History it seems like
01:23:18people have always managed to reapply
01:23:19their skills to a new exciting perhaps
01:23:21more interesting more fulfilling area so
01:23:24yeah that'll be fun to watch right
01:23:25that's the hope all right that's all for
01:23:27part one but look out for part two of
01:23:29our big ideas for 2023 where we cover
01:23:31fintech American dynamism and my own
01:23:33health and if you're looking for the
01:23:35full list of 40 plus ideas you can head
01:23:37over to a16c.com and you'll find it on
01:23:39the homepage see you soon
01:23:42thanks for listening to the a16z podcast
01:23:45if you like this episode don't forget to
01:23:47subscribe here on YouTube to get our
01:23:48exclusive video content we'll see you
![How To Succeed As A NEW & YOUNG Realtor [Deals Every Month + Luxury Listings]](/_next/image?url=https%3A%2F%2Fi.ytimg.com%2Fvi%2FrtrRghgTayc%2Fhqdefault.jpg&w=3840&q=75)
