March 8, 2025 • 47 mins
How can every single person build a personal AI protégé and then accumulate (and share) a host of other assistants? In this episode, we dive into the world of no-code AI with Scott Meyer from Chipp.ai. We discuss AI tooling for people that can't code, the cultural shift that needs to happen for widespread AI adoption in businesses, and the predicted growth trajectory of AI assistant that you can own.
Featuring:
- Chris Benson – Website, GitHub, LinkedIn, X
- Daniel Whitenack – Website, GitHub, X
- Scott Meyer - LinkedIn, X
Sponsors:
- Domo – The AI and data products platform. Strengthen your entire data journey with Domo’s AI and data products.
Show Notes:
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Jared (00:04):
Welcome to Practical AI, the podcast that makes
artificial intelligencepractical, productive, and
accessible to all. If you likethis show, you will love the
changelog. It's news on Mondays,deep technical interviews on
Wednesdays, and on Fridays, anawesome talk show for your
weekend enjoyment. Find us bysearching for the changelog
(00:24):
wherever you get your podcasts.Thanks to our partners at
fly.io.
Launch your AI apps in fiveminutes or less. Learn how at
fly.io.
Daniel (00:44):
Welcome to another episode of the Practical AI
podcast. This is DanielWitenack. I'm CEO at Prediction
Guard, and I'm joined as alwaysby my cohost, Chris Benson, who
is a principal AI researchengineer at Lockheed Martin. How
are doing, Chris?
Chris (01:00):
Oh, I'm feeling pretty chipper today. It's a a good day
to talk about AI.
Daniel (01:05):
Yeah. Yeah. I I feel feel quite quite chipper as
well, especially as we we've gotour guest today, Scott Meyer
with us, who's, founder and CEOat Chip, which is, you can find
at chip.ai, I believe, is is thelink. But, yeah, Chip is
awesome. Also, Scott is awesome.
And, also, Scott is is along agood friend because he's a a
(01:30):
fellow member of the the SiliconPrairie, not living on the on
the coast, but out here in themiddle somewhere where AI is
really blossoming, if you didn'tknow.
Scott (01:40):
It is. And it gives an unfair advantage for those of us
in non metro areas. The abilityto leverage AI to have the power
of 10 people in a place thatdoesn't have enough people to do
the job, it's perfect. It's aperfect solution. So it's great
to be here live from Fargo, justlike the movie.
It's it's fantastic to see youall and, be heard by all of you
(02:01):
listening.
Daniel (02:02):
Yeah. Yeah. Scott, we'll we'll we'll get into all the
cool stuff, you're doing withChip and some of the things
you've learned through that. ButI'm wondering if you work in the
space of, I guess we might putit like low code, no code AI
assistant builders. So for maybeaudience members that aren't as
(02:24):
familiar with that space, ormaybe they're just kind of
wondering what's out there, asof today, could you paint a
little bit of a picture for usfor kind of what sorts of tools
are out there?
And then maybe that would kindof motivate some of the unique
things that you thought shouldbe out there but weren't, which
(02:45):
would maybe kind of highlightsome of the things you're doing
with Chip.
Scott (02:48):
Yeah. No, it's great to be here. I think the stat that
blows my mind is that almost 50%of Americans use AI every week,
but 7% of businesses use AI,which is obviously a lie because
50% of Americans are using AIevery week and they work at
those companies. So what'shappening is the businesses
aren't, they have no idea what'sgoing on. It's like the early
(03:08):
days of cell phones wheneveryone would come to work with
their own cell phone, their ownlaptop, do whatever they wanted
to.
And eventually we got to thispoint where you get a company
email, you get company apps, youget like the standard way to do
it. I think the risk right nowis that, and the opportunity is
those who are willing to haveagency and try stuff have unfair
advantage, right? So I can go domy work with AI. And if my
(03:28):
colleagues don't know, and Idon't have a culture of sharing,
like all of a sudden, I'm asuper, superhuman. The number
one thing I tell businesses whenI meet with them is you should
have a lunch and learn once amonth and just have people say
what they're doing.
Because just that horizontalsharing of AI practices and
ideas is all you need to build aculture of acceptance. What
(03:49):
makes AI so unique is it's nottop down. It's not the CIO or
CTO saying, I bought this thing.You guys all go use it. It's
each individual figuring out howthey can use it for their
specific tasks.
What I've seen is adminassistants, marketers, interns,
right? They're all gonna use itdifferently and often even know
better how to use it becausethey're the ones doing the
(04:09):
tasks. And that kind ofmotivated what we built with
Chip, which is how do we justmake AI as easy as possible to
use? Our motto is AI for all.And I think I've spent most of
my professional career workingon bridging a digital divide
because maybe like you, peoplethat work and live alongside me
in Fargo aren't always takingadvantage of the latest
technology, right?
(04:29):
And so I kind of feel like it'sboth a passion and mission to
bring what's happening and makeit accessible to those around
me. In 02/2009, I started myfirst company and I was trying
to tell businesses there's thisthing called social media they
should use, right? Before therewere Facebook pages and Facebook
ads. And it feels like that tome again, almost twenty years
later, where it's like thisamazing power is right here and
(04:51):
the best time to start learningis now. With tools like Chip and
others that we can talk about,it's actually better now than
ever for people who aren'ttechnical because it's not about
technical ability, it's aboutknowledge and agency.
And I think we all have that. Sohappy to give a landscape. Think
that already went off track fromyour question, but hopefully
that gives you the starting No,
Daniel (05:12):
that's awesome. What would you say are kind of some
of those things that might makeAI hard to use? And here, mostly
we're talking of course, we'vetalked about a lot of things in
the show, but mostly we'retalking about kind of what
typical people would consider AInow, which would be kind of
generative AI, language models,maybe vision models, etcetera.
(05:34):
So, like, what what can makethose difficult to use or or or
how might people getdisillusioned as they're
exploring the technology?
Scott (05:43):
I'll say almost almost every excuse people have not to
use AI tools is fear. They arescared of a blank page. This is
the same with technology fortwenty years. I taught
entrepreneurship. I startedentrepreneurship centers and all
these students with amazingideas.
And you know what? 90% of themdidn't do anything because they
(06:04):
had to actually go do something,right? And it's like, you just
have to start. And I'm convincedthe biggest challenge in AI is
change management. It's justgetting people to start.
And I think this happened whenGoogle first came out. It's a
blank screen, blank promptwindow. Like what do I say when
I can say anything? It'sactually quite intimidating. And
(06:26):
so that's the challenge I thinkwith AI is like anything's
possible.
So where do you start? I telleverybody the best place to
start is to create your digitalprotege. Like just tell AI what
you do and have it help you dothose things. AI is great at
what you hate. And so find thosethings that you hate doing or
that take a lot of time andstart there.
You've maybe seen that quote Ireally love that's, I want AI to
(06:47):
do my dishes and laundry so Ican do more art and music, not
AI to do art and music so I cando more dishes and laundry,
right? So I think we all havedishes and laundry in our day to
day life. And so let's use AIthere first, because that'll be
the you'll get more motivated todo fewer financial analyses or
fewer, I don't know, copyediting, because that's kind of
(07:09):
annoying than you would likemaking music, because maybe
that's fun for you, right? Sostart with things that you don't
like. One thing I findfascinating about research on AI
is actually having knowledgemakes you better positioned to
use AI.
I think about AI as like therebirth of the Renaissance
person. It's like, if I want tocreate a picture on AI that
(07:29):
looks like Picasso, but I don'tknow Picasso's name, it's really
hard to describe that, right? IfI want to make a blueprint of a
Georgian architecture building,like how do I explain that if I
don't know what Georgianarchitecture is? And so whatever
area you live in or work in orcare about, you have like
expertise, right? You can talkabout it all day.
And that's a great place tostart with AI because you can go
(07:52):
say those words like, give me, Idon't know, a hierarchy of
Pokemon characters and you canname all the things and have it
rank order it. Like, I have noidea what I would say for that,
right? But I can talk all dayabout saunas and have the AI
help me improve my sauna, findnew water buckets, look at
different ratios of time in thesauna, because I care about
that. So find some things thatyou know about, that you're
(08:14):
passionate about, and startasking AI about it so you can go
deeper. I love it.
Chris (08:17):
I'm curious, a quick follow-up on that. Because you
raised a point that I hadn'treally thought about, but I've
observed it many, many times.And and you've kind of you
brought it to the surface herewith I I see people who are
totally comfortable getting onthe search engine of their
choice and searching topics, andthey've been doing that for
years. But as soon as they pullup, you know, a, you know, a
(08:40):
chat with a given model, they'rereally struggling with that.
They're really that that's what,you know, like from a I'm just
curious, as you've if you'veclearly thought about this quite
a lot, what is the difference?
And why are people so easy to goto search and yet struggling
with that model that has thesame text box in front of it?
Part of
Scott (09:00):
its exposure, right, just history. But I also think
there's something quitevulnerable about AI where it's
really a two way conversation.Search engine is, you know, very
much like the old card catalogs.Know, I remember my first year
of elementary school, I learnedcard catalog and then the next
year was told never have totouch that again. But it's the
same, that worked the same,right?
I'm just gonna go findsomething. But with AI, it's
(09:22):
probing back and forth andactually you can get pushback
and it kind of identifies howyou're thinking about things. So
I think there's somevulnerability around that and
plenty of like blank pageproblem of just not knowing
where to start. So start bycreating protege, start by
diving into areas you careabout. And I always tell people
a great framework to get startedis what I call the RIPE
framework.
(09:42):
R I P E. And it's just a way oflike four sentences to put into
AI to get good answers, which isthe role. Like you are an
expert, I don't know, copyeditor, the instruction, like
read through my paper andimprove it, parameters. So make
sure it's very concise and don'trepeat a lot of the same points
and examples. Like here's apaper I wrote before that shows
(10:04):
my kind of tone.
You know, if you just do thosefour things, a a role
instruction parameter example,like, you're gonna get awesome
output that's personalized andmuch more effective and less
robotic than just going thereand saying, write me a paper.
Daniel (10:17):
Yeah. I've had this kind of hypothesis, I guess, going
around in my mind. I'm curious,Scott, on your on your take on
this because you've seen a lotof people now. You you're always
interacting with people onDiscord or wherever, you know,
trying to get their assistanceto do this or that. What have
you found to be kind of thequalities that make up someone
(10:41):
who is just really proficient atkind of honing in the
instructions, the dataintegration, the configuration
of AI systems.
My hypothesis is sort of this isalmost like a I I think if we
took a bunch of hostagenegotiators and had them log in
(11:05):
to to AI systems to try to, youknow, either get them to do
things that they wanted them todo or to jailbreak them, think
they would be amazing at this.Because a lot of times it seems
to me, you know, not that I feelin danger physically or
something, but it's like peoplecan get disillusioned with this,
it's not quite what I want. Howdo I get you to do what I want
(11:29):
you to do? How do I warm you upto this idea? So, yeah, I'm
curious on the qualities thatyou've seen in terms of people
that have become good atconfiguring these systems,
prompting, understanding how to,you know, pull in integrations
or when and where to do that?
Any thoughts?
Scott (11:49):
Yeah. I mean, people who are great at this are
kindergarten teachers or parentsof three year olds. Maybe also
hostage negotiators. Also, it'sbasically the same job title.
Daniel (12:00):
There's some similarity there maybe.
Scott (12:03):
Yeah. I mean, think about talking I mean, people always
say an intern, but that's eventoo too experienced. Think about
talking to my three year oldSebastian. If I tell him three
things to go do in order,there's no way he's going to get
all three of them done, right?Go to the bathroom, pick out
some shoes, grab your snack, goto the car.
Like that's not happening. Ihave to be like, Go to the
(12:23):
bathroom. Good. Now this, right?And now this.
It's very step by step. And Ithink what's interesting is
there's two models or two typesof models emerging in AI. And
you guys maybe have your ownlanguage for this, but I think
about linear models like four o,Claude Sonnet 3.5, and we have
reasoning models now like othree, DeepSeq, and now Sonnet
(12:44):
3.7. It's like the reasoningmodels actually, that's like
talking to an intern who you cangive a ton of stuff and you just
let it go. But if you're doing alinear model, that's very much
need to do that step by step.
First do this, then do this,then do this. Because the
biggest, I think, frustrationpeople have is that AI too
quickly tries to get to ananswer before it has all the
(13:04):
details and things get lost. Andso with Chip, you can prompt
your AI tool and then anyone canuse it. And so what we found is
flipping the relationship isreally powerful where the AI
prompts you to get what it needsand then gives an answer. So you
can even, you know, on chip, youcan build this in so you don't
have to type it every time.
(13:25):
But on any AI tool, you mightsay like, before you write the
paper, before you create thestrategy, before you create the,
I don't know, the press release,make sure to ask me these three
things, right? And force it toget all of that information step
by step, just like you do with athree year old, and then you go
to school, and then you writethe paper, and then you do the
thing, right? So I think that'sreally fascinating, seeing that
(13:46):
divergence with reasoning, whichis like, don't go step by step,
just give all the context andit's going to work through it on
its own versus the three yearold linear that's needs that
guidance. So yeah, I think atthe end of the day, hostage
negotiator and parents, you gotthis.
Sponsor (14:16):
Well, friends, today's ever changing AI landscape means
your data demands more than thenarrow applications and single
model solutions that mostcompanies offer. Domo's AI and
data products platform is a morerobust all in one solution for
your data. It's not justambitious, it's practical and
(14:38):
adaptable so your business canmeet those new challenges with
ease. With Domo, you and yourteam can channel AI and data
into innovative uses thatdeliver measurable impact, and
their all in one platform bringsyou trustworthy AI results
without having to overhaul yourentire data infrastructure,
secure AI agents that connect,prepare, and automate your
(15:01):
workflows, helping you and yourteam to gain insights, receive
alerts, and act with easethrough guided apps tailored to
your role, and the flexibilityto choose which AI models you
wanna use. Domo goes beyondproductivity.
It's designed to transform yourprocesses, helping you make
smarter and faster decisionsthat drive real growth, all
(15:22):
powered by Domo's trust,flexibility, and their years of
expertise in data and AIinnovation. Data is hard. Domo
is easy. Make smarter decisionsand unlock your data's full
potential with Domo. Learn moretoday at ai.domo.com.
Again, that's ai.domo.com.
Daniel (15:50):
So, Scott, maybe we'll come back to kind of the the the
tooling itself. Could could youmaybe kinda circle back and
describe some of the the maybepeople aren't familiar with some
of the kinds of tools that areout there, especially, you know,
maybe there's programmers thathave interacted with with APIs
(16:10):
that are listening to the show.Maybe there are people that have
explored one tool or another.Maybe there's people that
haven't explored anything yet.So could you maybe just help us
kind of form a mental model forthe kinds of AI tools that are
out there?
And then maybe that would leadinto a discussion about kind of
some of the things that werereally on your mind in terms of
(16:33):
needs that weren't beingaddressed in that ecosystem?
Scott (16:37):
Yeah. If you want to think of a simple two by two
matrix, I think there's a reallyclear vertical versus horizontal
and closed versus opendichotomy. So you can think
about horizontal tools doing alot of things across modes.
JatGPT can write, it can createimages, it can code. It's really
good at all of those, but if youwant to just make images,
(16:59):
Midjourney is probably better.
It's a vertical image generationtool. Pika is really good at
video generation, which some ofthe general horizontal tools
aren't as good at. And, youknow, my sense is like
horizontal is gonna win, butthere's always gonna be a need
for people who want the Maseratiof AI, right? If you're only
doing code, like you're gonnaprobably be in cursor going deep
(17:22):
into like using these tools,whereas someone like me, I'm
gonna do the vibe coding where Ican use a tool like lovable or
bolt and just try stuff orreplet, right? So I think
horizontal, vertical, and then Ithink kind of open close.
So there are tools that let youuse it on their platform and you
don't necessarily know what'shappening. That would be
obviously like ChatGPT orClaude. You can't change rules
(17:44):
underpinning it. Also, you haveto go to their website. You
can't brand it.
You don't really have muchcontrol over the privacy. And
then more open tools are onesthat you could put on your own
site. You could add privacy intoit. You could brand it. So
that's what Chip is.
We wanna bring the power of AItools like ChatGPT and Claude to
your website, add privacy so thefiles stay locally, add your own
(18:05):
branding, you can see the chatlog. So just a lot more control,
obviously like PredictionGuard,same thing, right? Where you can
bring AI into your own cloud. Soa little bit more work,
obviously, with an open toolwhere you more power, also you
get more options. So I thinkthat's kind of like the lay of
the land.
I think it's just like when youlook at the internet broadly,
like it started with textbecause that was easy to send
(18:28):
across wires and then musicbecause MP3s were smaller than
video and then video. See thesame thing with AI, right? Where
it started with text and codebecause that's text heavy,
starting to get pretty goodimages now. Video is still
coming, not quite there yet, butgetting better every day. So I
kind of see that evolutionhappening.
Yeah. I think what maybe is asurprise is that people thought
(18:49):
the value was in the largelanguage models. I think what's
become really clear the lastmonth or two is it's actually
going to be in the customerrelationship and making this
stuff easier to use. DeepSeek isthe model that came out of China
a couple of weeks ago. If I lookat what Chip's cost per API call
is, it's gone down 90% ineighteen months, right?
(19:10):
So just think about the value ofthese large language models
becoming more commoditized. Thenwhat people are signing up for
is the experience of signing upand creating. I can go to Replit
and say, want an app that istracking my to dos and get it in
a few minutes. It's all on topof the same power, right? It's
all on top of ChatGPT or Claw.
(19:32):
Just like Chip, you can use anymodel underneath, but it's that
end user experience, which maybeisn't so different than the web.
Right? There's protocolsunderneath, but you still use
the browser that you like or theweb app you like because of how
it works, not necessarily thatit uses FTP versus something
else. Right?
Chris (19:49):
Could you talk a little bit more about that end user
experience, both the good andthe bad? Because Yeah. I I
think, you know, kinda goingback to to what we were talking
about before, it it's one ofthose barriers. And, you know,
there's a set of people that aretotally bought in across a whole
bunch of different industries,but there's also a very large
segment of the population thatstill really hasn't engaged. You
(20:12):
know, they're hearing about itevery day in the news and
everything, but they're justintimidated and haven't done.
So could you talk a little bitabout the landscape of being on
both sides of that barrier fordifferent people?
Scott (20:23):
I mean, the biggest increase in use that we see with
AI is putting it where peoplealready are so they don't have
to learn a new interface. Right?So if they can engage with AI
via a Slack channel or viaWhatsApp or via text message,
way easier. And so I think it'sfascinating to see there's a lot
of amazing UIs out there, butit's still getting people there.
(20:47):
It seems like time to value isreally important with the tools.
So like the faster you can showsomebody an outcome, and that's
I think where a lot of the newkind of text to app tools like
Lovable and Volt are reallyexciting for people because they
can get something quick, whichmakes sense. I think that's kind
of like how all UI is, is likehow do you get someone to the
value quickest? I actually thinklike the default UI we are
(21:11):
accustomed to with ChatGPT isnot great. You know, like for
someone to come in there anduse, it's interesting that
ChatGPT was a research project.It was not supposed to be a
consumer app and it just becamethat on accident.
So I think there's a lot ofimprovements to the UI to come
to make it easier for people touse. You see those already
coming into play where there'spre built ideas, autofill,
(21:33):
connect to data sources. Themost common way people use chip
is by duplicating an existingapp. It's like solving that
blank page problem is reallyimportant, I think, for any AI
tool. The easier you get peopleto motion is key.
Daniel (21:48):
I'm intrigued. You made me think of something. So, like,
for those that haven't seen Chipand what Scott and team are
building, you can go in andcreate individual assistance
that, as Scott mentioned, youcan kind of control and
configure, make the way youwant, connect the data sources
you want. And often, I think inmy conversations in the past
with Scott, I've heard him talkabout how people are creating
(22:10):
sort of proliferating these.Right?
You create one to do this and,like, one to do that and one to
do this, and you clone this oneto do that because it's not
quite that, which is a differentit's a different paradigm than
the sort of like, here's a chatinterface. This chat interface
is gonna do everything that wethat we want it to do. Could you
(22:32):
talk about that that element ofit a little bit and what you've
seen there? Because I I also seethis on the business side, like
when we engage customers, thekind of tendency it seems from
my perspective is to say, hey,how are we gonna build, like,
our internal AI, right, and getit to do all the things that we
(22:54):
want it to do. But it's like asingle in their mind, it's a
single thing, right?
It's like, this is our tool andit's gonna be the tool to sort
of rule them all. They'rethinking very singularly in that
way, which definitely does notseem to be kind of how people
are engaging in the way they'rebuilding assistance in your
tooling. Any thoughts there?
Scott (23:14):
I mean, think the high level thought is the concept of
software is getting turned onits head where software is now
an individual sport, not a teamsport. You know, you think about
if you're the CTO even a fewyears ago, it's like, I have to
do a lot of research, buy theright thing because everyone's
gonna use this. It has to fitthe most use cases. We have to
(23:36):
squeeze everything we can intoone thing. And now it's flipped
where every single person canbuild custom software within, we
say sixty seconds, So you wouldnever build software to, I don't
know, write a betterintroduction paragraph to a
grant.
But now someone on Chip will gobuild an app that just does
introductory paragraphs forgrant applications because it
(23:56):
takes sixty seconds and it savesthem three minutes every single
time and they do 10 a day and soit's thirty minutes. And we're
seeing the average admin personsaving sixty minutes a day on
chip, going from ninety minutesto thirty minutes on admin work
because they're buildingspecific apps for their specific
tools. Today I was looking atone that was getting IRS status
(24:17):
from the IRS website, right, andputting it onto a spreadsheet.
It's like nobody is going to gobuild a SaaS tool that just does
that because the market is maybe100 people or something, but
with AI you can. And so there'sdefinitely no need to have this
laborious top down purchasecycle when you can say, Just try
it.
Does this solve one problem, twoproblems, five problems, 10
(24:39):
problems? Great. Imagine thepower of every single person in
your org being a web developeror a coder. That's what it is
now, right? And so now we don'thave to bother our IT people or
our developers.
They can go do the hard stuffintegrating with antiquated
systems, right? Getting ourbilling to talk to our web to
talk to this. But for my job, Ijust have a file and I need to
(25:01):
get something done. And like,I'm not going to bother our
developer, but I'm going be myown developer. I don't know,
that's a total flip, right?
We're now we're not makingdecisions for the org, we're
making decisions for Scott and Ican just build it myself. So the
only limiter again is agency.Like just go, you have to go do
it. Most people still won't,even though the tool is right
(25:21):
there. But if they can at leasttry once, it's not as hard as
they might think.
Chris (25:25):
So it's a fascinating point you're making there with
it, but it does change eventhough you're talking about
flipping the model over, youknow, from kind of catering to
the the business as a whole tobeing able to cater it to each
individual contributor in thebusiness. By doing that, I'm
curious, you know, that thatopens up a lot of possibilities
for how you might run thebusiness going forward. Do you
(25:48):
have any thoughts on, like, whatthat does to the business if
assuming what's in a in ahypothetical world that you
could get your entire workforceto engage in that way? What do
you think that does for abusiness and how how might if
you were the CEO of a business,how might you operate in such a
way to change that if you werejust everyone's empowered with
(26:08):
AI agents that they can make insixty seconds? What does that do
for them?
Scott (26:12):
Yeah. And this is what Chip's trying to build. This is
really my Arzondetra, like wherewe think work is going is we
need an umbrella of safety sothat our employees can do
whatever they want withoutfeeling like they're going to
break something. Right now, thefear of messing up is greater
than the fear of missing out.And so we need to get rid of
that fear of messing up.
(26:32):
So I always say, the FOMO isgreater than the FOMO. We got to
get rid of the FOMO becausepeople aren't taking action
because they're scared. And so Ithink if I'm a company, what I'm
doing is I have my five to 10core apps. This is how we work.
When you start at Scott Inc,you're gonna go through the
onboarding chatbot.
You're gonna get the contentcreator that writes everything
in our voice. You're going to,you know, get the data analysis
(26:55):
that's gonna analyze thespreadsheets in the same way. So
these are the apps everybodyuses. This is company standard.
This is getting the laptop withprebuilt software.
And then underneath that now,you can duplicate or build your
own to how you work, right? Soyou have this layer of company
wide apps and then I have myScott apps and maybe they're
only visible to me. And a lot oftimes I might even cross
(27:15):
personal and professionalpotentially, right? Where it's
like, here's my workout scheduleand my agenda builder for work
and my, I don't know, grantwriter tool. But since it's
underneath this umbrella, weknow that it's going to adhere
to privacy.
Any personal information will beremoved so it doesn't violate
any problems. And then the finalpiece is, yeah, we have the
tools, but then we need thatmonthly or biweekly lunch and
(27:37):
learn where like, Hey Scott,what did you build this week?
Oh, cool. Let's just duplicatethat one click and now send it
to Dan. And Dan has similar workor new employee starts.
They can look over my shoulder,already the bots already trained
on all the history, it knowswhat to do so they can jump in.
And, you know, I always, Ialways say that AI really raises
the floor, you know, like everynew employee could start at
(27:59):
average or slightly aboveaverage, You still need to raise
the ceiling yourself, add thatspecial spice, right? Your own
ideas, but it's going to makeeveryone on a whole quicker to
get to work and higher, I guess,like higher average across the
board. And I always tell, youknow, the framework I always
recommend is like the AIsandwich. Like, just think about
you the the AI interactionstarts with you, the human, the
(28:22):
bread on top.
Then the AI is gonna dosomething. That's the meat in
the middle. But then you stillhave to be the human on the
bottom to take that output andto improve it, to share it, to
repurpose it. And so I think alot of new people get the bread
and the meat, but they forgetthe bottom piece of bread. And
so that'd be like the work Iwould do as a leader is, here's
our tools, you can all use it,and you're all going to be good.
(28:44):
Like you're not going to havespelling mistakes, it won't be
wordy, it'll make sense. But nowhow do you get better? And it's
gonna be like adding your ownspice on that last piece of
bread. So that's what I would dofor Scott Inc. So I think home
run.
Daniel (28:57):
And part of that too is like developing the muscle
memory. So like for me, forexample, we've been going
through fundraising recently.There's always the same set of
questions that come up indiligence and in questions about
the product and all this. Andmost of those have been answered
(29:20):
like 3,000,000 times now in someform. And, you know, now looking
back, like and, you know, we'vestarted to do this actually, but
really what would be best is ifwe just had a little chat that
had all of that preloaded intoit and could chat over that.
But at the time it's like, oh,well, I'll just answer this
(29:42):
email that's asking these 10questions. Right? I can bang
that out really quick. But that,I guess there's a muscle memory
thing there, and then there issome barrier to overcome to
configure the system for futurebenefit, right, that you might
not see there. So I don't know.
Yeah. Any suggestions even inyour own personal life where
(30:04):
you've kind of come over?
Scott (30:05):
Yeah. Mean, did the same thing, right? Like, we did a
raise with Chip and we built aChip Chat and it was trained on
all of our slides andeverything. People still want to
talk to you. It doesn't meanthat they don't get a human,
Daniel (30:19):
it
Scott (30:20):
gives them the option. The data we're seeing for our
users using Chip for customersupport, like a chat bubble sort
of use case, 70% of them are notclicking the talk to a human
button. They just want to knowwhat are your opening hours? How
much does it cost? Who are you?
Just give me the facts. As abusy parent, I get that. I don't
want to make phone calls becauseI know it'll take five to ten
(30:42):
minutes versus a minute if I'mdoing it myself. I think there's
that aspect of time efficiencyand it is changing habits of
going somewhere else or like yousaid, taking core info and
putting it into a repository.What we found most helpful is we
have something called dynamicknowledge sources.
So if it's a spreadsheet or afolder on Google Drive or
OneDrive, anything that getsadded into those places is
(31:05):
automatically added into youragent. And so I think with
businesses, it's important tothink about that flow of
information and minimizing asmuch like documentation work as
you can. So we always puteverything into notion or
confluence or Google Sheets orGoogle Docs, make that your hub
that is fed into the AI. Soeverything that you put in that
(31:27):
place gets automatically addedinto your FAQ bot or your, you
know, marketing assistant bot orwhatever. So I think that's,
that's key is like, you can askpeople to do it, but even better
is like not to require more workor even changing behavior
because we know that's thehardest part.
So maybe it's a BCC email thatgoes into a spreadsheet that's
(31:48):
automated, right? Or you know,something like that. You can
kind of decide. Way we do it iswe actually look at our chat
logs of people engaging withChip and find the answers that
are going unanswered or don'thave a great answer. And then we
add those things in once a weekinto our chat so that it
improves for the things peopleare asking for rather than
trying to solve for hypotheticaledge cases.
Daniel (32:09):
Well, Scott, we've talked a little bit about Chip.
I've described it a little bit.I'm wondering maybe for you've
been on this journey of kind oftrying to build this easy to use
AI tool. Along that journey,have you found I'm sure you
tried various things that didwork and didn't work, and
(32:32):
certain things have beendifficult and certain things
have been easier. As you reflecton that kind of as a founder of
of an AI company trying to buildan AI tool, any things that
you'd wanna highlight in termsof things that were kind of key
insights or bumps along the roadthat in retrospect you look at
(32:53):
and kind of make sense oranything like that?
Because I think there are a lotin our audience that have maybe
ideas for things out there.Yeah.
Scott (33:03):
No, that's amazing. There's so many. Think I'll take
like a non obvious one, which iswe were pretty early on focused
on building community. So wehave over 20 chip chapters
around the world, peopleteaching one another AI, fairly
active Discord. That's beeninvaluable because those are the
people who are bringing backproblems and ideas and being
(33:25):
able to build towards actualcustomer questions is so
important.
And a lot of times customersdon't have time or interest
giving you feedback, which youneed. And so what we've done is
every two weeks or so, havingfree workshops to try to educate
our users and anybody. Andthat's really built a
relationship, I think, where weknow these people by name, we
(33:46):
know where they live, what theydo, and it makes it a lot easier
for them to be like, Yo, can youbuild this thing? I need it for
a pitch on Friday. We're like,Yeah, for you, of course,
because you're contributing.
So it's building relationships.And it doesn't have to be
hundreds, right? This can bedozens of people who love you
and that's how you really start,is like a strong foundation. So
I think that one's non obvious.I think technically something
(34:08):
that we found maybe an accidentand we're trying to lean into
now is riding the wave of otherpeople's innovation.
Know, like you can only build somany unique pieces and you need
to be on top of other parts ofthe tech stack. Is built on top
of large language models. So asAnthropic and OpenAI build
(34:28):
better models, Chip gets better.For a lot of our users, they
think Chip is doing that becausewe are their front door to AI.
As the models get better, Chipgets better and their experience
gets better.
We partner with folks likePredictionGuard who help us
provide better privacy andsecurity. We could go spend six
months trying to build that, butnow we've lost the whole point
(34:48):
of what we're doing, right? Andso what is your forte is really
important. One thing that hasreally recently that we kind of
focused on is Anthropic has anew protocol called, what is it,
Model context protocol. It'sbasically an easy way to connect
APIs into AI tools.
And so that's another example oflike, we've been building one
(35:09):
off APIs to all these differenttools and now it's like, wow,
there's this whole world that'sbuilt towards this standard. And
if we just tap into that, now wecan, again, get better the more
the open source communitycontributes. So I think that's
really interesting to look outwhere are the areas that will
move quickly that you can ridethat wave and then where do you
wanna be a differentiator? Andyou can kind of draw your line
(35:31):
wherever the right place is, butprobably don't try to draw it on
all of them. Pick the onesyou're best at.
Yeah, I think those are a fewand I think just the power of
small teams now. I mean, youread that a lot of places, but
you know our CTO Hunter, who isjust like a beast with AI
coding. It's like, I know ouroutput compared to some legacy
(35:51):
teams is just vastly greater.And so I wouldn't underestimate
if you're a solo founder, yougot a team. We have a couple of
chip users.
There's a guy named Chuck inColorado who he's building a
million dollar one person agencyand he's almost there, right?
And it's all built with AIautomations and he's conducting
everything. There's a lot ofpotential out there. So I would
encourage anyone listening,finding a co founder or a team
(36:13):
is really, really hard, but youdon't have to wait. Like you can
do a lot on your own.
Chris (36:17):
I'm curious, you actually started to get in for a second
to the next question I was gonnaanswer. And that was, you
mentioned like privacy andsecurity and and partnering with
prediction guard for that. Asyou're thinking about these
these different concerns thatweigh in on various industries,
and, you know, there'll be, youknow, legal concerns, things
(36:39):
like, you know, HIPAA in themedical world. And every
industry has its own set ofconcerns that are kind of
external, but are binding thework in those areas. And as you
are kind of kind of unleashingpeople's potential with the work
that you're doing, those kind ofhave to find some sort of
balance.
How are you thinking about theconstraints versus the
(37:02):
unleashing that we talked aboutand finding a balance so that
people are unleashed whilethey're still having to be held
to account, you know, bywhatever those constraints in
their industry is?
Scott (37:12):
Right. Yeah. I mean, I think regulation is always
innovation. And so I would sayas a company, as an individual,
like, look at yourself firstbefore worrying about the
regulatory environment. Youknow, I think about privacy
pyramid as what we tell ourcustomers like the bottom of the
pyramid, the first thing youshould do is just think about
(37:33):
what are you okay sharing andnot sharing and just tell
people.
Again, FOMO is greater thanFOMO. People will not take
action if they think they'regoing to get in trouble. Even if
it's hypothetical, not real. Idon't know, that fear from
elementary school sticks withus. So the first thing you have
to do is remove the fear.
And the best way to do that isjust to say what the rules are.
(37:55):
As long as people know therules, they'll work within them.
But if they don't know what theyare, they're afraid that
whatever they do will get themin trouble, right? So, hey, just
don't upload customer data. Likethat's our rule.
Great. That's a great place tostart. Now go do anything else.
Or no customer data and don'tintegrate with these files.
Great.
And the second level of thepyramid after kind of just best
(38:17):
practices internally is thengonna be like human protection
air, I call it, which one thingPrediction Guard offers as well,
which is like encrypting piecesof information that get added
that shouldn't be. So if I add aphone number or a social
security number or something, itgets removed for me because I
made a mistake. That's fine. Wemake mistakes, best practices
(38:37):
and then cover other people'smistakes up as they make them.
And then I think the top of thepyramid is where you actually
say, You know what?
Let's put it in our ownenvironment. So that way, if we
can share whatever we wantwithout having to worry, and
that's where you can run an opensource large language model in
your own cloud infrastructure,whatever you share is in your
cloud infrastructure. So somebusinesses have to do that. So
(39:00):
if you are in finance,healthcare, you're probably
gonna wanna do that anyway, justfor regulatory reasons. Some
people wanna do that becausethey know they're gonna be
sharing data that might besensitive.
But I think for most of us, toget started, just follow that
basic best practice of like,think about it before you share
it. And if you're working with ateam that might make mistakes or
contractors who aren't followingyour rules, like add in that
(39:21):
second level of like human errorprotection.
Daniel (39:24):
Scott, as we kind of get near to the end here, I'm
wondering if you can maybe sharejust a few standout use cases of
maybe things that you've seenpeople do with chip that have
either surprised you or stoodout in way like, oh, I didn't
expect people would do this,that, you know? Or things that
(39:47):
are like, oh, I didn't even knowyou know, I built the platform,
but I didn't even know that waspossible.
Scott (39:53):
Every day. That's my favorite part of chip and AI
generally is we really arebuilding the tools and we don't
know how people will use them.And it's so crazy to see what
people do with it. I mean, themost common use cases, I would
say there's five areas thatpeople use all the time. It's
operations, marketing, sales.
I call it company search,finding stuff in your Google
(40:15):
Drive basically. And what's thelast one? Data analysis,
reviewing financials and thingslike that. So those are the most
common. But in terms of fun,weird ones, we had somebody who
launched a Canadian tariffchecker.
And so as the tariffs on Canadawere released, you could
actually search any product andit would source where they were
coming from and tell you whatthe change in price would be.
(40:37):
Was totally interesting. One ofmy favorite use cases, a guy
named Tyler Hansen, he's inSioux Falls, South Dakota, and
he runs an HVAC company. He putin all of the training manuals
for all of the equipment thatthey service. So then his
technicians are on the ground.
And instead of having to be inthe bathroom watching a YouTube
video, which I know has happenedwhen my HVAC guy comes, right?
(40:58):
He's actually learning how to dothe thing that I asked him to
do. They can actually pull upthe specific model via their
chitchat and get instructions onwhat to do and how to service it
and parts. That one's reallyfun. There's a contractor out in
Washington.
He uses it to create supplylists. He just puts in square
footage and what people aregoing to build. And then it'll
spit out how much wood he needs,how many nails, whatever else.
(41:21):
Again, things I know nothingabout. A lot of people doing it
for finding HR policies, findinglet's see, there's a car dealer
that's using it to find cars topurchase, to then resell, right?
So it searches through AutoTrader and Craigslist or
wherever else to find vehicles.It's just so many things, right?
(41:42):
And every day I'm encounteringnew ones that are so
fascinating. The fun part is weintegrate with APIs and
webhooks. Really any tool canget pulled in.
A lot of times chip ends upbeing a front end to an AI tool
that's talking to theirsoftware. So chip becomes the
way they communicate, but thenit's pulling their own data. So
(42:02):
that's super fun. Personally, Ihave a Scott Bot. That's the one
I use every single day.
And so I can write things veryquickly and remember people that
I've talked to, so it brings inpast conversations. So that
helps me quite a bit. So yeah,those are a few random ideas. I
haven't built the West Lafayettetour guide yet, but we do have
some travel AI tools out there.So I bet we could do that too.
Chris (42:26):
So very cool. And while you're building that tour guide,
I might give you a location ortwo as well.
Scott (42:31):
Okay, there you go. Yeah, that's awesome.
Chris (42:33):
So really cool use cases there. That's gotta get you
thinking about thepossibilities. So you come at it
with your own mindset and thethings that you have. Your your
customers are are teaching youevery day about what the new
possibilities and boundariesmight be. So where does that
(42:53):
where does that take you?
Like, when you are are you know,you're kinda done for the
workday, your brain'sdecompressing, and but you're
still kind of, you know, justworking on things. What what are
what's going through your headabout, like, where could things
go with this? You know, you youtake what you're driving and and
the and the folks you're workingwith are driving. You're taking
(43:13):
what your customers are showingyou that you never thought
about. And that's got to leaveyou with some pretty cool ideas
about what the future mighthold.
But can you share some of thoseideas with us?
Scott (43:22):
Yeah, I think I mean, I I reflect at the end of the day in
a lot of ways, because I havefour kids that are 11, nine,
seven, three. And I just reallytry to think about like, what
does society look like when thisis more present? And, you know,
what does education look like?Spent a lot of my life in
education, we work with a lot ofschools who use it for tutors
(43:43):
and advisors. And, what's thevalue of a credential saying,
you know, something when thepace of change is like way
faster than four years, right?
I think ultimately, I imaginethis technology has to fade away
from being AI and just being apart of what we use. And it
helps us lean into the thingsthat make us weird. I think
(44:05):
about AI is the world's bestcover band and it needs like the
originals to cover. So I thinkit really forces us to be more
unique as individuals and createsomething new rather. We're
going to use AI for a lot of thequick answers and it's going to
be average.
It's going to be the middle ofthat bell curve and that'll be
fine for most work. But again,we have to raise the ceiling
(44:27):
ourselves. And so I think itmakes me feel like I want my
kids and hopefully myself tolike, just get good, really good
at whatever weird, interestingthing we care about. Yeah. And
I, man, I don't know, I thinkagency again, like I keep coming
back to that.
But how do you instill a lacklike a fearlessness in people?
Because it feels like, first ofall, most people aren't aware of
(44:49):
the pace of change. And as theybecome aware of it, it's either
I'm scared, I'm going to backaway, or I'm going to lean into
it. And I think we just reallyneed to lean into it. And I
don't know, I think it'sexciting because I'm in Fargo,
and I couldn't, you know, learnto be a nuclear physicist in
Fargo, right?
But now I could like I caneasily go down that path and
(45:10):
learn what I need to connectwith the resources, you know,
showcase my work. And this haskind of been my dream since my
first company in 2009 of like,really given, give giving anyone
wherever they are a chance tobuild. And AI is just like the
next step in that process. And Iknow a lot of people still will
find reasons not to but it'sgoing to be just on that agency
(45:32):
piece like you can. So I don'tknow.
I think a society whereeverybody has a chance to build
and create is incrediblyexciting. It's going to be more
competitive. Everyone around theworld has equal access to the
same models as NASA and thedefense department. It's kind of
wild that you can log into thesethings for free and have the
same power as everyone else. Sothat's an opportunity if you
(45:55):
take it.
I think I saw there was a recentstudy the World Bank did in
Nigeria and students who wereusing ChatGPT as a tutor for six
weeks had the equivalent of twoyears of education. And it's
just so many of our problems areproblems of access. And I think
a lot of those access problemsgo away. And then what happens
when another 1,000,000,000people come online with
(46:17):
education who don't have it now,like that's just better for us
all. We can come up with reallyexciting solutions to our
problems.
Daniel (46:22):
Well said. Yeah. That's a that's a great way to end.
Thanks for thanks for joining,Scott. I encourage everyone to
go, create your first chip chaton, chip, chipp.ai, and and have
some fun.
Explore those that weirdness asas Scott put it. I I love that.
Thanks for joining, Scott. It'sbeen great
Scott (46:43):
to chat. Great to be here, guys.
Jared (46:51):
Alright. That is our show for this week. If you haven't
checked out our changelognewsletter, head to
changelog.com/news. There you'llfind 29 reasons. Yes.
29 reasons why you shouldsubscribe. I'll tell you reason
number 17. You might actuallystart looking forward to
(47:11):
Mondays.
Scott (47:12):
Sounds like somebody's got a case of the Mondays.
Jared (47:15):
28 more reasons are waiting for you at
changelog.com/news. Thanks againto our partners at fly.io to
Brakemaster Cylinder for theBeats and to you for listening.
That is all for now, but we'lltalk to you again next time.
Welcome to Practical AI, the podcast that makes
artificial intelligencepractical, productive, and
accessible to all. If you likethis show, you will love the
changelog. It's news on Mondays,deep technical interviews on
Wednesdays, and on Fridays, anawesome talk show for your
weekend enjoyment. Find us bysearching for the changelog
(00:24):
wherever you get your podcasts.Thanks to our partners at
fly.io.
Launch your AI apps in fiveminutes or less. Learn how at
fly.io.
Daniel (00:44):
Welcome to another episode of the Practical AI
podcast. This is DanielWitenack. I'm CEO at Prediction
Guard, and I'm joined as alwaysby my cohost, Chris Benson, who
is a principal AI researchengineer at Lockheed Martin. How
are doing, Chris?
Chris (01:00):
Oh, I'm feeling pretty chipper today. It's a a good day
to talk about AI.
Daniel (01:05):
Yeah. Yeah. I I feel feel quite quite chipper as
well, especially as we we've gotour guest today, Scott Meyer
with us, who's, founder and CEOat Chip, which is, you can find
at chip.ai, I believe, is is thelink. But, yeah, Chip is
awesome. Also, Scott is awesome.
And, also, Scott is is along agood friend because he's a a
(01:30):
fellow member of the the SiliconPrairie, not living on the on
the coast, but out here in themiddle somewhere where AI is
really blossoming, if you didn'tknow.
Scott (01:40):
It is. And it gives an unfair advantage for those of us
in non metro areas. The abilityto leverage AI to have the power
of 10 people in a place thatdoesn't have enough people to do
the job, it's perfect. It's aperfect solution. So it's great
to be here live from Fargo, justlike the movie.
It's it's fantastic to see youall and, be heard by all of you
(02:01):
listening.
Daniel (02:02):
Yeah. Yeah. Scott, we'll we'll we'll get into all the
cool stuff, you're doing withChip and some of the things
you've learned through that. ButI'm wondering if you work in the
space of, I guess we might putit like low code, no code AI
assistant builders. So for maybeaudience members that aren't as
(02:24):
familiar with that space, ormaybe they're just kind of
wondering what's out there, asof today, could you paint a
little bit of a picture for usfor kind of what sorts of tools
are out there?
And then maybe that would kindof motivate some of the unique
things that you thought shouldbe out there but weren't, which
(02:45):
would maybe kind of highlightsome of the things you're doing
with Chip.
Scott (02:48):
Yeah. No, it's great to be here. I think the stat that
blows my mind is that almost 50%of Americans use AI every week,
but 7% of businesses use AI,which is obviously a lie because
50% of Americans are using AIevery week and they work at
those companies. So what'shappening is the businesses
aren't, they have no idea what'sgoing on. It's like the early
(03:08):
days of cell phones wheneveryone would come to work with
their own cell phone, their ownlaptop, do whatever they wanted
to.
And eventually we got to thispoint where you get a company
email, you get company apps, youget like the standard way to do
it. I think the risk right nowis that, and the opportunity is
those who are willing to haveagency and try stuff have unfair
advantage, right? So I can go domy work with AI. And if my
(03:28):
colleagues don't know, and Idon't have a culture of sharing,
like all of a sudden, I'm asuper, superhuman. The number
one thing I tell businesses whenI meet with them is you should
have a lunch and learn once amonth and just have people say
what they're doing.
Because just that horizontalsharing of AI practices and
ideas is all you need to build aculture of acceptance. What
(03:49):
makes AI so unique is it's nottop down. It's not the CIO or
CTO saying, I bought this thing.You guys all go use it. It's
each individual figuring out howthey can use it for their
specific tasks.
What I've seen is adminassistants, marketers, interns,
right? They're all gonna use itdifferently and often even know
better how to use it becausethey're the ones doing the
(04:09):
tasks. And that kind ofmotivated what we built with
Chip, which is how do we justmake AI as easy as possible to
use? Our motto is AI for all.And I think I've spent most of
my professional career workingon bridging a digital divide
because maybe like you, peoplethat work and live alongside me
in Fargo aren't always takingadvantage of the latest
technology, right?
(04:29):
And so I kind of feel like it'sboth a passion and mission to
bring what's happening and makeit accessible to those around
me. In 02/2009, I started myfirst company and I was trying
to tell businesses there's thisthing called social media they
should use, right? Before therewere Facebook pages and Facebook
ads. And it feels like that tome again, almost twenty years
later, where it's like thisamazing power is right here and
(04:51):
the best time to start learningis now. With tools like Chip and
others that we can talk about,it's actually better now than
ever for people who aren'ttechnical because it's not about
technical ability, it's aboutknowledge and agency.
And I think we all have that. Sohappy to give a landscape. Think
that already went off track fromyour question, but hopefully
that gives you the starting No,
Daniel (05:12):
that's awesome. What would you say are kind of some
of those things that might makeAI hard to use? And here, mostly
we're talking of course, we'vetalked about a lot of things in
the show, but mostly we'retalking about kind of what
typical people would consider AInow, which would be kind of
generative AI, language models,maybe vision models, etcetera.
(05:34):
So, like, what what can makethose difficult to use or or or
how might people getdisillusioned as they're
exploring the technology?
Scott (05:43):
I'll say almost almost every excuse people have not to
use AI tools is fear. They arescared of a blank page. This is
the same with technology fortwenty years. I taught
entrepreneurship. I startedentrepreneurship centers and all
these students with amazingideas.
And you know what? 90% of themdidn't do anything because they
(06:04):
had to actually go do something,right? And it's like, you just
have to start. And I'm convincedthe biggest challenge in AI is
change management. It's justgetting people to start.
And I think this happened whenGoogle first came out. It's a
blank screen, blank promptwindow. Like what do I say when
I can say anything? It'sactually quite intimidating. And
(06:26):
so that's the challenge I thinkwith AI is like anything's
possible.
So where do you start? I telleverybody the best place to
start is to create your digitalprotege. Like just tell AI what
you do and have it help you dothose things. AI is great at
what you hate. And so find thosethings that you hate doing or
that take a lot of time andstart there.
You've maybe seen that quote Ireally love that's, I want AI to
(06:47):
do my dishes and laundry so Ican do more art and music, not
AI to do art and music so I cando more dishes and laundry,
right? So I think we all havedishes and laundry in our day to
day life. And so let's use AIthere first, because that'll be
the you'll get more motivated todo fewer financial analyses or
fewer, I don't know, copyediting, because that's kind of
(07:09):
annoying than you would likemaking music, because maybe
that's fun for you, right? Sostart with things that you don't
like. One thing I findfascinating about research on AI
is actually having knowledgemakes you better positioned to
use AI.
I think about AI as like therebirth of the Renaissance
person. It's like, if I want tocreate a picture on AI that
(07:29):
looks like Picasso, but I don'tknow Picasso's name, it's really
hard to describe that, right? IfI want to make a blueprint of a
Georgian architecture building,like how do I explain that if I
don't know what Georgianarchitecture is? And so whatever
area you live in or work in orcare about, you have like
expertise, right? You can talkabout it all day.
And that's a great place tostart with AI because you can go
(07:52):
say those words like, give me, Idon't know, a hierarchy of
Pokemon characters and you canname all the things and have it
rank order it. Like, I have noidea what I would say for that,
right? But I can talk all dayabout saunas and have the AI
help me improve my sauna, findnew water buckets, look at
different ratios of time in thesauna, because I care about
that. So find some things thatyou know about, that you're
(08:14):
passionate about, and startasking AI about it so you can go
deeper. I love it.
Chris (08:17):
I'm curious, a quick follow-up on that. Because you
raised a point that I hadn'treally thought about, but I've
observed it many, many times.And and you've kind of you
brought it to the surface herewith I I see people who are
totally comfortable getting onthe search engine of their
choice and searching topics, andthey've been doing that for
years. But as soon as they pullup, you know, a, you know, a
(08:40):
chat with a given model, they'rereally struggling with that.
They're really that that's what,you know, like from a I'm just
curious, as you've if you'veclearly thought about this quite
a lot, what is the difference?
And why are people so easy to goto search and yet struggling
with that model that has thesame text box in front of it?
Part of
Scott (09:00):
its exposure, right, just history. But I also think
there's something quitevulnerable about AI where it's
really a two way conversation.Search engine is, you know, very
much like the old card catalogs.Know, I remember my first year
of elementary school, I learnedcard catalog and then the next
year was told never have totouch that again. But it's the
same, that worked the same,right?
I'm just gonna go findsomething. But with AI, it's
(09:22):
probing back and forth andactually you can get pushback
and it kind of identifies howyou're thinking about things. So
I think there's somevulnerability around that and
plenty of like blank pageproblem of just not knowing
where to start. So start bycreating protege, start by
diving into areas you careabout. And I always tell people
a great framework to get startedis what I call the RIPE
framework.
(09:42):
R I P E. And it's just a way oflike four sentences to put into
AI to get good answers, which isthe role. Like you are an
expert, I don't know, copyeditor, the instruction, like
read through my paper andimprove it, parameters. So make
sure it's very concise and don'trepeat a lot of the same points
and examples. Like here's apaper I wrote before that shows
(10:04):
my kind of tone.
You know, if you just do thosefour things, a a role
instruction parameter example,like, you're gonna get awesome
output that's personalized andmuch more effective and less
robotic than just going thereand saying, write me a paper.
Daniel (10:17):
Yeah. I've had this kind of hypothesis, I guess, going
around in my mind. I'm curious,Scott, on your on your take on
this because you've seen a lotof people now. You you're always
interacting with people onDiscord or wherever, you know,
trying to get their assistanceto do this or that. What have
you found to be kind of thequalities that make up someone
(10:41):
who is just really proficient atkind of honing in the
instructions, the dataintegration, the configuration
of AI systems.
My hypothesis is sort of this isalmost like a I I think if we
took a bunch of hostagenegotiators and had them log in
(11:05):
to to AI systems to try to, youknow, either get them to do
things that they wanted them todo or to jailbreak them, think
they would be amazing at this.Because a lot of times it seems
to me, you know, not that I feelin danger physically or
something, but it's like peoplecan get disillusioned with this,
it's not quite what I want. Howdo I get you to do what I want
(11:29):
you to do? How do I warm you upto this idea? So, yeah, I'm
curious on the qualities thatyou've seen in terms of people
that have become good atconfiguring these systems,
prompting, understanding how to,you know, pull in integrations
or when and where to do that?
Any thoughts?
Scott (11:49):
Yeah. I mean, people who are great at this are
kindergarten teachers or parentsof three year olds. Maybe also
hostage negotiators. Also, it'sbasically the same job title.
Daniel (12:00):
There's some similarity there maybe.
Scott (12:03):
Yeah. I mean, think about talking I mean, people always
say an intern, but that's eventoo too experienced. Think about
talking to my three year oldSebastian. If I tell him three
things to go do in order,there's no way he's going to get
all three of them done, right?Go to the bathroom, pick out
some shoes, grab your snack, goto the car.
Like that's not happening. Ihave to be like, Go to the
(12:23):
bathroom. Good. Now this, right?And now this.
It's very step by step. And Ithink what's interesting is
there's two models or two typesof models emerging in AI. And
you guys maybe have your ownlanguage for this, but I think
about linear models like four o,Claude Sonnet 3.5, and we have
reasoning models now like othree, DeepSeq, and now Sonnet
(12:44):
3.7. It's like the reasoningmodels actually, that's like
talking to an intern who you cangive a ton of stuff and you just
let it go. But if you're doing alinear model, that's very much
need to do that step by step.
First do this, then do this,then do this. Because the
biggest, I think, frustrationpeople have is that AI too
quickly tries to get to ananswer before it has all the
(13:04):
details and things get lost. Andso with Chip, you can prompt
your AI tool and then anyone canuse it. And so what we found is
flipping the relationship isreally powerful where the AI
prompts you to get what it needsand then gives an answer. So you
can even, you know, on chip, youcan build this in so you don't
have to type it every time.
(13:25):
But on any AI tool, you mightsay like, before you write the
paper, before you create thestrategy, before you create the,
I don't know, the press release,make sure to ask me these three
things, right? And force it toget all of that information step
by step, just like you do with athree year old, and then you go
to school, and then you writethe paper, and then you do the
thing, right? So I think that'sreally fascinating, seeing that
(13:46):
divergence with reasoning, whichis like, don't go step by step,
just give all the context andit's going to work through it on
its own versus the three yearold linear that's needs that
guidance. So yeah, I think atthe end of the day, hostage
negotiator and parents, you gotthis.
Sponsor (14:16):
Well, friends, today's ever changing AI landscape means
your data demands more than thenarrow applications and single
model solutions that mostcompanies offer. Domo's AI and
data products platform is a morerobust all in one solution for
your data. It's not justambitious, it's practical and
(14:38):
adaptable so your business canmeet those new challenges with
ease. With Domo, you and yourteam can channel AI and data
into innovative uses thatdeliver measurable impact, and
their all in one platform bringsyou trustworthy AI results
without having to overhaul yourentire data infrastructure,
secure AI agents that connect,prepare, and automate your
(15:01):
workflows, helping you and yourteam to gain insights, receive
alerts, and act with easethrough guided apps tailored to
your role, and the flexibilityto choose which AI models you
wanna use. Domo goes beyondproductivity.
It's designed to transform yourprocesses, helping you make
smarter and faster decisionsthat drive real growth, all
(15:22):
powered by Domo's trust,flexibility, and their years of
expertise in data and AIinnovation. Data is hard. Domo
is easy. Make smarter decisionsand unlock your data's full
potential with Domo. Learn moretoday at ai.domo.com.
Again, that's ai.domo.com.
Daniel (15:50):
So, Scott, maybe we'll come back to kind of the the the
tooling itself. Could could youmaybe kinda circle back and
describe some of the the maybepeople aren't familiar with some
of the kinds of tools that areout there, especially, you know,
maybe there's programmers thathave interacted with with APIs
(16:10):
that are listening to the show.Maybe there are people that have
explored one tool or another.Maybe there's people that
haven't explored anything yet.So could you maybe just help us
kind of form a mental model forthe kinds of AI tools that are
out there?
And then maybe that would leadinto a discussion about kind of
some of the things that werereally on your mind in terms of
(16:33):
needs that weren't beingaddressed in that ecosystem?
Scott (16:37):
Yeah. If you want to think of a simple two by two
matrix, I think there's a reallyclear vertical versus horizontal
and closed versus opendichotomy. So you can think
about horizontal tools doing alot of things across modes.
JatGPT can write, it can createimages, it can code. It's really
good at all of those, but if youwant to just make images,
(16:59):
Midjourney is probably better.
It's a vertical image generationtool. Pika is really good at
video generation, which some ofthe general horizontal tools
aren't as good at. And, youknow, my sense is like
horizontal is gonna win, butthere's always gonna be a need
for people who want the Maseratiof AI, right? If you're only
doing code, like you're gonnaprobably be in cursor going deep
(17:22):
into like using these tools,whereas someone like me, I'm
gonna do the vibe coding where Ican use a tool like lovable or
bolt and just try stuff orreplet, right? So I think
horizontal, vertical, and then Ithink kind of open close.
So there are tools that let youuse it on their platform and you
don't necessarily know what'shappening. That would be
obviously like ChatGPT orClaude. You can't change rules
(17:44):
underpinning it. Also, you haveto go to their website. You
can't brand it.
You don't really have muchcontrol over the privacy. And
then more open tools are onesthat you could put on your own
site. You could add privacy intoit. You could brand it. So
that's what Chip is.
We wanna bring the power of AItools like ChatGPT and Claude to
your website, add privacy so thefiles stay locally, add your own
(18:05):
branding, you can see the chatlog. So just a lot more control,
obviously like PredictionGuard,same thing, right? Where you can
bring AI into your own cloud. Soa little bit more work,
obviously, with an open toolwhere you more power, also you
get more options. So I thinkthat's kind of like the lay of
the land.
I think it's just like when youlook at the internet broadly,
like it started with textbecause that was easy to send
(18:28):
across wires and then musicbecause MP3s were smaller than
video and then video. See thesame thing with AI, right? Where
it started with text and codebecause that's text heavy,
starting to get pretty goodimages now. Video is still
coming, not quite there yet, butgetting better every day. So I
kind of see that evolutionhappening.
Yeah. I think what maybe is asurprise is that people thought
(18:49):
the value was in the largelanguage models. I think what's
become really clear the lastmonth or two is it's actually
going to be in the customerrelationship and making this
stuff easier to use. DeepSeek isthe model that came out of China
a couple of weeks ago. If I lookat what Chip's cost per API call
is, it's gone down 90% ineighteen months, right?
(19:10):
So just think about the value ofthese large language models
becoming more commoditized. Thenwhat people are signing up for
is the experience of signing upand creating. I can go to Replit
and say, want an app that istracking my to dos and get it in
a few minutes. It's all on topof the same power, right? It's
all on top of ChatGPT or Claw.
(19:32):
Just like Chip, you can use anymodel underneath, but it's that
end user experience, which maybeisn't so different than the web.
Right? There's protocolsunderneath, but you still use
the browser that you like or theweb app you like because of how
it works, not necessarily thatit uses FTP versus something
else. Right?
Chris (19:49):
Could you talk a little bit more about that end user
experience, both the good andthe bad? Because Yeah. I I
think, you know, kinda goingback to to what we were talking
about before, it it's one ofthose barriers. And, you know,
there's a set of people that aretotally bought in across a whole
bunch of different industries,but there's also a very large
segment of the population thatstill really hasn't engaged. You
(20:12):
know, they're hearing about itevery day in the news and
everything, but they're justintimidated and haven't done.
So could you talk a little bitabout the landscape of being on
both sides of that barrier fordifferent people?
Scott (20:23):
I mean, the biggest increase in use that we see with
AI is putting it where peoplealready are so they don't have
to learn a new interface. Right?So if they can engage with AI
via a Slack channel or viaWhatsApp or via text message,
way easier. And so I think it'sfascinating to see there's a lot
of amazing UIs out there, butit's still getting people there.
(20:47):
It seems like time to value isreally important with the tools.
So like the faster you can showsomebody an outcome, and that's
I think where a lot of the newkind of text to app tools like
Lovable and Volt are reallyexciting for people because they
can get something quick, whichmakes sense. I think that's kind
of like how all UI is, is likehow do you get someone to the
value quickest? I actually thinklike the default UI we are
(21:11):
accustomed to with ChatGPT isnot great. You know, like for
someone to come in there anduse, it's interesting that
ChatGPT was a research project.It was not supposed to be a
consumer app and it just becamethat on accident.
So I think there's a lot ofimprovements to the UI to come
to make it easier for people touse. You see those already
coming into play where there'spre built ideas, autofill,
(21:33):
connect to data sources. Themost common way people use chip
is by duplicating an existingapp. It's like solving that
blank page problem is reallyimportant, I think, for any AI
tool. The easier you get peopleto motion is key.
Daniel (21:48):
I'm intrigued. You made me think of something. So, like,
for those that haven't seen Chipand what Scott and team are
building, you can go in andcreate individual assistance
that, as Scott mentioned, youcan kind of control and
configure, make the way youwant, connect the data sources
you want. And often, I think inmy conversations in the past
with Scott, I've heard him talkabout how people are creating
(22:10):
sort of proliferating these.Right?
You create one to do this and,like, one to do that and one to
do this, and you clone this oneto do that because it's not
quite that, which is a differentit's a different paradigm than
the sort of like, here's a chatinterface. This chat interface
is gonna do everything that wethat we want it to do. Could you
(22:32):
talk about that that element ofit a little bit and what you've
seen there? Because I I also seethis on the business side, like
when we engage customers, thekind of tendency it seems from
my perspective is to say, hey,how are we gonna build, like,
our internal AI, right, and getit to do all the things that we
(22:54):
want it to do. But it's like asingle in their mind, it's a
single thing, right?
It's like, this is our tool andit's gonna be the tool to sort
of rule them all. They'rethinking very singularly in that
way, which definitely does notseem to be kind of how people
are engaging in the way they'rebuilding assistance in your
tooling. Any thoughts there?
Scott (23:14):
I mean, think the high level thought is the concept of
software is getting turned onits head where software is now
an individual sport, not a teamsport. You know, you think about
if you're the CTO even a fewyears ago, it's like, I have to
do a lot of research, buy theright thing because everyone's
gonna use this. It has to fitthe most use cases. We have to
(23:36):
squeeze everything we can intoone thing. And now it's flipped
where every single person canbuild custom software within, we
say sixty seconds, So you wouldnever build software to, I don't
know, write a betterintroduction paragraph to a
grant.
But now someone on Chip will gobuild an app that just does
introductory paragraphs forgrant applications because it
(23:56):
takes sixty seconds and it savesthem three minutes every single
time and they do 10 a day and soit's thirty minutes. And we're
seeing the average admin personsaving sixty minutes a day on
chip, going from ninety minutesto thirty minutes on admin work
because they're buildingspecific apps for their specific
tools. Today I was looking atone that was getting IRS status
(24:17):
from the IRS website, right, andputting it onto a spreadsheet.
It's like nobody is going to gobuild a SaaS tool that just does
that because the market is maybe100 people or something, but
with AI you can. And so there'sdefinitely no need to have this
laborious top down purchasecycle when you can say, Just try
it.
Does this solve one problem, twoproblems, five problems, 10
(24:39):
problems? Great. Imagine thepower of every single person in
your org being a web developeror a coder. That's what it is
now, right? And so now we don'thave to bother our IT people or
our developers.
They can go do the hard stuffintegrating with antiquated
systems, right? Getting ourbilling to talk to our web to
talk to this. But for my job, Ijust have a file and I need to
(25:01):
get something done. And like,I'm not going to bother our
developer, but I'm going be myown developer. I don't know,
that's a total flip, right?
We're now we're not makingdecisions for the org, we're
making decisions for Scott and Ican just build it myself. So the
only limiter again is agency.Like just go, you have to go do
it. Most people still won't,even though the tool is right
(25:21):
there. But if they can at leasttry once, it's not as hard as
they might think.
Chris (25:25):
So it's a fascinating point you're making there with
it, but it does change eventhough you're talking about
flipping the model over, youknow, from kind of catering to
the the business as a whole tobeing able to cater it to each
individual contributor in thebusiness. By doing that, I'm
curious, you know, that thatopens up a lot of possibilities
for how you might run thebusiness going forward. Do you
(25:48):
have any thoughts on, like, whatthat does to the business if
assuming what's in a in ahypothetical world that you
could get your entire workforceto engage in that way? What do
you think that does for abusiness and how how might if
you were the CEO of a business,how might you operate in such a
way to change that if you werejust everyone's empowered with
(26:08):
AI agents that they can make insixty seconds? What does that do
for them?
Scott (26:12):
Yeah. And this is what Chip's trying to build. This is
really my Arzondetra, like wherewe think work is going is we
need an umbrella of safety sothat our employees can do
whatever they want withoutfeeling like they're going to
break something. Right now, thefear of messing up is greater
than the fear of missing out.And so we need to get rid of
that fear of messing up.
(26:32):
So I always say, the FOMO isgreater than the FOMO. We got to
get rid of the FOMO becausepeople aren't taking action
because they're scared. And so Ithink if I'm a company, what I'm
doing is I have my five to 10core apps. This is how we work.
When you start at Scott Inc,you're gonna go through the
onboarding chatbot.
You're gonna get the contentcreator that writes everything
in our voice. You're going to,you know, get the data analysis
(26:55):
that's gonna analyze thespreadsheets in the same way. So
these are the apps everybodyuses. This is company standard.
This is getting the laptop withprebuilt software.
And then underneath that now,you can duplicate or build your
own to how you work, right? Soyou have this layer of company
wide apps and then I have myScott apps and maybe they're
only visible to me. And a lot oftimes I might even cross
(27:15):
personal and professionalpotentially, right? Where it's
like, here's my workout scheduleand my agenda builder for work
and my, I don't know, grantwriter tool. But since it's
underneath this umbrella, weknow that it's going to adhere
to privacy.
Any personal information will beremoved so it doesn't violate
any problems. And then the finalpiece is, yeah, we have the
tools, but then we need thatmonthly or biweekly lunch and
(27:37):
learn where like, Hey Scott,what did you build this week?
Oh, cool. Let's just duplicatethat one click and now send it
to Dan. And Dan has similar workor new employee starts.
They can look over my shoulder,already the bots already trained
on all the history, it knowswhat to do so they can jump in.
And, you know, I always, Ialways say that AI really raises
the floor, you know, like everynew employee could start at
(27:59):
average or slightly aboveaverage, You still need to raise
the ceiling yourself, add thatspecial spice, right? Your own
ideas, but it's going to makeeveryone on a whole quicker to
get to work and higher, I guess,like higher average across the
board. And I always tell, youknow, the framework I always
recommend is like the AIsandwich. Like, just think about
you the the AI interactionstarts with you, the human, the
(28:22):
bread on top.
Then the AI is gonna dosomething. That's the meat in
the middle. But then you stillhave to be the human on the
bottom to take that output andto improve it, to share it, to
repurpose it. And so I think alot of new people get the bread
and the meat, but they forgetthe bottom piece of bread. And
so that'd be like the work Iwould do as a leader is, here's
our tools, you can all use it,and you're all going to be good.
(28:44):
Like you're not going to havespelling mistakes, it won't be
wordy, it'll make sense. But nowhow do you get better? And it's
gonna be like adding your ownspice on that last piece of
bread. So that's what I would dofor Scott Inc. So I think home
run.
Daniel (28:57):
And part of that too is like developing the muscle
memory. So like for me, forexample, we've been going
through fundraising recently.There's always the same set of
questions that come up indiligence and in questions about
the product and all this. Andmost of those have been answered
(29:20):
like 3,000,000 times now in someform. And, you know, now looking
back, like and, you know, we'vestarted to do this actually, but
really what would be best is ifwe just had a little chat that
had all of that preloaded intoit and could chat over that.
But at the time it's like, oh,well, I'll just answer this
(29:42):
email that's asking these 10questions. Right? I can bang
that out really quick. But that,I guess there's a muscle memory
thing there, and then there issome barrier to overcome to
configure the system for futurebenefit, right, that you might
not see there. So I don't know.
Yeah. Any suggestions even inyour own personal life where
(30:04):
you've kind of come over?
Scott (30:05):
Yeah. Mean, did the same thing, right? Like, we did a
raise with Chip and we built aChip Chat and it was trained on
all of our slides andeverything. People still want to
talk to you. It doesn't meanthat they don't get a human,
Daniel (30:19):
it
Scott (30:20):
gives them the option. The data we're seeing for our
users using Chip for customersupport, like a chat bubble sort
of use case, 70% of them are notclicking the talk to a human
button. They just want to knowwhat are your opening hours? How
much does it cost? Who are you?
Just give me the facts. As abusy parent, I get that. I don't
want to make phone calls becauseI know it'll take five to ten
(30:42):
minutes versus a minute if I'mdoing it myself. I think there's
that aspect of time efficiencyand it is changing habits of
going somewhere else or like yousaid, taking core info and
putting it into a repository.What we found most helpful is we
have something called dynamicknowledge sources.
So if it's a spreadsheet or afolder on Google Drive or
OneDrive, anything that getsadded into those places is
(31:05):
automatically added into youragent. And so I think with
businesses, it's important tothink about that flow of
information and minimizing asmuch like documentation work as
you can. So we always puteverything into notion or
confluence or Google Sheets orGoogle Docs, make that your hub
that is fed into the AI. Soeverything that you put in that
(31:27):
place gets automatically addedinto your FAQ bot or your, you
know, marketing assistant bot orwhatever. So I think that's,
that's key is like, you can askpeople to do it, but even better
is like not to require more workor even changing behavior
because we know that's thehardest part.
So maybe it's a BCC email thatgoes into a spreadsheet that's
(31:48):
automated, right? Or you know,something like that. You can
kind of decide. Way we do it iswe actually look at our chat
logs of people engaging withChip and find the answers that
are going unanswered or don'thave a great answer. And then we
add those things in once a weekinto our chat so that it
improves for the things peopleare asking for rather than
trying to solve for hypotheticaledge cases.
Daniel (32:09):
Well, Scott, we've talked a little bit about Chip.
I've described it a little bit.I'm wondering maybe for you've
been on this journey of kind oftrying to build this easy to use
AI tool. Along that journey,have you found I'm sure you
tried various things that didwork and didn't work, and
(32:32):
certain things have beendifficult and certain things
have been easier. As you reflecton that kind of as a founder of
of an AI company trying to buildan AI tool, any things that
you'd wanna highlight in termsof things that were kind of key
insights or bumps along the roadthat in retrospect you look at
(32:53):
and kind of make sense oranything like that?
Because I think there are a lotin our audience that have maybe
ideas for things out there.Yeah.
Scott (33:03):
No, that's amazing. There's so many. Think I'll take
like a non obvious one, which iswe were pretty early on focused
on building community. So wehave over 20 chip chapters
around the world, peopleteaching one another AI, fairly
active Discord. That's beeninvaluable because those are the
people who are bringing backproblems and ideas and being
(33:25):
able to build towards actualcustomer questions is so
important.
And a lot of times customersdon't have time or interest
giving you feedback, which youneed. And so what we've done is
every two weeks or so, havingfree workshops to try to educate
our users and anybody. Andthat's really built a
relationship, I think, where weknow these people by name, we
(33:46):
know where they live, what theydo, and it makes it a lot easier
for them to be like, Yo, can youbuild this thing? I need it for
a pitch on Friday. We're like,Yeah, for you, of course,
because you're contributing.
So it's building relationships.And it doesn't have to be
hundreds, right? This can bedozens of people who love you
and that's how you really start,is like a strong foundation. So
I think that one's non obvious.I think technically something
(34:08):
that we found maybe an accidentand we're trying to lean into
now is riding the wave of otherpeople's innovation.
Know, like you can only build somany unique pieces and you need
to be on top of other parts ofthe tech stack. Is built on top
of large language models. So asAnthropic and OpenAI build
(34:28):
better models, Chip gets better.For a lot of our users, they
think Chip is doing that becausewe are their front door to AI.
As the models get better, Chipgets better and their experience
gets better.
We partner with folks likePredictionGuard who help us
provide better privacy andsecurity. We could go spend six
months trying to build that, butnow we've lost the whole point
(34:48):
of what we're doing, right? Andso what is your forte is really
important. One thing that hasreally recently that we kind of
focused on is Anthropic has anew protocol called, what is it,
Model context protocol. It'sbasically an easy way to connect
APIs into AI tools.
And so that's another example oflike, we've been building one
(35:09):
off APIs to all these differenttools and now it's like, wow,
there's this whole world that'sbuilt towards this standard. And
if we just tap into that, now wecan, again, get better the more
the open source communitycontributes. So I think that's
really interesting to look outwhere are the areas that will
move quickly that you can ridethat wave and then where do you
wanna be a differentiator? Andyou can kind of draw your line
(35:31):
wherever the right place is, butprobably don't try to draw it on
all of them. Pick the onesyou're best at.
Yeah, I think those are a fewand I think just the power of
small teams now. I mean, youread that a lot of places, but
you know our CTO Hunter, who isjust like a beast with AI
coding. It's like, I know ouroutput compared to some legacy
(35:51):
teams is just vastly greater.And so I wouldn't underestimate
if you're a solo founder, yougot a team. We have a couple of
chip users.
There's a guy named Chuck inColorado who he's building a
million dollar one person agencyand he's almost there, right?
And it's all built with AIautomations and he's conducting
everything. There's a lot ofpotential out there. So I would
encourage anyone listening,finding a co founder or a team
(36:13):
is really, really hard, but youdon't have to wait. Like you can
do a lot on your own.
Chris (36:17):
I'm curious, you actually started to get in for a second
to the next question I was gonnaanswer. And that was, you
mentioned like privacy andsecurity and and partnering with
prediction guard for that. Asyou're thinking about these
these different concerns thatweigh in on various industries,
and, you know, there'll be, youknow, legal concerns, things
(36:39):
like, you know, HIPAA in themedical world. And every
industry has its own set ofconcerns that are kind of
external, but are binding thework in those areas. And as you
are kind of kind of unleashingpeople's potential with the work
that you're doing, those kind ofhave to find some sort of
balance.
How are you thinking about theconstraints versus the
(37:02):
unleashing that we talked aboutand finding a balance so that
people are unleashed whilethey're still having to be held
to account, you know, bywhatever those constraints in
their industry is?
Scott (37:12):
Right. Yeah. I mean, I think regulation is always
innovation. And so I would sayas a company, as an individual,
like, look at yourself firstbefore worrying about the
regulatory environment. Youknow, I think about privacy
pyramid as what we tell ourcustomers like the bottom of the
pyramid, the first thing youshould do is just think about
(37:33):
what are you okay sharing andnot sharing and just tell
people.
Again, FOMO is greater thanFOMO. People will not take
action if they think they'regoing to get in trouble. Even if
it's hypothetical, not real. Idon't know, that fear from
elementary school sticks withus. So the first thing you have
to do is remove the fear.
And the best way to do that isjust to say what the rules are.
(37:55):
As long as people know therules, they'll work within them.
But if they don't know what theyare, they're afraid that
whatever they do will get themin trouble, right? So, hey, just
don't upload customer data. Likethat's our rule.
Great. That's a great place tostart. Now go do anything else.
Or no customer data and don'tintegrate with these files.
Great.
And the second level of thepyramid after kind of just best
(38:17):
practices internally is thengonna be like human protection
air, I call it, which one thingPrediction Guard offers as well,
which is like encrypting piecesof information that get added
that shouldn't be. So if I add aphone number or a social
security number or something, itgets removed for me because I
made a mistake. That's fine. Wemake mistakes, best practices
(38:37):
and then cover other people'smistakes up as they make them.
And then I think the top of thepyramid is where you actually
say, You know what?
Let's put it in our ownenvironment. So that way, if we
can share whatever we wantwithout having to worry, and
that's where you can run an opensource large language model in
your own cloud infrastructure,whatever you share is in your
cloud infrastructure. So somebusinesses have to do that. So
(39:00):
if you are in finance,healthcare, you're probably
gonna wanna do that anyway, justfor regulatory reasons. Some
people wanna do that becausethey know they're gonna be
sharing data that might besensitive.
But I think for most of us, toget started, just follow that
basic best practice of like,think about it before you share
it. And if you're working with ateam that might make mistakes or
contractors who aren't followingyour rules, like add in that
(39:21):
second level of like human errorprotection.
Daniel (39:24):
Scott, as we kind of get near to the end here, I'm
wondering if you can maybe sharejust a few standout use cases of
maybe things that you've seenpeople do with chip that have
either surprised you or stoodout in way like, oh, I didn't
expect people would do this,that, you know? Or things that
(39:47):
are like, oh, I didn't even knowyou know, I built the platform,
but I didn't even know that waspossible.
Scott (39:53):
Every day. That's my favorite part of chip and AI
generally is we really arebuilding the tools and we don't
know how people will use them.And it's so crazy to see what
people do with it. I mean, themost common use cases, I would
say there's five areas thatpeople use all the time. It's
operations, marketing, sales.
I call it company search,finding stuff in your Google
(40:15):
Drive basically. And what's thelast one? Data analysis,
reviewing financials and thingslike that. So those are the most
common. But in terms of fun,weird ones, we had somebody who
launched a Canadian tariffchecker.
And so as the tariffs on Canadawere released, you could
actually search any product andit would source where they were
coming from and tell you whatthe change in price would be.
(40:37):
Was totally interesting. One ofmy favorite use cases, a guy
named Tyler Hansen, he's inSioux Falls, South Dakota, and
he runs an HVAC company. He putin all of the training manuals
for all of the equipment thatthey service. So then his
technicians are on the ground.
And instead of having to be inthe bathroom watching a YouTube
video, which I know has happenedwhen my HVAC guy comes, right?
(40:58):
He's actually learning how to dothe thing that I asked him to
do. They can actually pull upthe specific model via their
chitchat and get instructions onwhat to do and how to service it
and parts. That one's reallyfun. There's a contractor out in
Washington.
He uses it to create supplylists. He just puts in square
footage and what people aregoing to build. And then it'll
spit out how much wood he needs,how many nails, whatever else.
(41:21):
Again, things I know nothingabout. A lot of people doing it
for finding HR policies, findinglet's see, there's a car dealer
that's using it to find cars topurchase, to then resell, right?
So it searches through AutoTrader and Craigslist or
wherever else to find vehicles.It's just so many things, right?
(41:42):
And every day I'm encounteringnew ones that are so
fascinating. The fun part is weintegrate with APIs and
webhooks. Really any tool canget pulled in.
A lot of times chip ends upbeing a front end to an AI tool
that's talking to theirsoftware. So chip becomes the
way they communicate, but thenit's pulling their own data. So
(42:02):
that's super fun. Personally, Ihave a Scott Bot. That's the one
I use every single day.
And so I can write things veryquickly and remember people that
I've talked to, so it brings inpast conversations. So that
helps me quite a bit. So yeah,those are a few random ideas. I
haven't built the West Lafayettetour guide yet, but we do have
some travel AI tools out there.So I bet we could do that too.
Chris (42:26):
So very cool. And while you're building that tour guide,
I might give you a location ortwo as well.
Scott (42:31):
Okay, there you go. Yeah, that's awesome.
Chris (42:33):
So really cool use cases there. That's gotta get you
thinking about thepossibilities. So you come at it
with your own mindset and thethings that you have. Your your
customers are are teaching youevery day about what the new
possibilities and boundariesmight be. So where does that
(42:53):
where does that take you?
Like, when you are are you know,you're kinda done for the
workday, your brain'sdecompressing, and but you're
still kind of, you know, justworking on things. What what are
what's going through your headabout, like, where could things
go with this? You know, you youtake what you're driving and and
the and the folks you're workingwith are driving. You're taking
(43:13):
what your customers are showingyou that you never thought
about. And that's got to leaveyou with some pretty cool ideas
about what the future mighthold.
But can you share some of thoseideas with us?
Scott (43:22):
Yeah, I think I mean, I I reflect at the end of the day in
a lot of ways, because I havefour kids that are 11, nine,
seven, three. And I just reallytry to think about like, what
does society look like when thisis more present? And, you know,
what does education look like?Spent a lot of my life in
education, we work with a lot ofschools who use it for tutors
(43:43):
and advisors. And, what's thevalue of a credential saying,
you know, something when thepace of change is like way
faster than four years, right?
I think ultimately, I imaginethis technology has to fade away
from being AI and just being apart of what we use. And it
helps us lean into the thingsthat make us weird. I think
(44:05):
about AI is the world's bestcover band and it needs like the
originals to cover. So I thinkit really forces us to be more
unique as individuals and createsomething new rather. We're
going to use AI for a lot of thequick answers and it's going to
be average.
It's going to be the middle ofthat bell curve and that'll be
fine for most work. But again,we have to raise the ceiling
(44:27):
ourselves. And so I think itmakes me feel like I want my
kids and hopefully myself tolike, just get good, really good
at whatever weird, interestingthing we care about. Yeah. And
I, man, I don't know, I thinkagency again, like I keep coming
back to that.
But how do you instill a lacklike a fearlessness in people?
Because it feels like, first ofall, most people aren't aware of
(44:49):
the pace of change. And as theybecome aware of it, it's either
I'm scared, I'm going to backaway, or I'm going to lean into
it. And I think we just reallyneed to lean into it. And I
don't know, I think it'sexciting because I'm in Fargo,
and I couldn't, you know, learnto be a nuclear physicist in
Fargo, right?
But now I could like I caneasily go down that path and
(45:10):
learn what I need to connectwith the resources, you know,
showcase my work. And this haskind of been my dream since my
first company in 2009 of like,really given, give giving anyone
wherever they are a chance tobuild. And AI is just like the
next step in that process. And Iknow a lot of people still will
find reasons not to but it'sgoing to be just on that agency
(45:32):
piece like you can. So I don'tknow.
I think a society whereeverybody has a chance to build
and create is incrediblyexciting. It's going to be more
competitive. Everyone around theworld has equal access to the
same models as NASA and thedefense department. It's kind of
wild that you can log into thesethings for free and have the
same power as everyone else. Sothat's an opportunity if you
(45:55):
take it.
I think I saw there was a recentstudy the World Bank did in
Nigeria and students who wereusing ChatGPT as a tutor for six
weeks had the equivalent of twoyears of education. And it's
just so many of our problems areproblems of access. And I think
a lot of those access problemsgo away. And then what happens
when another 1,000,000,000people come online with
(46:17):
education who don't have it now,like that's just better for us
all. We can come up with reallyexciting solutions to our
problems.
Daniel (46:22):
Well said. Yeah. That's a that's a great way to end.
Thanks for thanks for joining,Scott. I encourage everyone to
go, create your first chip chaton, chip, chipp.ai, and and have
some fun.
Explore those that weirdness asas Scott put it. I I love that.
Thanks for joining, Scott. It'sbeen great
Scott (46:43):
to chat. Great to be here, guys.
Jared (46:51):
Alright. That is our show for this week. If you haven't
checked out our changelognewsletter, head to
changelog.com/news. There you'llfind 29 reasons. Yes.
29 reasons why you shouldsubscribe. I'll tell you reason
number 17. You might actuallystart looking forward to
(47:11):
Mondays.
Scott (47:12):
Sounds like somebody's got a case of the Mondays.
Jared (47:15):
28 more reasons are waiting for you at
changelog.com/news. Thanks againto our partners at fly.io to
Brakemaster Cylinder for theBeats and to you for listening.
That is all for now, but we'lltalk to you again next time.
Popular Podcasts
United States of Kennedy
United States of Kennedy is a podcast about our cultural fascination with the Kennedy dynasty. Every week, hosts Lyra Smith and George Civeris go into one aspect of the Kennedy story.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
Bookmarked by Reese's Book Club
Welcome to Bookmarked by Reese’s Book Club — the podcast where great stories, bold women, and irresistible conversations collide! Hosted by award-winning journalist Danielle Robay, each week new episodes balance thoughtful literary insight with the fervor of buzzy book trends, pop culture and more. Bookmarked brings together celebrities, tastemakers, influencers and authors from Reese's Book Club and beyond to share stories that transcend the page. Pull up a chair. You’re not just listening — you’re part of the conversation.