April 26, 2023 • 10 mins
In this episode, Brick and Caleb discuss ChatGPT - its current capabilities, shortcomings, and anticipated impact on BI. They offer some advice about preparing your company data to leverage these advancing technologies.
Click here to watch this episode on our YouTube channel.
Blue Margin helps private equity owned and mid-market companies organize their data into dashboards to execute on strategy and create a culture of accountability. We call it The Dashboard Effect, the title of our book and podcast.
Visit Blue Margin's library of additional BI resources here.
For a free, downloadable copy of our book, The Dashboard Effect, click here, or buy a hardcopy or Kindle version on Amazon.
#AI #datastrategy #datalake #BI
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Brick Thompson (00:04):
Welcome to The Dashboard Effect podcast. I'm
Brick Thompson.
Caleb Ochs (00:07):
I'm Caleb Ochs.
Brick Thompson (00:10):
So today we're gonna be talking about ChatGPT.
Caleb Ochs (00:14):
Yeah, here we go.
Brick Thompson (00:16):
It's funny- for some people, it seems like this
is almost a constant source ofconversation and
experimentation. Yet, we're bothrunning into people who haven't
used it, maybe have heard of it,but don't really know what it
is. So I thought we'd starttoday just by kind of talking
about what is ChatGPT? Thenwe'll bring it back around to it
(00:41):
might have some bearing on data.
Caleb Ochs (00:43):
Yeah. I mean, in our sphere of social tech nerds,
this is everywhere, right? AndIt's like, everybody's talking
about This. But then you gooutside of that tech circle, and
you talk to people, and they'relike, "maybe I've heard of it",
or haven't heard of it or don'tknow it's free to use or some of
(01:03):
the implications. We're notgoing to get into all that
stuff. I think It's interestingto connect it back to what we do
and kind of what we see comingdown the road.
Brick Thompson (01:12):
So for listeners who may not be familiar with
ChatGPT- ChatGPT is what theycall a large language model.
It's a form of AI. ChatGPT isactually a product produced by a
company called OpenAI. Microsoftis a big investor in them and
actually provides a lot ofinfrastructure for them. On the
(01:34):
back end behind sort of the theUI, there's a tool called GPT-4
which OpenAI built, which isquite amazing when you first
experience it.
Caleb Ochs (01:47):
Well, nuts and bolts it's just code, but It's put
together like various layers.
Again, we won't get into all thedetails. It's really, really
cool what it's able to produceand what it's able to understand
from the prompts and texts youput into it.
Brick Thompson (02:02):
So again, not to get technical, not that I'm an
expert anyway, but the largelanguage model is a neural
network that seems to understandquestions that you're asking it
and gives, sometimes veryinsightful and smart, answers.
You can ask it to do things likewrite an email, summarize an
(02:23):
article, write an article, writea podcast outline. It's quite
amazing the amount of time itcan save, because it does such a
good job. I find when I use itfor things like writing an
email, I still have to do anedit on that, but it gets that
first draft out, which isfantastic and often very
(02:44):
insightful. One of the toolsthat we're using around the
office quite a bit is based onGPT-4, which is the underlying
technology, behind ChatGPT. It'scalled Bing chat. You can
actually access it just by goingto the Bing search engine and
looking for the chat icon at thetop of the page, or using the
(03:06):
Edge browser. There's a big chatsidebar that you can use and
we're using it in our companyquite a lot to just shorten that
writing process.
Caleb Ochs (03:18):
I mean, it is really interesting. I think chat GPT-3
came out late last year. What alot of people don't know is
that's what really blew it up.
This was ChatGPT has been athing for a while,and they had
different different models.
Brick Thompson (03:36):
GPT-1, GPT-2...
Caleb Ochs (03:39):
And they left a lot to be desired. So they've been
working on this for a long timeand not just open AI either.
There's a lot of the models thatare out there that have been
working on it. Yeah. The shortof it is is that ChatGPT is
really the first one that'sreally starting to integrate
itself with productivity tools.
You can obviously see why withMicrosoft's heavy hand in it.
(04:02):
These these AI models, and theselarge language models are not
new. They're all over the place,and they're going to become more
and more part of what people doat work, and they're going to
open up a lot of opportunity.
Brick Thompson (04:20):
I think that's right. It's kind of early days,
in terms of them getting to wideadoption. There are things you
have to be careful of. They'llgive you the wrong answer, and
make it sound very convincing.
Also, at least internally hereat our company, we're being very
careful not to put anythingproprietary or confidential out
on them, because we're not sureexactly where that might go. So
(04:42):
we're trying to be supercareful, but we can see how
quickly it's evolving. There's athere's a thing that open AI is
doing right now, which is toallow add ons to ChatGPT which
will connect this large languagemodel to other technologies like
Wolfram Alpha, or the internet,or all sorts of different
(05:05):
things. This sort of brings usto our topic today, which is
when and how will it allow us toconnect to data?
Caleb Ochs (05:14):
I mean, there's an API that OpenAI has, that we've
done some playing around with,and tried to throw some data at
it and see what it can come upwith. You can past an Excel
sheet or a table of data in theChatGPT right now, and it'll
give you an analysis of it. Sothose things are happening. And
(05:36):
it's gonna going to be here,before we know it.
Brick Thompson (05:39):
Well, it'll write pretty credible SQL code
for you. Not always correct butdarn close. That'll keep getting
better. I have a feeling, wellyou and I both have a feeling,
that it won't be too long beforeyou'll be able to point chat
GPT, or a tool like it, at adata source like a data lake and
(05:59):
have access to semantic layerthat tells it how the data
relates to the different partsof the data lake, and do that
analysis and figure it outitself.
Caleb Ochs (06:11):
I'm sure at some point it'll get there. First
step will probably be somethingmore along the lines of "Point
me as something like a datamodel, and I'll give you some
answers about it." I do thinkthat's essentially what it has
done with all the data it's beenfed, right? There's just a bunch
of data, and it's just kind oflearned all these things. Now it
(06:35):
can give you intelligentresponses when you asked you a
question. You're just going tobe able to do that with your
business's data.
Brick Thompson (06:42):
So natural language queering on data is
available. I talked with Katelast week about this. Power BI
has a q&a function. It's been ofquestionable utility, just
because it takes a lot of finetuning to make it work. I have a
feeling with these largelanguage models, It's going to
get to where it has hugeutility, and may become one of
(07:04):
the main interfaces that dataanalysts and especially
executives or business peopleare using to query their data.
Caleb Ochs (07:12):
Imagine just being able to ask a question of your
of your data and it just givesyou an answer. The Power BI one,
like you said, a lot to bedesired there. I would bet
that's going to come alongpretty quickly, especially on
the like, kind of your curateddata models already. That q&a
(07:33):
feature is gonna get awesome.
Then broader picture, gettingyour data into a spot to where
that thing can analyze it.
Brick Thompson (07:42):
You saw an example that you sent over to me
recently of a guy that built apower app that uses the OpenAI
API, to interface with datainside of a Power BI model, and
tell you smart things. It'sstill early, and maybe a little
crude. I didn't go deep on it,but It's a very, very obvious
(08:04):
glimpse into things that arecoming soon.
Caleb Ochs (08:08):
We've talked about it before- that we think this is
going to become a must to haveyour data in a spot to where
when these models are ready forit, they can actually go do the
analysis and kind of pulltogether the answers that you're
gonna need from it. If you don'thave it, and everything's still
(08:28):
kind of siloed away, you'regonna be scrambling to get
there.
Brick Thompson (08:32):
So that's leading us to a recommendation
for our clients, which is tomake sure you're consolidating
your data, integrating yourdata, probably into a data lake,
because that's inexpensive andquick to do, so that you're
ready when these tools becomeavailable, to start applying
them immediately. I think itcould be a competitive
(08:54):
advantage, especially if you'rein a competitive market. If you
have a good handle on your data.
It's going to give you theability to figure things out
easier than your competition.
Caleb Ochs (09:05):
Just imagine sitting in a meeting and having somebody
ask a question you don't knowthe answer to. "Let me find out
right now" and just go type inyour your question, and it gives
you back the answer. Extremelypowerful. We're obviously not
there yet, but It's going to bethere. Then that's such a key
foundational piece to have yourdata ready to go.
Brick Thompson (09:24):
I think actually, there's good arguments
for having your data in a singlerepository anyway. Even with
natural language querying,you're still going to want
standard reports that people arelooking at, that are up on
screens around the office, thattype of thing. The data lake
enables that. Then just has youready for this extra- I was
(09:46):
going to call it icing, but itmay become sort of the main
thing.
Caleb Ochs (09:51):
I wouldn't be surprised!
Brick Thompson (09:54):
All right. What else on this?
Caleb Ochs (09:55):
Plenty more to come,. I think we're gonna have
a few more episodes about thistopic, because it's important
and I think it is a shaker.
Brick Thompson (10:07):
Yeah, I think It's gonna be important to all
businesses. Our typical clientsare mid market companies usually
PE owned. It could have reallyparticular benefit there and
particular utility there. Somaybe we'll talk about that next
week.
Caleb Ochs (10:25):
Yeah, for sure.
Let's do it. Thanks!
Brick Thompson (10:27):
Thanks!
Welcome to The Dashboard Effect podcast. I'm
Brick Thompson.
Caleb Ochs (00:07):
I'm Caleb Ochs.
Brick Thompson (00:10):
So today we're gonna be talking about ChatGPT.
Caleb Ochs (00:14):
Yeah, here we go.
Brick Thompson (00:16):
It's funny- for some people, it seems like this
is almost a constant source ofconversation and
experimentation. Yet, we're bothrunning into people who haven't
used it, maybe have heard of it,but don't really know what it
is. So I thought we'd starttoday just by kind of talking
about what is ChatGPT? Thenwe'll bring it back around to it
(00:41):
might have some bearing on data.
Caleb Ochs (00:43):
Yeah. I mean, in our sphere of social tech nerds,
this is everywhere, right? AndIt's like, everybody's talking
about This. But then you gooutside of that tech circle, and
you talk to people, and they'relike, "maybe I've heard of it",
or haven't heard of it or don'tknow it's free to use or some of
(01:03):
the implications. We're notgoing to get into all that
stuff. I think It's interestingto connect it back to what we do
and kind of what we see comingdown the road.
Brick Thompson (01:12):
So for listeners who may not be familiar with
ChatGPT- ChatGPT is what theycall a large language model.
It's a form of AI. ChatGPT isactually a product produced by a
company called OpenAI. Microsoftis a big investor in them and
actually provides a lot ofinfrastructure for them. On the
(01:34):
back end behind sort of the theUI, there's a tool called GPT-4
which OpenAI built, which isquite amazing when you first
experience it.
Caleb Ochs (01:47):
Well, nuts and bolts it's just code, but It's put
together like various layers.
Again, we won't get into all thedetails. It's really, really
cool what it's able to produceand what it's able to understand
from the prompts and texts youput into it.
Brick Thompson (02:02):
So again, not to get technical, not that I'm an
expert anyway, but the largelanguage model is a neural
network that seems to understandquestions that you're asking it
and gives, sometimes veryinsightful and smart, answers.
You can ask it to do things likewrite an email, summarize an
(02:23):
article, write an article, writea podcast outline. It's quite
amazing the amount of time itcan save, because it does such a
good job. I find when I use itfor things like writing an
email, I still have to do anedit on that, but it gets that
first draft out, which isfantastic and often very
(02:44):
insightful. One of the toolsthat we're using around the
office quite a bit is based onGPT-4, which is the underlying
technology, behind ChatGPT. It'scalled Bing chat. You can
actually access it just by goingto the Bing search engine and
looking for the chat icon at thetop of the page, or using the
(03:06):
Edge browser. There's a big chatsidebar that you can use and
we're using it in our companyquite a lot to just shorten that
writing process.
Caleb Ochs (03:18):
I mean, it is really interesting. I think chat GPT-3
came out late last year. What alot of people don't know is
that's what really blew it up.
This was ChatGPT has been athing for a while,and they had
different different models.
Brick Thompson (03:36):
GPT-1, GPT-2...
Caleb Ochs (03:39):
And they left a lot to be desired. So they've been
working on this for a long timeand not just open AI either.
There's a lot of the models thatare out there that have been
working on it. Yeah. The shortof it is is that ChatGPT is
really the first one that'sreally starting to integrate
itself with productivity tools.
You can obviously see why withMicrosoft's heavy hand in it.
(04:02):
These these AI models, and theselarge language models are not
new. They're all over the place,and they're going to become more
and more part of what people doat work, and they're going to
open up a lot of opportunity.
Brick Thompson (04:20):
I think that's right. It's kind of early days,
in terms of them getting to wideadoption. There are things you
have to be careful of. They'llgive you the wrong answer, and
make it sound very convincing.
Also, at least internally hereat our company, we're being very
careful not to put anythingproprietary or confidential out
on them, because we're not sureexactly where that might go. So
(04:42):
we're trying to be supercareful, but we can see how
quickly it's evolving. There's athere's a thing that open AI is
doing right now, which is toallow add ons to ChatGPT which
will connect this large languagemodel to other technologies like
Wolfram Alpha, or the internet,or all sorts of different
(05:05):
things. This sort of brings usto our topic today, which is
when and how will it allow us toconnect to data?
Caleb Ochs (05:14):
I mean, there's an API that OpenAI has, that we've
done some playing around with,and tried to throw some data at
it and see what it can come upwith. You can past an Excel
sheet or a table of data in theChatGPT right now, and it'll
give you an analysis of it. Sothose things are happening. And
(05:36):
it's gonna going to be here,before we know it.
Brick Thompson (05:39):
Well, it'll write pretty credible SQL code
for you. Not always correct butdarn close. That'll keep getting
better. I have a feeling, wellyou and I both have a feeling,
that it won't be too long beforeyou'll be able to point chat
GPT, or a tool like it, at adata source like a data lake and
(05:59):
have access to semantic layerthat tells it how the data
relates to the different partsof the data lake, and do that
analysis and figure it outitself.
Caleb Ochs (06:11):
I'm sure at some point it'll get there. First
step will probably be somethingmore along the lines of "Point
me as something like a datamodel, and I'll give you some
answers about it." I do thinkthat's essentially what it has
done with all the data it's beenfed, right? There's just a bunch
of data, and it's just kind oflearned all these things. Now it
(06:35):
can give you intelligentresponses when you asked you a
question. You're just going tobe able to do that with your
business's data.
Brick Thompson (06:42):
So natural language queering on data is
available. I talked with Katelast week about this. Power BI
has a q&a function. It's been ofquestionable utility, just
because it takes a lot of finetuning to make it work. I have a
feeling with these largelanguage models, It's going to
get to where it has hugeutility, and may become one of
(07:04):
the main interfaces that dataanalysts and especially
executives or business peopleare using to query their data.
Caleb Ochs (07:12):
Imagine just being able to ask a question of your
of your data and it just givesyou an answer. The Power BI one,
like you said, a lot to bedesired there. I would bet
that's going to come alongpretty quickly, especially on
the like, kind of your curateddata models already. That q&a
(07:33):
feature is gonna get awesome.
Then broader picture, gettingyour data into a spot to where
that thing can analyze it.
Brick Thompson (07:42):
You saw an example that you sent over to me
recently of a guy that built apower app that uses the OpenAI
API, to interface with datainside of a Power BI model, and
tell you smart things. It'sstill early, and maybe a little
crude. I didn't go deep on it,but It's a very, very obvious
(08:04):
glimpse into things that arecoming soon.
Caleb Ochs (08:08):
We've talked about it before- that we think this is
going to become a must to haveyour data in a spot to where
when these models are ready forit, they can actually go do the
analysis and kind of pulltogether the answers that you're
gonna need from it. If you don'thave it, and everything's still
(08:28):
kind of siloed away, you'regonna be scrambling to get
there.
Brick Thompson (08:32):
So that's leading us to a recommendation
for our clients, which is tomake sure you're consolidating
your data, integrating yourdata, probably into a data lake,
because that's inexpensive andquick to do, so that you're
ready when these tools becomeavailable, to start applying
them immediately. I think itcould be a competitive
(08:54):
advantage, especially if you'rein a competitive market. If you
have a good handle on your data.
It's going to give you theability to figure things out
easier than your competition.
Caleb Ochs (09:05):
Just imagine sitting in a meeting and having somebody
ask a question you don't knowthe answer to. "Let me find out
right now" and just go type inyour your question, and it gives
you back the answer. Extremelypowerful. We're obviously not
there yet, but It's going to bethere. Then that's such a key
foundational piece to have yourdata ready to go.
Brick Thompson (09:24):
I think actually, there's good arguments
for having your data in a singlerepository anyway. Even with
natural language querying,you're still going to want
standard reports that people arelooking at, that are up on
screens around the office, thattype of thing. The data lake
enables that. Then just has youready for this extra- I was
(09:46):
going to call it icing, but itmay become sort of the main
thing.
Caleb Ochs (09:51):
I wouldn't be surprised!
Brick Thompson (09:54):
All right. What else on this?
Caleb Ochs (09:55):
Plenty more to come,. I think we're gonna have
a few more episodes about thistopic, because it's important
and I think it is a shaker.
Brick Thompson (10:07):
Yeah, I think It's gonna be important to all
businesses. Our typical clientsare mid market companies usually
PE owned. It could have reallyparticular benefit there and
particular utility there. Somaybe we'll talk about that next
week.
Caleb Ochs (10:25):
Yeah, for sure.
Let's do it. Thanks!
Brick Thompson (10:27):
Thanks!
Popular Podcasts
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
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.
Therapy Gecko
An unlicensed lizard psychologist travels the universe talking to strangers about absolutely nothing. TO CALL THE GECKO: follow me on https://www.twitch.tv/lyleforever to get a notification for when I am taking calls. I am usually live Mondays, Wednesdays, and Fridays but lately a lot of other times too. I am a gecko.