Speaker:
00:00:00
In today's episode, Frank and Andy sit down with special guest
Speaker:
00:00:03
Max Sklar to delve into the world of artificial intelligence, data
Speaker:
00:00:07
science, and data engineering. Max, a
Speaker:
00:00:11
trailblazer in location data and machine learning, shares
Speaker:
00:00:14
insights from his extensive experience at Foursquare, including his
Speaker:
00:00:18
work on local search and bias correction. Get ready
Speaker:
00:00:22
for a thought provoking discussion about groundbreaking projects, tech
Speaker:
00:00:26
drama, and the ever evolving landscape of technology.
Speaker:
00:00:30
So sit back, relax, and prepare to be amazed.
Speaker:
00:00:38
Hello, and welcome to Data Driven, the Podcast, we explore the
Speaker:
00:00:42
emergent fields of artificial intelligence, data science, data engineering,
Speaker:
00:00:46
and all that good stuff. With me on this ever present,
Speaker:
00:00:51
journey down the information superhighway is Andy Leonard. How's it going, Andy?
Speaker:
00:00:55
Good, Frank. How are you doing? I'm doing alright. It's been a wild 24
Speaker:
00:00:59
hours in, Maison Levin, or or
Speaker:
00:01:02
Maison, Lavinia, depending on your, how you wanna pronounce it.
Speaker:
00:01:07
At one point, we will share those crazy
Speaker:
00:01:11
details, but it's been good most of the part. I I am,
Speaker:
00:01:14
recovering from a, a bout of COVID,
Speaker:
00:01:19
that hit the entire house. I think we all picked it up on the cruise.
Speaker:
00:01:23
Again, there are worse places to pick up COVID, and there's worse
Speaker:
00:01:27
things that could happen. I've I've sneezed quite a
Speaker:
00:01:31
bit, I've coughed quite a bit, but The thing that's bugging me the most is
Speaker:
00:01:34
this headache I had now for 48 straight hours.
Speaker:
00:01:38
But it's okay, like I'm kind of living, learning to live with it, and I've
Speaker:
00:01:41
actually given it a name, I call it Charlie. So, you
Speaker:
00:01:45
know, Charlie is is is gonna be on the show today as
Speaker:
00:01:48
well. I see. Yeah. You do sound a little different, but not much.
Speaker:
00:01:53
Nah. That's why I need to clone my voice, and, maybe how they
Speaker:
00:01:56
do the avatar stays. As I was saying in the virtual dream room, this is
Speaker:
00:02:00
the 1st time I've I've felt fit for camera In about
Speaker:
00:02:03
a week. Perhaps you can use some of that good
Speaker:
00:02:07
AI voice modulation. There you go. We are
Speaker:
00:02:11
pleased to present, with us, today is Max Sklar.
Speaker:
00:02:14
Max is a, not only a fellow, data
Speaker:
00:02:18
guy, but also a Fellow podcaster.
Speaker:
00:02:22
Welcome to the show, Max. Thank you so much, Frank and Andy,
Speaker:
00:02:26
for, for having me on. I'm looking forward to it. And,
Speaker:
00:02:30
yeah. Thank you. Cool. So you you've been at you were at a
Speaker:
00:02:34
company for a number of years. You were talking about this in the virtual dream
Speaker:
00:02:37
room. You were at Foursquare for, oh, quite some time. Right.
Speaker:
00:02:41
That's unusual, actually. 10 years at a at a company as as, you
Speaker:
00:02:44
know, building software is pretty unusual. But I was really there For, I
Speaker:
00:02:48
think, like, 3 different phases, so I kind of break it up into 3
Speaker:
00:02:52
jobs. Interesting. So, you know,
Speaker:
00:02:55
Foursquare is one of those companies that
Speaker:
00:03:00
On when they broke up into 2 different parts, it kind of I didn't
Speaker:
00:03:04
understand it. Like, this is just user. I'm not I'm not I know 1 I
Speaker:
00:03:07
know that sounds terrible, but despite being my headache, Charlie talking, but, like,
Speaker:
00:03:13
what, you know, obviously, that was a decision
Speaker:
00:03:16
that was made in the marketing department, and I don't think people probably thought that
Speaker:
00:03:19
through. But I love that location. Like, I I think, you know, like, I was
Speaker:
00:03:23
the mayor of, like, 5 6 different places. It was such a cool
Speaker:
00:03:26
creative, concept that if you go there long
Speaker:
00:03:30
enough, you're the mayor, and some places would offer the mayor, like, a free coffee
Speaker:
00:03:34
and a donut or something like that. Like, it was really clever.
Speaker:
00:03:38
I know. I used to I mean, that was one of the things that got
Speaker:
00:03:41
me really excited about it back in the day. And this is already 10 years
Speaker:
00:03:43
ago. I assume you're talking about the, The app split back in
Speaker:
00:03:47
2014, which I guess is That sounds about right. I I guess
Speaker:
00:03:51
almost 10 years. I guess it's It it's actually been 10 years since we started
Speaker:
00:03:55
working on that, which was also how it worked. But,
Speaker:
00:03:58
no. I I mean, I actually remember a lot of what was going on
Speaker:
00:04:02
at the time, and, we could talk about that a little more in a
Speaker:
00:04:06
second. I I I I I'm happy to talk about that.
Speaker:
00:04:10
I I think the the whole gamification thing was really
Speaker:
00:04:14
exciting. And I really loved the idea that I can go into
Speaker:
00:04:18
a place and be like, hey. I was here for 5 times. Oh, we're gonna
Speaker:
00:04:20
give you some free, you know, free, you know, dessert or
Speaker:
00:04:24
free chicken nuggets or whatever. Maybe these days, you know, kind of trying to watch
Speaker:
00:04:27
what I eat a little bit more, it might not be as exciting. I I
Speaker:
00:04:30
it was really sad that that didn't scale as a business. I think what
Speaker:
00:04:34
would happen was, and and this I both experienced
Speaker:
00:04:38
this personally. And also from talking to the leadership there, it was clear
Speaker:
00:04:42
that they felt it it, You know, it it it wasn't a
Speaker:
00:04:45
sustainable thing because you'd often go into a place and say, hey, you know,
Speaker:
00:04:49
I checked in here all these times. I got all these rewards. You know,
Speaker:
00:04:53
I'm supposed to get, like, a dollar off, or I'm supposed to get some
Speaker:
00:04:56
some free dessert or something, which is It's always exciting
Speaker:
00:05:00
to get a little reward, even if it's not, you know, even if it's just
Speaker:
00:05:02
a dollar or whatever. It's always but I think what would end up happening is
Speaker:
00:05:06
like the the man, the The management who worked there or the or the
Speaker:
00:05:10
bartender or whatever would know about it. And, you know,
Speaker:
00:05:14
once a few times that, you know, you got that thing. Oh, oh, let me
Speaker:
00:05:17
go in the back and check. And then they come back in, like, 20 minutes,
Speaker:
00:05:19
and it's like a whole big deal. They're like, maybe I don't wanna do this
Speaker:
00:05:22
again. That's interesting, because, like, from my point of view,
Speaker:
00:05:26
it was always very It was very fun, like, so I was mayor of,
Speaker:
00:05:30
there used to be a ferry service between, in the
Speaker:
00:05:34
suburbs of DC, between Poolesville, Maryland and
Speaker:
00:05:37
Leesburg, Virginia. And, I would take that ferry a lot so
Speaker:
00:05:41
much. So I was actually the mayor of the ferry. It didn't get
Speaker:
00:05:45
me anything, like, in that case, but it was this kind of cool, like I
Speaker:
00:05:48
I love to I I always imagined, like, there would be, You know, like
Speaker:
00:05:52
the ferry could have some kind of a a little like TV screen that could
Speaker:
00:05:55
like show who the mayor was. And then you can, you know, and and and
Speaker:
00:05:59
then then then we can like kind of scale up those fights, But alas,
Speaker:
00:06:03
society didn't go in that direction. Interestingly,
Speaker:
00:06:07
Foursquare, what, like, I have been interested in that this like local
Speaker:
00:06:11
local search space since since well before forest grabbing even my, you
Speaker:
00:06:14
know, my my senior project as an undergrad, back,
Speaker:
00:06:18
though, starting in 2005, was this, like, you know, this this website
Speaker:
00:06:22
called sticky map, where people would post little, Icons all
Speaker:
00:06:25
over Google Map. It's kind of inspired by Wikipedia. You can add messages.
Speaker:
00:06:29
And I just thought it was pretty cool. People started marking up the, the campus,
Speaker:
00:06:32
and then people started, you know, marking up. It was like, well, it's based on
Speaker:
00:06:36
Google Maps API, so you could just mark up anything you want. And
Speaker:
00:06:40
I think in that first project, I noticed All the problems
Speaker:
00:06:44
that still exist with that kind of data today. Okay. What happens when you
Speaker:
00:06:48
add duplicates? It's the first thing that happened. As, you know, As an undergrad, I
Speaker:
00:06:51
was like, oh, I'm so excited. Let me show something like this. They're like, okay,
Speaker:
00:06:54
let me create a marker. And I'm like, don't create that one. It's already been
Speaker:
00:06:58
created. And then it's like, okay, now we have duplicates Right off the bat. And
Speaker:
00:07:01
that is still something that, you know, Foursquare
Speaker:
00:07:05
deals with and I'm sure Google deals with and Apple Maps deals with and then
Speaker:
00:07:09
they they all deal with it. So,
Speaker:
00:07:14
Yeah. It it it it I I think when I
Speaker:
00:07:17
discovered Foursquare, you know, several years later,
Speaker:
00:07:22
it was the it was innovative in several ways. It was, first of
Speaker:
00:07:25
all, it was based on on mobile apps actually being at the place when you're
Speaker:
00:07:29
commenting on it, which is exciting. It was the gamification of it. And the
Speaker:
00:07:33
fact And and for those listening Yeah. For those listening, that was new. Like,
Speaker:
00:07:36
that was brand new. So sorry. I didn't mean to cut you off, but, like,
Speaker:
00:07:40
the context is important because All I had was was something on a on
Speaker:
00:07:43
a website. You know, I I wasn't thinking of the I the iPhone didn't
Speaker:
00:07:47
exist, for actually, pre Foursquare,
Speaker:
00:07:51
there there was something called dodgeball, which was kind of the the
Speaker:
00:07:55
predecessor to Foursquare, Which I wasn't involved with, but it but, it
Speaker:
00:07:59
it was based on, like, you know, SMS kind of text messages where people were
Speaker:
00:08:02
messaging on. If you remember, You know, those, you know,
Speaker:
00:08:06
the the the what was it? The t nine texting where people were Oh,
Speaker:
00:08:10
God. Yeah. Yeah. Yeah. But people would use that and and Foursquare and,
Speaker:
00:08:14
Google bought that. And then the the the founders,
Speaker:
00:08:19
Dennis Nautz went it went in and started Foursquare after that.
Speaker:
00:08:23
So, very, there there's a very
Speaker:
00:08:27
interesting history of kind of, like, Local search, city guide,
Speaker:
00:08:31
and basically sort of social kinda local applications. We're
Speaker:
00:08:35
we're very big at the time. Nowadays, I think we need to find a new
Speaker:
00:08:38
take on it, but, that when I joined Forescord
Speaker:
00:08:42
in 2011, it was very exciting. I'm glad you mentioned that because there was
Speaker:
00:08:46
a a real so I started my career in New York City. Right?
Speaker:
00:08:49
So I worked at Barnes and Noble com. So I was there in the fairly
Speaker:
00:08:52
early days of Silicon Alley, and that was a huge thing. It was
Speaker:
00:08:56
Microsoft. I think it was Microsoft had something called Sidewalk.
Speaker:
00:09:01
And then there was there was maybe it was maybe it wasn't Microsoft. Maybe
Speaker:
00:09:05
somebody else had something called sidewalk AOL had something, going and the fact
Speaker:
00:09:09
you can't remember it, I think says it all. Right? Like, it was like in
Speaker:
00:09:12
in, you know, anybody that could register a .com could spell and could
Speaker:
00:09:16
spell HTML To get funding back in those days, a
Speaker:
00:09:19
bit like the way the AI startup ecosystem is kinda
Speaker:
00:09:23
today. But, no, there was I mean,
Speaker:
00:09:26
people there was a time, young children
Speaker:
00:09:30
out there, That when, you know, people saw the online
Speaker:
00:09:34
world is slightly different than the real world, and they saw this as an opportunity.
Speaker:
00:09:38
Right? But no, I'd like That takes me back. As soon as
Speaker:
00:09:42
you said to, like, the local kind of connection guides, I was like, wow, it
Speaker:
00:09:45
takes me back. Yeah. Yeah. Coming back to
Speaker:
00:09:48
the app split, and I wasn't expecting to talk about this today. I'm sorry if
Speaker:
00:09:52
it brings up If for what it's worth, I'm a former Windows For what it's
Speaker:
00:09:56
worth, I'm a former Windows phone developer, and I wrote a book on silver light.
Speaker:
00:09:59
So I understand the pain of working on
Speaker:
00:10:03
It'll fade and I worked at barnes and noble.com. Right? So there's my trifecta of
Speaker:
00:10:07
ill fated technology projects. Yeah. I I think it's
Speaker:
00:10:12
A lot of technology companies, in order to become successful, actually have
Speaker:
00:10:16
to go through big changes where people yell at them.
Speaker:
00:10:20
And so it's like, how do you know whether you're breaking things or whether you're
Speaker:
00:10:23
actually doing what you're supposed to do? And so that's kind of a Right. That's
Speaker:
00:10:26
that's kind of a tough decision. I I think For Foursquare
Speaker:
00:10:30
at the time, there was always, like, kind of a design,
Speaker:
00:10:34
and product, like, tension between the people who
Speaker:
00:10:38
wanted to be there As essentially like a Yelp replacement, kind of
Speaker:
00:10:41
like a a local search city guide. And then there and then versus the people
Speaker:
00:10:45
who were there for the the life logging, the check ins, the game. And I
Speaker:
00:10:48
think, I think the separation could have been
Speaker:
00:10:52
done. I mean, my personal thing is I think there could have been a
Speaker:
00:10:56
Separation, I could think, could have been executed a little bit better. I
Speaker:
00:10:59
think technically we did a good job. I think the apps that
Speaker:
00:11:03
we ended up with were well designed, but I think, I think
Speaker:
00:11:07
we needed to do is take into account how the how people were using the
Speaker:
00:11:10
apps, at the time and not just, like, Kick all the
Speaker:
00:11:14
people who are checking in, which was which was Foursquare's kind of bread and butter
Speaker:
00:11:18
and just, like, kick them to the side with this other app. And then it's
Speaker:
00:11:20
like, well, what is this? I'm calling this something different. Sorta. That was that was,
Speaker:
00:11:23
I think, too much. But, again,
Speaker:
00:11:27
there there were people saying it at the time. But
Speaker:
00:11:31
The problem is, I I guess, you know, there whenever you make a
Speaker:
00:11:35
change, there's always a great many people saying a great many things. So
Speaker:
00:11:40
We could wax this we could wax nostalgic about Foursquare because I used to
Speaker:
00:11:44
like, when I I was travelling a lot at the time when I worked I
Speaker:
00:11:47
worked at Microsoft about 10 years ago. But I remember Sometimes I would actually
Speaker:
00:11:50
choose different connecting airports, so I could get, like, the the
Speaker:
00:11:54
jet that was it, the the jet set tag, like level up in my achievement
Speaker:
00:11:58
there. Right. Which is kinda sad, but, we could
Speaker:
00:12:02
we could wax nostalgic about that all day, and I would love to. But I
Speaker:
00:12:05
think what What was the role of AI and ML in that
Speaker:
00:12:09
space? Right? Because you're obviously collecting a lot of data. No. Like, I'm just curious,
Speaker:
00:12:12
like, because how how was that being used? How was that,
Speaker:
00:12:17
leverage. In in in a lot of ways,
Speaker:
00:12:20
and, you know, many of which I I worked on over all those years.
Speaker:
00:12:25
You know, one of them was, I mean, just, you know, search
Speaker:
00:12:28
ranking in general, which, you know, Foursquare had a lot of ex Google engineers.
Speaker:
00:12:32
So I learned directly from them so they they knew what to do.
Speaker:
00:12:36
But search ranking, search ranking was a was a was a big project.
Speaker:
00:12:40
This is kind of more of a statistical Problem where you were kinda trying
Speaker:
00:12:44
to weight different attributes, like, is this related to the search the
Speaker:
00:12:47
person put in? Is this related to how much do we wanna, You know,
Speaker:
00:12:52
score things that are, you know, maybe someone's
Speaker:
00:12:56
friends went to. So something like that. I think that
Speaker:
00:13:00
The biggest well, I'll talk about the one that I think is the biggest deal
Speaker:
00:13:03
and then the one that that I worked on the most. The one I think
Speaker:
00:13:06
was the biggest deal for for Foursquare, which I did work on a little
Speaker:
00:13:10
bit, is basically trying to figure out where
Speaker:
00:13:13
someone is given. So we know where someone is given there that long
Speaker:
00:13:17
from their phone. But it's like, what are they actually in a particular
Speaker:
00:13:20
store? Like, are they in the Starbucks? Are they in the, you know, are they
Speaker:
00:13:24
in the office, over there? Are they Are they just walking down
Speaker:
00:13:28
the street? And so using the fact that people were were
Speaker:
00:13:31
giving us training data, essentially, which was a big theme there, which is, you know,
Speaker:
00:13:35
I think, something that,
Speaker:
00:13:40
data scientists and data entrepreneurs need to need to look in closely, which is
Speaker:
00:13:43
like, How can you get people to give you training data? Because it is really
Speaker:
00:13:46
useful. So if you have people giving you where they are and
Speaker:
00:13:50
then you could see the information from their phone, not just a lot long, but
Speaker:
00:13:53
like what, You know, things like what Wi Fi's can you see? What, you know,
Speaker:
00:13:56
other sensors from your phone, can you figure out where they are? And then there's
Speaker:
00:14:00
the whole stop detection, problem. And so, Yep.
Speaker:
00:14:04
Foursquare essentially can kinda figure out, you know, where you went day
Speaker:
00:14:08
to day, and it's actually pretty good. Like, you know, if I Don't tell Foursquare
Speaker:
00:14:11
where I went. Even today, I still look at it and, you know, it tells
Speaker:
00:14:15
me what, what actual stores I was in. Now maybe there's a question of,
Speaker:
00:14:19
you know, whether whether our apps are knowing too much about us, but
Speaker:
00:14:22
that's that's a whole another question. But that was a very important,
Speaker:
00:14:27
a resource for the business. And the one that I worked on the most, that
Speaker:
00:14:30
was the most exciting though for me was the natural language
Speaker:
00:14:34
processing Pipeline. And, of course, you know, text text
Speaker:
00:14:38
data today is is is having such a a resurgence
Speaker:
00:14:42
with, You know, I don't need to tell your audience with AI and all that.
Speaker:
00:14:45
But, you know, it it it back then it was like, well, people were giving
Speaker:
00:14:49
us, you know, several sentences called tips on Foursquare Venues,
Speaker:
00:14:53
which would often be like, here's, you know, here's what you should do here. Here's
Speaker:
00:14:56
what you should try. Here's a little review, something like that. People were
Speaker:
00:15:00
leaving text with their check-in. So there's a bunch of texts, there's
Speaker:
00:15:03
menus, things like that. So there's a bunch of texts in the system. And so
Speaker:
00:15:07
it's like, what do we do with all of that? And, one of the things
Speaker:
00:15:11
that we did was we pulled out key terms, you know, noun phrase detection.
Speaker:
00:15:14
This is all kind of standard natural language
Speaker:
00:15:18
processing. You know, not
Speaker:
00:15:22
you know, you know, people often ask, oh, you know, I I think
Speaker:
00:15:26
Nowadays, I'm often thinking everyone's thinking, oh, you were probably using, like,
Speaker:
00:15:30
generative AI or something. No. It was just kind of standard NLP that had
Speaker:
00:15:34
been developed over the last, you know, several decades. But,
Speaker:
00:15:38
we did sentiment analysis and we used that to come up with the ratings for
Speaker:
00:15:41
the venues, which which are used today. So you could tell
Speaker:
00:15:45
how good something is. And,
Speaker:
00:15:49
you know, I did some things that were a little bit
Speaker:
00:15:53
more interesting that, you know, maybe get overlooked, but they're
Speaker:
00:15:57
they're kind of unique to To to what we did there, which was sort of
Speaker:
00:16:00
like timeliness and seasonality, which is so, like, if you check into
Speaker:
00:16:04
a diner in the morning versus in the afternoon, It'll statistically
Speaker:
00:16:08
give you different suggestions based on how timely it thinks each
Speaker:
00:16:11
each suggestion is. Because with every Check-in where someone is doing
Speaker:
00:16:15
something in real time. We have the timestamp. We know what time of day. We
Speaker:
00:16:19
know what time of week. We know what time of year. And so it's kind
Speaker:
00:16:22
of cool to to put that all together. And some of
Speaker:
00:16:26
the some of the, some of the
Speaker:
00:16:29
models, got pretty,
Speaker:
00:16:34
you know, it was it was pretty neat how it all turned out. I think
Speaker:
00:16:37
that one I you know, I still talk about that one is one of my
Speaker:
00:16:39
That's my favorite one after being in the industry so long, even though it was
Speaker:
00:16:42
like 10 years ago, because it was like, okay, we had training
Speaker:
00:16:46
data again from on these tips where, You know, we
Speaker:
00:16:50
could tell if the person liked the venue or disliked the venue and because
Speaker:
00:16:54
they they told us, and they also left the tip. There were a lot of
Speaker:
00:16:56
people who did that. So that just gave us training data for sentiment analysis.
Speaker:
00:17:00
And at the time, I'm sure the tools now are much more sophisticated at the
Speaker:
00:17:03
time when we use pretrained sentiment analysis tools, Didn't really
Speaker:
00:17:06
work well on our data because it's just it was just a different kind of
Speaker:
00:17:09
text. People wrote on Foursquare differently than they did on Twitter, for
Speaker:
00:17:13
example. So, so that gave us training data. Give
Speaker:
00:17:17
us training data for every language. And so that was nice. We got kind of,
Speaker:
00:17:20
like, you know, 90 languages for free just by just by
Speaker:
00:17:24
using that Strategy of Oh, wow. Using the data that people gave us.
Speaker:
00:17:27
Probably not probably didn't work very well in all 90, but certainly worked
Speaker:
00:17:31
well. Well, the beauty of it is It ends up working
Speaker:
00:17:35
well so long as we have good language detection, it ends up working well
Speaker:
00:17:38
in, any language that has any Particular,
Speaker:
00:17:43
you know, any particular popularity in Foursquare.
Speaker:
00:17:47
So for example, if, the Turkish was very popular. Okay. Well, that
Speaker:
00:17:51
means we have a lot of Turkish training data. That means that the
Speaker:
00:17:55
the model, which trains monthly, is Is going to use all that training data. That
Speaker:
00:17:59
means it's going to work very well. And so, and
Speaker:
00:18:02
so that the fact that the models were always regenerating
Speaker:
00:18:07
And they were always regenerating based on the latest data
Speaker:
00:18:11
was was really cool because oftentimes you think these think about
Speaker:
00:18:16
ML teams kind of building a model, and then they kind of throw
Speaker:
00:18:20
it over a wall. They they productionize it. And then you have to
Speaker:
00:18:24
work on the next one, but you have to you have to do some work.
Speaker:
00:18:27
It's not automated, you know. So it's like, well, this is this is gonna start
Speaker:
00:18:30
going downhill If we don't, if we don't interact. And the fact that we were
Speaker:
00:18:33
able to set it up where it was just constantly getting smarter was,
Speaker:
00:18:37
was pretty neat. So MLOps and pipelines before they were called
Speaker:
00:18:41
MLOps. Well, they might have been called pipelines, but yeah. Interesting. Yeah.
Speaker:
00:18:45
Pipeline was a big Big big key phrase. So what
Speaker:
00:18:48
what did the data what did the back it because like one of the one
Speaker:
00:18:51
of the jokes that we have, and in fact, it's a domain name that I
Speaker:
00:18:54
registered. 1st, you get the data, is a phrase that
Speaker:
00:18:58
a lot of data scientists will often use, much to the chagrin of a lot
Speaker:
00:19:01
of data engineers, because a lot of data,
Speaker:
00:19:05
you have to get the data in a certain way to to format it and
Speaker:
00:19:08
and and to get it trained. And if you go to first you get
Speaker:
00:19:12
the data.com, it should redirect you to our website,
Speaker:
00:19:15
hopefully. God only knows if it works. 1st, I think
Speaker:
00:19:19
yeah. I'm gonna I'm gonna try to get the data.com.
Speaker:
00:19:23
I'm shattered to think that okay. Good. It does work. Okay. DNS
Speaker:
00:19:27
and me have a long history. Yes. It's going back.
Speaker:
00:19:30
Good. I I it's always good to start off a week with a win with
Speaker:
00:19:33
DNS. What did it what did the
Speaker:
00:19:37
because I'm curious, like, Foursquare was one of those early, kind
Speaker:
00:19:41
of mobile first, kind of success stories. I'm
Speaker:
00:19:45
always curious, what did the back end data platform look like? Right?
Speaker:
00:19:49
Because, and again, going back 10 years, I
Speaker:
00:19:52
mean, I mean, did you use what was the name of the,
Speaker:
00:19:57
gosh. Can't think of the name of the platform, but what sorts of technologies did
Speaker:
00:20:00
you did you guys use? Yeah. I'm I mean, I'm sure
Speaker:
00:20:04
it works similar today in in at at Foursquare. Mhmm.
Speaker:
00:20:08
Well, we were using data pipe I assume. But, yeah,
Speaker:
00:20:11
if I remember correctly, we had, you know, our transactional database, our Mongo
Speaker:
00:20:15
database that was sort of like, Every once in a
Speaker:
00:20:19
while. And so that was kinda like the baseline. And then there'd be a series
Speaker:
00:20:23
of jobs that, like, you know, built it up, that that
Speaker:
00:20:26
that kind of Calculated things off of that, and that
Speaker:
00:20:30
would, in the at the end of that pipeline, you know, release
Speaker:
00:20:34
a, a dataset that would then be kind of,
Speaker:
00:20:38
Automatically, deployed and then read by,
Speaker:
00:20:42
read by the server in real time. So, if I can think of, like, the
Speaker:
00:20:46
technologies, I think the, the pipeline technology, the
Speaker:
00:20:50
pipeline, what was it? It was
Speaker:
00:20:53
like Luigi. It was written in pipe. Python. I don't know if that's too interesting.
Speaker:
00:20:57
There's a lot of different ones you could use these days.
Speaker:
00:21:02
It's an interesting question of, like, you know, which one do you use? I,
Speaker:
00:21:07
it's It's probably,
Speaker:
00:21:10
you know, from from my point of view, it's always like, well, the company kinda
Speaker:
00:21:14
chooses it. You don't really have much of a say.
Speaker:
00:21:19
And then then it's like, well, well, how do I know how to compare? But
Speaker:
00:21:22
let's see. Like, you know, we were using MapReduce jobs. We're using Hadoop At the
Speaker:
00:21:26
time, I think scalding, was was one that's that's maybe kind of out of fashion
Speaker:
00:21:29
now. That was a, a scala based framework for for
Speaker:
00:21:33
some of these jobs which were, which which
Speaker:
00:21:37
was based on abstract algebra. So it's actually pretty cool. I wish it
Speaker:
00:21:40
was. It was kinda hard to to reason about sometimes if you
Speaker:
00:21:45
It kind of went too far, to the side of, okay,
Speaker:
00:21:49
you know, I love abstract algebra, but I don't want everybody
Speaker:
00:21:53
who I don't I don't want that to be a barrier to entry for people
Speaker:
00:21:56
who are we're working on this. But, I'm
Speaker:
00:22:00
just trying to remember, like, some of the, You know, some of the some of
Speaker:
00:22:03
the tech bud buzzwords. But if you have any specific questions, maybe they'll jog jog
Speaker:
00:22:06
my memory. I don't know. Like, one of the things that was popular about that
Speaker:
00:22:09
time was, HBase. Oh,
Speaker:
00:22:13
yes. I, were we using interest? I think we were using,
Speaker:
00:22:18
Yeah, I remember that Term, but I I know. I know. I was as you
Speaker:
00:22:21
were talking, I'm like using it or if we wanted to use it. It was
Speaker:
00:22:24
one of those 2. Now from that if memory serves, I think Facebook is the
Speaker:
00:22:27
one who pioneered h base because it was really it was a right once read
Speaker:
00:22:31
many thing, and basically, the last one in when,
Speaker:
00:22:34
last last I can't talk, sorry. Last one wins.
Speaker:
00:22:38
Let Andy help might help me out if with the
Speaker:
00:22:42
technical term for that last one. Last one wins. What's the
Speaker:
00:22:46
Oh, yeah. So right. I remember they were called h files, so it must have
Speaker:
00:22:49
been yes. It must have been that. Yeah. Yeah. Yeah. That was one
Speaker:
00:22:53
of those sorry, Andy. Go ahead. That's okay. You were I was thinking,
Speaker:
00:22:57
you know, ChipLogic last in first out. Yeah.
Speaker:
00:23:01
Yeah. Something like that. Yeah. Know if that's what you were after or not. Last
Speaker:
00:23:04
one wins. Last one That was their concurrency strategy.
Speaker:
00:23:08
That's, I know there's a better term for that, but again,
Speaker:
00:23:13
it's a Monday and, I have
Speaker:
00:23:17
a headache. But no, it's it's it's it's
Speaker:
00:23:21
fascinating to kind of Almost like technology
Speaker:
00:23:25
archaeology. Like what worth the big projects
Speaker:
00:23:28
that were popular at the time. Right? You know, and it's
Speaker:
00:23:32
just, And it's scary to think that, you know, we're talking 10 years
Speaker:
00:23:36
ago. I mean, I mean, you Not even though. A lot of this stuff was
Speaker:
00:23:39
I mean, a lot of this stuff is probably still in place at Foursquare today.
Speaker:
00:23:43
Yeah. I mean, what's interesting to me is you you mentioned a lot of
Speaker:
00:23:47
the NLP, techniques that, You know, for lack
Speaker:
00:23:51
of better term, people would consider legacy now, right? Because they're pre
Speaker:
00:23:54
transformers, right? They're pre GPT, right? Sentiment
Speaker:
00:23:58
analysis, a lot of, you know, I I speak with a lot of people with
Speaker:
00:24:01
varying degrees of technology skills, and they
Speaker:
00:24:05
assume that this field of research didn't exist prior to
Speaker:
00:24:09
last year. And, very
Speaker:
00:24:12
much not the case. It's just that radically changed about a year ago.
Speaker:
00:24:17
Right. I mean and and this is something that I'm trying to figure out how
Speaker:
00:24:19
to do, which, I I might not be alone. It's like, okay. I did
Speaker:
00:24:23
all these things. How do I reinvent myself now in this new world?
Speaker:
00:24:27
And, you know, once you realize it could be exciting thing, then
Speaker:
00:24:31
it's maybe not so much of a drag, you know, because there's there's so many
Speaker:
00:24:34
opportunities out there. But it's like, but I can't be
Speaker:
00:24:38
the only one out there who's struggling with this being like, okay,
Speaker:
00:24:41
wow, I've got a, You know, I I've gotta,
Speaker:
00:24:45
you know, work or at least do projects for companies that
Speaker:
00:24:49
are at the cutting edge here in order to, In order to be,
Speaker:
00:24:53
you know, in order to be at the forefront. Yeah. It's funny, like, you miss,
Speaker:
00:24:57
like, I was, You know, offline for I tried to be offline, but for the
Speaker:
00:25:01
better part of a week and for vacation.
Speaker:
00:25:04
And like During that week, AMD announces that they are
Speaker:
00:25:08
producing their own, GPU LLM
Speaker:
00:25:12
type hardware. Gemini comes out and all
Speaker:
00:25:15
these other innovations that come out, and I'm like, I feel, like, hopelessly behind
Speaker:
00:25:19
now. I'm being offline for a week. Yeah.
Speaker:
00:25:23
Yeah. It's I mean, I I guess the
Speaker:
00:25:26
only, consolation there is everyone's dealing with that.
Speaker:
00:25:30
Right. You know? Yeah. Right. And Kinda like impostor
Speaker:
00:25:33
syndrome. Right? Yeah. Yeah. I think I think the
Speaker:
00:25:37
question is, especially in this new world of generative
Speaker:
00:25:41
AI. And and the question I'm asking I don't necessarily have answers. But it's
Speaker:
00:25:44
like, how do you so You wanna jump in the stream
Speaker:
00:25:48
and get all the latest stuff, but you also want to leverage your experience
Speaker:
00:25:52
and understanding. Cannot be leveraged. And
Speaker:
00:25:56
so what's the best way to to, you know, what
Speaker:
00:26:00
what's the best way to balance that? I think that's something that I would like
Speaker:
00:26:03
to see more people asking. And I would like guidance on this. I know I'm
Speaker:
00:26:06
the guest. I'm supposed to say what I know, but No. But try now. You
Speaker:
00:26:10
know, Mads, the, the thing is a lot of the stuff that's
Speaker:
00:26:13
That's new. I'm doing the air quotes here for people who are listening.
Speaker:
00:26:17
A lot of things that are new are really coming out of tech That was
Speaker:
00:26:21
developed. The math was developed, for instance, in the late sixties, seventies,
Speaker:
00:26:24
eighties, nineties. So a lot of that is just being reapplied
Speaker:
00:26:29
Back when the math was developed and the theorems and and such,
Speaker:
00:26:32
we didn't have machines fast enough to do it or at least do it
Speaker:
00:26:36
usefully. So I wouldn't feel bad at all about,
Speaker:
00:26:40
you know, having a bunch of, a bunch of experience that seems dated
Speaker:
00:26:43
right now because A couple of weeks to a couple of months. That might
Speaker:
00:26:47
be the new shiny. Right. That's true. When I went back to
Speaker:
00:26:51
a a university computer science program, You know, they're still studying
Speaker:
00:26:55
data structures and algorithms. It's still very relevant.
Speaker:
00:26:58
And, you know, I think a lot of outsiders think, oh, everything's
Speaker:
00:27:02
gonna Turnover in, in a year and a lot of things
Speaker:
00:27:06
do. But there are also a lot of kind of like universal,
Speaker:
00:27:10
kind of, there's a lot of universal Theory that's, good to know
Speaker:
00:27:14
about. Sure. The fundamentals don't change that often.
Speaker:
00:27:18
Nope. And it's a lot of reapplications. I see a lot of people
Speaker:
00:27:21
reapplying stuff 2 or 3 times. I mean, I've been I've been
Speaker:
00:27:25
around computing since 1975. So I've seen kinda like these meta
Speaker:
00:27:29
patterns flow, you know, through several generations,
Speaker:
00:27:33
and they kinda keep just resurfacing. One of one of the
Speaker:
00:27:36
interesting ones is, like, the, well,
Speaker:
00:27:40
both the chatbot and the text based interface versus the,
Speaker:
00:27:45
Graphical interface seems like we keep going back and forth. You know,
Speaker:
00:27:48
I I remember chatbots back in the, you know, AOL days.
Speaker:
00:27:52
AIM days probably way before that too.
Speaker:
00:27:56
And then, you know, and then There was kind
Speaker:
00:28:00
of a a a chatbot resurgence in, you
Speaker:
00:28:03
know, 2016, 2015, whenever when every company wanted
Speaker:
00:28:07
a chatbot and we're excited about that. Yeah. It didn't quite work. It
Speaker:
00:28:11
seemed to fizzle out. Then, you know, the
Speaker:
00:28:15
the then nowadays, we have So many chat interfaces,
Speaker:
00:28:18
chat GPT and and generative AI seems to be resurgent again.
Speaker:
00:28:22
So there are these weird sine waves, these weird
Speaker:
00:28:26
cycles, and I almost think of it as a coil where, you know, you're starting
Speaker:
00:28:29
at the bottom and you're cycling, but you're also moving up at the same time.
Speaker:
00:28:33
And so How do you how do you surf the wave? That's, that's,
Speaker:
00:28:37
something that's once you kind of, understand the
Speaker:
00:28:40
fact that that's what you're doing, then then then you can be excited about it.
Speaker:
00:28:44
I I think it's fair. Well, we're at that point in the show
Speaker:
00:28:47
where we transition to our, questions. And, we
Speaker:
00:28:51
dropped them into the chat here for you. Our very first one is how did
Speaker:
00:28:55
you find your way into data? Did data find you or did you find
Speaker:
00:28:59
data, Max? Interesting. Well, I
Speaker:
00:29:02
guess I was always interested in math and computer
Speaker:
00:29:06
science. You know, going back to undergrad, you
Speaker:
00:29:10
know, it was like there was a lot of different areas I could choose. I
Speaker:
00:29:13
had a hard time going into a field that, you know, where I
Speaker:
00:29:17
wasn't, using all different parts of my brain and
Speaker:
00:29:21
computer science department, it was it was not just the
Speaker:
00:29:25
mathematics. It was, you know, there was, you know,
Speaker:
00:29:29
there was a bunch of creativity in it as well. There was human computer interface.
Speaker:
00:29:33
There was it. So, So I was kind of, I gravitated to that field
Speaker:
00:29:37
as an undergrad. When I graduated, I I joined a company
Speaker:
00:29:40
called wireless generation, which, today is called Amplify.
Speaker:
00:29:45
And that's it was an education tech company. And I was
Speaker:
00:29:48
doing, you know, some simple kind of software engineering work. Actually, back
Speaker:
00:29:52
then, It was, which sounds really dated now, but, you know, they
Speaker:
00:29:56
were probably doing this up to, like, 2010, which was, you know,
Speaker:
00:29:59
writing c plus plus for the palm pilot. You know, we yeah.
Speaker:
00:30:03
Because it was they were assessing students and then it would sync to to
Speaker:
00:30:07
the web and all that. And Sure. It was a lot of, like, taking
Speaker:
00:30:10
stuff, Taking that information out of databases and putting it into a a
Speaker:
00:30:14
dashboard. And it was it was you know, I I felt like there
Speaker:
00:30:18
could be something more interesting I was doing even though I love kind of the
Speaker:
00:30:21
mission of that company there. So I ended up in grad school. I ended up
Speaker:
00:30:24
at NYU and I went there from I guess
Speaker:
00:30:28
2009 to 2011 really discovered,
Speaker:
00:30:32
you know, data mining, was the 1st related
Speaker:
00:30:36
class. Then I took, You know, machine learning, natural language processing. Actually,
Speaker:
00:30:39
the the machine learning class was with, Jan Lacun, who is,
Speaker:
00:30:43
a very well known machine learning researcher. He's Like The Lani.
Speaker:
00:30:47
The the the the the the the the the the the the the the the
Speaker:
00:30:49
the the the the the the the the the the the the the the the
Speaker:
00:30:51
the the the the the the the the the the the the the the the
Speaker:
00:30:53
the the the the the the the the the the the the the the the
Speaker:
00:30:53
the the the the the the the the the the the the the the the
Speaker:
00:30:53
the the the the the the the the the the the the the the the
Speaker:
00:30:53
the the the the the the the the the the the You know, all the
Speaker:
00:30:55
stuff that exists today. Like, even this was 2010. He would show us a camera
Speaker:
00:30:59
where he would point to different objects. He'd be like key, wallet,
Speaker:
00:31:02
chair, and it would like, the the the text would appear
Speaker:
00:31:06
on the the screen based on what he pointed at. So they knew how to
Speaker:
00:31:10
do all this stuff, that that you think of as as kind of
Speaker:
00:31:14
it it's it almost seems crazy that that was not, like,
Speaker:
00:31:19
and and turned into a product that anyone could use back then that it almost
Speaker:
00:31:22
seems crazy that it took you know so long to do it but they and
Speaker:
00:31:26
actually it it may have been Used by someone.
Speaker:
00:31:29
It's, sure. Maybe we just don't know about it.
Speaker:
00:31:35
Sorry. My paranoia. No. No. You're right. I'm sure it was used quite
Speaker:
00:31:39
a bit, but it it it's just like what it was that kind
Speaker:
00:31:43
of Sitting on his laptop was so much more sophisticated than anything that
Speaker:
00:31:46
that I I saw a year later. But,
Speaker:
00:31:50
Yeah. So it was That was kind of inspiring. And so it was
Speaker:
00:31:54
like, you know, it was
Speaker:
00:31:58
to me, it seemed like a much more interesting problem. Well, how do you How
Speaker:
00:32:01
does the machine learn? You know? How do you, you know, I don't I don't
Speaker:
00:32:05
wanna sit around writing code that's just dead. I want it to To
Speaker:
00:32:08
be alive, I wanted to to learn from experience. And so when you dive into
Speaker:
00:32:12
that question, well, then you get into machine learning, which is actually Pretty well
Speaker:
00:32:16
named. And then and, you figure, okay, well, you need
Speaker:
00:32:19
data to learn from, and then that that ends up being a statistical model
Speaker:
00:32:23
and so on and so forth. So, you know, when I
Speaker:
00:32:27
so Foursquare, was a company that that essentially came
Speaker:
00:32:30
out of NYU And, you know, it kind of intersected. So
Speaker:
00:32:34
and and they wanted to, to learn from from
Speaker:
00:32:38
their data. They wanted to kind of, sort of a
Speaker:
00:32:42
to build a data science team. And so I had already been
Speaker:
00:32:45
working on that sticky map project, And I was into local search. I
Speaker:
00:32:49
loved the the product aspect. I didn't have my new
Speaker:
00:32:53
interest in machine learning and LP in there. So it all kinda came together. And
Speaker:
00:32:56
so that's why I think that was such a good fit for me and probably,
Speaker:
00:33:01
probably would be very difficult to find such a fit again.
Speaker:
00:33:08
Our next question is, what's your favorite part of your current
Speaker:
00:33:12
gig? And that was, in The virtual green room, you said you
Speaker:
00:33:16
kinda had a good story about that.
Speaker:
00:33:19
Right. So I don't. Well, I don't exactly have a a
Speaker:
00:33:23
current gig right now. I have a bunch of different projects that I'm working on.
Speaker:
00:33:27
It was you know, I think It it was on one hand, it was
Speaker:
00:33:30
nice in Foursquare to be able to focus on one thing, and I'm gonna come
Speaker:
00:33:34
back to that. But I feel like you need these periods, almost like the same
Speaker:
00:33:37
as the grad school period That I had, back in 2010 where it was
Speaker:
00:33:41
like, well, you're working on a few different side projects, but let's see.
Speaker:
00:33:44
Hopefully, like, eventually it'll coalesce into something,
Speaker:
00:33:48
you know, something a little bit more long term and permanent. So I'm working on
Speaker:
00:33:52
several projects. One is with with the Foursquare founder, Dennis
Speaker:
00:33:55
Crowley. And we are Working on a new product, a new
Speaker:
00:33:59
kind of city guide where you walk around the city with your headphones in,
Speaker:
00:34:03
with your AirPods in or whatever. And We kinda know what you're passing,
Speaker:
00:34:07
by. We sort of are are using some of the Foursquare
Speaker:
00:34:11
tools that are publicly available that we know about, but also, You know, we're
Speaker:
00:34:15
kind of rigging up our our own thing because we've just done it so many
Speaker:
00:34:17
times. You know how to do it. We're okay. We know what
Speaker:
00:34:21
stores and stuff you're walking past. So what kind of sounds can we play? Right
Speaker:
00:34:24
now, it's a bunch of text to speech. Essentially, the way I've rigged it up,
Speaker:
00:34:28
the the old version 0, the alpha version is, you know,
Speaker:
00:34:32
we asked chat g p t or OpenAI API what to say. So it's
Speaker:
00:34:35
basically like you're you're walking down the street hearing,
Speaker:
00:34:39
content From OpenAI. Interestingly, OpenAI
Speaker:
00:34:43
seems to the the GPT seems to know,
Speaker:
00:34:46
stuff about Every place along the way, like, you don't
Speaker:
00:34:50
have to go into, like, location based database.
Speaker:
00:34:54
It seems to seems to know quite a bit. There is a question of the
Speaker:
00:34:57
all the content is there's some interesting content in there, but it all ends up
Speaker:
00:35:01
being kind of mediocre. So it's like, okay, well, how do we turn this into
Speaker:
00:35:04
something really cool? I think, you know, in the end, having,
Speaker:
00:35:09
you know, you know, maybe music and and and speeches and an art
Speaker:
00:35:13
project somehow in there, based on where you walk is an interesting
Speaker:
00:35:16
idea. So if I could That'd be cool. Yeah. I could be like a
Speaker:
00:35:20
platform that people can use, like a cultural version of
Speaker:
00:35:23
Foursquare. Yeah. Yeah. And or maybe it's just
Speaker:
00:35:27
like an enhancement of the the sounds of the city. Or maybe
Speaker:
00:35:31
it's, You know, I mean, a lot of people think, okay, maybe maybe a tour
Speaker:
00:35:34
guide. I I don't know. But, you know, it it's it feels like,
Speaker:
00:35:40
It feels like there needs to be, a variety
Speaker:
00:35:43
of use cases tried because there's there's a lot you could do with it. And
Speaker:
00:35:47
and Maybe, you know, if if you put this in the hands of more
Speaker:
00:35:51
creative or of of additional creative people, they would,
Speaker:
00:35:54
ultimately find something interesting. I'm also working
Speaker:
00:35:58
yeah. Oh, I could answer questions about that. But then my other project is my
Speaker:
00:36:02
other 2 projects are are kind of interesting as well. Well, I have the
Speaker:
00:36:05
podcast, The Local Maximum. So still doing that every week and, you know,
Speaker:
00:36:09
interviewing people. Talking about,
Speaker:
00:36:14
talking about data, talking about AI, you know, few episodes on the
Speaker:
00:36:18
whole. You know, all the drama around OpenAI recently.
Speaker:
00:36:24
I I never wanted to become kind of the the the,
Speaker:
00:36:27
the the the tech drama, you know, what's it called?
Speaker:
00:36:31
TMZ of technology? Yeah. Yeah. But but that's something that happened because
Speaker:
00:36:35
I remember, like, last year, a couple years ago, there was all this craziness coming
Speaker:
00:36:38
out of Google with, You know, there was 1 guy at Google who said,
Speaker:
00:36:42
you know, he thought that the LLM has come to life. And Oh, yeah. And
Speaker:
00:36:46
then and then there was a there was A whole
Speaker:
00:36:50
bunch of stuff with, like, the the AI safety, you
Speaker:
00:36:53
know, seemingly staffed
Speaker:
00:36:56
by, people who are a little nutty. And
Speaker:
00:37:00
so, it was a And they fired a bunch of
Speaker:
00:37:04
people From that team too. So, like, there's
Speaker:
00:37:08
definitely, it was something weird some weird mojo
Speaker:
00:37:11
was going around. That's for sure. Yeah. And when when I cover that, I
Speaker:
00:37:15
mean, it's hard to, you know, it's hard to hide the fact where it's like,
Speaker:
00:37:18
wow, everyone in this story seems kinda nutty. But I also try to, you know,
Speaker:
00:37:22
I try to take a step back and say, okay, this is what we know.
Speaker:
00:37:25
These are a few things that could be happening internally, but we don't know everything.
Speaker:
00:37:28
I'm not gonna jump to conclusions. But,
Speaker:
00:37:32
I I I I try when I'm covering a story in a local maximum to
Speaker:
00:37:35
give, like, a a balanced, a balanced version of
Speaker:
00:37:39
of whatever story I come across. You know, maybe it's my show as I try
Speaker:
00:37:42
to give my opinion. But, yeah, I I
Speaker:
00:37:46
I my attempts, which, you know, some people have have,
Speaker:
00:37:51
said I I I've captured that. But my my attempt is to sort
Speaker:
00:37:54
of, try to try to
Speaker:
00:37:58
approach each Story with a little bit of humility and try
Speaker:
00:38:02
to help people understand what's going on without the
Speaker:
00:38:05
raw emotion that you get often on on Twitter. Gotcha. That's a good
Speaker:
00:38:09
point. Yeah. So we have, go ahead. I'm
Speaker:
00:38:13
sorry. Oh, no. No. It's okay. Go ahead. Okay. So we got,
Speaker:
00:38:17
3 complete dishonest. And, the first is when
Speaker:
00:38:20
I'm not working, I enjoy blank. Right.
Speaker:
00:38:24
So now that I've moved to Connecticut, I feel like I
Speaker:
00:38:27
am such a a Connecticut stereotype where I kinda, like, Drive
Speaker:
00:38:31
around, going for walks in the woods and into
Speaker:
00:38:36
various malls and stuff. So it's like it's like it's When the
Speaker:
00:38:39
weather's good, you go into the woods. When the weather's bad, you go into the
Speaker:
00:38:42
mall. Yeah. So I I actually like enjoy doing
Speaker:
00:38:46
that. I enjoy listening to podcasts.
Speaker:
00:38:49
I, honestly, enjoy hanging out with friends. You know,
Speaker:
00:38:53
after, I used to live in New York City. I enjoyed
Speaker:
00:38:57
it a lot, and I sorta had this, situation where
Speaker:
00:39:01
I had this be careful what you wish for because, at the end of 2019,
Speaker:
00:39:05
I was like, oh my god. I'm going to, like, events every single day. It's
Speaker:
00:39:08
just it's just too much. How can we, like, how can we, you know,
Speaker:
00:39:12
cut back on that. And then COVID came. And then to me, it was just
Speaker:
00:39:16
it was the worst thing because it was like, okay, you stay in your apartment
Speaker:
00:39:19
in New York City all day and you don't go and and talk to anyone.
Speaker:
00:39:23
And it was just like it it it was just awful. It just
Speaker:
00:39:27
felt like a a prison. So I I
Speaker:
00:39:31
moved to New Hampshire for a couple years, then I came back. But, you
Speaker:
00:39:34
know, nowadays, when I get a chance to hang out with with friends and and
Speaker:
00:39:38
family, I just I try to do it, whenever I can because I'm
Speaker:
00:39:42
not like, you know, it's not like when I was living in New York in
Speaker:
00:39:45
the 2010s and got kind of overload on that.
Speaker:
00:39:50
Right. Right. So, yeah. That's my answer there. And we have another complete this
Speaker:
00:39:53
sentence. I think the coolest thing in technology is blank.
Speaker:
00:40:03
The the way I've been putting it recently Is this,
Speaker:
00:40:07
where, you know. It.
Speaker:
00:40:13
You know, back maybe 10 years ago, the story we
Speaker:
00:40:16
were getting that the hopeful story we were getting was that, okay, if you're an
Speaker:
00:40:20
engineer, you could Build anything you want at
Speaker:
00:40:24
a very low cost or if you're not an engineer for anyone because we
Speaker:
00:40:28
have access to social media. You know, you can,
Speaker:
00:40:31
you can put anything out there into the world that you want
Speaker:
00:40:35
and and have people read it if if if they want to, or have people
Speaker:
00:40:39
look at it if they want to. And so that was kind of the new
Speaker:
00:40:42
exciting world. I think today, The new exciting
Speaker:
00:40:46
world goes well beyond that, which is going to be like,
Speaker:
00:40:51
you you can create worlds. Any Any world that you wanna
Speaker:
00:40:55
build, any scenario that you can imagine, you
Speaker:
00:40:59
can just have a machine fill in all the gaps for you and, You
Speaker:
00:41:03
know, write the write the story, make the
Speaker:
00:41:06
image and maybe, like, you know, make the make the video, make the whole
Speaker:
00:41:10
world. So I think, I I I think the
Speaker:
00:41:14
idea with generative AI that I want people thinking about more that that I
Speaker:
00:41:18
I also wanna think about more is, like, Okay. If you could create any world
Speaker:
00:41:21
you want, to explore, to live in, just
Speaker:
00:41:25
to, you know, maybe it's something to to teach us about something. Maybe it maybe
Speaker:
00:41:28
it's just an artistic adventure
Speaker:
00:41:32
venture, you know, what kind of world do you want to create because that's
Speaker:
00:41:35
that's going to become very cheap very quickly. Yeah. I could
Speaker:
00:41:39
see that. So I'm gonna skip to,
Speaker:
00:41:43
share something different about yourself. But we remind our guests
Speaker:
00:41:46
to remember it's a family podcast.
Speaker:
00:41:52
Okay. And I'm, you know, I'm I'm trying to,
Speaker:
00:41:58
I'm I'm trying to think of an answer here, and it's not because,
Speaker:
00:42:02
it's it's not because of the, of the of the caveat there,
Speaker:
00:42:06
but it No. No. I get it. Well, you've already covered a lot
Speaker:
00:42:09
that's It's different. I did. It is good stuff. Yeah. I mean, I
Speaker:
00:42:13
think, I think one thing that, It's,
Speaker:
00:42:17
I I enjoy doing that that that I forgot to mention, because
Speaker:
00:42:21
I'm actually doing it again for the 1st time in in in 6 years was,
Speaker:
00:42:25
I was a member of the Yale Alumni Service Corps. It's not
Speaker:
00:42:28
a member. It's like you can do a a it was essentially we were doing
Speaker:
00:42:32
trips to underserved Communities around the world and,
Speaker:
00:42:35
you know, doing little, like, kinda, kind of service
Speaker:
00:42:39
trips where you'd ever Either build a structure or work with small business
Speaker:
00:42:43
owners or go, you know, teach in a school. And so I've been
Speaker:
00:42:47
to, Nicaragua and Ghana And I actually
Speaker:
00:42:50
got to lead one of their trips in 2017 and that was to the Fort
Speaker:
00:42:54
Mohave Indian reservation. Very different kind of a trip because
Speaker:
00:42:58
it was Within the United States here. And
Speaker:
00:43:02
so it was honestly a lot easier. Because that's very cool
Speaker:
00:43:05
flying to Vegas. But yeah. But we're actually going back there, in in a few
Speaker:
00:43:09
months after 6 years. And so I was so even though it was less
Speaker:
00:43:13
convenient this time around, I'm I'm very excited Do that. And so
Speaker:
00:43:17
I I don't know. I really like learning about different cultures,
Speaker:
00:43:21
different philosophies, different religions. I think A lot of people might
Speaker:
00:43:24
assume given the you know given the tenor of my
Speaker:
00:43:28
podcast that I'm very like you know rationalist and I talk about Bayesian
Speaker:
00:43:32
inference a lot. But, I I
Speaker:
00:43:36
sort of venture out of that a lot. I don't think that,
Speaker:
00:43:40
That raw math can, can explain everything in life. And I also love like the
Speaker:
00:43:43
diversity of, of cultures and stuff. So No. It's cool.
Speaker:
00:43:47
That so that's maybe a positive thing, so I I don't know. It's very positive.
Speaker:
00:43:50
Something different. No. It definitely is. It definitely is. So where can
Speaker:
00:43:54
folks find out more about you and what Sure. So you mentioned you have a
Speaker:
00:43:57
podcast, which I love the name, The Local Maximum.
Speaker:
00:44:01
Right. Yeah. Local maximum's triple entendre, because it's got my name, Max.
Speaker:
00:44:05
It's a local maximum is, of course, you know, in machine learning,
Speaker:
00:44:10
when you're when you're trying to find the, well, sometimes it's often the local minimum
Speaker:
00:44:13
if you're trying to, Minimize the the loss function, but the
Speaker:
00:44:17
in basing inference, if you're trying to maximize the probability, whatever, you're you you get
Speaker:
00:44:21
stuck stuck in one of your next which was Your name is Max. So Yeah.
Speaker:
00:44:24
Exactly. Right. Right. That's the first one. That's the second one. And then, you know,
Speaker:
00:44:28
I I worked on location data a lot, so it's kind of a a triple
Speaker:
00:44:30
meaning. And so, I've
Speaker:
00:44:34
been doing that for for quite a while. You can go back into kind of
Speaker:
00:44:38
a a really extensive library there. And, I have
Speaker:
00:44:41
the website local max radio.com. I have.
Speaker:
00:44:47
If you go, I have local maximum labs. If you go to local maximum local
Speaker:
00:44:51
max radio.com/labs, I have a
Speaker:
00:44:54
bunch of papers And, you know, kind
Speaker:
00:44:58
of works that I've done, which, you know, includes some
Speaker:
00:45:02
discussion of machine learning, like kind of the math mathematics behind bias correction,
Speaker:
00:45:05
but But also something kind of fun that I did, like, with the podcast last
Speaker:
00:45:09
year, which is, like, I just rewrote the US constitution, fixed a bunch of
Speaker:
00:45:13
things just because I I felt that was fun. I was Taken aback by how
Speaker:
00:45:17
mad people get when you when you do that, it's not like I was actually
Speaker:
00:45:19
trying to, you know, run a political campaign for it. I just thought it was
Speaker:
00:45:23
a a fun project, and I learned a lot. But Some once you venture into
Speaker:
00:45:27
the political, people start treating things different. People get angry pretty quick.
Speaker:
00:45:30
Yes. That's true. Yeah. Yeah. I I I The
Speaker:
00:45:34
I I I love to hear criticisms on it. I wanna hear what what what
Speaker:
00:45:38
people think. But The one criticism
Speaker:
00:45:42
that I get a lot, which I really hate is, like, how dare you spend
Speaker:
00:45:44
your free time on this, which I I just don't get at all.
Speaker:
00:45:48
Yeah. But, which, you know, whenever I put
Speaker:
00:45:52
out some kind of math paper, even if it's like and there there is one
Speaker:
00:45:55
called relative probability, which is, you know, sort
Speaker:
00:45:59
of an abstract paper where it's like, okay, a reimagined probability
Speaker:
00:46:03
theory as, okay, let's say you can't talk about The probability of something
Speaker:
00:46:06
happening. Let's say you can only talk about 1 probability relative
Speaker:
00:46:10
to another. What, what does that look like? And I just stated some basic facts
Speaker:
00:46:14
and, you know, Not that many people gonna use it. Maybe people
Speaker:
00:46:18
won't use it for for a while. I I feel like it's an interesting idea,
Speaker:
00:46:21
and I feel like it will have uses eventually. But,
Speaker:
00:46:25
You know, nobody criticized me for that. For like, how dare you spend your free
Speaker:
00:46:28
time on that? Exactly. They they pick on you for the other
Speaker:
00:46:32
stuff. Yeah. I mean, I I look at people. I mean, you spend your free
Speaker:
00:46:35
time yelling at people on Twitter. I mean, what's the difference? I was gonna say
Speaker:
00:46:38
you can you can look at TikTok, and you can find far more destructive uses
Speaker:
00:46:42
is of Exactly. Exactly. So that's so that's that's my
Speaker:
00:46:46
main thing. I I think maybe with the, with the Constitution, I think people have
Speaker:
00:46:50
their sort of ideal society in mind. And if If your thing doesn't wind up
Speaker:
00:46:54
with that, they they kind of perceive you as a threat. Like you're trying to,
Speaker:
00:46:57
like I was trying to revitalize democracy, but some people are saying, no, you're
Speaker:
00:47:01
backsliding on democracy. Alright. Like, let's talk about it. But, yeah, it's it's
Speaker:
00:47:05
people get you know, people get different. We need to have you back and talk
Speaker:
00:47:08
about that more. Yeah. For sure. For sure. Talk about that. Absolutely.
Speaker:
00:47:12
We'd love having you. Both Andy and I, however, do have a hard stop, and
Speaker:
00:47:16
I would love this This covers you to go out for a couple hours, and
Speaker:
00:47:18
we'll talk to you more. And I just had a a conversation last night with
Speaker:
00:47:22
my cohost that went a couple hours. I know how it goes. Yeah. Yeah. Yeah.
Speaker:
00:47:26
We ended at 1 AM, and I was like, oh my god.
Speaker:
00:47:31
Well, those 1 AM conversations. I know what you mean. You got it. So
Speaker:
00:47:35
With that, we'll definitely make sure. Send us all your links, and
Speaker:
00:47:39
we'll make sure we get them in the show notes, and we'll let Bailey, our
Speaker:
00:47:42
semi extension AI host, Co host, 3rd
Speaker:
00:47:45
host, wrap up the show. And thank you, dear
Speaker:
00:47:49
listener, for subscribing to our podcast. You
Speaker:
00:47:53
have subscribed to us, haven't you? Once you do,
Speaker:
00:47:56
please be sure to rate and review our podcast on iTunes, Stitcher,
Speaker:
00:48:00
or wherever you subscribe to us. Having high ratings
Speaker:
00:48:04
and reviews helps us improve the quality of our show and rank us more
Speaker:
00:48:08
favorably with the search algorithms. That means more
Speaker:
00:48:11
people listen to us, spreading the joy. And,
Speaker:
00:48:15
can't the world use a little more joy these days?
Speaker:
00:48:19
So, go do your part to make the world just a little better and be
Speaker:
00:48:22
sure to rate and review the show.
00:00:00
In today's episode, Frank and Andy sit down with special guest
Speaker:
00:00:03
Max Sklar to delve into the world of artificial intelligence, data
Speaker:
00:00:07
science, and data engineering. Max, a
Speaker:
00:00:11
trailblazer in location data and machine learning, shares
Speaker:
00:00:14
insights from his extensive experience at Foursquare, including his
Speaker:
00:00:18
work on local search and bias correction. Get ready
Speaker:
00:00:22
for a thought provoking discussion about groundbreaking projects, tech
Speaker:
00:00:26
drama, and the ever evolving landscape of technology.
Speaker:
00:00:30
So sit back, relax, and prepare to be amazed.
Speaker:
00:00:38
Hello, and welcome to Data Driven, the Podcast, we explore the
Speaker:
00:00:42
emergent fields of artificial intelligence, data science, data engineering,
Speaker:
00:00:46
and all that good stuff. With me on this ever present,
Speaker:
00:00:51
journey down the information superhighway is Andy Leonard. How's it going, Andy?
Speaker:
00:00:55
Good, Frank. How are you doing? I'm doing alright. It's been a wild 24
Speaker:
00:00:59
hours in, Maison Levin, or or
Speaker:
00:01:02
Maison, Lavinia, depending on your, how you wanna pronounce it.
Speaker:
00:01:07
At one point, we will share those crazy
Speaker:
00:01:11
details, but it's been good most of the part. I I am,
Speaker:
00:01:14
recovering from a, a bout of COVID,
Speaker:
00:01:19
that hit the entire house. I think we all picked it up on the cruise.
Speaker:
00:01:23
Again, there are worse places to pick up COVID, and there's worse
Speaker:
00:01:27
things that could happen. I've I've sneezed quite a
Speaker:
00:01:31
bit, I've coughed quite a bit, but The thing that's bugging me the most is
Speaker:
00:01:34
this headache I had now for 48 straight hours.
Speaker:
00:01:38
But it's okay, like I'm kind of living, learning to live with it, and I've
Speaker:
00:01:41
actually given it a name, I call it Charlie. So, you
Speaker:
00:01:45
know, Charlie is is is gonna be on the show today as
Speaker:
00:01:48
well. I see. Yeah. You do sound a little different, but not much.
Speaker:
00:01:53
Nah. That's why I need to clone my voice, and, maybe how they
Speaker:
00:01:56
do the avatar stays. As I was saying in the virtual dream room, this is
Speaker:
00:02:00
the 1st time I've I've felt fit for camera In about
Speaker:
00:02:03
a week. Perhaps you can use some of that good
Speaker:
00:02:07
AI voice modulation. There you go. We are
Speaker:
00:02:11
pleased to present, with us, today is Max Sklar.
Speaker:
00:02:14
Max is a, not only a fellow, data
Speaker:
00:02:18
guy, but also a Fellow podcaster.
Speaker:
00:02:22
Welcome to the show, Max. Thank you so much, Frank and Andy,
Speaker:
00:02:26
for, for having me on. I'm looking forward to it. And,
Speaker:
00:02:30
yeah. Thank you. Cool. So you you've been at you were at a
Speaker:
00:02:34
company for a number of years. You were talking about this in the virtual dream
Speaker:
00:02:37
room. You were at Foursquare for, oh, quite some time. Right.
Speaker:
00:02:41
That's unusual, actually. 10 years at a at a company as as, you
Speaker:
00:02:44
know, building software is pretty unusual. But I was really there For, I
Speaker:
00:02:48
think, like, 3 different phases, so I kind of break it up into 3
Speaker:
00:02:52
jobs. Interesting. So, you know,
Speaker:
00:02:55
Foursquare is one of those companies that
Speaker:
00:03:00
On when they broke up into 2 different parts, it kind of I didn't
Speaker:
00:03:04
understand it. Like, this is just user. I'm not I'm not I know 1 I
Speaker:
00:03:07
know that sounds terrible, but despite being my headache, Charlie talking, but, like,
Speaker:
00:03:13
what, you know, obviously, that was a decision
Speaker:
00:03:16
that was made in the marketing department, and I don't think people probably thought that
Speaker:
00:03:19
through. But I love that location. Like, I I think, you know, like, I was
Speaker:
00:03:23
the mayor of, like, 5 6 different places. It was such a cool
Speaker:
00:03:26
creative, concept that if you go there long
Speaker:
00:03:30
enough, you're the mayor, and some places would offer the mayor, like, a free coffee
Speaker:
00:03:34
and a donut or something like that. Like, it was really clever.
Speaker:
00:03:38
I know. I used to I mean, that was one of the things that got
Speaker:
00:03:41
me really excited about it back in the day. And this is already 10 years
Speaker:
00:03:43
ago. I assume you're talking about the, The app split back in
Speaker:
00:03:47
2014, which I guess is That sounds about right. I I guess
Speaker:
00:03:51
almost 10 years. I guess it's It it's actually been 10 years since we started
Speaker:
00:03:55
working on that, which was also how it worked. But,
Speaker:
00:03:58
no. I I mean, I actually remember a lot of what was going on
Speaker:
00:04:02
at the time, and, we could talk about that a little more in a
Speaker:
00:04:06
second. I I I I I'm happy to talk about that.
Speaker:
00:04:10
I I think the the whole gamification thing was really
Speaker:
00:04:14
exciting. And I really loved the idea that I can go into
Speaker:
00:04:18
a place and be like, hey. I was here for 5 times. Oh, we're gonna
Speaker:
00:04:20
give you some free, you know, free, you know, dessert or
Speaker:
00:04:24
free chicken nuggets or whatever. Maybe these days, you know, kind of trying to watch
Speaker:
00:04:27
what I eat a little bit more, it might not be as exciting. I I
Speaker:
00:04:30
it was really sad that that didn't scale as a business. I think what
Speaker:
00:04:34
would happen was, and and this I both experienced
Speaker:
00:04:38
this personally. And also from talking to the leadership there, it was clear
Speaker:
00:04:42
that they felt it it, You know, it it it wasn't a
Speaker:
00:04:45
sustainable thing because you'd often go into a place and say, hey, you know,
Speaker:
00:04:49
I checked in here all these times. I got all these rewards. You know,
Speaker:
00:04:53
I'm supposed to get, like, a dollar off, or I'm supposed to get some
Speaker:
00:04:56
some free dessert or something, which is It's always exciting
Speaker:
00:05:00
to get a little reward, even if it's not, you know, even if it's just
Speaker:
00:05:02
a dollar or whatever. It's always but I think what would end up happening is
Speaker:
00:05:06
like the the man, the The management who worked there or the or the
Speaker:
00:05:10
bartender or whatever would know about it. And, you know,
Speaker:
00:05:14
once a few times that, you know, you got that thing. Oh, oh, let me
Speaker:
00:05:17
go in the back and check. And then they come back in, like, 20 minutes,
Speaker:
00:05:19
and it's like a whole big deal. They're like, maybe I don't wanna do this
Speaker:
00:05:22
again. That's interesting, because, like, from my point of view,
Speaker:
00:05:26
it was always very It was very fun, like, so I was mayor of,
Speaker:
00:05:30
there used to be a ferry service between, in the
Speaker:
00:05:34
suburbs of DC, between Poolesville, Maryland and
Speaker:
00:05:37
Leesburg, Virginia. And, I would take that ferry a lot so
Speaker:
00:05:41
much. So I was actually the mayor of the ferry. It didn't get
Speaker:
00:05:45
me anything, like, in that case, but it was this kind of cool, like I
Speaker:
00:05:48
I love to I I always imagined, like, there would be, You know, like
Speaker:
00:05:52
the ferry could have some kind of a a little like TV screen that could
Speaker:
00:05:55
like show who the mayor was. And then you can, you know, and and and
Speaker:
00:05:59
then then then we can like kind of scale up those fights, But alas,
Speaker:
00:06:03
society didn't go in that direction. Interestingly,
Speaker:
00:06:07
Foursquare, what, like, I have been interested in that this like local
Speaker:
00:06:11
local search space since since well before forest grabbing even my, you
Speaker:
00:06:14
know, my my senior project as an undergrad, back,
Speaker:
00:06:18
though, starting in 2005, was this, like, you know, this this website
Speaker:
00:06:22
called sticky map, where people would post little, Icons all
Speaker:
00:06:25
over Google Map. It's kind of inspired by Wikipedia. You can add messages.
Speaker:
00:06:29
And I just thought it was pretty cool. People started marking up the, the campus,
Speaker:
00:06:32
and then people started, you know, marking up. It was like, well, it's based on
Speaker:
00:06:36
Google Maps API, so you could just mark up anything you want. And
Speaker:
00:06:40
I think in that first project, I noticed All the problems
Speaker:
00:06:44
that still exist with that kind of data today. Okay. What happens when you
Speaker:
00:06:48
add duplicates? It's the first thing that happened. As, you know, As an undergrad, I
Speaker:
00:06:51
was like, oh, I'm so excited. Let me show something like this. They're like, okay,
Speaker:
00:06:54
let me create a marker. And I'm like, don't create that one. It's already been
Speaker:
00:06:58
created. And then it's like, okay, now we have duplicates Right off the bat. And
Speaker:
00:07:01
that is still something that, you know, Foursquare
Speaker:
00:07:05
deals with and I'm sure Google deals with and Apple Maps deals with and then
Speaker:
00:07:09
they they all deal with it. So,
Speaker:
00:07:14
Yeah. It it it it I I think when I
Speaker:
00:07:17
discovered Foursquare, you know, several years later,
Speaker:
00:07:22
it was the it was innovative in several ways. It was, first of
Speaker:
00:07:25
all, it was based on on mobile apps actually being at the place when you're
Speaker:
00:07:29
commenting on it, which is exciting. It was the gamification of it. And the
Speaker:
00:07:33
fact And and for those listening Yeah. For those listening, that was new. Like,
Speaker:
00:07:36
that was brand new. So sorry. I didn't mean to cut you off, but, like,
Speaker:
00:07:40
the context is important because All I had was was something on a on
Speaker:
00:07:43
a website. You know, I I wasn't thinking of the I the iPhone didn't
Speaker:
00:07:47
exist, for actually, pre Foursquare,
Speaker:
00:07:51
there there was something called dodgeball, which was kind of the the
Speaker:
00:07:55
predecessor to Foursquare, Which I wasn't involved with, but it but, it
Speaker:
00:07:59
it was based on, like, you know, SMS kind of text messages where people were
Speaker:
00:08:02
messaging on. If you remember, You know, those, you know,
Speaker:
00:08:06
the the the what was it? The t nine texting where people were Oh,
Speaker:
00:08:10
God. Yeah. Yeah. Yeah. But people would use that and and Foursquare and,
Speaker:
00:08:14
Google bought that. And then the the the founders,
Speaker:
00:08:19
Dennis Nautz went it went in and started Foursquare after that.
Speaker:
00:08:23
So, very, there there's a very
Speaker:
00:08:27
interesting history of kind of, like, Local search, city guide,
Speaker:
00:08:31
and basically sort of social kinda local applications. We're
Speaker:
00:08:35
we're very big at the time. Nowadays, I think we need to find a new
Speaker:
00:08:38
take on it, but, that when I joined Forescord
Speaker:
00:08:42
in 2011, it was very exciting. I'm glad you mentioned that because there was
Speaker:
00:08:46
a a real so I started my career in New York City. Right?
Speaker:
00:08:49
So I worked at Barnes and Noble com. So I was there in the fairly
Speaker:
00:08:52
early days of Silicon Alley, and that was a huge thing. It was
Speaker:
00:08:56
Microsoft. I think it was Microsoft had something called Sidewalk.
Speaker:
00:09:01
And then there was there was maybe it was maybe it wasn't Microsoft. Maybe
Speaker:
00:09:05
somebody else had something called sidewalk AOL had something, going and the fact
Speaker:
00:09:09
you can't remember it, I think says it all. Right? Like, it was like in
Speaker:
00:09:12
in, you know, anybody that could register a .com could spell and could
Speaker:
00:09:16
spell HTML To get funding back in those days, a
Speaker:
00:09:19
bit like the way the AI startup ecosystem is kinda
Speaker:
00:09:23
today. But, no, there was I mean,
Speaker:
00:09:26
people there was a time, young children
Speaker:
00:09:30
out there, That when, you know, people saw the online
Speaker:
00:09:34
world is slightly different than the real world, and they saw this as an opportunity.
Speaker:
00:09:38
Right? But no, I'd like That takes me back. As soon as
Speaker:
00:09:42
you said to, like, the local kind of connection guides, I was like, wow, it
Speaker:
00:09:45
takes me back. Yeah. Yeah. Coming back to
Speaker:
00:09:48
the app split, and I wasn't expecting to talk about this today. I'm sorry if
Speaker:
00:09:52
it brings up If for what it's worth, I'm a former Windows For what it's
Speaker:
00:09:56
worth, I'm a former Windows phone developer, and I wrote a book on silver light.
Speaker:
00:09:59
So I understand the pain of working on
Speaker:
00:10:03
It'll fade and I worked at barnes and noble.com. Right? So there's my trifecta of
Speaker:
00:10:07
ill fated technology projects. Yeah. I I think it's
Speaker:
00:10:12
A lot of technology companies, in order to become successful, actually have
Speaker:
00:10:16
to go through big changes where people yell at them.
Speaker:
00:10:20
And so it's like, how do you know whether you're breaking things or whether you're
Speaker:
00:10:23
actually doing what you're supposed to do? And so that's kind of a Right. That's
Speaker:
00:10:26
that's kind of a tough decision. I I think For Foursquare
Speaker:
00:10:30
at the time, there was always, like, kind of a design,
Speaker:
00:10:34
and product, like, tension between the people who
Speaker:
00:10:38
wanted to be there As essentially like a Yelp replacement, kind of
Speaker:
00:10:41
like a a local search city guide. And then there and then versus the people
Speaker:
00:10:45
who were there for the the life logging, the check ins, the game. And I
Speaker:
00:10:48
think, I think the separation could have been
Speaker:
00:10:52
done. I mean, my personal thing is I think there could have been a
Speaker:
00:10:56
Separation, I could think, could have been executed a little bit better. I
Speaker:
00:10:59
think technically we did a good job. I think the apps that
Speaker:
00:11:03
we ended up with were well designed, but I think, I think
Speaker:
00:11:07
we needed to do is take into account how the how people were using the
Speaker:
00:11:10
apps, at the time and not just, like, Kick all the
Speaker:
00:11:14
people who are checking in, which was which was Foursquare's kind of bread and butter
Speaker:
00:11:18
and just, like, kick them to the side with this other app. And then it's
Speaker:
00:11:20
like, well, what is this? I'm calling this something different. Sorta. That was that was,
Speaker:
00:11:23
I think, too much. But, again,
Speaker:
00:11:27
there there were people saying it at the time. But
Speaker:
00:11:31
The problem is, I I guess, you know, there whenever you make a
Speaker:
00:11:35
change, there's always a great many people saying a great many things. So
Speaker:
00:11:40
We could wax this we could wax nostalgic about Foursquare because I used to
Speaker:
00:11:44
like, when I I was travelling a lot at the time when I worked I
Speaker:
00:11:47
worked at Microsoft about 10 years ago. But I remember Sometimes I would actually
Speaker:
00:11:50
choose different connecting airports, so I could get, like, the the
Speaker:
00:11:54
jet that was it, the the jet set tag, like level up in my achievement
Speaker:
00:11:58
there. Right. Which is kinda sad, but, we could
Speaker:
00:12:02
we could wax nostalgic about that all day, and I would love to. But I
Speaker:
00:12:05
think what What was the role of AI and ML in that
Speaker:
00:12:09
space? Right? Because you're obviously collecting a lot of data. No. Like, I'm just curious,
Speaker:
00:12:12
like, because how how was that being used? How was that,
Speaker:
00:12:17
leverage. In in in a lot of ways,
Speaker:
00:12:20
and, you know, many of which I I worked on over all those years.
Speaker:
00:12:25
You know, one of them was, I mean, just, you know, search
Speaker:
00:12:28
ranking in general, which, you know, Foursquare had a lot of ex Google engineers.
Speaker:
00:12:32
So I learned directly from them so they they knew what to do.
Speaker:
00:12:36
But search ranking, search ranking was a was a was a big project.
Speaker:
00:12:40
This is kind of more of a statistical Problem where you were kinda trying
Speaker:
00:12:44
to weight different attributes, like, is this related to the search the
Speaker:
00:12:47
person put in? Is this related to how much do we wanna, You know,
Speaker:
00:12:52
score things that are, you know, maybe someone's
Speaker:
00:12:56
friends went to. So something like that. I think that
Speaker:
00:13:00
The biggest well, I'll talk about the one that I think is the biggest deal
Speaker:
00:13:03
and then the one that that I worked on the most. The one I think
Speaker:
00:13:06
was the biggest deal for for Foursquare, which I did work on a little
Speaker:
00:13:10
bit, is basically trying to figure out where
Speaker:
00:13:13
someone is given. So we know where someone is given there that long
Speaker:
00:13:17
from their phone. But it's like, what are they actually in a particular
Speaker:
00:13:20
store? Like, are they in the Starbucks? Are they in the, you know, are they
Speaker:
00:13:24
in the office, over there? Are they Are they just walking down
Speaker:
00:13:28
the street? And so using the fact that people were were
Speaker:
00:13:31
giving us training data, essentially, which was a big theme there, which is, you know,
Speaker:
00:13:35
I think, something that,
Speaker:
00:13:40
data scientists and data entrepreneurs need to need to look in closely, which is
Speaker:
00:13:43
like, How can you get people to give you training data? Because it is really
Speaker:
00:13:46
useful. So if you have people giving you where they are and
Speaker:
00:13:50
then you could see the information from their phone, not just a lot long, but
Speaker:
00:13:53
like what, You know, things like what Wi Fi's can you see? What, you know,
Speaker:
00:13:56
other sensors from your phone, can you figure out where they are? And then there's
Speaker:
00:14:00
the whole stop detection, problem. And so, Yep.
Speaker:
00:14:04
Foursquare essentially can kinda figure out, you know, where you went day
Speaker:
00:14:08
to day, and it's actually pretty good. Like, you know, if I Don't tell Foursquare
Speaker:
00:14:11
where I went. Even today, I still look at it and, you know, it tells
Speaker:
00:14:15
me what, what actual stores I was in. Now maybe there's a question of,
Speaker:
00:14:19
you know, whether whether our apps are knowing too much about us, but
Speaker:
00:14:22
that's that's a whole another question. But that was a very important,
Speaker:
00:14:27
a resource for the business. And the one that I worked on the most, that
Speaker:
00:14:30
was the most exciting though for me was the natural language
Speaker:
00:14:34
processing Pipeline. And, of course, you know, text text
Speaker:
00:14:38
data today is is is having such a a resurgence
Speaker:
00:14:42
with, You know, I don't need to tell your audience with AI and all that.
Speaker:
00:14:45
But, you know, it it it back then it was like, well, people were giving
Speaker:
00:14:49
us, you know, several sentences called tips on Foursquare Venues,
Speaker:
00:14:53
which would often be like, here's, you know, here's what you should do here. Here's
Speaker:
00:14:56
what you should try. Here's a little review, something like that. People were
Speaker:
00:15:00
leaving text with their check-in. So there's a bunch of texts, there's
Speaker:
00:15:03
menus, things like that. So there's a bunch of texts in the system. And so
Speaker:
00:15:07
it's like, what do we do with all of that? And, one of the things
Speaker:
00:15:11
that we did was we pulled out key terms, you know, noun phrase detection.
Speaker:
00:15:14
This is all kind of standard natural language
Speaker:
00:15:18
processing. You know, not
Speaker:
00:15:22
you know, you know, people often ask, oh, you know, I I think
Speaker:
00:15:26
Nowadays, I'm often thinking everyone's thinking, oh, you were probably using, like,
Speaker:
00:15:30
generative AI or something. No. It was just kind of standard NLP that had
Speaker:
00:15:34
been developed over the last, you know, several decades. But,
Speaker:
00:15:38
we did sentiment analysis and we used that to come up with the ratings for
Speaker:
00:15:41
the venues, which which are used today. So you could tell
Speaker:
00:15:45
how good something is. And,
Speaker:
00:15:49
you know, I did some things that were a little bit
Speaker:
00:15:53
more interesting that, you know, maybe get overlooked, but they're
Speaker:
00:15:57
they're kind of unique to To to what we did there, which was sort of
Speaker:
00:16:00
like timeliness and seasonality, which is so, like, if you check into
Speaker:
00:16:04
a diner in the morning versus in the afternoon, It'll statistically
Speaker:
00:16:08
give you different suggestions based on how timely it thinks each
Speaker:
00:16:11
each suggestion is. Because with every Check-in where someone is doing
Speaker:
00:16:15
something in real time. We have the timestamp. We know what time of day. We
Speaker:
00:16:19
know what time of week. We know what time of year. And so it's kind
Speaker:
00:16:22
of cool to to put that all together. And some of
Speaker:
00:16:26
the some of the, some of the
Speaker:
00:16:29
models, got pretty,
Speaker:
00:16:34
you know, it was it was pretty neat how it all turned out. I think
Speaker:
00:16:37
that one I you know, I still talk about that one is one of my
Speaker:
00:16:39
That's my favorite one after being in the industry so long, even though it was
Speaker:
00:16:42
like 10 years ago, because it was like, okay, we had training
Speaker:
00:16:46
data again from on these tips where, You know, we
Speaker:
00:16:50
could tell if the person liked the venue or disliked the venue and because
Speaker:
00:16:54
they they told us, and they also left the tip. There were a lot of
Speaker:
00:16:56
people who did that. So that just gave us training data for sentiment analysis.
Speaker:
00:17:00
And at the time, I'm sure the tools now are much more sophisticated at the
Speaker:
00:17:03
time when we use pretrained sentiment analysis tools, Didn't really
Speaker:
00:17:06
work well on our data because it's just it was just a different kind of
Speaker:
00:17:09
text. People wrote on Foursquare differently than they did on Twitter, for
Speaker:
00:17:13
example. So, so that gave us training data. Give
Speaker:
00:17:17
us training data for every language. And so that was nice. We got kind of,
Speaker:
00:17:20
like, you know, 90 languages for free just by just by
Speaker:
00:17:24
using that Strategy of Oh, wow. Using the data that people gave us.
Speaker:
00:17:27
Probably not probably didn't work very well in all 90, but certainly worked
Speaker:
00:17:31
well. Well, the beauty of it is It ends up working
Speaker:
00:17:35
well so long as we have good language detection, it ends up working well
Speaker:
00:17:38
in, any language that has any Particular,
Speaker:
00:17:43
you know, any particular popularity in Foursquare.
Speaker:
00:17:47
So for example, if, the Turkish was very popular. Okay. Well, that
Speaker:
00:17:51
means we have a lot of Turkish training data. That means that the
Speaker:
00:17:55
the model, which trains monthly, is Is going to use all that training data. That
Speaker:
00:17:59
means it's going to work very well. And so, and
Speaker:
00:18:02
so that the fact that the models were always regenerating
Speaker:
00:18:07
And they were always regenerating based on the latest data
Speaker:
00:18:11
was was really cool because oftentimes you think these think about
Speaker:
00:18:16
ML teams kind of building a model, and then they kind of throw
Speaker:
00:18:20
it over a wall. They they productionize it. And then you have to
Speaker:
00:18:24
work on the next one, but you have to you have to do some work.
Speaker:
00:18:27
It's not automated, you know. So it's like, well, this is this is gonna start
Speaker:
00:18:30
going downhill If we don't, if we don't interact. And the fact that we were
Speaker:
00:18:33
able to set it up where it was just constantly getting smarter was,
Speaker:
00:18:37
was pretty neat. So MLOps and pipelines before they were called
Speaker:
00:18:41
MLOps. Well, they might have been called pipelines, but yeah. Interesting. Yeah.
Speaker:
00:18:45
Pipeline was a big Big big key phrase. So what
Speaker:
00:18:48
what did the data what did the back it because like one of the one
Speaker:
00:18:51
of the jokes that we have, and in fact, it's a domain name that I
Speaker:
00:18:54
registered. 1st, you get the data, is a phrase that
Speaker:
00:18:58
a lot of data scientists will often use, much to the chagrin of a lot
Speaker:
00:19:01
of data engineers, because a lot of data,
Speaker:
00:19:05
you have to get the data in a certain way to to format it and
Speaker:
00:19:08
and and to get it trained. And if you go to first you get
Speaker:
00:19:12
the data.com, it should redirect you to our website,
Speaker:
00:19:15
hopefully. God only knows if it works. 1st, I think
Speaker:
00:19:19
yeah. I'm gonna I'm gonna try to get the data.com.
Speaker:
00:19:23
I'm shattered to think that okay. Good. It does work. Okay. DNS
Speaker:
00:19:27
and me have a long history. Yes. It's going back.
Speaker:
00:19:30
Good. I I it's always good to start off a week with a win with
Speaker:
00:19:33
DNS. What did it what did the
Speaker:
00:19:37
because I'm curious, like, Foursquare was one of those early, kind
Speaker:
00:19:41
of mobile first, kind of success stories. I'm
Speaker:
00:19:45
always curious, what did the back end data platform look like? Right?
Speaker:
00:19:49
Because, and again, going back 10 years, I
Speaker:
00:19:52
mean, I mean, did you use what was the name of the,
Speaker:
00:19:57
gosh. Can't think of the name of the platform, but what sorts of technologies did
Speaker:
00:20:00
you did you guys use? Yeah. I'm I mean, I'm sure
Speaker:
00:20:04
it works similar today in in at at Foursquare. Mhmm.
Speaker:
00:20:08
Well, we were using data pipe I assume. But, yeah,
Speaker:
00:20:11
if I remember correctly, we had, you know, our transactional database, our Mongo
Speaker:
00:20:15
database that was sort of like, Every once in a
Speaker:
00:20:19
while. And so that was kinda like the baseline. And then there'd be a series
Speaker:
00:20:23
of jobs that, like, you know, built it up, that that
Speaker:
00:20:26
that kind of Calculated things off of that, and that
Speaker:
00:20:30
would, in the at the end of that pipeline, you know, release
Speaker:
00:20:34
a, a dataset that would then be kind of,
Speaker:
00:20:38
Automatically, deployed and then read by,
Speaker:
00:20:42
read by the server in real time. So, if I can think of, like, the
Speaker:
00:20:46
technologies, I think the, the pipeline technology, the
Speaker:
00:20:50
pipeline, what was it? It was
Speaker:
00:20:53
like Luigi. It was written in pipe. Python. I don't know if that's too interesting.
Speaker:
00:20:57
There's a lot of different ones you could use these days.
Speaker:
00:21:02
It's an interesting question of, like, you know, which one do you use? I,
Speaker:
00:21:07
it's It's probably,
Speaker:
00:21:10
you know, from from my point of view, it's always like, well, the company kinda
Speaker:
00:21:14
chooses it. You don't really have much of a say.
Speaker:
00:21:19
And then then it's like, well, well, how do I know how to compare? But
Speaker:
00:21:22
let's see. Like, you know, we were using MapReduce jobs. We're using Hadoop At the
Speaker:
00:21:26
time, I think scalding, was was one that's that's maybe kind of out of fashion
Speaker:
00:21:29
now. That was a, a scala based framework for for
Speaker:
00:21:33
some of these jobs which were, which which
Speaker:
00:21:37
was based on abstract algebra. So it's actually pretty cool. I wish it
Speaker:
00:21:40
was. It was kinda hard to to reason about sometimes if you
Speaker:
00:21:45
It kind of went too far, to the side of, okay,
Speaker:
00:21:49
you know, I love abstract algebra, but I don't want everybody
Speaker:
00:21:53
who I don't I don't want that to be a barrier to entry for people
Speaker:
00:21:56
who are we're working on this. But, I'm
Speaker:
00:22:00
just trying to remember, like, some of the, You know, some of the some of
Speaker:
00:22:03
the tech bud buzzwords. But if you have any specific questions, maybe they'll jog jog
Speaker:
00:22:06
my memory. I don't know. Like, one of the things that was popular about that
Speaker:
00:22:09
time was, HBase. Oh,
Speaker:
00:22:13
yes. I, were we using interest? I think we were using,
Speaker:
00:22:18
Yeah, I remember that Term, but I I know. I know. I was as you
Speaker:
00:22:21
were talking, I'm like using it or if we wanted to use it. It was
Speaker:
00:22:24
one of those 2. Now from that if memory serves, I think Facebook is the
Speaker:
00:22:27
one who pioneered h base because it was really it was a right once read
Speaker:
00:22:31
many thing, and basically, the last one in when,
Speaker:
00:22:34
last last I can't talk, sorry. Last one wins.
Speaker:
00:22:38
Let Andy help might help me out if with the
Speaker:
00:22:42
technical term for that last one. Last one wins. What's the
Speaker:
00:22:46
Oh, yeah. So right. I remember they were called h files, so it must have
Speaker:
00:22:49
been yes. It must have been that. Yeah. Yeah. Yeah. That was one
Speaker:
00:22:53
of those sorry, Andy. Go ahead. That's okay. You were I was thinking,
Speaker:
00:22:57
you know, ChipLogic last in first out. Yeah.
Speaker:
00:23:01
Yeah. Something like that. Yeah. Know if that's what you were after or not. Last
Speaker:
00:23:04
one wins. Last one That was their concurrency strategy.
Speaker:
00:23:08
That's, I know there's a better term for that, but again,
Speaker:
00:23:13
it's a Monday and, I have
Speaker:
00:23:17
a headache. But no, it's it's it's it's
Speaker:
00:23:21
fascinating to kind of Almost like technology
Speaker:
00:23:25
archaeology. Like what worth the big projects
Speaker:
00:23:28
that were popular at the time. Right? You know, and it's
Speaker:
00:23:32
just, And it's scary to think that, you know, we're talking 10 years
Speaker:
00:23:36
ago. I mean, I mean, you Not even though. A lot of this stuff was
Speaker:
00:23:39
I mean, a lot of this stuff is probably still in place at Foursquare today.
Speaker:
00:23:43
Yeah. I mean, what's interesting to me is you you mentioned a lot of
Speaker:
00:23:47
the NLP, techniques that, You know, for lack
Speaker:
00:23:51
of better term, people would consider legacy now, right? Because they're pre
Speaker:
00:23:54
transformers, right? They're pre GPT, right? Sentiment
Speaker:
00:23:58
analysis, a lot of, you know, I I speak with a lot of people with
Speaker:
00:24:01
varying degrees of technology skills, and they
Speaker:
00:24:05
assume that this field of research didn't exist prior to
Speaker:
00:24:09
last year. And, very
Speaker:
00:24:12
much not the case. It's just that radically changed about a year ago.
Speaker:
00:24:17
Right. I mean and and this is something that I'm trying to figure out how
Speaker:
00:24:19
to do, which, I I might not be alone. It's like, okay. I did
Speaker:
00:24:23
all these things. How do I reinvent myself now in this new world?
Speaker:
00:24:27
And, you know, once you realize it could be exciting thing, then
Speaker:
00:24:31
it's maybe not so much of a drag, you know, because there's there's so many
Speaker:
00:24:34
opportunities out there. But it's like, but I can't be
Speaker:
00:24:38
the only one out there who's struggling with this being like, okay,
Speaker:
00:24:41
wow, I've got a, You know, I I've gotta,
Speaker:
00:24:45
you know, work or at least do projects for companies that
Speaker:
00:24:49
are at the cutting edge here in order to, In order to be,
Speaker:
00:24:53
you know, in order to be at the forefront. Yeah. It's funny, like, you miss,
Speaker:
00:24:57
like, I was, You know, offline for I tried to be offline, but for the
Speaker:
00:25:01
better part of a week and for vacation.
Speaker:
00:25:04
And like During that week, AMD announces that they are
Speaker:
00:25:08
producing their own, GPU LLM
Speaker:
00:25:12
type hardware. Gemini comes out and all
Speaker:
00:25:15
these other innovations that come out, and I'm like, I feel, like, hopelessly behind
Speaker:
00:25:19
now. I'm being offline for a week. Yeah.
Speaker:
00:25:23
Yeah. It's I mean, I I guess the
Speaker:
00:25:26
only, consolation there is everyone's dealing with that.
Speaker:
00:25:30
Right. You know? Yeah. Right. And Kinda like impostor
Speaker:
00:25:33
syndrome. Right? Yeah. Yeah. I think I think the
Speaker:
00:25:37
question is, especially in this new world of generative
Speaker:
00:25:41
AI. And and the question I'm asking I don't necessarily have answers. But it's
Speaker:
00:25:44
like, how do you so You wanna jump in the stream
Speaker:
00:25:48
and get all the latest stuff, but you also want to leverage your experience
Speaker:
00:25:52
and understanding. Cannot be leveraged. And
Speaker:
00:25:56
so what's the best way to to, you know, what
Speaker:
00:26:00
what's the best way to balance that? I think that's something that I would like
Speaker:
00:26:03
to see more people asking. And I would like guidance on this. I know I'm
Speaker:
00:26:06
the guest. I'm supposed to say what I know, but No. But try now. You
Speaker:
00:26:10
know, Mads, the, the thing is a lot of the stuff that's
Speaker:
00:26:13
That's new. I'm doing the air quotes here for people who are listening.
Speaker:
00:26:17
A lot of things that are new are really coming out of tech That was
Speaker:
00:26:21
developed. The math was developed, for instance, in the late sixties, seventies,
Speaker:
00:26:24
eighties, nineties. So a lot of that is just being reapplied
Speaker:
00:26:29
Back when the math was developed and the theorems and and such,
Speaker:
00:26:32
we didn't have machines fast enough to do it or at least do it
Speaker:
00:26:36
usefully. So I wouldn't feel bad at all about,
Speaker:
00:26:40
you know, having a bunch of, a bunch of experience that seems dated
Speaker:
00:26:43
right now because A couple of weeks to a couple of months. That might
Speaker:
00:26:47
be the new shiny. Right. That's true. When I went back to
Speaker:
00:26:51
a a university computer science program, You know, they're still studying
Speaker:
00:26:55
data structures and algorithms. It's still very relevant.
Speaker:
00:26:58
And, you know, I think a lot of outsiders think, oh, everything's
Speaker:
00:27:02
gonna Turnover in, in a year and a lot of things
Speaker:
00:27:06
do. But there are also a lot of kind of like universal,
Speaker:
00:27:10
kind of, there's a lot of universal Theory that's, good to know
Speaker:
00:27:14
about. Sure. The fundamentals don't change that often.
Speaker:
00:27:18
Nope. And it's a lot of reapplications. I see a lot of people
Speaker:
00:27:21
reapplying stuff 2 or 3 times. I mean, I've been I've been
Speaker:
00:27:25
around computing since 1975. So I've seen kinda like these meta
Speaker:
00:27:29
patterns flow, you know, through several generations,
Speaker:
00:27:33
and they kinda keep just resurfacing. One of one of the
Speaker:
00:27:36
interesting ones is, like, the, well,
Speaker:
00:27:40
both the chatbot and the text based interface versus the,
Speaker:
00:27:45
Graphical interface seems like we keep going back and forth. You know,
Speaker:
00:27:48
I I remember chatbots back in the, you know, AOL days.
Speaker:
00:27:52
AIM days probably way before that too.
Speaker:
00:27:56
And then, you know, and then There was kind
Speaker:
00:28:00
of a a a chatbot resurgence in, you
Speaker:
00:28:03
know, 2016, 2015, whenever when every company wanted
Speaker:
00:28:07
a chatbot and we're excited about that. Yeah. It didn't quite work. It
Speaker:
00:28:11
seemed to fizzle out. Then, you know, the
Speaker:
00:28:15
the then nowadays, we have So many chat interfaces,
Speaker:
00:28:18
chat GPT and and generative AI seems to be resurgent again.
Speaker:
00:28:22
So there are these weird sine waves, these weird
Speaker:
00:28:26
cycles, and I almost think of it as a coil where, you know, you're starting
Speaker:
00:28:29
at the bottom and you're cycling, but you're also moving up at the same time.
Speaker:
00:28:33
And so How do you how do you surf the wave? That's, that's,
Speaker:
00:28:37
something that's once you kind of, understand the
Speaker:
00:28:40
fact that that's what you're doing, then then then you can be excited about it.
Speaker:
00:28:44
I I think it's fair. Well, we're at that point in the show
Speaker:
00:28:47
where we transition to our, questions. And, we
Speaker:
00:28:51
dropped them into the chat here for you. Our very first one is how did
Speaker:
00:28:55
you find your way into data? Did data find you or did you find
Speaker:
00:28:59
data, Max? Interesting. Well, I
Speaker:
00:29:02
guess I was always interested in math and computer
Speaker:
00:29:06
science. You know, going back to undergrad, you
Speaker:
00:29:10
know, it was like there was a lot of different areas I could choose. I
Speaker:
00:29:13
had a hard time going into a field that, you know, where I
Speaker:
00:29:17
wasn't, using all different parts of my brain and
Speaker:
00:29:21
computer science department, it was it was not just the
Speaker:
00:29:25
mathematics. It was, you know, there was, you know,
Speaker:
00:29:29
there was a bunch of creativity in it as well. There was human computer interface.
Speaker:
00:29:33
There was it. So, So I was kind of, I gravitated to that field
Speaker:
00:29:37
as an undergrad. When I graduated, I I joined a company
Speaker:
00:29:40
called wireless generation, which, today is called Amplify.
Speaker:
00:29:45
And that's it was an education tech company. And I was
Speaker:
00:29:48
doing, you know, some simple kind of software engineering work. Actually, back
Speaker:
00:29:52
then, It was, which sounds really dated now, but, you know, they
Speaker:
00:29:56
were probably doing this up to, like, 2010, which was, you know,
Speaker:
00:29:59
writing c plus plus for the palm pilot. You know, we yeah.
Speaker:
00:30:03
Because it was they were assessing students and then it would sync to to
Speaker:
00:30:07
the web and all that. And Sure. It was a lot of, like, taking
Speaker:
00:30:10
stuff, Taking that information out of databases and putting it into a a
Speaker:
00:30:14
dashboard. And it was it was you know, I I felt like there
Speaker:
00:30:18
could be something more interesting I was doing even though I love kind of the
Speaker:
00:30:21
mission of that company there. So I ended up in grad school. I ended up
Speaker:
00:30:24
at NYU and I went there from I guess
Speaker:
00:30:28
2009 to 2011 really discovered,
Speaker:
00:30:32
you know, data mining, was the 1st related
Speaker:
00:30:36
class. Then I took, You know, machine learning, natural language processing. Actually,
Speaker:
00:30:39
the the machine learning class was with, Jan Lacun, who is,
Speaker:
00:30:43
a very well known machine learning researcher. He's Like The Lani.
Speaker:
00:30:47
The the the the the the the the the the the the the the the
Speaker:
00:30:49
the the the the the the the the the the the the the the the
Speaker:
00:30:51
the the the the the the the the the the the the the the the
Speaker:
00:30:53
the the the the the the the the the the the the the the the
Speaker:
00:30:53
the the the the the the the the the the the the the the the
Speaker:
00:30:53
the the the the the the the the the the the the the the the
Speaker:
00:30:53
the the the the the the the the the the the You know, all the
Speaker:
00:30:55
stuff that exists today. Like, even this was 2010. He would show us a camera
Speaker:
00:30:59
where he would point to different objects. He'd be like key, wallet,
Speaker:
00:31:02
chair, and it would like, the the the text would appear
Speaker:
00:31:06
on the the screen based on what he pointed at. So they knew how to
Speaker:
00:31:10
do all this stuff, that that you think of as as kind of
Speaker:
00:31:14
it it's it almost seems crazy that that was not, like,
Speaker:
00:31:19
and and turned into a product that anyone could use back then that it almost
Speaker:
00:31:22
seems crazy that it took you know so long to do it but they and
Speaker:
00:31:26
actually it it may have been Used by someone.
Speaker:
00:31:29
It's, sure. Maybe we just don't know about it.
Speaker:
00:31:35
Sorry. My paranoia. No. No. You're right. I'm sure it was used quite
Speaker:
00:31:39
a bit, but it it it's just like what it was that kind
Speaker:
00:31:43
of Sitting on his laptop was so much more sophisticated than anything that
Speaker:
00:31:46
that I I saw a year later. But,
Speaker:
00:31:50
Yeah. So it was That was kind of inspiring. And so it was
Speaker:
00:31:54
like, you know, it was
Speaker:
00:31:58
to me, it seemed like a much more interesting problem. Well, how do you How
Speaker:
00:32:01
does the machine learn? You know? How do you, you know, I don't I don't
Speaker:
00:32:05
wanna sit around writing code that's just dead. I want it to To
Speaker:
00:32:08
be alive, I wanted to to learn from experience. And so when you dive into
Speaker:
00:32:12
that question, well, then you get into machine learning, which is actually Pretty well
Speaker:
00:32:16
named. And then and, you figure, okay, well, you need
Speaker:
00:32:19
data to learn from, and then that that ends up being a statistical model
Speaker:
00:32:23
and so on and so forth. So, you know, when I
Speaker:
00:32:27
so Foursquare, was a company that that essentially came
Speaker:
00:32:30
out of NYU And, you know, it kind of intersected. So
Speaker:
00:32:34
and and they wanted to, to learn from from
Speaker:
00:32:38
their data. They wanted to kind of, sort of a
Speaker:
00:32:42
to build a data science team. And so I had already been
Speaker:
00:32:45
working on that sticky map project, And I was into local search. I
Speaker:
00:32:49
loved the the product aspect. I didn't have my new
Speaker:
00:32:53
interest in machine learning and LP in there. So it all kinda came together. And
Speaker:
00:32:56
so that's why I think that was such a good fit for me and probably,
Speaker:
00:33:01
probably would be very difficult to find such a fit again.
Speaker:
00:33:08
Our next question is, what's your favorite part of your current
Speaker:
00:33:12
gig? And that was, in The virtual green room, you said you
Speaker:
00:33:16
kinda had a good story about that.
Speaker:
00:33:19
Right. So I don't. Well, I don't exactly have a a
Speaker:
00:33:23
current gig right now. I have a bunch of different projects that I'm working on.
Speaker:
00:33:27
It was you know, I think It it was on one hand, it was
Speaker:
00:33:30
nice in Foursquare to be able to focus on one thing, and I'm gonna come
Speaker:
00:33:34
back to that. But I feel like you need these periods, almost like the same
Speaker:
00:33:37
as the grad school period That I had, back in 2010 where it was
Speaker:
00:33:41
like, well, you're working on a few different side projects, but let's see.
Speaker:
00:33:44
Hopefully, like, eventually it'll coalesce into something,
Speaker:
00:33:48
you know, something a little bit more long term and permanent. So I'm working on
Speaker:
00:33:52
several projects. One is with with the Foursquare founder, Dennis
Speaker:
00:33:55
Crowley. And we are Working on a new product, a new
Speaker:
00:33:59
kind of city guide where you walk around the city with your headphones in,
Speaker:
00:34:03
with your AirPods in or whatever. And We kinda know what you're passing,
Speaker:
00:34:07
by. We sort of are are using some of the Foursquare
Speaker:
00:34:11
tools that are publicly available that we know about, but also, You know, we're
Speaker:
00:34:15
kind of rigging up our our own thing because we've just done it so many
Speaker:
00:34:17
times. You know how to do it. We're okay. We know what
Speaker:
00:34:21
stores and stuff you're walking past. So what kind of sounds can we play? Right
Speaker:
00:34:24
now, it's a bunch of text to speech. Essentially, the way I've rigged it up,
Speaker:
00:34:28
the the old version 0, the alpha version is, you know,
Speaker:
00:34:32
we asked chat g p t or OpenAI API what to say. So it's
Speaker:
00:34:35
basically like you're you're walking down the street hearing,
Speaker:
00:34:39
content From OpenAI. Interestingly, OpenAI
Speaker:
00:34:43
seems to the the GPT seems to know,
Speaker:
00:34:46
stuff about Every place along the way, like, you don't
Speaker:
00:34:50
have to go into, like, location based database.
Speaker:
00:34:54
It seems to seems to know quite a bit. There is a question of the
Speaker:
00:34:57
all the content is there's some interesting content in there, but it all ends up
Speaker:
00:35:01
being kind of mediocre. So it's like, okay, well, how do we turn this into
Speaker:
00:35:04
something really cool? I think, you know, in the end, having,
Speaker:
00:35:09
you know, you know, maybe music and and and speeches and an art
Speaker:
00:35:13
project somehow in there, based on where you walk is an interesting
Speaker:
00:35:16
idea. So if I could That'd be cool. Yeah. I could be like a
Speaker:
00:35:20
platform that people can use, like a cultural version of
Speaker:
00:35:23
Foursquare. Yeah. Yeah. And or maybe it's just
Speaker:
00:35:27
like an enhancement of the the sounds of the city. Or maybe
Speaker:
00:35:31
it's, You know, I mean, a lot of people think, okay, maybe maybe a tour
Speaker:
00:35:34
guide. I I don't know. But, you know, it it's it feels like,
Speaker:
00:35:40
It feels like there needs to be, a variety
Speaker:
00:35:43
of use cases tried because there's there's a lot you could do with it. And
Speaker:
00:35:47
and Maybe, you know, if if you put this in the hands of more
Speaker:
00:35:51
creative or of of additional creative people, they would,
Speaker:
00:35:54
ultimately find something interesting. I'm also working
Speaker:
00:35:58
yeah. Oh, I could answer questions about that. But then my other project is my
Speaker:
00:36:02
other 2 projects are are kind of interesting as well. Well, I have the
Speaker:
00:36:05
podcast, The Local Maximum. So still doing that every week and, you know,
Speaker:
00:36:09
interviewing people. Talking about,
Speaker:
00:36:14
talking about data, talking about AI, you know, few episodes on the
Speaker:
00:36:18
whole. You know, all the drama around OpenAI recently.
Speaker:
00:36:24
I I never wanted to become kind of the the the,
Speaker:
00:36:27
the the the tech drama, you know, what's it called?
Speaker:
00:36:31
TMZ of technology? Yeah. Yeah. But but that's something that happened because
Speaker:
00:36:35
I remember, like, last year, a couple years ago, there was all this craziness coming
Speaker:
00:36:38
out of Google with, You know, there was 1 guy at Google who said,
Speaker:
00:36:42
you know, he thought that the LLM has come to life. And Oh, yeah. And
Speaker:
00:36:46
then and then there was a there was A whole
Speaker:
00:36:50
bunch of stuff with, like, the the AI safety, you
Speaker:
00:36:53
know, seemingly staffed
Speaker:
00:36:56
by, people who are a little nutty. And
Speaker:
00:37:00
so, it was a And they fired a bunch of
Speaker:
00:37:04
people From that team too. So, like, there's
Speaker:
00:37:08
definitely, it was something weird some weird mojo
Speaker:
00:37:11
was going around. That's for sure. Yeah. And when when I cover that, I
Speaker:
00:37:15
mean, it's hard to, you know, it's hard to hide the fact where it's like,
Speaker:
00:37:18
wow, everyone in this story seems kinda nutty. But I also try to, you know,
Speaker:
00:37:22
I try to take a step back and say, okay, this is what we know.
Speaker:
00:37:25
These are a few things that could be happening internally, but we don't know everything.
Speaker:
00:37:28
I'm not gonna jump to conclusions. But,
Speaker:
00:37:32
I I I I try when I'm covering a story in a local maximum to
Speaker:
00:37:35
give, like, a a balanced, a balanced version of
Speaker:
00:37:39
of whatever story I come across. You know, maybe it's my show as I try
Speaker:
00:37:42
to give my opinion. But, yeah, I I
Speaker:
00:37:46
I my attempts, which, you know, some people have have,
Speaker:
00:37:51
said I I I've captured that. But my my attempt is to sort
Speaker:
00:37:54
of, try to try to
Speaker:
00:37:58
approach each Story with a little bit of humility and try
Speaker:
00:38:02
to help people understand what's going on without the
Speaker:
00:38:05
raw emotion that you get often on on Twitter. Gotcha. That's a good
Speaker:
00:38:09
point. Yeah. So we have, go ahead. I'm
Speaker:
00:38:13
sorry. Oh, no. No. It's okay. Go ahead. Okay. So we got,
Speaker:
00:38:17
3 complete dishonest. And, the first is when
Speaker:
00:38:20
I'm not working, I enjoy blank. Right.
Speaker:
00:38:24
So now that I've moved to Connecticut, I feel like I
Speaker:
00:38:27
am such a a Connecticut stereotype where I kinda, like, Drive
Speaker:
00:38:31
around, going for walks in the woods and into
Speaker:
00:38:36
various malls and stuff. So it's like it's like it's When the
Speaker:
00:38:39
weather's good, you go into the woods. When the weather's bad, you go into the
Speaker:
00:38:42
mall. Yeah. So I I actually like enjoy doing
Speaker:
00:38:46
that. I enjoy listening to podcasts.
Speaker:
00:38:49
I, honestly, enjoy hanging out with friends. You know,
Speaker:
00:38:53
after, I used to live in New York City. I enjoyed
Speaker:
00:38:57
it a lot, and I sorta had this, situation where
Speaker:
00:39:01
I had this be careful what you wish for because, at the end of 2019,
Speaker:
00:39:05
I was like, oh my god. I'm going to, like, events every single day. It's
Speaker:
00:39:08
just it's just too much. How can we, like, how can we, you know,
Speaker:
00:39:12
cut back on that. And then COVID came. And then to me, it was just
Speaker:
00:39:16
it was the worst thing because it was like, okay, you stay in your apartment
Speaker:
00:39:19
in New York City all day and you don't go and and talk to anyone.
Speaker:
00:39:23
And it was just like it it it was just awful. It just
Speaker:
00:39:27
felt like a a prison. So I I
Speaker:
00:39:31
moved to New Hampshire for a couple years, then I came back. But, you
Speaker:
00:39:34
know, nowadays, when I get a chance to hang out with with friends and and
Speaker:
00:39:38
family, I just I try to do it, whenever I can because I'm
Speaker:
00:39:42
not like, you know, it's not like when I was living in New York in
Speaker:
00:39:45
the 2010s and got kind of overload on that.
Speaker:
00:39:50
Right. Right. So, yeah. That's my answer there. And we have another complete this
Speaker:
00:39:53
sentence. I think the coolest thing in technology is blank.
Speaker:
00:40:03
The the way I've been putting it recently Is this,
Speaker:
00:40:07
where, you know. It.
Speaker:
00:40:13
You know, back maybe 10 years ago, the story we
Speaker:
00:40:16
were getting that the hopeful story we were getting was that, okay, if you're an
Speaker:
00:40:20
engineer, you could Build anything you want at
Speaker:
00:40:24
a very low cost or if you're not an engineer for anyone because we
Speaker:
00:40:28
have access to social media. You know, you can,
Speaker:
00:40:31
you can put anything out there into the world that you want
Speaker:
00:40:35
and and have people read it if if if they want to, or have people
Speaker:
00:40:39
look at it if they want to. And so that was kind of the new
Speaker:
00:40:42
exciting world. I think today, The new exciting
Speaker:
00:40:46
world goes well beyond that, which is going to be like,
Speaker:
00:40:51
you you can create worlds. Any Any world that you wanna
Speaker:
00:40:55
build, any scenario that you can imagine, you
Speaker:
00:40:59
can just have a machine fill in all the gaps for you and, You
Speaker:
00:41:03
know, write the write the story, make the
Speaker:
00:41:06
image and maybe, like, you know, make the make the video, make the whole
Speaker:
00:41:10
world. So I think, I I I think the
Speaker:
00:41:14
idea with generative AI that I want people thinking about more that that I
Speaker:
00:41:18
I also wanna think about more is, like, Okay. If you could create any world
Speaker:
00:41:21
you want, to explore, to live in, just
Speaker:
00:41:25
to, you know, maybe it's something to to teach us about something. Maybe it maybe
Speaker:
00:41:28
it's just an artistic adventure
Speaker:
00:41:32
venture, you know, what kind of world do you want to create because that's
Speaker:
00:41:35
that's going to become very cheap very quickly. Yeah. I could
Speaker:
00:41:39
see that. So I'm gonna skip to,
Speaker:
00:41:43
share something different about yourself. But we remind our guests
Speaker:
00:41:46
to remember it's a family podcast.
Speaker:
00:41:52
Okay. And I'm, you know, I'm I'm trying to,
Speaker:
00:41:58
I'm I'm trying to think of an answer here, and it's not because,
Speaker:
00:42:02
it's it's not because of the, of the of the caveat there,
Speaker:
00:42:06
but it No. No. I get it. Well, you've already covered a lot
Speaker:
00:42:09
that's It's different. I did. It is good stuff. Yeah. I mean, I
Speaker:
00:42:13
think, I think one thing that, It's,
Speaker:
00:42:17
I I enjoy doing that that that I forgot to mention, because
Speaker:
00:42:21
I'm actually doing it again for the 1st time in in in 6 years was,
Speaker:
00:42:25
I was a member of the Yale Alumni Service Corps. It's not
Speaker:
00:42:28
a member. It's like you can do a a it was essentially we were doing
Speaker:
00:42:32
trips to underserved Communities around the world and,
Speaker:
00:42:35
you know, doing little, like, kinda, kind of service
Speaker:
00:42:39
trips where you'd ever Either build a structure or work with small business
Speaker:
00:42:43
owners or go, you know, teach in a school. And so I've been
Speaker:
00:42:47
to, Nicaragua and Ghana And I actually
Speaker:
00:42:50
got to lead one of their trips in 2017 and that was to the Fort
Speaker:
00:42:54
Mohave Indian reservation. Very different kind of a trip because
Speaker:
00:42:58
it was Within the United States here. And
Speaker:
00:43:02
so it was honestly a lot easier. Because that's very cool
Speaker:
00:43:05
flying to Vegas. But yeah. But we're actually going back there, in in a few
Speaker:
00:43:09
months after 6 years. And so I was so even though it was less
Speaker:
00:43:13
convenient this time around, I'm I'm very excited Do that. And so
Speaker:
00:43:17
I I don't know. I really like learning about different cultures,
Speaker:
00:43:21
different philosophies, different religions. I think A lot of people might
Speaker:
00:43:24
assume given the you know given the tenor of my
Speaker:
00:43:28
podcast that I'm very like you know rationalist and I talk about Bayesian
Speaker:
00:43:32
inference a lot. But, I I
Speaker:
00:43:36
sort of venture out of that a lot. I don't think that,
Speaker:
00:43:40
That raw math can, can explain everything in life. And I also love like the
Speaker:
00:43:43
diversity of, of cultures and stuff. So No. It's cool.
Speaker:
00:43:47
That so that's maybe a positive thing, so I I don't know. It's very positive.
Speaker:
00:43:50
Something different. No. It definitely is. It definitely is. So where can
Speaker:
00:43:54
folks find out more about you and what Sure. So you mentioned you have a
Speaker:
00:43:57
podcast, which I love the name, The Local Maximum.
Speaker:
00:44:01
Right. Yeah. Local maximum's triple entendre, because it's got my name, Max.
Speaker:
00:44:05
It's a local maximum is, of course, you know, in machine learning,
Speaker:
00:44:10
when you're when you're trying to find the, well, sometimes it's often the local minimum
Speaker:
00:44:13
if you're trying to, Minimize the the loss function, but the
Speaker:
00:44:17
in basing inference, if you're trying to maximize the probability, whatever, you're you you get
Speaker:
00:44:21
stuck stuck in one of your next which was Your name is Max. So Yeah.
Speaker:
00:44:24
Exactly. Right. Right. That's the first one. That's the second one. And then, you know,
Speaker:
00:44:28
I I worked on location data a lot, so it's kind of a a triple
Speaker:
00:44:30
meaning. And so, I've
Speaker:
00:44:34
been doing that for for quite a while. You can go back into kind of
Speaker:
00:44:38
a a really extensive library there. And, I have
Speaker:
00:44:41
the website local max radio.com. I have.
Speaker:
00:44:47
If you go, I have local maximum labs. If you go to local maximum local
Speaker:
00:44:51
max radio.com/labs, I have a
Speaker:
00:44:54
bunch of papers And, you know, kind
Speaker:
00:44:58
of works that I've done, which, you know, includes some
Speaker:
00:45:02
discussion of machine learning, like kind of the math mathematics behind bias correction,
Speaker:
00:45:05
but But also something kind of fun that I did, like, with the podcast last
Speaker:
00:45:09
year, which is, like, I just rewrote the US constitution, fixed a bunch of
Speaker:
00:45:13
things just because I I felt that was fun. I was Taken aback by how
Speaker:
00:45:17
mad people get when you when you do that, it's not like I was actually
Speaker:
00:45:19
trying to, you know, run a political campaign for it. I just thought it was
Speaker:
00:45:23
a a fun project, and I learned a lot. But Some once you venture into
Speaker:
00:45:27
the political, people start treating things different. People get angry pretty quick.
Speaker:
00:45:30
Yes. That's true. Yeah. Yeah. I I I The
Speaker:
00:45:34
I I I love to hear criticisms on it. I wanna hear what what what
Speaker:
00:45:38
people think. But The one criticism
Speaker:
00:45:42
that I get a lot, which I really hate is, like, how dare you spend
Speaker:
00:45:44
your free time on this, which I I just don't get at all.
Speaker:
00:45:48
Yeah. But, which, you know, whenever I put
Speaker:
00:45:52
out some kind of math paper, even if it's like and there there is one
Speaker:
00:45:55
called relative probability, which is, you know, sort
Speaker:
00:45:59
of an abstract paper where it's like, okay, a reimagined probability
Speaker:
00:46:03
theory as, okay, let's say you can't talk about The probability of something
Speaker:
00:46:06
happening. Let's say you can only talk about 1 probability relative
Speaker:
00:46:10
to another. What, what does that look like? And I just stated some basic facts
Speaker:
00:46:14
and, you know, Not that many people gonna use it. Maybe people
Speaker:
00:46:18
won't use it for for a while. I I feel like it's an interesting idea,
Speaker:
00:46:21
and I feel like it will have uses eventually. But,
Speaker:
00:46:25
You know, nobody criticized me for that. For like, how dare you spend your free
Speaker:
00:46:28
time on that? Exactly. They they pick on you for the other
Speaker:
00:46:32
stuff. Yeah. I mean, I I look at people. I mean, you spend your free
Speaker:
00:46:35
time yelling at people on Twitter. I mean, what's the difference? I was gonna say
Speaker:
00:46:38
you can you can look at TikTok, and you can find far more destructive uses
Speaker:
00:46:42
is of Exactly. Exactly. So that's so that's that's my
Speaker:
00:46:46
main thing. I I think maybe with the, with the Constitution, I think people have
Speaker:
00:46:50
their sort of ideal society in mind. And if If your thing doesn't wind up
Speaker:
00:46:54
with that, they they kind of perceive you as a threat. Like you're trying to,
Speaker:
00:46:57
like I was trying to revitalize democracy, but some people are saying, no, you're
Speaker:
00:47:01
backsliding on democracy. Alright. Like, let's talk about it. But, yeah, it's it's
Speaker:
00:47:05
people get you know, people get different. We need to have you back and talk
Speaker:
00:47:08
about that more. Yeah. For sure. For sure. Talk about that. Absolutely.
Speaker:
00:47:12
We'd love having you. Both Andy and I, however, do have a hard stop, and
Speaker:
00:47:16
I would love this This covers you to go out for a couple hours, and
Speaker:
00:47:18
we'll talk to you more. And I just had a a conversation last night with
Speaker:
00:47:22
my cohost that went a couple hours. I know how it goes. Yeah. Yeah. Yeah.
Speaker:
00:47:26
We ended at 1 AM, and I was like, oh my god.
Speaker:
00:47:31
Well, those 1 AM conversations. I know what you mean. You got it. So
Speaker:
00:47:35
With that, we'll definitely make sure. Send us all your links, and
Speaker:
00:47:39
we'll make sure we get them in the show notes, and we'll let Bailey, our
Speaker:
00:47:42
semi extension AI host, Co host, 3rd
Speaker:
00:47:45
host, wrap up the show. And thank you, dear
Speaker:
00:47:49
listener, for subscribing to our podcast. You
Speaker:
00:47:53
have subscribed to us, haven't you? Once you do,
Speaker:
00:47:56
please be sure to rate and review our podcast on iTunes, Stitcher,
Speaker:
00:48:00
or wherever you subscribe to us. Having high ratings
Speaker:
00:48:04
and reviews helps us improve the quality of our show and rank us more
Speaker:
00:48:08
favorably with the search algorithms. That means more
Speaker:
00:48:11
people listen to us, spreading the joy. And,
Speaker:
00:48:15
can't the world use a little more joy these days?
Speaker:
00:48:19
So, go do your part to make the world just a little better and be
Speaker:
00:48:22
sure to rate and review the show.
