Latest / AI for Founders with Ryan Estes / AI That Predicts the Future

Transcript

AI That Predicts the Future · AI for Founders with Ryan Estes

0:00
Welcome back to AI for Founders. I'm Ryan Estes.
0:02
Today's episode starts with a problem every founder secretly
0:05
wrestles with. The future is moving faster than
0:08
your product. Road map models change monthly,
0:10
competitors ship weekly, and half the AI space seems to be
0:14
guessing their strategy on the fly.
0:16
But what if prediction wasn't a parlor trick, but a real
0:19
capability? And what if companies who master
0:21
it don't just survive the AI boom, they run the table.
0:25
That's exactly where we're going today with Ben Turtle, founder
0:28
of Lightning Rod AI Company, beating frontier models at
0:31
forecasting the real world using messy data, no humans in the
0:35
loop, and a training method inspired by how humans actually
0:38
learn. If you're a founder, CEO,
0:40
builder, or operator trying to understand where the puck is
0:44
going in AI, how to build a defendable product when software
0:47
is collapsing to 0, what predictive AI means to your road
0:50
map, your team, and the way you make decisions, this
0:53
conversation is built for you. By the end of this episode,
0:56
you'll see the future of AI through a totally new lens, and
0:59
you may walk away with the most important strategic insight for
1:02
the next 5 years. And if you want to think smarter
1:04
than the timeline, subscribe to the newsletter at AI for
1:07
founders.co. It's where the real playbooks
1:10
actually show up first. Every week I will break down the
1:13
founders, experiments and breakthroughs shaping this
1:16
entire ecosystem so you can build with unfair clarity.
1:20
Join 25,000 founders who don't wait for the future, they build
1:23
toward it. Sign up at AI for founders.co.
1:27
And if this episode stretches your brain in a way you didn't
1:29
expect, do me a favor. Leave a review so other founders
1:33
can find the show before the road map catches fire.
1:36
This episode is brought to you by Kitcaster, the podcast
1:39
booking service trusted by hundreds of founders and the one
1:42
I personally run. Here's the simplest truth in B
1:45
to B. The fastest way to drive demand,
1:47
authority and deal flow is to get in front of the audiences
1:51
who already care about the problem you solve.
1:53
And there's nothing that rivals the trust of podcast audiences.
1:57
Kitcaster puts you, the founder, the expert, the storyteller
2:00
directly onto the world's top podcast in your category.
2:04
SAS, AI, Venture, Health Tech, Climate tech, Fintech, all the
2:07
techs. We've booked more than 35,000
2:10
interviews and helped founders generate millions in pipeline
2:13
just by telling their story on the right shows.
2:15
If you're building something real, if you want to strengthen
2:18
your category position, or if GTM depends on trust and
2:22
authority, podcast guesting is the highest ROI channel you're
2:25
not using yet. So if you want your message in
2:28
front of the right ears, apply today at
2:30
kitcaster.com/application. Let's get you booked on the
2:34
shows your future customers already listened to.
2:36
That's kick Caster by MO Burst. Ben, welcome to the show.
2:54
Thank you so much for joining me.
2:55
Thanks for having me, Ryan. Absolutely.
2:57
I'm really excited to talk to you for multiple reasons.
3:00
We're going to talk about your product, but I'd like to start
3:02
with a great quote I saw and maybe you could unpack it a
3:05
little bit for us. So you said every decision is a
3:08
prediction problem. That's a provocative framing.
3:11
So what exactly do you mean by a prediction problem?
3:15
Yeah, I think when people think about prediction, we tend to
3:19
think of these really niche use cases, right?
3:22
Like professional forecasters predicting political outcomes or
3:27
geopolitical risk, things like that, right?
3:29
But at the end of the day, every decision really is a prediction
3:33
problem. You choose A / B because in your
3:36
mind you have a model of what path A is going to lead to and
3:40
what path B is going to lead to, and you think path A is better.
3:43
You predicted the outcome of A is better, and your decision
3:46
reflects your ability to predict what those two paths will lead
3:50
to. And this also just applies to
3:52
just how we think about like the world, right?
3:54
Like science itself is about prediction.
3:57
Your theory only holds water if it helps you predict from one
4:00
observation what's going to happen next, right?
4:03
And I think this is a principle that has been pulled up, pulled
4:06
in quite enough to the way we're training LMS today, and there's
4:09
a lot more that we could do with it.
4:11
Yeah, I'm really excited to talk about predictive AI because
4:14
ultimately if I I'm jazzed about a lot of stuff, but that realm,
4:19
you know, is really what is really what kind of piques my
4:23
interest. Now prediction is probably a
4:27
uniquely human experience and maybe maybe not exactly, but I
4:31
think it gives us a biological advantage against the Saber
4:34
toothed cats. You know, I know if I'm out
4:36
hunting, you know, that like chances are if those cats are,
4:40
you know, in the rut or whatever, like I probably don't
4:43
want to mess with them. So you're you're predicting to
4:44
give you better outcomes based on that.
4:46
Some people might call that anxiety, right, because you're
4:49
projecting into the future. Now, anxiety gets a bad rap
4:53
because we're all riddled with it.
4:54
But like the the biological origin of that has a lot of
4:58
evolutionary benefits. So, you know, maybe if we're
5:02
thinking about LLMS, like how can, how can LLMS help us
5:06
predict the future? And this is really at the core
5:09
of your company, which is lightning rod dot AI.
5:12
So kind of frame it out for us. Yeah.
5:14
I mean, I, I like the way that you're framing that.
5:16
I think the place to start is, you know, the way we train LLMS
5:20
today requires labeled training data, right?
5:23
So you don't need to show the examples of what you want it to
5:25
do, or you need to at least keep up problems where you know the
5:28
right answer so you can check the answer, right?
5:31
And the question is, why don't humans need labeled data to
5:34
learn, right? Humans just, we just move
5:37
through the world and we're constantly making predictions
5:40
about what we expect to happen next.
5:42
And when we feel surprised, that's because we expected
5:45
something. And that's not what just
5:47
happened, right? If I walk into a room and
5:49
there's a guy sitting there with a clown face, I'm surprised
5:52
because I predicted something other than that when I opened
5:54
the door, right? And we're constantly updating
5:57
our beliefs about the world just by moving to the world and
5:59
realizing when our predictions are right and when they're wrong
6:02
and making our world model a little bit better every time,
6:05
right? And AI doesn't learn that it
6:07
needs this curated training data set to learn.
6:10
So our core insight was basically, can we make training
6:13
LMS a bit more like the way humans learn?
6:17
LLMS and specifically. Well, it really could applied to
6:21
any AI, but today it you know, LMS are all the rage and the way
6:24
most of our frameworks are built are are centered on LMS.
6:28
OK. I'm going to pitch a product to
6:29
you and I want to know if if Lightning Rod can make this for
6:33
me. If this is going to, it could be
6:35
way off base, but OK, let's say in the future that the recent
6:39
future, five years from now, the LO Ms. know that it's going to
6:46
be my son's birthday and he, they know that like for my son's
6:49
birthday, he likes to get 3 giant cheese pizzas from Little
6:53
Caesars. Now they also know that I can't
6:56
eat a Little Caesars because it, it, it basically ruins my entire
7:01
day. It could ruin two days now it
7:04
not while you're eating it. Now, I love Little Caesar's
7:06
pizza, but there it will like, hit me like a tranquilizer dart
7:09
right in the neck now. But it also knows that I'm
7:12
powerless against the fumes of Little Caesar's pizza.
7:15
And I will tell myself maybe this time it'll be different.
7:18
So knowing all those things, what it's going to do is
7:20
preemptively 24 hours, it's going to give me a micro dose of
7:24
some Ozempic or something to curb my, my desire for Little
7:29
Caesars. It's going to know all this
7:30
ahead. I'm thinking I'm not eating
7:32
Little Caesars. But then once I get there, I'm
7:34
I'm going to look at that pizza. I'm going to be attracted to it.
7:37
But because my Apple Watch injected me with the GOP ones,
7:41
then I'm going to probably abstain.
7:43
Now can you make this product for me?
7:46
Yeah, I mean, I think, you know, that is probably a little bit of
7:49
a ways all, but I think where you're going with this is
7:52
exactly why we think this is so powerful, right?
7:55
When you think about like the way humans make decisions, we
7:58
only have so much focus. We only have so much attention,
8:00
right? So you're you're really deciding
8:03
between the couple options that pop into your head at any given
8:05
moment. And it's even hard to think
8:07
ahead to like, I'm going to be at this birthday party and this
8:09
is going to be the situation. And so this is what I'm going to
8:11
do to prepare for it, right? Only really, you know, type A
8:14
organized people think about those kind of things in advance.
8:17
For the most part, AI doesn't have those limitations, right?
8:20
So AI was good at constantly imagining every possible
8:24
scenario that might happen and how that might work out for you
8:27
or how that might, you know, screw things up for you and
8:30
could make decisions on your behalf.
8:31
Or at least, you know, tell you, hey, this is an option you might
8:34
want to consider. I think that would be really
8:37
powerful and that would be a lot more powerful than us having to
8:40
go to the AI with our questions that we think of, right?
8:43
It's, it's better. It's completely measuring all
8:45
these paths and telling us which ones we we might want to
8:48
consider. Yeah, but you know the one thing
8:51
and that's why I made the case study so that it's non
8:54
consensual. It's going to just inject me
8:56
largely because because it knows me well, but also it because
9:02
you're kind of coming from a platform of no humans in the
9:05
loop. So like where ultimately do we
9:08
want autonomy to like be like, hey, hey, check with me first.
9:12
I'm going to let you know or we're going to be like, now I
9:14
trust you. You let me know what's up, you
9:17
know? Yeah, I mean, I think that's a
9:19
question that's probably a bit beyond my pig, right?
9:22
Where we are is we're helping developers and teams create high
9:26
quality training data for their AI models in this way, right,
9:29
where there's no humans in the loop, but it's also not
9:32
synthetic slop or it's garbage and garbage out because they got
9:35
to the next level and start making some moral decisions
9:37
about how much that AI should be controlling my decision.
9:40
I don't know if if I'm your guy for that.
9:44
Well, I have to take a pause now because see, I get really
9:47
excited when I start thinking about predicting the future and,
9:50
and robots and AIII jumped straight to like some kind of
9:54
like robotic, you know, injection from your Apple Watch.
9:59
Let's back up a little bit. So what we're talking about not
10:03
the near future, we're talking about today.
10:05
So what what folks need now is they need their data basically
10:10
arranged in such a way that it can fit LLMS the best and they
10:13
can utilize it in in a productive way.
10:17
So talk a little bit about exactly like who's your kind of
10:19
like best customer and what are the best use cases and case
10:23
studies that you found for Lightning rod?
10:25
Yeah. So today it's it's hard to get
10:28
high quality training data for your AI models, whether you want
10:31
to train your own model or fine tune something or even just
10:34
evaluate how good a model is for your use case.
10:37
And most teams building with AI today and they don't actually
10:40
struggle with the models or the compute anymore.
10:42
The bottleneck is this high quality training data.
10:44
That's what's hard. And there's already a big
10:47
industry around this, right. So there are a handful of multi
10:49
billion dollar companies like Scale AI and Label Box or
10:53
Snorkel, But what these companies do is they're paying
10:55
humans to label data at scale for you to put together these
10:59
data sets for you. And they work with all the big
11:02
AI companies. But this is slow and it's
11:05
expensive and certainly not an option for individual developers
11:08
or for smaller teams, right? So on the other hand, you have
11:12
synthetic data, you can use AI to generate a ton of training
11:15
data, but it's really hard to get AI to produce information
11:19
that AI already doesn't know. Of course, right?
11:22
You're not adding anything new that you're teaching me.
11:24
I would right. So it's a it's a garbage in
11:26
garbage out scenario. Our unique insight is that we
11:29
created a method to create what we call grounded training data
11:33
directly from your messy dev. So every company already has
11:36
these streams of data that have a lot of gems within them,
11:39
right? So if you gave me access to like
11:42
your company Slack, right? Company Slack probably has every
11:45
important thing that happens to your company within the Slack,
11:48
right? Every big customer that you
11:50
close, every product feature that you want to launch, there's
11:53
probably a record of that. If an AI model could predict
11:57
what's going to happen tomorrow in your Slack channel, it
11:59
probably understands your business and your customers and
12:02
your product really, really well.
12:05
So what we do is we make it easy for you to take these kind of
12:07
messy internal data sources, whether it's a pile of PDFs or
12:11
your Slack messages or emails or whatever, or define data like
12:15
that on the open web from news or from just public domain
12:19
documents and things like that. And transform that into high
12:22
quality training data that you can get AI to do what you're
12:25
looking to make it do with. Amazing.
12:27
OK. And then it predicts.
12:29
What is it predicting then? Yeah.
12:32
So I mean, 1 core insight here is basically that the nature of
12:35
real world data is chronological, right?
12:38
So most, most things happen at a point in time.
12:42
So if you have internal documents, they tend to be time
12:44
stamped, right? Or if you want to predict
12:46
something that happens in the news or understand a domain that
12:49
has real world information, but you typically have a, a source
12:52
of data that happens in time that with chronological time
12:56
stamps, right? So we basically train AI by
12:59
playing this game of from point in time A, can you predict
13:03
what's going to happen in point in time day B And we can do this
13:06
over and over again over all of your messy data.
13:09
And by doing this, we kind of force the AI to really learn
13:12
what are the causal factors here that are going to tell me what's
13:14
going to happen at the next time step, right?
13:17
And if the AI gets good at this, you have an AI that really
13:19
understands what you wanted to understand.
13:22
Interesting. OK.
13:23
And so this is built on top of the LLM model.
13:26
So would you would do you, do you first decide on like we want
13:29
to use Grok or we want to use Claude, we want to use
13:33
Perplexity, we want to use Open AI?
13:35
Is that kind of the first step? We do use AI models in in kind
13:40
of playing this game and generating this data at
13:42
different points in the process. But we got to make it really
13:45
easy for you to swap out whatever models you want to use.
13:48
So typically if our companies work with are training their own
13:50
model, they're using one of the open source models often like
13:53
the Quinn models or, or different smaller, you know, 14
13:59
billion to 70 billion parameter models that are open source.
14:03
And we'll use a combination of things generated by those open
14:05
source models and questions and answers generated from these
14:09
bigger frontier models to put together the perfect data set.
14:13
Yeah, OK. So maybe a case study if that
14:15
would be appropriate for the audience.
14:16
We're all founders here and one thing we all have is the the
14:19
speed and pace of AI. You know, evolution is such that
14:24
like every quarter you really have to re evaluate your
14:26
product. Like, hey, do I have a Moat
14:28
here? Do I have something sellable?
14:30
Like what what's going on? It the, the talk that, you know,
14:33
software is basically reduced to 0 like there, there's so much
14:37
fear. So that will lead us to be like,
14:39
OK, what's the next thing? What do I, what's my next idea?
14:42
What do I need to do next? So if I'm, I'm coming to
14:45
lightning rod AI, my data set is going to be like, what are the
14:49
ideas that are kind of percolating in the, in the
14:54
ether, so to speak, right now? And how can I capitalize?
14:57
So I choose, I could go to grok because grok has all that data
15:00
that the primary source is going to be Twitter or X or whatever.
15:04
Or do I want to go to Gemini because they have all the search
15:07
data, all the trend data, they can see kind of where the puck
15:10
is going. And then I want to say like,
15:12
these are the my data sets. I'm coming to lightning Rod
15:14
because I want to make sense of this and I want to have
15:17
predictive value of like what the next big ideas are going to
15:20
be that kind of fits within the other data set, which is like,
15:24
you know, what are my core strengths?
15:25
What do I believe in? Like am I the right person to
15:27
bring it forward at this time? So would would that be an
15:30
opportunity that you could kind of build a tool that's going to
15:33
predictably give us some outcomes that we can work on?
15:37
Yeah. So typically, our customers fall
15:39
into, I'd say like 2 buckets, right?
15:41
One bucket. We have a bunch of data and we
15:44
don't know how to turn this into line.
15:45
We can train an AI model. Yeah, we're working with one
15:49
customer that wants to predict financial metrics, right?
15:51
So they have all this data on these different companies and
15:54
what they achieved in different quarters, right?
15:56
But these are like dense company reports.
15:59
This is not something you want to just feed into an AI model
16:02
for supervised fine tuning and have it talk like that, right?
16:04
So we can take that kind of data and help you turn it into
16:07
something that you can train the app.
16:09
The other bucket is companies that, you know, I want to train
16:12
a model that does whatever, right?
16:14
Maybe make predictions in the entertainment space, right?
16:18
But I don't have any data whatsoever.
16:19
How can I just generate data out of thin air that trains a model
16:23
to be good at that, right? And in those cases, what we'll
16:26
do is we'll make it really easy to pull data for you from, for
16:29
example, public news sources where we could look over all the
16:33
news over the last couple years and pull all the articles about
16:36
the different things happening in the entertainment world and
16:39
use that to create a training data set that may say I learn.
16:42
What's? Going on in this space and
16:43
really understand that domain. What's up everybody?
16:47
Let me hit you with some math that will blow your mind.
16:50
A wealth manager using Warm Start dot AI grew their assets
16:54
under management by 8.5 million in three months.
16:57
Their monthly fee? 500 bucks.
17:00
That's a 1700% return on investment in three months.
17:04
And I get it. Another AI tool, another monthly
17:06
subscription. But here's what makes Warm Start
17:09
different. It's not trying to find you new
17:11
people to bug. It's helping you reconnect with
17:13
people who already know you. As a founder, your biggest asset
17:17
isn't your product or your website.
17:19
It's your relationships. But most of us are terrible at
17:22
nurturing them systematically. Warm Start connects your e-mail,
17:26
identifies warm contacts, and helps you craft personalized
17:29
outreach. That doesn't sound like a robot
17:31
wrote it. The best part?
17:32
You review every message before it goes out.
17:35
No automation without permission.
17:37
If you're ready to stop leaving money in your inbox, try Warm
17:41
Start dot AI. Use code AIFF for 50 bucks off
17:46
your first month because the ROI on your existing relationships
17:50
is probably better than any investment you'll make this
17:52
year. That's Warmstar dot AI and use
17:55
Code AI F Now you and I know startups are built in
18:01
boardrooms. They're built in the wild during
18:03
3:00 AM. Bug fixes, botched launches, and
18:06
that one meeting where everything almost fall apart.
18:08
That's what Code Story is all about.
18:10
It's a podcast where tech founders breakdown the make or
18:13
break moments behind their products, what went wrong, what
18:16
went right, and what nearly wrecked everything.
18:19
Hosted by NOAA Labhart, Code Story goes past the highlight
18:23
reel and into the honest stuff, late pivots, brutal bugs, and
18:26
somehow coming out alive with a product.
18:28
People actually love if you're in the build stage and the scale
18:32
stage or the please Let this feature work stage, I can
18:35
relate. This show will speak to your
18:36
soul. Listen to codestory@codestory.co
18:40
or wherever you get your podcast.
18:42
Just search code story. It'll be right at the top
18:44
because great startups aren't born.
18:46
They're forged one hard moment at a time.
18:48
Code story.co. Nice.
18:51
So it's either people have the data or they need you to find it
18:54
first and before you can create the operating tool.
18:58
Had a great conversation with Mihu and he does something
19:01
similar, but largely what he's doing is like the data sets are
19:05
literally locked in filing cabinets.
19:07
Do do you find something similar where it's like it's like if
19:11
it's a legal firm or something and they're like, we have 50
19:13
years of case data and it's all on paper.
19:16
I mean, is that something you run into?
19:19
We do not help with the part where we scan the paper.
19:24
Well, we definitely deal with all sorts of messy digital
19:26
formats. And yeah, that's something we we
19:28
help our enterprise customers out with.
19:30
Sure. OK, great.
19:32
So when we're talking about data sets, maybe bring a little
19:35
something more concrete. Is this like sales and marketing
19:38
information? Is it Slack channels?
19:40
Is it, you know, your Google Workspace emails?
19:43
What are some popular data points?
19:46
Yeah. So I mean, the model that we
19:47
kind of came into the scene with was a forecasting model or we
19:50
just wanted to show that this works really well even when we
19:54
don't have some sort of special data advantage.
19:56
So using mostly just public news data, data that, you know, all
20:00
the big frontier models have been trained on and have access
20:03
to, we were able to kind of play this game, right.
20:06
We're given one point in time, make AI try to predict what's
20:09
going to happen by the next point in time, and use that that
20:12
process to train a model that was actually better.
20:15
And making predictions about the future then all the big frontier
20:17
models, right? And we measure this by having
20:20
mall place a bunch of bets on polymarket and our model was
20:23
able to be more accurate and produce more profit and produce
20:26
better calibrated probabilities then all the models we compared
20:29
it to, right. And this is a tiny model just
20:32
trained on this public data. So that's, that's one example,
20:35
some concrete examples of companies that we're working
20:37
with. There's the financial metrics
20:39
company that I just mentioned. For another one, we're trying to
20:42
predict alcohol clinical trials. So given this proposal and given
20:45
this experiment set up, you know, what's the likelihood that
20:48
this drug is going to make it to market in the next, let's say 24
20:51
months or something like that? We're working on predicting the
20:54
likelihood of flights being disrupted by getting a delay
20:56
that's going to be at least a couple hours long.
20:59
We're working with other healthcare companies to just
21:01
train models that they won't actually be making predictions
21:05
in their usage, but we're just going to use this kind of
21:08
predictive process to train a model that really understands
21:11
whatever specific healthcare problem that company is.
21:15
That's, that might be the evidence of like the, the
21:19
intelligence of the, the, the training day itself is that it's
21:24
able to predict so that it really understands it
21:27
essentially. Yeah, that's right.
21:29
I mean, at the end of the day, prediction really is like the
21:32
best test of intelligence, right?
21:34
But you have an expert that can explain to you why something
21:37
happened the way it happened, or if the market's crashing, you
21:39
know, everyone has a reason why the market's crashing, right?
21:43
But if you really understand what's going on, if you actually
21:45
have a better idea of the causal factors, you're better than me,
21:48
you should be able to predict. Yeah, exactly.
21:51
Everyone can explain it after the fact.
21:54
Should I? Should I think of this as like
21:56
MCPS? We, I mean, we use MCPS so we
22:00
can, we can trade models to not just reason over the information
22:04
that they're given, but learn how to ask for the right
22:08
information, how to use different tools to, to get to
22:12
the, the best prediction. I mean, the nice thing about
22:13
this game is it's, it's quite general, right?
22:16
So you can kind of train models to learn whatever patterns help
22:19
you make better predictions. Yeah.
22:22
You know, you brought up Polymarket and that you're
22:26
you're beating Polymarket. What's your take on that?
22:29
This is a little bit of a, a divergent here, but like what's
22:32
your take on polymarket and just this like I'll just leave it
22:36
there. What do you think about it?
22:39
Yeah, I mean, I think I have mixed feelings about it.
22:42
I I kind of love prediction markets personally.
22:46
Like, I think it's, it's a great way to kind of train yourself,
22:51
right? Like we're even making small
22:53
bets. If you don't, you know, take it
22:54
too far, it really forces you to put your beliefs down on paper
22:58
and then realize when you're right or when you're wrong,
23:01
which is great for calibrating your own confidence about
23:04
things. I think you learn a lot by
23:06
paying attention to how well do I actually do when I put my
23:08
money down on things. On the other hand, you know, I
23:11
do think a lot of people have a gambling problem and explosion
23:14
of a new way to gamble is probably not going to help with
23:17
that. Yeah, it's it's kind of, I mean,
23:19
that's just the the kind of downside.
23:21
You're right. It's like put your money where
23:22
your mouth is and like, are you really interested in learning?
23:25
Well, here's a great incentive to do it, but it seems to be,
23:30
and I'll put it out there. I'm a Chauncey Billups fan
23:32
because I'm a Denver guy. So to see what's going on with
23:35
the NBA and like all of that in that this is just going to
23:37
extend indefinitely in all directions now, particularly
23:40
because all those gaming platforms are getting into kind
23:43
of polymarket type stuff too, because why not, you know, why
23:47
not just bet on sports when I can bet on, you know, is my
23:50
newspaper going to get delivered today or whatever?
23:52
You know, it's, it's, it's I, I, I really like it.
23:57
You know, I, I don't have a gambling problem.
23:59
So it's just like a novelty. If I pop something open and I
24:02
want to, you know, look at like, you know, who's going to have a
24:05
larger November, NVIDIA or Apple or, you know, like, all right,
24:08
it's it's kind of silly and fun, but it seems to it seems to be
24:12
detrimental to to culture in some way.
24:16
For sure. I mean, the incentives that make
24:18
prediction markets work well are both kind of a blessing and a
24:21
curse, right? It's a blessing in that like,
24:24
there's such a strong incentive that if someone has information,
24:27
they should place a bet and share that information with the
24:29
rest of us, right? We can all look at Polymarket
24:32
and get the best odds on who's going to win an election.
24:35
But it's also a curse because now a lot of people have an
24:37
incentive to place bets on things that they probably
24:41
shouldn't. Yeah, exactly.
24:43
Oh man. Interesting.
24:45
OK, We're going to talk about you a little bit.
24:47
One thing that's kind of striking about your website is
24:50
there's you have your hero section.
24:52
Then right under there is just the founders and it's you and
24:55
Danny. And both of you guys have had
24:56
like looks like several exits and really solid careers and
25:00
have a lot of experience in machine learning, automation,
25:04
AI. You, you've been working on this
25:06
for some time. When you're thinking about like
25:10
developing this and it, would it be accurate to say this is kind
25:13
of like a service as a software? Do you feel like this is a
25:16
productized service? How would you cut it?
25:18
What's the business model? Yeah.
25:20
So until now, these have been pretty much internal tools that
25:23
we've been using to train models for our enterprise customers and
25:27
these have been mostly like bigger enterprise deals.
25:30
What I'm really excited about now is we're actually about to
25:32
launch an SDK and basically make this a self-serve developer,
25:36
developer first product that anyone can use the train better
25:40
right out, amazing, but it'll be like swipe or credit card, just
25:44
pay for what you used. And this framework that we've
25:47
found to be really effective for generating high quality training
25:49
data, we're going to bring it to everyone.
25:51
So I guess that is to say our business model is changing
25:54
quickly, but in the near future I hope for it to be very much
25:57
developer first and a much more scaled model.
26:00
Incredible. So yeah.
26:02
And then what does go to market look like for that?
26:04
Do you rebrand? Do you what?
26:06
What are exactly are you planning?
26:08
Yeah. I mean, so one insight here was
26:10
just that our champions at all of our big customers were people
26:15
who understood machine learning and had some Python experience
26:18
themselves. And I think you just kind of you
26:19
need that experience to understand the differentiation
26:23
here and like what we're bringing to the table.
26:25
And so when we realized that, we just realized it probably makes
26:28
a lot more sense to go to developer 1st and make, you
26:30
know, attack the communities where these people are, right?
26:33
So hug me, face GitHub, go after the Python Reddits and and
26:38
groups like this. And so, yeah, I mean that that
26:41
kind of informed our strategy. We're kind of packaging up all
26:44
of our best practices into this framework that we're hoping to
26:47
release in the next couple weeks.
26:49
Cool. Do you do you raise around that
26:51
or how or have you raised for the project so far?
26:57
So we raised a bit over two years ago.
26:59
Our most recent round, we haven't raised like on, on that
27:04
kind of new developer first direction.
27:08
You know, my approach is if we don't need to raise, I'd rather
27:10
first launch and improve the traction and then raise on that.
27:13
And we'll, we'll probably do a much better round if we can make
27:16
that happen. And right now we're in a good
27:17
place from a runway perspective. So that seemed to be the right
27:20
approach. Tremendous cool moving into kind
27:23
of this this SDK kind of format. Who do you see is going to be
27:27
your your first clients or who could benefit most you think
27:30
from this? I mean, I think Python
27:33
developers either within big enterprises or at start-ups or
27:37
even just indie developers will kind of love a product like
27:40
this. You know, one thing I think is
27:43
Python And like machine learning engineers are very different
27:46
than your average software and share.
27:48
And so if you pitch Hugging Face to your average software and
27:50
share a couple years ago, yeah, the average software engineer
27:53
would be like, this is just cloud storage.
27:55
Like why would this be a product?
27:58
What about Python guys? And no, I know because this is
28:00
my background. This is what I did at at Google
28:02
before starting and start up. We don't really like dealing
28:06
with infrastructure and and we don't really like worrying about
28:09
servers or about, you know, signing up for different APIs.
28:13
Do you want to work in our notebooks and play with the data
28:15
and just, you know, be able to measure what we want to measure
28:18
as quickly as possible. So a lot of what we're doing
28:20
here is abstracting away a lot of the work of having to build a
28:25
complicated data infrastructure yourself and making it really
28:28
easy as a Python machine learning guide to just get
28:31
going. Just give me the data I need to
28:33
train the model that I want to train.
28:35
Yeah. Which do you think will be kind
28:37
of the bigger winner, the actually going and retrieving
28:40
the data or incorporating existing data?
28:45
That's a good question. I think it will depend on the
28:47
kind of developer. I think for startups and indie
28:52
developers often don't have, you know, new new startups and indie
28:55
developers often don't have data sets and be able to generate
28:59
something on the fly from like public news or SCC documents or
29:03
some something like that is really valuable just for getting
29:06
a project off the ground. Big enterprises tend to have the
29:09
opposite problems that they actually have tons of data.
29:11
It's just not at all in a good format for training or
29:15
evaluating LMS. And if you can just drop all
29:18
this data somewhere and have that happen for you, that's kind
29:20
of magical A. 100%, man, I can't stop thinking about medical.
29:29
I think actually like the personal data is what I'm most
29:33
interested in. Yeah.
29:34
And I suppose because it's most valuable to me.
29:38
If you're thinking about data sets, what are the most valuable
29:41
data sets right now across the board?
29:46
If you already said unplugged, I thought.
29:49
I mean, look, I also love the medical use case and if there's
29:52
anything I, if I could make a medical data set appear out of
29:57
nowhere that we could play with, like I would love that just
30:00
because I think predicting medical outcomes be such a
30:05
powerful kind of modeling. You'd probably make you'd
30:07
probably make scientific discoveries in that process of
30:10
even destroying a model that was better at predicting medical
30:14
outcomes. And then what we can do today,
30:17
that's going to be a really hard data set to get.
30:19
And I don't, you know, I don't even know all the regulations
30:21
that would have to be broken to to make that happen.
30:25
But yeah, I mean, I I would love to to make that happen one day.
30:29
Yeah, you know, it it the, the medical applications too.
30:32
It's and I've spoke to some founders, they're doing amazing
30:35
things. And the problem is, is that it
30:37
comes with increased data, comes with increased liability.
30:41
So you have all this data that's available now, you have
30:44
potentially predictive or you know, prescriptive like
30:49
recommendations from LLMS definitely into the future.
30:52
But if a doctor receives that information and then doesn't
30:57
make the necessary or appropriate applica or
31:00
prescriptions, then they're held liable.
31:03
You know, they could be sued for that, You know, So we're, this
31:06
is just kind of like one of the weird kind of compliance, like
31:09
growing pains that that's kind of compiling very quickly.
31:13
And maybe where it's most concerning to me is because
31:16
medical industry is just notoriously slow, you know, on
31:20
everything. And maybe for good reason.
31:22
But like, if we're sitting on like a gold mine of information,
31:26
let's say there's some kind of gold cap tooth I could put into
31:30
my mouth and it smells my breath and it knows exactly what's what
31:34
my iron is or what whatever. You know, there's going to be
31:36
all kinds of massive data. And the doctors are like, we
31:39
cannot do anything with that because if you send me that
31:43
information now, suddenly I'm liable for any potential outcome
31:47
down the road. You know, so maybe there's
31:49
opportunity there where you have people that aren't necessary
31:52
doctors and you get into kind of like the, the kind of like
31:57
medical care for the, the healthy, you know, actual
32:01
healthcare, not sick care, You know, where it's like, OK, we,
32:04
we can take these bio stamps, we can predict that these could are
32:08
some potential outcomes. They're pretty good at that
32:10
already, you know, And then like, here's some things you can
32:13
do to supplement you, not essentially buy your Apple
32:17
Watch. I think it's a real, I think
32:20
what you're pointing out is like a real problem, right?
32:21
Which is that the most important domains are where we kind of ask
32:25
for the most regulation, right? Because it's important.
32:28
So we need to, you know, have guard Reds, make sure things
32:30
don't go off the rails, but they're also important, which
32:33
means we need the most innovation to help solve those
32:35
problems, right? And so Healthcare is definitely
32:37
a place where we have all these regulations.
32:39
A lot of them are there for for good reasons, but especially now
32:44
with the power that AI is bringing to really disrupt the
32:47
way we solve a lot of problems, Man, it would be a shame if,
32:52
yeah, I couldn't come in and do the things that we know it could
32:54
do just because all these regulations get in the way.
32:57
So there might be a New Balance or some new way when you start
33:01
thinking about these problems. Yeah, Elon will figure it out.
33:05
In your hopes. In order to get like hit your
33:10
Tesla medical prescription is going to be some stack of papers
33:13
that you're like signing away you're your life to some doctor
33:17
in the Cayman Islands or whatever.
33:19
I mean, there'll be some kind of weird work around like that.
33:22
How about you? When you're looking in the
33:24
future of the five years down the road, what's really exciting
33:26
to you? I mean, I think there's a ton to
33:30
be excited about. There's a ton to be worried
33:32
about. I think just we cannot possibly
33:34
begin to understand how much, you know, things are going to
33:37
change over the next couple years as a founder and a
33:41
creator. Like I already feel like the
33:44
ability of AI to just make execution so much easier is like
33:48
incredible. Like the gap between having an
33:51
idea and making it materialize in some way or trusting it is
33:55
just getting smaller and smaller.
33:57
So I just think there's going to be this explosion of creativity
34:02
in education, right, people? You know, there's such a barrier
34:05
now to learn about all these things that you used to need to
34:07
go to school to take all these classes or get the right
34:10
professors, to educate yourself on different subjects.
34:13
And I think that's all disappearing.
34:15
So I just think we're going to see a lot more people doing a
34:18
lot more interesting things because the friction involved
34:21
has gone away. Yeah.
34:23
And it there's a curiosity. This is kind of a predictive
34:27
kind of element too. That's like really important
34:29
with LLMS when I'm really like kind of in the pocket and really
34:32
having fun with these models, which is usually like, you know,
34:35
Gemini 3 comes out and they're like, oh, beat all the
34:37
benchmarks. I'm like, all right, well, I'm
34:39
going to hit it with my shtick and just start like talking to
34:42
it, asking it weird questions have asking it to bring
34:45
disparate things together and some kind of narrative that
34:47
makes sense, you know, and that's really fun.
34:51
You know, I, I think that like one thing I, I talked to my kids
34:54
about is, you know, they're great for using LLMS as a tool.
34:58
Like, Hey, how do I do this math problem?
35:00
And it'll walk you through step by step, which is great.
35:03
How do I translate this into French?
35:04
But only can do it in two seconds, which is great too.
35:07
But like, then it, it just becomes like, I have a problem.
35:11
Give me the solution as opposed to going it with kind of a
35:14
curious mind of like, what problems should I be thinking
35:17
about? You know what I mean?
35:19
Like kind of giving up creative agency because it's really good
35:23
at kind of like presenting you with kind of an uncarved block
35:27
and allowing you to kind of chisel away.
35:29
If you're not so certain with the answer that you're trying to
35:31
get out of it. You're not like it's not that
35:33
this to this, to this, you know, So I, I wonder too, like how
35:37
much better it's going to get, you know, because it, it really
35:40
seems like the LO Ms. are, are still in kind of an infancy.
35:45
I mean, maybe not that they might be like 3rd or 4th grade
35:48
now, but like once they're seniors in high school, to just
35:52
hold the metaphor here, or they're in grad school and like
35:55
now they're way smarter than me. And it, you know, it's, it feels
35:59
like the, the way I want to interact with the LL Ms. is
36:04
going to need to change, you know, in order to get the best
36:06
out of it to, in order for it to, to create all these really
36:09
interesting creative solutions, you know, because it just, it
36:12
has the ability to like pull so much together.
36:17
I don't know. That's right.
36:18
I mean, one thing that I love doing is kind of flipping it
36:20
around and asking the AI to ask me questions.
36:24
So it's really easy to like listen to a podcast or an audio
36:26
book and think you understand things unless you're being
36:30
quizzed on it. Kind of I love like taking some
36:34
idea and telling the ad, like quiz me to check my
36:36
understanding. Let's find the whole and then
36:39
explain to me like what I don't really understand.
36:41
I mean, that's a much more effective way for me to learn
36:43
than just, hey, tell me about this thing and me acting like I
36:47
registered all those tokens. Yeah, I know somebody else told
36:52
me about that like to always follow your prompts with like
36:55
and ask me 3 questions to like better understand what I'm
36:58
trying to get at here because a lot of times it's through the
37:01
back door that you get the best answers, you know.
37:05
Cool. I want to talk to you about your
37:06
Co founder because it looks like both of your previous companies,
37:09
you were involved with them. This is a a new company, what's
37:13
the the dynamic of relationship like with your Co founder?
37:17
Yeah. I mean, we've worked together
37:19
for a really long time, almost eight years now and we've always
37:22
worked on these kind of mission driven AI first products
37:26
together. So the first thing we worked on
37:29
together was actually while part of Google, I founded a company
37:32
called the Brivet with an Area 120 that was a reading app for
37:36
kids. They used the ad to listen while
37:37
kids real out to the app and give them feedback on the
37:39
reading. Danny came on board to lead the
37:42
engineering team and we kind of hit it off and I think we made a
37:46
great team and we kind of just kept working together since
37:49
then. When I decided to leave Google
37:50
and start start up, he was the first person I asked to be my
37:53
CTO and he was excited to come on board.
37:56
And so, yeah, I mean, we've been working together for a long time
37:59
and it's it's a good dynamic. What what do you think is the
38:02
attributes of that dynamic? What makes it good?
38:04
What makes it challenging I. Think you know, we really we
38:10
balance each other out in in important ways.
38:14
I think I can be very creative and be a little all over the
38:18
place and be very quick to abstract ideas and Danny's very
38:22
good at like him and bring it back down to earth.
38:23
Like OK, concretely what should like the input here be and and
38:27
what are we going to get on the other side kind of thing.
38:30
We disagree really well, I think that's that's really important
38:34
for Co founders. I think we're just emotionally
38:39
mature. And if there's a disagreement,
38:41
like I genuinely want to understand why he disagrees with
38:44
me. And he tends to do the same.
38:46
And so, you know, we're never just like yelling at each other
38:49
or something like that. And, you know, we've never
38:52
really had something where we didn't end up aligning after
38:55
talking about it, like where we had to be like, I'm just, I'm
38:58
making the call for us that I don't think that's ever
39:00
happened. So yeah, I don't know.
39:04
I mean, I think we balance each other out well and we just, we
39:07
seem to be able to disagree in a productive way to get on the
39:11
same page I'm. Always interested if there's
39:14
some great secret, but ultimately I think it's just
39:16
like any good relationship. It's like, oh, it's just kind of
39:18
easy. I don't know.
39:19
Yeah. I think that's right.
39:22
And I think, you know, people have different superpowers and I
39:25
think we both kind of respect each other's abilities that that
39:30
we bring to the table, right. So I've, I've never seen an
39:33
engineer like Danny. He's just able to whip things up
39:35
and understand how you could tie things together to make
39:38
something happy and happen really quickly.
39:41
And yeah, yeah, I don't know, it just works.
39:45
Cool, amazing. We are about to enter the
39:49
lightning round. Are you ready?
39:51
I guess so let's do it. Let's go.
39:53
OK, so what's most unsettling about AI for you right now?
39:57
And you've been amazing guests to letting me like going all
40:01
these weird directions. But like, if, if there's
40:03
something unsettling, what what sticks out for you right now?
40:09
Well, I mean, I think there's going to be a lot of disruption
40:11
and it will be uncomfortable, to say the least, I think.
40:17
I don't think as a society we're having the conversations we
40:21
probably need to be having about like what society look like when
40:25
there's literally nothing I can do better than a machine, right?
40:28
If that's not happening next year, it's probably happening in
40:31
in 20 years. And I don't think that's like
40:34
the industrial revolution. A lot of people like to make
40:36
that comparison, but there were plenty of things the machines
40:39
still couldn't do at the end of the Industrial revolution.
40:43
And so I think, you know, that's going to be a problem.
40:47
And we're really going to need to think about like, what gives
40:49
us value as members of the society and what do we want to
40:53
do with our time to make this a worthwhile experience when when
40:57
we don't need to work and kind of earn our keep that way.
41:01
It's interesting, I kind of think that like we'll have to
41:03
learn from like the classic aristocracy.
41:06
And I only say that because I remember was watching some
41:09
stupid dating show years ago and there it was like this Italian
41:13
Duke or something like that. It's beautiful man, unbelievable
41:16
hair. Like just he was incredible and
41:19
he was always getting so frustrated because he's like
41:22
Americans, They they all want to know, like, what do you do?
41:26
Like we really like it. It it doesn't matter who you are
41:28
in the social strata. It's like, oh, what do you do?
41:30
And he was offended by that because he didn't have anything
41:33
to do. You know, he's getting fitted
41:35
for suits and like going to balls or whatever.
41:37
I mean, I don't know what, what aristocracy does, but what they
41:42
do is they, they, they have more of a refined taste, I suppose.
41:45
I mean, so it's like if, and I would love to think it's
41:49
optimistic to think we're moving in that direction where what we
41:53
have to do is it necessarily what defines us as people rather
41:56
as kind of like the attitude and culture we we cultivate, you
42:00
know? Yeah, I mean, that sounds nice.
42:03
I hope that. Sounds nice, 100% OK like like
42:11
like I said, like the the audience loves business ideas.
42:15
Do you have any like, interesting ideas rattling
42:18
around that you'd like to to bring to life at some .0?
42:27
Yeah, I always have a bunch of random ideas.
42:29
I don't know if they're any good.
42:31
The bad ones are the best. Yeah, sure.
42:34
I mean, I think we still don't have an AI experience.
42:38
That's that's kind of where the AI sets the stage, right.
42:44
It's very the user prompt first still.
42:47
And I would love personally something where I put on my
42:50
headphones and the AI is just going to talk to me about things
42:54
I want to talk about. But it's an interactive
42:56
experience and I'm learning and just like TikTok or something
42:59
like that, it's kind of adjusting the algorithm based
43:02
on, you know, when I seem bored or when I seem uninterested or,
43:05
or whatever. I think someone should nail that
43:08
experience and that'll be I'll. I'll use that when someone does
43:12
that. Yeah, yeah, I'm really
43:15
interested in that too. You know, like, and I wonder if
43:18
if voice is the ultimate application, you know, typing.
43:22
I'm kind of like, I don't want to type if I don't have to.
43:24
It seems, it seems half baked talking to it.
43:28
Maybe I I really am just waiting for the day when I can kind of
43:32
use like maybe involuntary things that that I do physically
43:38
or somatically, but and just like create shortcuts with
43:41
those. There's like if I wiggled my
43:43
ears, it would make a sandwich or something like that.
43:46
I think we're, we're, we are working into kind of some kind
43:48
of God realm where we're going to be able to program the world
43:52
around us to our suiting, you know, to the, I'm excited about
43:56
that. Yeah, when it's easier to kind
44:01
of maybe input and output easier than even talking or like
44:05
typing. Totally.
44:07
Read our brains or something like that.
44:08
That'll be both exciting and scary.
44:12
Yeah. What would be a game changer
44:14
client if you could, if you could have the the client that
44:17
has the most incredible data set you'd love to work with them.
44:20
Who would that be? Well, we've been working on
44:23
getting clients in the DoD and we'd love to make some headway
44:28
there. It's obviously a tough time to
44:30
be doing that for a number of reasons.
44:33
But you know, I think like when it comes to like forecasting,
44:36
forecasting risk and geopolitical risk is one of the
44:38
most interesting things that we can do.
44:39
And we've shown that our models are, are really good at that.
44:42
And so I'd love to put that to use in a way that actually makes
44:45
people safer and maybe makes our our our government more
44:49
responsive to threats. Doesn't that seem like
44:52
unbelievably scary? Yeah, maybe.
44:56
Yeah, we might as well just go all the way in if we're talking
45:02
about the DoD. And if you had some classified
45:05
information about the Uaps, what would be your recommendation or
45:11
actually, how could, how could the government, who seems like
45:13
they're going Full disclosure at some point, what do you think
45:16
the predictive model would be to release disclosure documents?
45:23
Like what's the likelihood that those get disclosed?
45:26
No. Like how should they do it?
45:27
So you'd you'd need a predictive model to be like, well, if you
45:30
do it like this, then that, then this, then that.
45:32
Because it seems like they, whatever model they're running
45:35
on disclosure, it seems to be doing a pretty good job because
45:38
it just seems to be seeping in. I don't know.
45:41
Right. Like, yeah, like somehow they
45:45
released information in a way where it's no longer mind
45:49
blowing that there are these Uaps everywhere and they're
45:51
real, like clear sites and military sites.
45:53
And we've kind of accepted that without going like, holy shit,
45:56
what? What's going on?
45:58
Yeah, yeah, that's interesting. I don't know someone.
46:01
Yeah, maybe someone's really thinking about that.
46:03
It was like 4 or five years ago where the Pope was like, you
46:05
know what? If there's things outside of the
46:07
earth, Jesus loves them too. I'm like, whoa, wait, what?
46:11
Why are you saying that? I that seems a little out of the
46:13
blue, You know, I don't know. Nice.
46:18
How about other entrepreneurs or CE OS that you follow closely or
46:22
admire? Anybody come to mind?
46:24
Well, I think I have to say Benjamin Franklin, because he
46:27
invented the lightning rod, obviously exactly that kind of
46:31
scientist, innovator, entrepreneur that I think we we
46:35
like to be more like. 100% plus, you know, he he might have had
46:41
some of the original podcasts. You know, I always think about
46:43
Ben Franklin in the French salons with his Coon skin hat,
46:47
you know, hammered it up for the gals over there, you know?
46:50
Yeah, that's right. I love that guy.
46:53
Amazing. How about founder LED marketing?
46:55
I kind of got halfway through that question.
46:57
You know when you get to your website, first thing you see is
47:00
the 2 founders. Why is it important that you're
47:02
so front and center in your business right now?
47:04
Well, I meant you're. I meant you're looking at the
47:06
adult page. Oh, am I on the about page?
47:11
I think so. Oh, I am.
47:13
How dare I? Yeah, I don't know.
47:18
I, I actually think that's an interesting line to walk.
47:22
It's definitely, you know, something we've been trying out.
47:25
I try to post on LinkedIn and be a bit more active on, on places
47:32
like Twitter and things like that.
47:33
And that's definitely something some founders have a lot of
47:36
success with. I also think it can definitely
47:38
be a distraction and it can make it more about you and less about
47:42
the product sometimes. So I don't know, I think it's
47:45
something we're experimenting with.
47:47
It's a line that I think we we have to learn to walk.
47:50
But it's certainly nice to be able to, you know, put
47:53
information out there. And when you get a group of
47:55
supporters that is kind of a fan of what you're doing, which
47:57
we've definitely found in the forecasting community, that's a
48:01
huge asset. 100% cool. As we're kind of wrapping it up
48:05
here, we like to give shout outs to charitable organizations and
48:08
you mentioned one that I really like too, which is MAPS, the the
48:12
Multi Association of Psychedelic Studies, Multidisciplinarian
48:17
Association of Psychedelic Studies.
48:18
Why? Why is that interesting to you?
48:21
Oh, you know, I just think that early studies on using
48:24
psychedelics for mental health are incredibly promising and
48:29
we've kind of only demonized and made these things legal for
48:33
weird historical reasons that I think everyone agrees now makes
48:36
no sense. So there's one thing that I
48:40
think it alleviate a ton of suffering in the world really
48:42
quickly and also help normal people learn more about
48:47
themselves and lead and lead more fulfilling lives.
48:50
It's probably making psychedelics go mainstream.
48:54
So, you know, I don't know, I'm just a big fan of what they're
48:56
doing over there. It's really interesting how
48:59
psychedelics have reemerged into the culture in the last 1015
49:02
years in like a very clinical way.
49:05
You know, I don't think psychedelics ever went anywhere
49:08
too far. Well, I mean, I live next to Red
49:10
Rocks, so I know they didn't. But it, it's interesting that
49:14
the therapeutic aspects, particularly as we're, we're
49:16
talking about what you had said, which is like, we're all going
49:18
to have to reassess, like what's, what's my meaning if
49:22
machines can do everything better than I can.
49:25
So it's, we have to shift kind of like our personal identity to
49:28
what I do to like how I am in the world.
49:32
And that's one thing that like the role of the shaman or, and
49:35
with their satchel of psychedelics.
49:38
I think you've always brought humanity.
49:40
And I mean, there's examples, whether it's Soma or it's, you
49:43
know, Kaichian or like there's all these like kind of like
49:46
mythological substances that cultures like big civilizations
49:50
were using to, to help people connect with themselves in
49:54
certain ways. Experience.
49:55
We're all a little bit different, maybe find some new
49:58
appreciation. And I, I don't think it's
50:00
necessarily like coincidence that like, as we're kind of
50:05
bringing these machines that can maybe greatly like enhance or,
50:10
or take away the beast of burden from all the tasks we do.
50:13
That also is emerging these like different applications,
50:16
different techniques for all kinds of different religious
50:18
backgrounds or medical backgrounds to, to help people
50:22
connect to how they are in the world, because this will be a
50:25
skill I think they'll really need.
50:27
You know, so I mean, for folks that don't know about MAPS and
50:31
Rick Doblin, like the studies they're putting out, you know,
50:34
these are clinical, you know, I mean, placebo based tests with
50:38
psychedelics is kind of ridiculous.
50:41
Yeah. Like I'm pretty sure I got the.
50:45
Yeah. Very hard to ping it, but you
50:48
know, particularly for for people that aren't known for
50:52
substances, you know, which would be like, let's say
50:55
religious people or military people, just the, the, the
50:58
results are like unbelievable. So, you know, they're doing
51:01
really good work and they're really fighting the good fight.
51:04
So shout out to those guys. Yeah, no, definitely.
51:09
And I think you're on to something there.
51:12
There seems to be some connection with where we're
51:14
going in society and with technology and AI.
51:18
And if you look through history, like a lot of cultures use these
51:22
things for coming of age ceremonies and finding your
51:24
place and adapting to change and accepting changes in your life
51:28
that you know that that accepting whatever changes in
51:33
your life you need to learned to settle with.
51:36
And I think we're going to go through a lot of change in the
51:38
next couple of years. We're going to Start learning
51:41
meeting to figure out how we reincorporate those kind of
51:44
ceremonies and and find our place.
51:45
So. Yeah.
51:47
The age of the psycho dogs again.
51:49
Yeah. And, you know, one of the, I
51:51
think the most prominent studies that they did were people at end
51:56
of life. You know, you got a terminal
51:58
case. You only got so many weeks,
52:00
days, months to live. And the, the anxiety and stress
52:05
and depression that goes along with that is it's, it's like an
52:08
onion, you know, there because on one side, oh, we're all kind
52:11
of scared to die, but on the other side, you're like, oh, I'm
52:13
deteriorating and I'm a burden on my family.
52:15
And you have like all of these different layers.
52:17
So that, you know, basically what they found is that, you
52:21
know, people that were using some of these methodologies were
52:25
more comfortable. They're comforted as they were.
52:28
I'm dying now. That's a very visceral the end
52:30
of life, which we'll all face someday.
52:32
But we are kind of coming to the end of, you know, how we used to
52:36
live. You know, there's going to be a
52:37
death and a rebirth. So it'll be interesting if if
52:41
kind of like there's going to be a similar onion of pain, you
52:44
know, because as you know, each of our industries are disrupted
52:48
to some place or another. Maybe it, it takes 10 years for
52:53
for me and maybe it took one year for somebody else.
52:55
We'll all have to kind of unpack all of these like various pain
53:00
and suffering to get to a place where you're potentially at
53:03
peace. So I'm really glad that you
53:04
brought up maps and people can check that out.
53:06
It's maps.org. Awesome.
53:09
Cool, the website is lightning rod dot AI Ben, what's the best
53:16
place for people to connect with you online?
53:19
I'm on LinkedIn, I'm on Twitter at B Turtel and I have a sub
53:24
stack Turtles all the way down. Love it.
53:29
Probably my favorite song from Oh boy.
53:31
What's his name? Oh man, I got Turkey brain.
53:37
He shouldn't have eaten that Little Caesar's pizza, I tell
53:39
you what. Country guy.
53:42
You know what? I'm talking.
53:45
About Oh my gosh, turtles all down country.
53:49
This is embarrassing. Sturgill Simpson, of course.
53:54
Great song. I'll.
53:56
Have to give it all. Great principle.
53:58
Cool. Ben, thank you so much for
53:59
joining me. I really appreciate your time.
54:02
Thanks for having me, great chatting with you.
54:06
This episode is brought to you by MO Burst, the AI first growth
54:10
engine trusted by iconic startups like Calm, Robin Hood,
54:14
Reddit, Uber and Google. Moburst is a digital, full
54:18
service, mobile first marketing agency that helps companies to
54:21
scale and become category leaders.
54:24
Whether you're finding product market fit or ready to
54:27
accelerate growth, Moburst delivers creative, data-driven
54:30
solutions that help with scale and efficiency and impact.
54:34
At Moburst, we're not just guessing what works.
54:37
We're predicting it. As an AI first agency, Moburst
54:40
uses cutting edge tools to analyze user behavior, adapt
54:44
creative in real time, and cover hidden growth opportunities.
54:48
The result? Marketing that's faster,
54:50
smarter, and built to convert from early stage startups to
54:54
global brands. We've helped hundreds of
54:56
companies overcome complex challenges, unlock new
54:59
audiences, and optimize every metric that matters.
55:03
At Moeburst, we believe that creative thinking, advanced
55:06
technology, and data are the core drivers of success at
55:10
scale. If you're serious about scaling
55:12
your business with precision and purpose, it's time to talk to
55:15
Moeburst. Visit moeburst.com and let's
55:18
build something category defining together.