Michael Levitz (00:00)
Hello and welcome to the forecasting podcast. Episode one, which is so fresh, we don't even have a name How are you doing Robin?

‍

Robin (00:08)
How are you doing?

‍

Michael Levitz (00:09)
Good. So why don't we start out with a quick round of intros. Do want to lead off?

‍

Robin (00:14)
I'm Robin Telly at Forecasting. data scientist and AI engineer. My background is largely in federal consulting and political campaign work, and a lot of that background attaches to what we're doing today at Forecasting.

‍

Michael Levitz (00:26)
And Michael Levitz, I spent my time at advertising agencies, mainly one called RGA, where I worked on marketing projects for clients like Samsung and Verizon.

Robin and are the co-founders of Forecasting. Forecasting is a platform that builds content strategy briefs for marketers using data science and AI. This is our one year anniversary.

So we had a goal of building out loud through our first year, which of course we did not do. So we're trying to make amends before it's too late and kick off the podcast and also blog and a newsletter where we will try to hold true to building out loud. So today we wanted to take 30 minutes and look back at the highlights of our first year. So Robin, where it all began started more or less on the Andrew Yang campaign.

‍

Robin (01:16)
so I was the director of data and analytics on Andrew Yang's presidential campaign in 2020. And if you are unaware, Andrew Yang was a relatively unknown entrepreneur who founded programs like venture for America and wanted to launch a presidential campaign. That was a bit unorthodox. He's kind of almost in the beginning, a single policy candidate on the topic of universal basic income.

So it was early hire there and was tasked with how do I message universal basic income and a relatively progressive suite of policies to the nation and how do we get this guy to get brand, know, name recognition for candidacy, which is very important for political campaigns. And also because we had small amounts of funding and were...

bit of a black sheep in the electoral cycle. We had to be pretty scrappy with our resource allocation and use intelligent marketing to try to find our supporters. And the campaign slogan was math and we all wore math hats and we tried to be as data-driven as possible. So one of the kind of core responsibilities of this campaign was trying to figure out which areas were we most likely to find supporters and donors. And when we were tasked with that,

My team and I of data scientists came up with an idea for product called Bob, which was this war room map of kind some of the key states, Iowa, Nevada, Carolinas.

and which precincts and caucuses and districts were we most competitive in. And with the background I had in data scientists and that my team had, we had this kind of viewpoint that an ensemble of weak classifiers often outperforms a strong classifier. And we can look for these kind of hidden signals of what might be indicative of both support in terms of the people that live there, but also kind of viability for our ability to play in that space.

So we tried to ensemble together a bunch of these different signals, you know, a hundred of these different signals, including kind of some underappreciated things. How far away is that caucus from one of our campaign offices? Do we have a volunteer that already lives there? there's some of the kind of more niche signals, some of the stronger, more conventional ones, socio-demographic information about the residents, historic campaign voting. So just ensemble all these things together and build

kind of a intelligent platform to try to figure out what was in play.

And one of the other kind of big underpinnings of this was this Bayesian statistical idea of we just want to play in areas that are most likely to be successful for us. So it was less relevant to say, hey, we think there is a 52 % chance that we will get three supporters if we go here. And we thought it was more relevant to say,

across these two different distributions saying if we have, you know, area A that we can go into or area B, we think that across a large number of simulated results, area B will be more effective. So you'll probably hear a bit more about kind of some of the implicit understandings of like Bayesian inference and what it actually means as we kind of talk about the rest of forecasting.

‍

Michael Levitz (04:15)
And there was, you had sent me a fairly extensive, like multi-part blog post that went through this Yang campaign and some of Bob.

‍

Robin (04:22)
Yeah, we can put that in the show notes, but one of my teammates wrote a blog in real time about his experiences on the campaign and it does go into the makings of Bob and also just kind of the broader dynamics of what it's like to work on a presidential campaign. But yeah, you know, when we were sitting there on the...

primary night in Iowa, Bob was the thing on the TV that the whole campaign was looking at to try to get the real-time results.

‍

Michael Levitz (04:49)
You know, we were trying to basically do what you were doing, but kind of failing. You know, we had this, hyper kind of targeting model.

but it was very manual. So we had kind of huge spreadsheets of people that liked to use their phone camera for food pictures And then when I saw that blog post, it just kind of opened up this whole new set of possibilities where you can actually like map out the things that we were doing in a very manual process. Not to mention kind of understanding and making a real closed loop around

what people would respond to. So you talked a bit about how do I get a person to go caucus for me here and you're actually driving a behavior and then measuring that

‍

Robin (05:34)
Well, one of the angles of that too is when I was doing that for the campaign or when you were doing that in previous work as well, like this was kind of the pre-gen AI world. So as we kind of move more and more into automated content, it's gonna be kind of increasingly significant to create the relevant content. How do you rise above the sea? So just doing this kind of individualized content prediction and.

playing the hands of what's gonna be the most likely to succeed is just gonna be more and more relevant as we see more and more mass content generation.

‍

Michael Levitz (06:05)
Then we form not forecasting, but Jackalope Studios. You shot me, I think a slack message and said, hey, how about we call our company Jackalope? Sure. No idea why you called it that. And then we started building a Streamlit app.

‍

Robin (06:22)
Yeah, like many, I mean, our company did not begin with a napkin at a bar, but it is named after a bar.

‍

Michael Levitz (06:28)
But I didn't know that at the time.

‍

Robin (06:29)
Yeah. So we did our best to do the kind of startup prototyping of how do we pay off some of these ideas about predicting content relevance. And Streamlit is a tool that allows you to prototype pretty quickly. So we put together just a basic version, in which we were just putting together all the data behind the scenes for.

prospective clients and we had a few initial meetings showing this off and trying to do initial validation where we were just showing kind of hey here's surfacing topics that we see in social media news conversations pairing that in with some external signals of SEO volume and all these things that we just thought would be this ensemble of what this audience zeitgeist would be talking about as Michael kind of had said like

Brands kind of found themselves feeling blinded or unaware of what the actual audience that they were trying to interact with was saying. So the first version of what we tried to build was just a surfacing tool to start showing some of those signals.

‍

Michael Levitz (07:25)
And I think that that's super interesting. Like for me, you know, working on brands like a Samsung or a Verizon, you know, there's just so much stuff, you know, they're in so many lines of business. There's so many kind of segments of customers, so many competitors.

At a certain point, think you kind of stop looking at what's going on around you. You just have to, to continue. And you really kind of lose touch at a certain point with, you know, just what the trends are. You know, you've seen so many devices launch, you've seen so many plans come and go, and you kind of get this confidence that you just know what you're doing. but at the same time, the world is changing around you, customers are changing.

what I really wanted was just this morning briefing, just tell me what's going on and all the things I need to know so I can make the right decisions. And I think what we came together to do was hand over the reins to the data trends, just let the data tell me what's important instead of me trying to do deep research.

‍

Robin (08:21)
Yeah, and allow for the automation of that research every night. Always just be pulling in the most recent trends that'll resonate with the audience. The indefatigable machine can just crunch through all the research.

‍

Michael Levitz (08:34)
So we then start doing customer interviews, And then we build V2,

‍

Robin (08:39)
Yeah, V2 was fairly modular in how we were approaching the problem space. So we had started with these initial interviews, but hadn't really solidified yet on exactly what we want to build or what angle of this problem space we're trying to solve. And V2 became a fairly configurable platform where

you would be able to pull in those different signals I was mentioning earlier and then kind of configure what you want to do with them. So an example of this could be, you know, based off of my social media analysis, write the header for a product promotional page. So it was a very high level, you know, open-ended platform.

And as we took that to more and more users in these initial interviews, it showed a little bit of the power of these topics, but one of the stumbling blocks there was really this kind of core question of just not presenting what we, you know, not presenting enough of an opinionated view, not showing enough of the power of what these could be used for. And as we see the marketing world change through a bunch of recent events, it becomes harder and harder to

promote a product that requires that level of investment of time and thought and configuration. So we wanted to become more opinionated over time and just as Michael said, give people more of that morning brief, more of that just readiness to write and create content.

‍

Michael Levitz (10:00)
I think in a lot of ways this is kind of like the product version of like social media ghost town.

You know, it's wouldn't it be great if like a million people were using this, everything would be awesome. And from a product perspective is like, if everybody took the amount of time to really figure out how to get the most out of this thing, it would be great. But then we got hit in the face with reality that people are like super busy and they don't care. Like you tell us what superpower this is going to give me. Don't make it my problem to figure that out.

‍

Robin (10:29)
Right, so then we moved on to trying to be as declarative as we could and just only showing Google Sheet powered results of just here's the pieces of content we want you to push, here's the content recommendations and all of that and just stripping the whole thing back down to recommendations through Google Sheets rather than expecting people to learn a platform to configure this stuff and create it.

‍

Michael Levitz (10:54)
Yeah, so then like a couple of things, interesting things are happening here. think, yeah, so we go in on kind of like, can we predict, you know, the best thing for you to say the next best action. And, know, I think that's where we start using the line money ball for content marketing. You're just really like all in on, we're going to do one thing really well, which is tell you the very next thing you should say to a customer.

And we have, you like you said, we have this Google sheet and that's kind of our pitch deck. And basically when the meeting goes badly, we just go back and change the Google sheet. I'm like, all right, they didn't like that. Let's put something else in there and see if they like it. You know, and sometimes they did. And then, you know, then we kind of build out this Google sheet a little bit more. And at that time we then join Rob Snyder's PMF camp.

And it was the night before and we felt like we didn't have anything like sticky enough to pitch to him,

so that births Moneyball for content marketing. think it also at that time we had a logo created for Jackalope and then it was finished. I remember I uploaded it to Rob's Slack channel and then we changed the name to forecasting like that maybe the very next day.

‍

Robin (12:00)
going all in on content prediction necessitated the name change.

‍

Michael Levitz (12:05)
we launched that you then come to New York and we, I sent you an $11 domain that I thought we should buy, which was like content prediction.io. think we still own it and we own like 10 permutations of it because I wanted to defend that. And then you came back with something else.

‍

Robin (12:24)
Yep, I came back with forecast.ing. Just having our, trying to have our name just be what we want to do, try to own this space of content prediction. I guess another one of these elements of our background too that we didn't discuss a lot earlier is.

Both Michael and I are fans of Nate Silver and his work. And actually, one of the kind early inspirations I had both with Bob and forecasting was the presidential forecasts on 538. And just trying to kind of do this both.

Bayesian simulation demonstration, but also just kind of data storytelling. And we thought that by naming ourselves forecasting, we can kind of embed in some of that data storing, telling ideas and the moneyball ideas and really just push that as the front and center idea of content relevance.

‍

Michael Levitz (13:11)
All right, so we were making no money. We buy a logo and basically just shred it. And then the next day, buy a $3,000 domain that we rent that we don't even own.

‍

Robin (13:24)
You gotta spend money to make money.

Michael Levitz (13:26)
All right, so then we start moving from V2, which is, you know, it's open kind of like you tell me what you want to do to this much more opinionated platform. And through Rob's tutelage, we kick off real sales outreach, where we're basically stopping discovery of, just tell me your pain points. And now we're actually trying to pitch a Google Sheet.

of what you should say next.

‍

Robin (13:48)
Yeah. So I mean, I do think we start seeing some, as Michael said, we had the ability to be pretty flexible with the Google sheet, but we start getting some of this initial feedback and people kind of reacting positively to some of the topics that we're suggesting and generating marketers saying, I hadn't thought of that, or I didn't know that that was so common in the audience voice. So.

Based off of that, we wanna start putting together kind of a more robust offering, not just the Google Sheets, something that can be a bit more visual and a bit more, you know, set up and repeatable for people. And the team at, you know, another kind of co-occurring thing.

is that at the same time that this is going on, the team at Answer AI releases an open source library called Modern BERT, which is a encoder which...

to make it quick and encoder can take words, text and put it into math. It can embed it into a vector space. And the strength of that is we're able to do a bunch of kind of classic prediction and really use the actual kind of pure data science to start prioritizing which of these topics will be relevant to a market or not and to an audience.

‍

Michael Levitz (14:50)
And this, I think, starts the raging debate that is probably never going to be resolved of do we tell people what's trending and what they should say, or do we ask them what they want to say and we'll go tell you what the trends are in that space.

‍

Robin (15:05)
Yeah, we've seen both sides of that coin and it's interesting because on the kind of, when we talk to marketers, they have editorial calendars, they have editorial schedules, they have white papers coming out, they exist in the context of the company that they're working for.

And then there's this audience voice and audience signal that is relatively unbridled in terms of what it's talking about and what's trending there. So trying to find out how to kind of bridge the gap between those audience signals and what the brand can connect with and is valuable to the brand is kind of the core question. But as Michael said, there's the two sides of.

How data-driven can somebody actually be? Can you just create a piece of content on a Tuesday because we tell you that a blog post about a topic will be the most successful? Or do we want to come up with a position where a brand might already have an idea of what they want to write for a newsletter or a landing page and we just want to slot in some of that external signal to validate it? So I don't think we'll ever have an answer to the raging debate.

but our company is kind of squarely rooted in assisting people through navigating that raging debate.

‍

Michael Levitz (16:11)
And around this time, we come in contact with my old boss, David Dewey, who puts an interesting angle on this, which was basically, if people are going to believe you, you have to tell them why they should believe you. So we start a work stream around.

validating our sources and showing people kind of how we arrived at these conclusions. Do want to talk about that?

‍

Robin (16:32)
Yeah, David, you the initial conversations with David Dewey were very illuminating about all of that and about trying to kind of figure out the level of citations needed and proofing needed and a little bit of that kind of thing. As Michael said earlier, they're like, you know, I'm ready. It's Tuesday and I want to write something and I want to know that what I write is valid and supported.

and I want to know that citations for all of the things you're saying exist if I want to go look for them, but I don't need to be drowning in them to even know where to start. So part of this kind of move towards trying to be a more opinionated platform overall is pushing these recommendations ahead and the narrative storytelling, data-driven storytelling ahead, and then having the citations where they are needed and as supportive things.

rather than presenting somebody with a full Excel sheet of all of our findings and all the things that could possibly be learned from the data and then just telling the marketer, good luck, right? Just coming up with what are the most actionable topics? What are the most actionable mechanisms for those topics and kind of mediums to push those themes through?

‍

Michael Levitz (17:37)
Yeah, you just brought me back. I think one of the key themes of our kind of first year is reducing, taking elements away. And you brought me back to that list of citations where

we would just give people like 50, 100, citations on the kind of topic that was trending, which in a way was helpful. It was this kind of like spark notes of everything that's related, but getting that down to like three or one, became the real thing that people needed. they just didn't want to read a hundred links.

You know, they wanted us to and choose the like one or two, like the one that's best and maybe give them a backup if they needed it.

‍

Robin (18:19)
Well, one of, you know, the history of our company is squarely set in the world of chat GPT. And in the chat GPT world, the marketers are already fairly well versed in going to chat GPT to get the kind of immediate answer to a question. So if they can go to chat GPT to get what they think, you know, the immediate answer to a question, but then if they go to our platform, they have to do work.

That doesn't work very well, so we want to be able to come up with just kind of the fastest ways to be able to present this information and distill it to what's most actionable.

‍

Michael Levitz (18:52)
And what is, can you talk a little bit about like, what is the difference when I asked Shachipati, like what should I say next? And you tell me what to say next. What's the difference?

‍

Robin (19:01)
So I would say that the main difference is that if you talk to ChatGPT and you ask it, hey, what should I do? The core thing that ChatGPT does is it, the technical answer is it predicts the next token in the sequence. So it would say, you you would ask it, if it's trying to answer a question of what is your favorite color, it'll say, well, the most common response to that sentence is your.

favorite color is blue. Every little word here is kind of evaluated. It's just the next most likely word in this sequence of things. And it does that by looking at this massive data set that it's pulled from all data ever. But it's only ever looking at things in that kind of like linguistic context of just this word is essentially most likely to occur after this other word. The difference with what we're doing is that...

We route our kind of citations and signals in some of these things that are external. They're both specific to the brand, specific to the moment, specific to the marketing project, but also rooted in actual kind of quantifiable predictions. So we can look at social media conversations and say, this topic has this number of likes, this number of favorites, this topic has this number of search volume and SEO, this topic occurs in

this number of preeminent news articles and publications. And we can use that to start predicting and scoring which of the topics are most likely to succeed based off of kind of these real quantifiable data signals. So rather than just saying like.

The best topic is the word that would appear most often answering what the best topic is. It's really squarely rooted in kind of the brand's context, the brand's vision, and the merging of that with external voice of the customer and what we kind of refer to as this zeitgeist of culture that the customer exists in.

‍

Michael Levitz (20:47)
both are prediction engines, really, right? So chat GPT is basically just predicting what's the next thing you want to hear, the next token you want to hear. Like how do I get closest to kind of that sweet spot? Whereas, you know, we're basically predicting,

this topic has showed up in a certain amount of volume and relevance across all these different data sources, that's most likely important for you right now. Is that right? ⁓

‍

Robin (21:11)
Right. Yeah,

I would say so. mean, chat GPT or any other LLM is just going to say, in most situations, this word, this topic satisfies being an answer to this question. And as you just said, it predicts the likelihood of a word, it doesn't predict the likelihood of that word being a successful campaign directly. So rather than just kind of predict what's the next word in a sequence, we say,

Well, hey, give me the actual kind of quantifiable metrics related to how this word exists in the context of the brand and the audience. And then using that score, how likely this campaign is to succeed. And then rather than just giving me the next word that you think is likely in the sentence, give me which of these topics has the highest potential score.

Yeah, like, you know, which of these topics has the highest expected payout?

‍

Michael Levitz (22:04)
we're selling, you know, we don't know in the world we're selling, but we're selling stuff when it doesn't go well, or we're just like meeting for hours, you know, late at night, kind of changing our Google Sheet. We're using V2 in the background to basically produce these topic predictions.

And then we realized, you we need to get this thing in the hands of customers. It can't just be this Wizard of Oz thing anymore. And we feel like we have about a week of work left. You know, so this is probably like January. We've got a week of work left. You know, let's just get that over the hump and then we'll start sending demo links. And we start telling people this on the calls. Like, we're going to get you login credentials next week. Stay tuned. And then what happens?

‍

Robin (22:43)
We tell them next week that it'll be next week. The infinite scope creep, the infinite startup fears about is our thing ready? There's the classic advice of you should be embarrassed by what you put in people's hands. And we might have been embarrassed of being embarrassed, but yeah, I mean the...

the nature of being that kind of as open-ended as we were with V2 where you could kind of just configure it to do most of these, any of these things you wanted really just put us in this position where scope creep was always going to be the only thing that would have happened and we would find ourselves, you you want to create the best topics for a header on a product promotion page and you want it to be in the brand voice specifically.

And we would generate some content for that. And we would hear back that the content, you know, wasn't exactly in the brand voice in some, in some way. So we wanted to kind of dial back eventually and just get back to the validity of the topics themselves

‍

Michael Levitz (23:39)
Yeah, and this January through April for me was just like going into like a narrowing tunnel. You know what I mean? It was just this kind of punch list of stuff we wanted to get done before we launched. It was always getting longer. I just remember like...

It was all of sudden Friday. I had no idea what it had, I don't think I like, I could tell by the smell of my hoodie sleeve, like, whether what day it was. How bad did that smell? How close was I needed to do laundry? Working through every single weekend, I remember like, I came out, we had an awesome session in LA working together as the last day.

You know, we met up at like 10.30 in the morning and I didn't see you again until like midnight. You know, we just like, we were like, we don't even have time to talk to each other. Like, let's just work.

‍

Robin (24:29)
Yeah, the big joke I remember of that time was, it's the weekend. The perfect time to catch up on work.

just sitting in meetings, Monday to Friday, and then, yeah, you finally, like, it's Friday afternoon, Friday evening, and it just finally feels like you can actually do any focus work at all. It's like Friday afternoon was the Monday morning of, I finally finished checking my email. I can just sit down and do all the stuff I have to do.

‍

Michael Levitz (24:54)
So then around May 1st, we launched the MDP.

‍

Robin (24:58)
We launch the MVP, we follow up with all of the people that we had sent the, it'll be ready next week messages to.

and we start getting on some of the calls. A very interesting moment is always that thing where Michael and I had just spent so much time staring at this thing and being the only people to look at this thing and tweaking the UI here and trying to make this thing better and just finally having some of those initial moments where we get on a call with somebody and...

basically just start the call by saying, hey, we sent you login credentials. Can you just open it up and tell us what you see? Walk through it. So biting our tongues about trying to say, hey, you need to do this or this is what that is. And just start really getting that purest form of user feedback.

‍

Michael Levitz (25:43)
And that pretty much brings us up to date where we are now.

‍

Robin (25:46)
year later.

‍

Michael Levitz (25:47)
one year later. So yeah, so a couple final topics to check in on. One is, so while we're spending a year figuring out what people want, building it, building it again, people are like, you know, vibe coding their application over the weekend and launching it with like 100,000 users. What's, like, we're feeling like crap and like, what are we doing wrong? What are you thinking doing all that stuff?

‍

Robin (26:11)
I'm just thinking that like, we're fighting the good fight.

Yeah, there's the whole industry of startups and startup success and telling people about startup success and all of that. But we made the decision pretty early on to try to do it the bootstrapped way to try to put in a good, honest days of work. So yeah, I mean, we unfortunately did not sell our company two weeks in for trillions of dollars, but.

I think when you look at the statistics about startups and all of that and the failure rates, well, we've lasted for a year and that's all there could be. We are still onboarding users, doing the sales calls, building the product, iterating. we hadn't said this earlier, but.

You know, both Michael and I had kind of worked in agile environments in the past and just that idea of, well, first you build the scooter and then you build the skateboard and then you build the bicycle. So I don't know what level of vehicle we're at right now, but just, know, that actual iterative feedback loop, we're still pretty squarely entrenched in. And that'll be the, that'll be likely the next year too. I mean, hopefully more customers and more success, but just keep keeping our.

noses to the grindstone and working. One of the other kind of hallmarks of this first year has been, Michael had said this, of like, there's kind of two jobs to be done, cell and code. So throughout this whole year, you know, we've largely tried to stick to those two jobs.

‍

Michael Levitz (27:34)
And so, okay, so other memorable moments, think the one I think about is, you know, just in terms of being embarrassed, we're on a sales call and the person says, by the way, your website's down, which had happened to us before on calls because, you know, we were kind of...

We didn't have users that were just updating the entire site and product whenever we needed to. So his website's down, I was freaking out. I said, well, can you send, I went to the website and it was up. And I said, can you send me a screenshot of the site? And he sends it to me, he's like, yeah, it's definitely down. And it was our actual live website, which was just so sparse and crappy that he honestly thought it was down.

and we changed the site shortly after that.

And yeah, maybe last thing, shout out to Michael Mantel. Do you want to talk a little bit about learnings there?

‍

Robin (28:26)
Yeah, mean, Michael Mantell has been a fabulous advisor to us through this whole journey.

He's done a few startups himself. So he's been there and just gives us pretty, you know, very candid, frank advice and really function as, as we said, you know, us trying to kind of bootstrap our way through this. It's very easy to just bounce around the rails of what should we be doing? very recently we try to create the kind of sprint board and does, you know, the joke now of like, the task I'm doing is actually on the sprint board instead of in the backlog. So just that kind of rapid prioritization of.

what actually matters right now is always difficult. So having Michael Mantell around to give us this feedback and keep us a bit accountable to the, well, last month, you said you were gonna do this and you gotta do this. And also his phrasing and opinions about what startups are and how they succeed are very memorable to me about like.

some of these things of building a product, giving it to users, selling it, figuring out the price point, figuring out experiments about how are you gonna target people, what vertical are you gonna go to, what needs to be built for these people, and just being very scientifically.

methodical about what are you doing and test and doing that experimentation and refining and iterating is what being a company is. It's not just implicitly you're a company because you're funded, you're a company because you have, you know, ex good idea on a napkin. It's just getting out there and doing the hard work of selling and building and

getting out of your comfort zone where me as a technical person didn't love always being on sales calls, but you got to do it. You're a founder and Michael has to put up with crappy demos and MVPs and all of that. So it's just like, yeah, you got to kind of.

you know, that quote about like, eat the bullfrog first thing in the morning, do the hard thing first, you're of, you're kind of always doing that. But you know, I think Michael Mantell's advice about being conscious about what the bullfrog you're eating is, has been extremely valuable to us.

‍

Michael Levitz (30:28)
One other thing we should talk about a little bit is Answer AI Stack. So we are a full, vertically integrated Answer AI company.

‍

Robin (30:36)
Yeah, Answer AI is the company that was created by Eric Reese and Jeremy Howard, who are both people that Michael and I have a tremendous professional respect for. So Answer AI has created a few of these different open source libraries of Monster UI, Fast HTML, Burt Topic. Anytime one of these things comes out, they've always been...

very kind of relevant to how we want to build these things and this kind of simplified web development flow, this like, HTMLX based things, just trying to build things that work that are simple. Answer AI, you even goes so far as like, having some of these offerings for the kind of startup side of things as well, so.

Big shout out to the answer AI team too. And they have, you know, a vibrant discord community and book clubs and all of this stuff that keeps us up to date with what's going on in the data science and LLM.

‍

Michael Levitz (31:31)
And those guys are both, I'm sure it's unrelated to the fact that they're brilliant, they're both ridiculously quotable.

So I feel like in my day-to-day life, I'm constantly like inspired and haunted by what would Eric and Jeremy say right now? those North Star quotes are always somewhere in my brain.

‍

Robin (31:49)
It's always interesting to

like when you have this startup that all these different things that do pop up as inspiring, you know, the acquired podcast, Nate Silver's podcast, all these things you can just, you're just driving somewhere and you'll hear something and it's just, you can just apply it to your business, right? There's no, you're the founder, you can just kind of do it, right? You can be inspired by it and that's all there is.

‍

Michael Levitz (32:13)
I think that is one of the big learnings of startup land. you read all this stuff when you don't have your own startup and you kind of think you know how to do it or you understand it. And then you start a company and

these things take on a whole different meaning. It's like, wow, actually like doing these things and living by them and not doing them and forgetting them and you know, it's just completely, it's a completely different level of knowledge, you know, of the applying it versus the just, yeah, sure, I know that, I know that's the way to do something versus I'm actually doing it.

‍

Robin (32:45)
Yeah.

Yep.

‍

Michael Levitz (32:47)
So assignment one is we will have a name for the podcast. And I think two, maybe we'll get into a bit of like the game theory of content marketing. So I think that's a super interesting area that we haven't touched on really yet.

‍

Robin (33:01)
We haven't said multi-armed bandits once.

‍

Michael Levitz (33:03)
Alright, well thank you Robin, this was episode one.