Select text to annotate, Click play in YouTube to begin
00:00:00
foreign [Music] ERS linkers and mappers we are really happy to host a series of conversations around the topic of tools for thinking our longer term goal is to spark a
00:00:20
diverse connected shared memory that will help us make important decisions together our near-term goal with these podcasts is to blow more oxygen on the growing tools for thinking sector addressing key issues and talking with the people who
00:00:32
are doing the work this podcast is created by Beta Works a New York city-based startup Studio I'm Jerry mikulski your interlocutor and obsessive mind mapper and our topic today is funding tools for thinking our
00:00:46
guests are Andrew Sutherland founder of Quizlet James Cham partner at Bloomberg beta and John Borthwick founder of beta works and we've got a lovely set of issues in front of us we've been we've
00:00:59
been exploring great things on the different podcast episodes including some of what was coming up in our prep conversation here about hey our tools for thinking possible without changing the OS and the hardware and we'll sort of get there but I think a common
00:01:12
interest here for us is what is a path to market for tools for thinking and these tools for thinking are sometimes quirky I don't know that any of them looks like the next Facebook or the next
00:01:24
LinkedIn or some major platform or a big unicorn and yet the need for these things appears to me to be pretty big like we sort of need to figure out how to think together better we're kind of stuck in some early phase
00:01:38
of of what the Visions for a collaborative thought and Collective sense making and all that might be um so let me just go uh James Andrew John for the same sort of question what
00:01:51
what are some either role models for making it to Market or possible paths that you see for this sector what have you imagined as as possible or what stories do you play for yourself that are sort of cautionary tales for how
00:02:04
this Market ever emerges if you just look at what's happened so far I think the the most interesting example has been the adoption of Rome uh I'm going to take the taking tool and uh it's sort
00:02:17
of flies the story sort of flies in the face of a lot of uh ideas about early adoption and and uh like it started as this tool that was
00:02:30
incredibly hard to use uh it had no onboarding experience no uh like explanations of how to use it um but it's really powerful and all of
00:02:44
its adoption basically came through evangelists and Word of Mouth who uh tweeted about a lot or or just like sat their friends down and gave them 30 minute demos which I've done to many of
00:02:57
my friends um and so you know they're not they're not running ads they're not like doing anything like that and it's very hard to use and even even with the demos like it
00:03:08
takes weeks to figure out what how to make it work for you and yet you know it achieved some pretty large scale and some uh a lot of
00:03:21
user laws and uh and then a lot of people who like didn't get it and um you know maybe Snapchat was was similar in the early days like the interface was like bewildering but like people
00:03:34
uh were so compelled by the like core value prop that they were willing to go through it and so I think that's um at least like one important example
00:03:45
because to shift people's uh usage of considered fundamentally like new action like it might not need to be easy to use which is interesting yeah because
00:03:58
there's so much emphasis on ease of use everywhere and so many things uh get stuck trying to hurdle that bump in a way John did you want to jump in yeah I think that the um
00:04:10
I I agree I agree with the way that Andrew framed that um and I think the the analogy of Snapchat's really interesting because Snapchat you know I think used a uh one
00:04:24
of the first apps to use a truly native mobile swipe based experience and I think the Rome did something similar with a sort of graph with the
00:04:36
graph database where it sort of exposed that and you sort of fell into it and felt like okay here's a entire new world that I can start to construct something I think that the generally in these
00:04:48
tools for thinking uh We've sort of been trapped by both the success and also the limitations of the sort of of the classic Word document and uh and that
00:05:00
has that has meant that for a majority of people they'll do their thinking and either a cat based Word document or some simple note-taking app and they um try
00:05:12
these new things and then they compare and contrast them to the Simplicity of those uh those forms and getting people into a sort of radically
00:05:24
new workflow and a new set of user behaviors is really challenging right it's just like it's a it's a big ask uh and you have to be able to you have to
00:05:36
be able to both uh get people there and also show the payoffs or the benefits of it uh and so you know I'm heavy roam user and I have
00:05:50
um I still feel like after three years or so that uh the benefits uh you know sort of the the the investment versus the benefits
00:06:01
is still not in the right ratio uh you know like I wish there was like better search within there because I'm constantly trying to discover stuff right I grow them up right now um and I made some notes in advance of
00:06:14
this podcast and then I you know I made the notes like two weeks ago two weeks ago I had to find them and just finding that this is tricky uh James you know as you talk about this I think of two sort of like I guess two concepts
00:06:28
one is this idea of legibility and the other is this idea of status I think one of the things that's interesting about these new tools for thought is how important the context is and how
00:06:41
important like the initial set of initial set of folks is not just the idea that Rome promises these things but it's also the fact that like you know sort of John Borthwick told me that Rome promises these things in in in in some
00:06:52
way like sort of you know for uh I think that that ability of Rome before Rome started made a huge difference and I think that that's another part of the angle um
00:07:04
um there's almost a way in which the discussion around there's no accident that some of these things start in these tight communities of committed Believers right and I think that that's that's the
00:07:17
other that's the other part of I think what's interesting and then and then of course you know sort of I I think about the promise of these tools for thought and I do Wonder in some ways there's a delicious
00:07:29
vagueness in what the end product is or the end promises and in that way you know there's a little bit of like sort of like the right analogies might be early religions rather than you know
00:07:43
sort of how to copy the next SAS application because the changes that folks are proposing or hoping for are so much bigger and so much more exciting yeah so interesting the two of you are
00:07:55
both you know investors uh who who want to see their their companies go to a large consumer scale like the early story of Rome is sort of one thing but
00:08:07
if you sort of Bridge to everyone using something like this like what do you what do you think needs to happen you know so I think that is the second part which is that legibility part where it has to look like something that
00:08:21
someone that people already understand or the use case you know there's that next phase after which folks understand what are the boundaries of a new product or let's say in the case of Rome as folks understand the boundaries of it and what it's good for and then there's
00:08:34
almost a willing which that that sort of set of values and bundles or features that become legible to the next slightly bigger audience and I think like that sort of probably the right way to think about it in the same way that you know
00:08:47
early days of the browser you know sort of they make you know browsers or make no sense to ignore me but then one day they're able to see their ESPN sports scores and they're very happy this has been a consistent theme through
00:08:59
our podcast calls and I'm referring to it sort of as the the tractor beam of the desktop UI and the office suite that new and interesting pioneering products kind of get dragged back to look more familiar to people so that there'll be
00:09:12
some uptake because a lot of these exotic tools have a real hard problem getting people trained up so we can sort of include that by reference and some of the other calls but I'll Point here also to like uh inspired by which you just
00:09:23
said James Facebook start which notoriously was you had to have a harvard.edu email address and then you had to have an Ivy League address and then you had to have a college address and then all human beings and their dogs
00:09:36
could sort of join Facebook but but the path taken was one of hey I'd like to be in that group too sort of early on and then a story that came up in episode three was about hypercard which I think
00:09:49
a lot of us are fans of and the story was that bill Atkinson invents hypercard and says here like ship this with the mac and Jean Louis gasei is so worried that people will pick up the mac and
00:10:01
just code their own stacks and not need software that he tries to block hypercard shipping with the Mac because he's trying to stand up a software development Community that's actually going to write a lot of software to make
00:10:13
the Mac cooler and better and more interesting than the competing machines and so hypercard is almost doesn't make it out Atkinson sort of has to Blackmail Apple to say hey you don't ship this with the Mac I'm going to make it
00:10:25
available openly somehow and this is before the internet before open source software is so easily moved around but but there's there's this hard thing and then the third thing I want to put in the mix is
00:10:36
and yet now and then new interfaces are born out of whole cloth like Siri Alexa and Google Google Assistant which are now so commonplace as to be ridiculous right I mean you know anybody with any
00:10:49
of these phones can call out at any moment and there's a voice UI that's actually pretty good and and it's very different from it is not restricted uh by the desktop metaphor or any of those things I just mentioned and it's like
00:11:01
okay okay so and we're not using those voice assistants for Thinking Tools much that I know of right okay so now you've got one of my hobby horses excellent I think that like voice interfaces are
00:11:14
both the future and totally squandered right now it is a little shocking to me how trapped sort of the existing big players have been with ideas around how the architecture and the value of voice
00:11:27
applications should fit and I think that's partly because so many of the core teams that built terrific voice models came from a world where that was the actual value rather than the actual
00:11:39
application and so we're stuck in this world where voice really is meant to be thrown in passovers like little instructions that kind of work but offer an incredible amount of value and I
00:11:51
think that's exactly where startups end up being successful and that is certainly one of the places where I continue to search and dream [Music] yeah so I I I agree and I mean I think
00:12:04
James you're making a companion argument that we need or at least pushing towards that we need new interfaces or new uh potentially new hardware new devices to be able to like think you know to get
00:12:17
these tools for thinking really working I think that uh you know I I break it up I break thinking up into staple things but a few of them are is a lot of what
00:12:29
thinking is for people in this context is uh is a personal memory base or just like a place to like store things to be able to hold them
00:12:41
and so I think that that's one part of thinking and I think that that actually is voice and subject to Jerry your track to beam because that sort of is a it's a
00:12:54
it's a workflow and a structure that people are like okay I store my data and then I just want to be able to find it again I want to make sure it's discoverable right um a second kind of thinking is what
00:13:07
we're actually doing right now which is conversation and so we're having a conversation together and I think that that is a much harder place to actually track that thinking that maybe voice
00:13:20
interfaces can do something there right I mean those I I played around with a lot of these tools that do will do a voice transcription what I really want is a voice transcription for which surfacing the concepts that we're
00:13:31
talking about and helping me remember and both think socially in this context right so that's the second kind of thinking a certain kind of thinking this past weekend I was dipping back into Hannah
00:13:44
rent and some of her work and I you know she talks a lot about uh that's the oppressing no no it's not all depressing um but um she talks a lot about that
00:13:57
second voice in your head and being able to have a conversation with yourself and the importance of being able to do that and that to me is you know Central to thinking
00:14:09
and I and I do wonder whether large language models and some of the stuff that's starting to come out of the sort of generative space can help with that sort of conversation in our heads and
00:14:24
give us a place where we can have that conversation in our heads on a Computing surface and I'm not sure what surface that is if it's a voice circus or if it's a you know uh a messaging based UI
00:14:35
or what it is but I just wonder if there's something there that can sort of help with that so kind of thinking so those are at least three kinds of thinking that I think are worth um unpacking a bit and exploring that's
00:14:48
fascinating I have to admit this sits on a couple of things that I've been doing you know I've been going to a few conferences recently and mostly because I don't really I'm too socially awkward to come up with small talk the thing I've been doing is I've been trying to
00:15:01
get people to sign up for open Ai and use the playground and I'm sort of surprised and use gpt3 directly right and I've been sort of surprised how few people have actually done it and also
00:15:13
how delightful it is once you get people to go through all the little steps to like provide your email and their phone number and actually get them into the playground rather than some other tool um you know some intermediate some mediating tool or some applications when
00:15:26
you actually get them to use it and they start playing with it it is sort of stunning to me the sorts of conversations or the sort of things people are able to start doing and you know one of the things I've been doing John which is along your lines is I've
00:15:38
been having imaginary conversations or I will really right write a one-act play between Martin Luther and Richard Nixon about politics and they're like just do a little and then it'll generate like this short you
00:15:52
know sort of like five to 50 line play in which they have these conversations or I will do you know sort of like explain you know sort of like uh I think the best one was the conversation
00:16:04
between Marx and Santa Claus right and and it's just interesting because in those conversations you see how good GT3 is at simulating you know plays but also having ideas
00:16:16
converge right and that I think is the key part of tool for thought yeah you also having fun right and I think that the fact that you're doing that for fun and you're
00:16:30
enjoying it you know when we talk about like establishing new behaviors and breaking through the existing patterns of workflows it's like fun giving somebody something to do that's fun is
00:16:42
like a sure-fire way to like you know going back to where Andrew was before Snapchat was just fun right you could say as a swipe first UI and all this other stuff but it was just kind of like
00:16:53
it felt like it was just a fun space and so when you use it directly you end up in a very different place than the theoretical thing right you know a lot of my work has been around machine learning and artificial intelligence and
00:17:07
there's all this great theoretical thought around how dangerous it could be but when you actually play with the incredibly powerful model it's kind of like it's kind of like getting a hammer for the first time you realize what it's
00:17:20
good at yeah you also realize you shouldn't stub your toe right you know like you use the hammer and you bang your toe it really hurts right and so like that kind of feedback mechanism and understanding of what these things are good at and bad at I think is the sort
00:17:33
of intuition that's available to us right now Andrew you look like you want to jump in yeah I mean I think the um the thing on my mind with these language
00:17:44
models is with Quizlet we we spent a while thinking about you know AI tutoring and having having models that could uh you know pose a question to you and then sort of
00:17:57
help you work through it uh in a way that you know understood what you were thinking and stuff like that and I think there's there's a lot of value um that can be created in building
00:18:12
something like that the thing that I'm um the thing that gives me pause on both like voice um uis and stuff like that is um
00:18:24
when you have a great teacher that you have a relationship with that you care about that is part of what that's sort of a major motivation to uh to respond or to
00:18:37
try um if I was having this conversation with the three of you and you are all GPT three Bots um I might not be applying myself in the
00:18:49
same way as like the complex dynamics of the three of us and uh the fact that I respect all of you and you know might do a deal with you in the future and like all these other things um and so
00:19:02
uh yeah it just sort of I don't know what the answer is but it makes me wonder how fire Will Go On
00:19:12
on language model uis when they're not embodied so Andrew this might trivialize your question a little bit but you know the movie her uh your hockey and feel like
00:19:26
Phoenix falls in love with an AI which is fully embodied and very much sounds like a person and it's that seems very feasible so so could there be a quick sort of work around where it's like hey
00:19:38
you're talking to Socrates right now and you get to ask him anything you want it's not good enough yet but just wait it'll like once it seduces me then I'll be uh I'll be fine
00:19:50
or not I mean the movie doesn't turn out all that well yeah yeah um but but there's all sorts of interesting possibilities there and and this idea of having a relationship with a great teacher is really important here because
00:20:03
um all these language models can assume different kinds of roles in our lives one of them could just be the diligent Note Taker during our call as John was saying a little earlier like hey if any of us mentioned a book a site a a paper
00:20:15
an experiment it could show up and say do you mean this one and put the you know put the link in the chat and that would be perfectly fine use for for an AI that would make our lives a little bit easier but it's not the thing we're talking about here which is more
00:20:27
challenging more interesting yeah I mean I think it would be more more fun uh to have like your little AI children or like things that you have trained from
00:20:41
scratch uh and taught what you care about and like you're talking about the sort of second voice in your head uh like if you have a bunch of these little Bots and they're like
00:20:52
things that have come up not knowing anything but you've sort of trained them in your thinking and then you could have conversations with them like then your relationship to them is
00:21:05
is not so it's like you you feel a sense of Pride or ownership of when they think um and I think that could be a path forward and then this is also a dynamic though
00:21:19
of how these ideas when in the abstract sound either foreign or crazy like the idea of AI children in the abstract seem crazy but then the reality is that we feel a great deal of ownership over a
00:21:31
lot of things that we create right and I think there's almost a way in which the challenge for great opportunities or product people is to make those sort of abstract ideas Concrete in some way that
00:21:43
like you know sort of a normal person or even like an early adopter might be listed like get a glimpse of that and start to feel that feel that little bit of emotional tie or that that first step right and I think that that's part of
00:21:56
what's exciting about right now yeah I think that the as as the embodiment question is really interesting question um and and you can sort of see the relationship I mean I think that the
00:22:09
names that uh companies have chosen for these uh sort of uh AI children um or that I remember right at the beginning of hugging face right the first iteration of hanging Pace they
00:22:21
added as a pet right it was meant to be like an AI bet and Clem the founder was like his his belief was that everybody is going to have just like we have pets that we
00:22:33
acknowledge that we love that we care for that we acknowledge at some degree of intelligence that uh can help us as humans uh we're going to have ai pets in the same way
00:22:44
and so AI friends AI pets um I think that the names of these things like Syrian or Siri is such a name right it's just like because it's almost as a human embodiment but I've
00:22:57
never met anybody called Syria I've certainly met people called Alexa um and um so yeah I I think it is I and I'm I'm
00:23:09
not sure I mean Andrew I don't think that these large language models uh I don't think that they are on a path to becoming sentient or to becoming like full
00:23:21
thinking Partners I mean I think that the sort of limitations of the structural limitations of the models we're going to run into at some point um four or five I just do I I just think
00:23:36
that aren't they just like ever increasing like uh there's still that that's still not thinking they're talking back at you
00:23:47
and the ability to be able to discern patents and language is extraordinary uh but I think it's different from thinking and uh so
00:24:02
yeah okay I don't know what's the embodiment yeah I don't want to get too metaphysical on a Monday morning although you're already an East Coast but I would propose that actually we
00:24:15
already have superhuman intelligences that we sort of like live with and live around all day long and these sorts of things are like organizations or nation states or corporations and in some ways
00:24:28
like like I like and then look I'm traditionally religious right but in some ways for me it's not that hard to ascribe intelligence to you know like there's Elon Musk trying to run Twitter right
00:24:40
now and then there's this weird thing this amalgamation but people and rules and processes and software called Twitter that he can't even control it because in some ways it's actually alive sort of the hive mind or the collective
00:24:53
intelligence or whatever you want to call it in in lots of different ways I mean I think the social I I was I can't believe we got some musk and Twitter um so uh there's now many forms but but
00:25:06
I think that social software has a life in of itself and has a you know this and and takes on uh uh takes a shape of itself
00:25:17
um but I think that I I still view that as being separate and distinct from the thinking from a thinking being um but let's maybe pause this Rabbit
00:25:31
Hole um because uh before we get to parents um I so going back to where we were a few minutes ago uh I think that the the need
00:25:43
for new interfaces I think is critical here um you know the term embodiment a uh you know I was struck when James you were talking before when you got really passionate you suddenly your hands start
00:25:56
like moving around I do the same thing right and I just think that we need to be able to like have interfaces that can can actually where we can the embodiment of our thinking is reflected in those
00:26:09
interfaces and I don't know what exactly what that looks like yet is it a mixed reality interface is it a uh is it some sort of sensor kind of Data Tracking in
00:26:21
space but but I think that there's a whole interface issue here uh which is not the freaking keyboard right and a mouse it's like and voice may be part of it but I think it it's we need to get
00:26:33
our bodies involved in thinking in some cases I mean it's it's like are you step into data sets or whatever else and there's there's you know some work on augmented reality where you actually can walk through a data set and see patterns
00:26:45
and so forth and if the data has dimensionality it makes a lot of sense um I think in a lot of other cases we're watching Zuckerberg say hey everybody come into my metaverse and I and a bunch of other people I know
00:26:58
are just sort of referring our brow going I don't see anything interesting in your metaverse that was not already there in Second Life many years ago that we're not all addicted to and using heavily and there's like plenty of
00:27:10
plenty of interesting reasons why so so this space of voice interfaces 3D interfaces AR VR XR whatever else are and tools for thinking is just young
00:27:23
we need some we need some more plays and more experimentation I'm also interested back in the in the conversation about how what is what are the paths to Market or paths and path to Market it for me is maybe a limiting way
00:27:35
of framing the question it's paths to Broad use and applicability of these tools for thinking and a piece of that seems to me to be this this difference between the commons and proprietary
00:27:46
tools or data sets and how do we how do we motivate more tools to write toward each other and how do we motivate more people to share their notes because one way this plays out is that everybody as
00:27:58
individuals just has a better note assistant and a better personal assistant and they get smarter uh you know like Iron Man or whatever they become try to become little individual Tony Starks the other way this this
00:28:11
plays out is that we build a better collective intelligence and this this then is an emergent phenomenon between us that is hard to control like social media but very I think really really promising in terms of the complexity of
00:28:24
the problems we're all facing but James um I I I will admit I think partly in a lot of those terms and partly in terms of economic legibility I think there's a way in which people once people are able
00:28:42
to grasp onto something there's a there's there's a there's both like the is it useful to me but there's another part of it which is like do I know how much it costs do you know what it means can other people play with me on this
00:28:54
and I think that economic legibility is part of what we're seeing with large language models right now right for like it's not that the tools suddenly started working there have been great demos for two years at GPD three right but what
00:29:06
changes like for the first time a couple different players for different reasons you know have made it economically legible so now you know either I can download this model and suddenly use it myself and do all sorts of Consultants
00:29:19
of Mischief or I know how much it's going to cost and because of that that creates all these interesting opportunities um like I think in that way the right metaphor might be you know
00:29:31
sort of thinking about adoption of Unix versus adoption of Linux right and then you think about sort of all the things that are you're able to build when it's you know sort of owned by when there's a big fight between digital and ATT and
00:29:43
all those guys and you're not able to get anything done at a mass level and then when Linux comes around like because one crazy guy decides to go through all the trouble to make this operating system and make it free like it's like I mean it's one guy convincing
00:29:55
a bunch of people right but like like that one thing something unlocks all this opportunity you know and and all these different web services so that's uh that's that's partly the that's that's my own mental models I'm watching
00:30:08
all these very big players spend incredible amounts of money trying to figure out how to like build these and Linus torvall's project is not a for-profit startup it's just say hey everybody I want to make this Unix thing
00:30:20
run on my laptop and the plausible promise as clay Shirkey sort of phrased it so long ago I think really nicely was whoever works on this this is going to to be under GPL and any effort you put in here will be a benefit to everybody
00:30:33
and that turned out to be a great way to recruit a tremendous number of participants and we're seeing the effects of that and then and then critically though this was not just an idealistic thing a bunch
00:30:45
of people made a bunch of money holding a bunch of things very private into themselves right that sort of this was one layer that became commodified and became economically then thus became economically legible and then people did
00:30:57
a bunch of stuff on top of that precisely and I was lucky enough to interview Bob LeBlanc back in the day he was an IBM exec who told me the story of how they adopted stuff and I'm a friend of mine is John Patrick who was one of the Mavericks inside of IBM Who
00:31:10
convinced them to go look at Apache then Linux and then other sorts of things saving the company because I was a tech industry Trends Analyst at the time and IBM was about to go under son and Apollo were eating their lunch and IBM had no
00:31:23
answers and IBM's own internal Architects that spent 10 years trying to make their five platforms talk to each other with no success and Along Comes tcpip Apache Linux and all these things right so so that's a that's a really
00:31:35
nice model because shortly IBM was selling two billion dollars in service revenues on top of Open Source software and so now what are the analogs in this space with large learning models to get us there
00:31:48
so uh so before I was given a list so I gave a list of three different ways of thinking uh there's a fourth one which I I'll add to that list which is developing Frameworks right and we're
00:32:02
doing that right now right and so I think that once you start to develop these Frameworks around open source closed around the economic legibility around legibility of stacks of things you start to like you you help people
00:32:14
think and you help yourself think and you also sort of create these information Stacks which I think can both can both serve you and also can actually thinking
00:32:26
um the I I think that this opened uh the the amount of or the unlock that's been achieved with these large language models or and particularly the visual models with stability uh and being able
00:32:41
to you know access and download and find models and hunting days and being able to actually run them yourself I think is you know we've seen in the last like in the last two months you
00:32:54
know million and a half people messing around with stability and uh it's truly extraordinary and I think what it does is that it sort of it it makes there's something really fun about it there's something
00:33:07
really creative about it I think it's opening the new door to our sort of imagination and how we can think with these tools um and it feels to me very much like the
00:33:19
sort of the beginning of a new interface right just like the browser felt like something new and the phone kind of like somebody new this feels like something new and then I also think there's something so incredibly empowering about it because these models have been used
00:33:33
uh essentially uh on us for the last you know five ten years right probably the most sophisticated AI that we talk to every day is Tick Tock right and uh and
00:33:46
now suddenly we can talk to or we can have some control and some agency and talk to it and construct with it um uh so I think that that's incredibly
00:33:57
powerful and and also fun okay Andrew you spent a bunch of time with a bunch of very smart thinkers recently like you hear all this like what do you think yeah yeah so I just I
00:34:12
recently ran a conference uh in France in the countryside for six days and uh sort of got a bunch of people together who are thinking about this stuff and uh the sort of independent researchers
00:34:25
thinking about future of computer interfaces and um no there are a bunch of different interests but I think uh large language models obviously a big one and
00:34:38
uh another is just like spatial interfaces um they're embodied Computing tangible Computing um and yeah what struck me from that is like
00:34:51
how how much um for a long time we've we've been excited about spatial interfaces or being able to interact
00:35:03
with computers in 3D space which is how our minds work and yet like there isn't really um you know much happening there um I guess with the exception of sort of
00:35:17
Facebook and and meta XR um stuff like no one no one is really building um Tools For Thought stuff that allows you
00:35:31
to access your 3D faculties of your brain um and uh Dynamic land is maybe the best example we had a couple people who were used to
00:35:44
um work on Dynamic land there and you know that project is is kind of dormant and so um I think that that may be another space where like it's sort of pre-roam mode
00:35:58
where like if you there's a everyone sort of thinks that there's a bunch of important stuff that could happen um with like embodied Computing or physical computers
00:36:12
um but no one there's no killer demo yet there's no one nothing that like you can see and you like we need one of these in every office or every home um and but I think it'll get there
00:36:26
um you know when you talk about Dynamic land which of course was something that I think both of us were obsessed with um the interesting question to me is why
00:36:38
it seems so obvious as the right solution and why hasn't it worked yet is sort of one of those delicious questions and one answer might be that it's just not the adjacent possible yet there's a
00:36:49
way in which Rome only makes sense because we had years and years of outliners right I was an investor in a company called WorkFlowy um where I'm still an investment in a company called WorkFlowy and then sort of Realm build some Innovation on top of
00:37:03
that right and I think that there's that question around I thought that with the pandemic maybe people would buy lots of projectors and then and then have projectors in the room I actually bought a couple projectors to stick in my room
00:37:15
and I thought that that would lend itself to towards you know something like Dynamic land or the fact that we have way too many cameras like I have three phones on my desk alone right you know and and the fact that that was all
00:37:28
true might lend itself to that but it hasn't quite coalesced yet and I do wonder what's the step before Dynamic land and um and for people for people not familiar with Dynamic land that was the environment where there's basically
00:37:40
ceiling mounted cameras that can see where everybody is and then there's physical objects that you can use and place on tables and so forth that mean that are meaningful to the system and you compose uh things you want to do or
00:37:53
whatever in the environment is that right and Brett Victor and a bunch of others were kind of involved in this well it it has both cameras and projectors so right right you can sort of all be standing
00:38:04
around the table and it can see what's sort of on the table you can adjust the things on the table you can add new code uh to the table and it sort of instantly recognizes and then changes what's being
00:38:16
projected onto the table it's like dance floors with four annoy diagrams except more useful but yeah I mean one of the one of the things that I think is incredibly compelling about that is that it's the best demo of like a shared
00:38:30
Computing environment that I've seen where like um all of us the four of us right now are on Personal computers there's no way for uh me to meaningfully use your computer
00:38:42
at the same time that you're using your computer and like humans are very social creatures humans like think collaboratively humans like want to work in a shared environment but
00:38:53
like you know obviously like me having a mouse and you having a mouse on the same computer is like not the the answer and so like we've just done like not enough thinking and not enough building on like
00:39:07
uh multiple people be able to compute together Andrew what else came out of the uh the event you hosted in France what what
00:39:18
other sorts of things showed up in the conversation um I mean lots of uh really cool language model demos uh people building um
00:39:30
yeah just sort of like interfaces on top of language models like the the tricky thing uh or the sort of problem for a long time has been like well yeah you can
00:39:42
have a prompt which you type into and it responds or you can have some sort of conversational thing but uh a conversational UI is is
00:39:54
super limiting and doesn't expose all the power of a language model and so there are several people that were showing sort of yeah you user interfaces or like graphical things that had
00:40:08
underlying them uh or games that had underlined them language models but you could like play a lot more because uh because they were powered by by models
00:40:22
underneath it's very interesting because we tend to limit the UI to like oh this is a speech interface so we're just going to speak to this thing and then then you're in the realm of well how much can you
00:40:33
communicate only through speech and we're doing very little sort of combined arms Warfare on thinking together where where you have Sensor Fusion data Fusion AI Fusion of different kinds that lets
00:40:44
you move to the kind of interaction that is best suited to the thing you're trying to do right now yeah I think that that's uh having having the ability to be able to fluidly move into two spaces because right now it's so it requires
00:40:59
you know if all of us said we wanted to use the shared whiteboard in Zoom right when you were talking before James I was thinking damn that button is now on my Chrome and zoom and I've just never used
00:41:10
it right and uh so if all of us want to do that there would be so much startup cost and switching costs so we forget what we'd forget what we were thinking about or why we wanted to use it right and so much of the time I think that the
00:41:25
software actually hinders the process of of thinking or the process of doing the work that you want to do and that you end up just like you you try it once or twice because it's fun to do new things but then you realize actually it's not
00:41:39
actually helping us think um going back to Dynamic land I think it's it's totally fascinating because we have a uh in the accelerator program uh there's one company that had a physical piece of Hardware exactly similar to
00:41:53
Dynamic land and then they ended up making that into a software-based campus that incorporates large language models that's come out and uh and I think that
00:42:04
they but they a lot of that thinking was informed by a a physical you know uh testing environment that they created uh similar to Dynamic land so it's like
00:42:17
where does the UI for this stuff live right is that is it a is it a is it an assistive conversational bot that like sits on the side and that you know is
00:42:28
your AI child or your AI Pat is it in the space is it is does it have memory that moves across spaces does it you know how much control do do I have on it
00:42:40
does it sit at the uh you know does it the operating system level and there's you know Siri grow up into this uh or you know where where does it live or is it a separate piece of Hardware
00:42:53
right because you speak to some of the people who built Hardware in this space and they're just like the only way to actually create a new experience is to introduce a new piece of hardware I mean I think the
00:43:07
a real conundrum at the heart of this tools for thinking stuff is like computers are so powerful but also so distracting like you sit down at your computer to do one thing you end up doing like 10 other things and not the
00:43:20
thing you originally intended and everyone does that and uh so part of what needs if we want good tools for thinking we actually need to turn down the volume on our computers make them like quieter
00:43:34
make them like more subtle make them you know maybe not have a screen that's always there and so like maybe yeah it like there's a projector like you know appears when you need to have some
00:43:46
interaction but then it like goes away or you're like you're when you walk into your office like you don't see anything any computer and it doesn't like prompt you to like sit down and work on it but when you need it you
00:43:58
uh you can activate it and I think that there's a tension with like that sort of let's turn down the volume and go with like the like I love computers computers are
00:44:12
really powerful I want them to be more powerful uh and uh and I want to be sort of immersed more so those two hearts are are things that are hard to hold in
00:44:24
tension because because yeah I want like I do my best thinking often you know writing on a piece of paper or going on a walk or in the shower or uh yeah I
00:44:38
mean I've got paper notes right here you've got paper notes you just showed them like um so like clearly computers of today are like not helping
00:44:49
us think and getting like pulling away from them using paper using like a whiteboard using uh like a walk with a friend like those are often much more
00:45:02
conducive to thinking than whatever the digital equivalent is so we're sort of stumped by the simple question of how do we improve thinking
00:45:14
right John you started with what is thinking and and we're pointing to a variety of experiments around improving thinking improving interfaces
00:45:26
riffing on these sorts of things but it's funny because you know in 1968 Doug engelbart does the mother of all demos and boy that fuels 30 40 years worth of innovation and design that were a little bit prisoners of right now when we
00:45:39
talked about the tractor beam and the desktop metaphor that's kind of that's that's that string of innovation which ate the world in some sense and it ate our brains also to the point where it's hard to imagine other things I know that
00:45:50
before the before the iPhone is introduced in 2007 I had not envisioned a little slab of unobtainium with no real buttons on it just like a couple little soft buttons kind of that could turn itself into just
00:46:04
about everything I hadn't pictured that hadn't occurred to me and now it's hard to unsee that I can't I can't imagine the world without you know the touch screen slab that is connected to everything
00:46:16
um and and so somewhere on the tables and in the rooms and the conversations we've been having some were in those already is the thing that we're going to take for granted in 50 years because historically when each of these
00:46:28
little revolutions happens the thing was around and somebody was shouting from the sidelines not being heard because their demo was too early underpowered whatever whatever right and and so uh it it's in the mix right
00:46:40
now the thing that we're going to all assume well why could nobody see that yeah no I mean it also sort of uh that point gives me a sense of
00:46:50
urgency on like investing and working on this because um the sort of early people who shape a lot of these things has
00:47:03
enormous impact and the designers that I found sort of set the rules they made a bunch of probably arbitrary decisions in the very early on that were all sort of living with yeah that were very
00:47:16
consequential and I think we're sort of in in the early parts of a phase transition where like we'll probably not be looking at a desktop metaphor 10 15 years from
00:47:29
now um but so the like the values of the system the privacy of a system so like how it shapes your thinking uh those are totally
00:47:41
uh to be defined and who defines them and what reasons they use to define them will be you know shaped in some some room somewhere in the world we don't know where uh yeah like three years
00:47:54
maybe and I'll put in that I'll put in a small plug here for the podcast that's going to drop this Friday so by the time this episode airs it will be available and it's basically uh episode seven rethinking data structures with Paul
00:48:08
Roney and uh Alex obenauer um and we had this we we went really deep into some of these sorts of things including the history of of the metaphors and all that and one of the things that Paul was saying strongly was
00:48:19
hey we actually sort of kept drifting away from the usefulness of the of the metaphor we kept breaking the metaphors which are in fact in many cases really useful and he was pointing out that the latest rev of the of Mac OS which is
00:48:32
Ventura further weekends sort of the finders adherence to the good old the good old metaphor that which in some cases made it more useful and then we started talking about how does do tools for thinking have to
00:48:46
be embedded in the OS right and and does this mean some fundamental changes to the way we think about data shared data shared ideas it's really quite fascinating um undercover by the way is the guy who
00:48:58
developed the the gspeak interface that uh Tom Cruise uses in Minority Report when he's busy looking for pre for pre-crime so that that's not a flashy UI bit of CGI that's actually a working
00:49:12
operating system that he can that he can demo that allows different people to throw things up on screen and drag them onto their machines it's very much a collaborative space for thinking uh so that so so that's kind of on the table
00:49:25
as well it's one of the piece Parts go ahead John literally on the table right yeah exactly on the on on the metaphorical table so I I think that going back to look if we assume that let's just play this game out right so
00:49:38
let's assume that the uh that the the the patterns or these new light seedlings and things already there right and so let's try and uh sort of like discern what where we think they are
00:49:50
right so I think the one is clearly and spatial right I mean I think that you know what what matter is doing and what you know the quest request Pro Quest Pro Plus Quest Pro
00:50:03
um has is is clearly a piece of it but um I think it is much more profound than that right you know I remember when I first went down to the magic leap and did a demo of that it was like one of those moments that sort of like changed
00:50:17
the way I thought about Computing interfaces things Faithful is one I think auditory is one right I do think that there's like I you know I walk around the streets here in
00:50:28
Manhattan and you know 10 20 of the people have their have their airports in they're not actually in this world right now they are partially in another world right they're on a conversation with their
00:50:40
friends they're on a podcast but they are like their their Consciousness and their thinking is happening in multi-dimension sure they're crossing the street but they're also like living somewhere else and I think that audio is
00:50:53
such an important part of this and that is like if the quest is like a visual interface this is like an amazing audio interface that's really pretty uh uh out there
00:51:04
so I I think those are two I think larger language models and Ai and AI interfaces and another one what are other ones that we see right
00:51:16
now and I just sort of did the cheat shot of somewhere I see a lot of stuff we already talked about so let's push let's push to the next edge of it now I have else I have trouble with the
00:51:29
spatial thing so so part my take on this I sort of borrow remember the view Master the little thing that you put a little Carousel of pictures into and you go click click I haven't found a better uh metaphor for this uh but how do I
00:51:43
click and switch into the next appropriate interface for the thing I'm trying to do now so I just wandered into talking about covid patterns and uh or looking at Hans rosling's data about progress you know in mortality or
00:51:57
whatever I could really use augmented reality stepping into data and manipulating it now with a with a physical interface the spatial makes a lot of sense there but now I step back into this conversation and thinking about what what references are we making
00:52:10
and what happened before kind of want to be back in 2D Land and I want some active way to share 2D insights with you all right in a way that matches the tools that you're note taking with in some other sense so I'm trying to figure
00:52:22
out yeah I think what you're saying there is is that there's some other thing that captures maybe the fluidity between these spaces yes that that these spaces can coexist in an arena that we
00:52:36
get used to so that which allows us to elegantly shift between the right tool for the right task as opposed to being dedicated to learning one interface and trying to make everything fit one of these metaphors because they don't years
00:52:49
ago I was at uh David eisenberg's freedom to connect conference and there was a demo that was going to happen showing 3D avatars you know collaborating and and I had to I had to go take a call so I put into the chat I
00:53:02
said hey watch it's going to take 10 minutes to set this up and for the Avatar to walk over to the Whiteboard and get properly connected to the Whiteboard to try to start doing this demo and sure as hell that's exactly what happened it was really awkward
00:53:14
really clumsy no better than Google Docs which we now take for granted right and so so how do we begin to map tasks and design and intentions to the tools that we're talking about that gets really
00:53:27
interesting and then if you if you then fold machine learning into the equation there's a dozen different ways that machine learning could be useful and we're not thinking very fruitfully about what are those different kinds of roles
00:53:39
and how can we actually make these things our helpers our assistants in different ways and sometimes our thinking Partners in different ways and you talk about this Jerry My head goes into two places one is I do think
00:53:50
everyone should be playing with these models right now because I think you'll be surprised at what it's good and bad at yeah but it's possible that it's only about reducing friction it's possible that the language models will not do anything sexy that they will just make
00:54:04
it easier for you to configure things right now that would be a miracle right it'd be a miracle that actually was you know sort of made that part easier and if it did that then it might add a lot of value to the university it literally
00:54:16
got rid of the 10 minutes of setup right because it was flexible enough to understand and catch Dynamics and figure out lighting whatever all those sorts of things so that's one possible that's one possibility but the other thing is John was talking about sort of these
00:54:29
different modes of interaction it did strike me that one thing that we do assume is that we do assume that these things are bottom up that we do assume that these things are adopted by individuals right and that individuals
00:54:41
buy these things and then create value and then because they create more value the organization sort of adopts it I'm pretty sure that's how it's going to work but it's worth imagining what happens if like let's say
00:54:54
let's say that real estate has decimated the next five years and then people who own a lot of real estate have to innovate in dramatic ways right it's possible that in those cases Tools For Thought come top down right that they
00:55:06
end up becoming features of real estate in the same way that we expect to have air conditioning it's possible that we'll expect to have I don't know name your sort of like collaboration environment and that's the only reason
00:55:18
you're willing to go into someone's office or willing to do something like that because they're decked out to do these things um I don't think that's going to happen but it's possible right yeah I think the interesting thing at least to me is to
00:55:30
like twist around the way we think about Tools For Thought around like what are other ways that they might this might be attacked from uh from an economic point of view so if there's a piece of this that writing tools the writing AIS I
00:55:43
think are already good at which is when you're busy thinking about writing an essay a document of any kind sometimes you get into a rut and there's only one path you see through and these models are really good at hey give me six different ways of saying this thing I'm
00:55:56
trying to say and all of a sudden in front of you are six different paths into the topic and some of them will be things you didn't even think about as a framing or way of explaining the thing you're trying to do that is doable today
00:56:08
and could be transformative for writers you guys have done this like I will literally like I'll have a very thoughtful conversation with someone at the end of it I'll be like oh I'll write a thank you note to them or something
00:56:20
and then I'll ask gpd3 to generate a thank you note and it oftentimes will do a better job than I will the things that I thought were so original turn out to be just the average of everything else on the web right that's amazing super
00:56:34
depressing yeah like I turn out to be um but yes I think that those are the sorts of opportunities available and then just the example I point to a lot when thinking about computer creativity is uh when alphago basically beat Lisi
00:56:49
doll in the game of go then Google went and created Alpha zero which was not trained on historic go games but rather trained just on the rules of Go and then there's a there's a really lovely chart where you can see how alpha zero quickly
00:57:03
beats Lisi doll then it beats alphago then it goes up into no man's land and that that little margin of of new moves that nobody had conceived of and go go Scholars are now studying because that
00:57:15
was creativity because because Alpha zero had no preconceptions of 4 000 years of playing go or however many years the the game has been around and so we could go where no man has gone
00:57:27
before uh ironically um so I I think there's a lot of narrow narrow opportunities for exactly that kind of creativity and Innovation because general purpose AGI ain't happening real soon I'm convinced of
00:57:41
that but I think extraordinary help in narrow domains is here now is like and we and we have to figure out how to fold it back into this Suite of tools for thinking in a productive way and that's going to take a heck a lot of
00:57:54
cooperation across vertically focused often siled startups and projects and initiatives I think right because because it's the combined it's the combinations that are fruitful it's
00:58:07
the linkages and the The Leverage that the different tools give us together that are fruitful um and that's that space is hard to hard to figure out so um I'm watching we've got a chat Channel
00:58:20
going which uh and I'm watching that so James is asking uh a couple of questions here so so what are we missing in the conversation so one thing I think we we haven't talked about is the social side
00:58:33
of thinking and um you know you know I think Twitter has been incredibly important part of you know sort of public thinking yeah and I I think Twitter threads were
00:58:46
like one of the great product breakthroughs that their company executed on successfully um and or pretty successfully and and I think that the one of the things that
00:58:58
going back to where we started with Rome one of the things that you know sort of keyed me into the wrong world was oh so just like I try to I compose a tweet as
00:59:10
a discreet thought now I'm using blocks as discrete thoughts and I can connect them as discrete blocks and then I link them up into a page so this there were
00:59:22
these basic sort of stacking metaphors that sort of became available um but let's talk a bit about the social side of this because I do think that Twitter is like being you know such an
00:59:34
important part of this uh I I don't know if that will continue to be the case um or like how that will evolve but um I think that it is uh the date has been sort of a place where you could
00:59:46
basically think in public and we should note for posterity's sake that this is the week after Elon Musk took control of Twitter so we're at this very strange cusp and Twitter and Twitter's history it's a lot yeah
00:59:59
there's been a lot of interesting conversation about that yeah I mean it's it's fun to think about Twitter as like you know a bunch of people at this conference that I just ran had only known each other through
01:00:10
Twitter has sort of were excited to meet each other uh for the first time because they like respected their thinking that they sort of showed on Twitter and I think um you know a number of uh my friend Ed
01:00:23
cook like I met uh because he tweeted that he was at a bar and I wanted to go meet him and uh that was like 10 years ago and now we're you know best friends
01:00:37
um and uh and so you know yeah Twitter is is maybe underrated in this sort of tools for thinking world where uh or like you know Ed would say he's he's an
01:00:53
Enthusiast of parties um and like he's like parties are basically the the greatest invention or technology of humankind where like per
01:01:04
unit time more uh ideas are shared more romances started more like new thoughts created uh more friendships created uh in a party than like any other human
01:01:18
activity we can think of um and so um he for example has been working on and showed at the conference a demo of like you know online party technology and that
01:01:30
has a ways to go I think but uh I think it's a really interesting point that um yeah we uh the the social dimension of
01:01:42
of online stuff is one of the uh great accelerators of thinking yeah I think also it's just um I think making this stuff fun
01:01:57
right is like some so much like this I think it exists it exists in my head in the sort of the the like you know it's sort of like this stuff needs to be work and it needs to be hard and it shouldn't
01:02:10
have to be hard right it should be fluid it should be like it should be fun and I think we have made too many of these tools to be like task driven and then you know going back to those
01:02:23
those models yeah productivity orientated and going back to the models before is that I think that the idea and assumption that thinking is about uh is about acquiring information and
01:02:36
controlling information I think that it is thinking is about being able to have that conversation in your head about a particular thing and then hopefully having it with another human
01:02:48
being on a walk at a party do something and that to me is the active part of the the active process of thinking it's not about being able to be really smart and saying I knew exactly what happened on
01:03:01
January 4th 1976 right it's like a computer can do that way better than me and so just clock it's like how can we make that active right James also asked like really interesting things
01:03:15
we've seen in the accelerator program that was beyond the midst of this excel in the program um and and I think that the large language models stop I'm really passionate about I totally think that like James has seen earlier and
01:03:29
everybody else is keying off this stuff is fun it's fascinating it it you know and we've seen a bunch of startups like starting to like expose different ways
01:03:41
to be able to uh it to be able to connect with these with these models is it in a special workspace where you basically can have ideas that you can then bounce off the models and the
01:03:55
models are sort of fine-tune to specific areas like informat is it in a uh gems like space where you basically have all of your ideas and then the models that we use to connect your existing ideas
01:04:07
what what what is the way that these models actually sort of like in How do they how do they function in the space and um and and how do you know going
01:04:19
back to that first question of like how do we get to large-scale adoption of a new workflow and a new tool for thinking I think that is at the core of you know what we're trying to do with this excel
01:04:31
in the program and I Andrew like you I mean I believe that we're this really sort of exciting precious time where all this stuff is starting to Bubble Up and it's um it it's one of those times
01:04:45
that you know I've Had The Good Fortune to live through a few of them but it's like we we have the we have the opportunity and I I think I think some responsibility to actually
01:04:58
see if we can make these damn machines think better with us because just because we made all the information World accessible it doesn't mean we've made ourselves better thinking about it
01:05:09
absolutely that's just scrolling back in what you said just a little bit um in in education kids education there's a term there's a little phrase that I really like called hard fun that kids don't want to be spoon fed they actually like hard fun they like difficult thorny
01:05:23
things they like to feel like they've mastered something and then we don't apply that really to grown-ups we don't apply that to adults but I think what we're talking about here is that is that chewing on the issues in the world is
01:05:35
hard fun and and that we adults are drawn to hard fun as long as there's a payoff because just chewing on something hard that has no payoff is not interesting but wow if you can kind of start to see that something might dissolve or improve
01:05:48
because of your efforts and your efforts together that's huge I think that's gigantic and I think that's a big motivator for a lot of us doing this work and for people coming into this field and it hasn't translated into like
01:06:01
what's the onboarding process for some of these tools or techniques it's like the onboarding thing is just like four Fields away in the hey here's some Arcane software here's how to start using it and it's not
01:06:14
here's a fun thorny problem where we've gotten somewhere on jump jump into the conversation with your tool yeah I I think it's totally fascinating right we were talking about new interfaces and what are the seedlings we keep coming
01:06:27
back to this metaphor of the conversation right just like and being able to you know being able to just have these very lightweight so going back to yours
01:06:39
to the question or the question before what it you know what's the what are the inclination coming off this call you know I think these conversational interfaces you know you know sort of you know
01:06:52
something like that is could be the place this this new thing sort of starts to emerge and I think the along with conversations comes
01:07:05
flexibility and variation I think about you know sort of like a few months ago my daughter like found Quizlet for the first time and she's like Papa I found a secret you know I'm able to like make
01:07:19
these flash cards you know about like you know sort of studying with flashcards and I think huh it is interesting to look at something like that which felt like for her so revelatory right and she's like this is
01:07:31
the most amazing thing possible I can do this anywhere now right and then I compare and I think what would that how would that what would that Revelation feel like in a com with a conversational interface I don't really know but like but I do think it's interesting to think
01:07:44
about like a lot of these we know all the itches right all the all the desires of humanity kind of remain static we're not that different but like how it gets solved or the opportunity or what
01:07:56
becomes easy what's frictionless which what's like even more capable like that's sort of the exciting thing that's happening right now but that's my excuse for answering to say something about both Quizlet and about like that as a
01:08:08
tool for thought that was actually that was a prompt right yeah yeah love that yeah I mean Quizlet um I I'm I'm glad to hear that that magic moment I mean uh it that's been
01:08:22
the reason the reason for requisite success is people sort of uh realize that they can uh help themselves learn and we give them the tools and like they can run with them as far as they want uh
01:08:34
that's a really like exciting empowering feeling for a lot of people part of the reason I got into this sort of adjacent World tools for thinking being adjacent to education is like after I left Quizlet I was thinking about what I
01:08:47
wanted to work on next and I was doing all this research doing all this reading and trying to write about it trying to sort of map it out and I also found myself like very very frustrated
01:09:00
with the tools themselves like the ways of like how I use my own computer and how distracting it is and how interact driven it is and uh
01:09:12
and so um yeah that that's sort of like the big part of the motivation for me is just like how do we reclaim our own
01:09:24
minds uh and be able to think for ourselves and and not be so sort of plugged in or so like oriented towards solving other people's problems or like
01:09:37
just being in the sort of Borg of other people's things um and just be able to like think for ourselves um because it it like sort of shocked me
01:09:50
with how hard it was for myself even to to do that after when I when I had the time and space to do it yeah thank you um any other concluding thoughts because that was a really nice place to sort of
01:10:02
sit in and tantalize and imagine if they're not then let me uh let me take us out um thanks everybody for listening to tools for thinking a new podcast that might just help you with your thinking if you're part of a startup in the
01:10:15
sector please knock on our door at betaworks.com and thank you all for being here really appreciate it thank you [Music]
End of transcript