September 11, 2023 • 23 mins
CrushBank CTO, David Tan, connects the dots between early days of fantasy football and the ChatGPT hype of today. Along the way, he co-founds a company to help bring AI sanity to MSPs.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:01):
Hello everyone and welcome to the Crush Bank AI for
MSPs podcast, a podcast wherewe talk about everything related
to artificial intelligence,machine learning, managed
services and really anythingelse that I think about that I
think you might find interestingthat I feel like talking about.
My name is David Tan, I'm yourhost and I'm really excited
(00:22):
about today's episode because Ithink it's going to be a lot of
fun.
We're going to talk about whatis arguably the hottest
technology on the planet rightnow generative AI.
But that's not it.
We're going to talk abouteverything from financial
reporting to high schoolfootball to fantasy football and
a bunch of things in between,and if you have no idea how
those things can be connected,strap in, because this is going
(00:44):
to be a bit of a ride.
We're going to have a littlebit of story time.
I'm going to talk about how Igot interested in AI.
I'm going to talk about thepower of generative AI a bit.
I'm going to talk about some ofthe dangers of generative AI.
But, to get started, the ideabehind this podcast is not to
talk down to my audience.
I fully expect that most of youwill know what generative AI is
(01:08):
, understand the technology,have a pretty good idea of what
I'm talking about, so I'm reallygoing to try and skip the
basics where possible.
That being said, I also thinkthere's a really good
possibility the only person thatlistens to these episodes is
going to be my mom.
She doesn't really know whatgenerative AI is, so, for my
mom's sake and anyone else,that's not technical.
(01:28):
Let's start with a little bit ofbackground.
Generative AI is, as it sounds,a type of artificial
intelligence where largelanguage models, or large models
, I should say, in general, areused to create original content.
Now, that content can beanything from text to images, to
(01:49):
computer code and a lot ofthings in between.
Obviously, generative AI hitthe mainstream back in about
November of 2022, when OpenAIreleased their ChatGPT product,
gave people the ability to logon for free and chat with a
computer bot that could doanything from answer questions,
(02:12):
create original content, writestories, write narratives,
create outlines, build recipesyou name it.
You can do it with ChatGPT, butbefore ChatGPT, openai was
getting a little bit ofpublicity around their Dolly
product.
Dolly is also a generative AIengine.
(02:34):
The difference there is agraphic generative engine.
What that means is you candescribe a picture and Dolly
will spit out what that picturewould look like.
So, in other words, you couldtell Dolly to create you a
picture of an Eskimo riding amotorcycle through Times Square
on New Year's Eve and it wouldcome back with something that
probably doesn't accuratelyrepresent that, but at least in
(02:54):
name represents that.
In other words, it will haveall of those elements.
It will be recognizable as whatyou asked for.
When OpenAI released ChatGPT,like I said, it got incredibly
popular, incredibly mainstream,and everyone started using it
and since then probably the yearalmost a year since not quite a
(03:15):
lot of other companies havecome out with similar
technologies that people arestarting to leverage in all
aspects of their business.
I mentioned, you can use it tocreate code.
There's a lot of low codeplatforms built on generative AI
.
People are using it to createcontent.
Like I said, people are usingit to create images, slideshow,
decks, things like that.
(03:36):
You name it.
There are really powerfulmodels that can be used to
create that.
And again, I'm also not going toget too deeply technical on
some of these episodes.
There may be some where I am,but this one's a little bit more
about having fun, a little bitmore story time, so I thought it
might be fun to take a stepback and give you guys a little
bit of a background of why Ifirst ever got interested in
(03:56):
artificial intelligence andwhere my fascination with some
of this technology came from.
So I'm going to go back,flashback to 1994.
My business partner, evan, whoyou will hear on a bunch of
these episodes, and I had juststarted our IT company.
We were two snot-nosed kidsliving at our parents' homes, no
(04:16):
expenses, really no idea whatwe wanted to do.
But we got a great opportunityand we started an IT company.
That's a long story for anotherday.
Maybe it'll come up on anotherepisode.
That's not important.
What's important was we reallydidn't have much to do.
We had a couple of customers,we were trying to build a
business, but again, thispredated the internet in a lot
(04:37):
of ways That'll become relevantin just a minute.
It predated what we know of asmanaged services.
It certainly predated the cloud.
We got our start back then,just to give you an idea from a
time frame standpoint grippingout Novel Network and installing
Windows NT for businesses.
I remember vividly having tosell our clients on email so not
(04:59):
to pay to myself as someonethat's a million years old, but
you get the idea of when thiswas.
So one of our friends had abrilliant idea of starting a
fantasy football league.
We thought it was great becausewe love football.
Listen, I'm not going to say wewere like the guys sitting in
the back of a diner that createda rotisserie baseball back in
the late 70s or early 80s Ithink it was but we were fairly
(05:22):
early on in the fantasy footballdays and again, we're talking a
time before mainstream internetas you think of it today, no
worldwide web as you know it.
So all of this stuff was donemanually.
So basically there was 12 of usin the league, a bunch of guys
that all grew up together, wentto high school together, great
friends, 12 of us in the leagueand Evan and I were the
commissioners.
We ran the league.
(05:42):
We were responsible essentiallyfor managing the lineups and
the scoring and basically theway it would work back then was
on Sunday mornings, everyone inthe league would call me, leave
a message on my answeringmachine and leave their lineup,
and then Sunday night I woulddownload the box scores from
copy server, prodigy, whateverdial up service I was using, and
(06:05):
we would manually or withspreadsheets we'd compile the
scores and then Tuesday morningafter the Monday night football
game we would fax the finalscores out to all of our friends
.
And again, like I said, wedidn't have much business to do
at the time, so we certainly hada bunch of spare time.
So I also took it upon myselfto start writing newsletters
(06:26):
Really, quite frankly, it was away to make fun of each other.
But it was also good fun.
We called it the PigskinChronicles and it was a great
time.
I would write articles aboutthe individual games I would
write about.
I'd have some sort of aninterest piece, like talk about
a made-up story about someone'slife.
It was fun.
It sounds a little bit childishand probably nonsensical to you
(06:48):
now, but trust me, at the timeit was a good time.
So that was 1994.
Flash forward to about 1999,2000.
The internet as we know itstarted to come more into shape.
Obviouslycom was all the rageback then and a bunch of
websites to run fantasy footballleagues started to sprout up.
We chose CBS Sportsline forsome reason.
(07:11):
I don't remember if it was theonly one or if we liked it, if
it was free, if it was thecheapest, don't remember, but
it's what we chose.
We still use it to this day andyou could do all of everything
online.
You could do it electronically.
I didn't have to manuallycalculate scores, I didn't have
to listen to my answeringmachine to get lineups.
(07:32):
Everyone could handle itthemselves.
They could create their ownlineups.
And what would happen was, astime went on, cbs would send out
an email every Tuesday morningwith the results of the games,
the final scores, but also a bitof a narrative.
Now, at this point our businesshad started to grow so I didn't
(07:52):
have time to write the pigs inchronicles anymore, and I can
assure you CBS Sportsline, whohad probably hundreds of
thousands, if not millions, ofleagues by this point in the
early 2000s, certainly didn'thave staff writers that were
writing fantasy box scorestories about the junkies that
was my team playing thesledgehammers that was Evan's
(08:13):
team.
But every Tuesday we would geta newsletter from CBS that had a
fairly well-written narrativeabout the outcome of the game
and it would talk about coachingmistakes and sitting and
starting the wrong player andbasically just describe the
games and the outcomes of theweek, and then they would come
out with another one which is apreview for the next week, and I
was fascinated to find out howthey did it.
(08:34):
So I dug into it and it turnsout they were using a very early
form of generative AI.
So basically they fed detailsabout the scores of the game,
the rosters, the possiblestarters and sit and players
that could be set free agents,things like that.
They'd feed all that into themachine I'm oversimplifying it,
(08:55):
obviously and the machine wouldspit out a narrative that looked
like an article about theparticular game and, like I said
, you couldn't necessarily tellthat it was written by a
computer.
It was interesting.
As time went on you started tosense a little bit that was
written by a computer, but stillvery cool.
And, like I said, I was superfascinated with the technology
(09:17):
and I found out where it camefrom.
So a few years prior to that,reuters, who had to cover
financial reporting news, foundthemselves in a bit of a dilemma
.
So if you think about publiccompanies, they report earnings
quarterly, obviously, and bigcompanies like Apple, google,
microsoft, you know, caterpillar, you name it the large
(09:41):
companies have teams of analyststhat listen in on earnings
calls and they dig through thefilings and they look inside the
P&Ls and the balance sheet andthey write long narrative
stories about the performance ofthese companies.
That's all well and good forthese large companies, but if
you look at a small company thatmaybe trades over the counter,
(10:02):
they're still responsible forreporting their financials every
quarter and they're stillinterested to the stockholders
to read about those reportings.
So what Reuters did was theybasically fed the same type of
financial reporting data into alanguage model.
I'm not going to call it alarge language model because it
very much predated what we havetoday.
So they would feed these into alanguage model and they would
(10:26):
write a financial news storyabout some over the counter
stock.
The company reported earnings.
Again.
The stock was probably tradingat 12 cents.
So I have no idea who wasreading it probably the same
people that are going to listento my podcast but it was a great
solution for Reuters and thatkind of grew into, like I said,
(10:47):
the technology that was used byfantasy football and that was
really the precursor of what wesee as generative AI now and
again.
It fascinated me.
I've always been fascinatedwith technology, emerging
technology specifically.
I wouldn't say that I sat downand started to try and figure
out business uses of it.
It was the type of thing thatwas kind of in the back of my
(11:07):
mind.
I've told that story many times.
It's like I said, it's lit thespark around AI specifically for
me and when we started CrushBank initially back in 2017, one
of the reasons I was sofascinated with AI was kind of
where that came from.
So I'm going to fast forward nowand this is going to get much
more relevant in a minute.
So thank you for indulging meand learning a little bit of my
(11:29):
background.
I'm going to fast forward nowto a newspaper article from the
Columbus Dispatch.
It's a newspaper in Columbus,ohio, a date line August 18,
2023.
It is a article about a highschool football team game
between two rival high schoolsand I'm going to read it.
(11:49):
So please indulge me.
I'm not going to read theentire thing, but so indulge me
for a minute.
I'm going to read a little bitof it to you.
So the headline is WestervilleNorth escapes Westerville
Central in thin wind in Ohio,high School football action, and
the byline of that is lead AIL-E-D-E-A-I.
Obviously, lead AI.
(12:10):
Well, not, obviously, but leadAI is a service that newspapers,
news stations, newsrooms ingeneral any news reporting
service can subscribe to thatwill essentially quote, unquote,
replace journalists and writeAI articles for you Fairly well
known product.
I know a lot of newsrooms useit, so this is not just not
(12:35):
meant to be critical of them.
It's more of a conversationaround generative AI and some of
the things we should bethinking of when we embrace this
technology.
So that headline again Westescapes Westerville Central and
thin wind in Ohio not what yourtypical sports journalist would
write, obviously right.
I grew up reading Mike Lupica inthe New York Post, right.
(12:58):
I used to wake up early everymorning to get the post to read
Mike Lupica's article about theRanger game from the night
before.
When I was in college I was inMichigan and Mitch Albom was the
journalist of Note in Michiganand he wrote stories about the
fab five or Michigan's footballgate teams or things like that,
and we couldn't wait to readthem.
Anything Mitch Albom wrote wasgold.
(13:19):
This is not the type ofheadline that someone like a
Mitch Albom or Mike Lupica wouldwrite.
Obviously, journalism ingeneral has changed
significantly in the last 30 to35 years, but you can tell
reading that that it is not,that it is written by AI and to
the Columbus Dispatches creditthey.
The byline is the AI productthat they use to write this
(13:41):
article.
I'm gonna read the first coupleparagraphs, so please indulge me
for another few seconds.
The Westerville North Warriorsdefeated the Westerville Central
War Hawks 2112 in an Ohio highschool football game on Friday.
Westerville North edgedWesterville Central at 2112 in a
close encounter of the athletickind at Westerville North High
(14:02):
on August 18th In Ohio footballaction.
Those are the first twoparagraphs.
The first paragraph of a newsarticle, if you don't know, is
called the lead LED and the ideaof it is to grasp your
attention and make youinterested in the article.
Both of those paragraphs wouldqualify as a lead.
So in other words they're bothmeant to be first paragraph
(14:26):
Pieces or parts of a newspaperarticle.
So that's the first strikethere.
The second strike is it justsounds very awkward, right?
If you read it again.
It's not what a ambitiousJournalist, what journalists,
would write, especially when andI'm extrapolating out now here
but especially when you considerthat someone covering high
school football in Ohio Probablyhas dreams of covering the
(14:47):
Browns or the Bengals or evenOhio State, not Westerville
North versus Westerville Central.
But you need to start somewhereand you would assume they would
put their best foot forward.
So we'll kind of leave all ofthat for a minute.
I want to focus on one sentencethere.
I read it to you.
Hopefully you caught it,hopefully you picked up on it
when I was talking about it, butif not, just in case, I'm gonna
(15:09):
read it again Westerville Northedged Westerville Central 2112
in a close encounter of theathletic kind at Westerville
North High.
That sentence Jumps right outof the page at you.
That is absolutely notsomething that a sportscaster
would write.
Our sports journalist wouldwrite.
I should say again it's a I,it's meant to sound like a human
(15:31):
, some ones and zeros gotcrossed someplace.
So it thinks that that is aCompelling sentence in a sports
article.
It is not.
But let's go a step further.
So I find that sentence veryinteresting.
So I want you to play along athome with me now and I want you
to open up Google and I want youto type a close Encounter of
(15:51):
the athletic kind into yoursearch bar and as soon as you do
that, I want you to switch tothe news tab.
So, in other words, you'regetting news articles and you
will notice dozens and dozens ofarticles that come up and if
you click on any one of them,let's see.
I'm gonna go back.
I'm gonna go to the fourth pagehere I'm gonna go back to let's
(16:17):
see how about January 10th of2023?
I'm gonna click on the articleand I'm gonna read it for you.
Raymond Lincoln would edgeverdant North Mac 41 34 in a
close encounter of the athletickind on January 10th in Illinois
boys high school basketball.
(16:39):
And every one of them is thesame.
You can find that sentence noexaggeration in 50 articles
without the blink of an eye, andprobably even way more than
that.
And again, which is actuallypretty funny, if you were to go
to lead a eyes website, whatthey tout is original and
creative content that soundslike it was written by a real
(17:01):
person.
I'm paraphrasing there.
The original and creative partis what strikes out.
That's funny to me.
Now, again, I think that lastarticle I picked a random was
from like the Tennessee and orsomething, and the other one I
read you, like I mentioned, wasfrom the Columbus dispatch.
So certainly neither of themare gonna lose any readers over
relying on a generative AIengine that uses the same silly
(17:24):
sounding sentence over and overagain.
But we need to take a step backand we need to think about this
and how it affects usspecifically.
Again.
Managed services, it companiesin general, but really any
enterprise, any business that'srelying on this technology.
So people often ask me myopinion on generative AI and the
uses of it, and open AIspecifically, and some other
(17:47):
vendors and again, I'm very highon the technology.
I think the promise of it isincredible.
In a lot of ways, it's notready for prime time.
The biggest thing and we'll talkabout this probably on future
episodes, the biggest thing isthat it's not required to be
accurate.
So, in other words, if you ask,if you go to open AI or chat,
you can ask it to writesomething.
(18:08):
It'll write something thatsounds real and sounds realistic
but doesn't necessarily meanit's true and I have a bunch of
great stories that we'll shareabout that at some other time.
So that's the biggest problemin it.
So one of the things I oftenencourage people to use it for
is to create content fromsomething that you know is real.
So, in other words, feeding ininformation about the final
(18:30):
score of a high school footballgame and asking it to generate a
notepad article about it.
In that scenario, under thoseconstraints, it will generally
create an article that isaccurate and truthful and, again
, it will use the technology,will use the language that it
finds appropriate, that thinksit's interesting or that it that
is usable.
In that case, the reason I bringthis up and the reason I talk
(18:52):
about it is because one of themost common uses that I
recommend for things like thisis around sales and marketing
pieces.
So, in other words, you're asmall upstart MSP and you need
to do a marketing blitz or anemail campaign and you want to
get a bunch of new clients tosign up for some new managed
security plan that you're aboutto launch, but you just don't
have the expertise on writingmarketing pieces around managed
(19:15):
services or managed securityservices in this case.
Now you can go out and you canhire someone, or you can
contract someone, or you can useone of the multitude of great
partners there are on thechannel to do stuff like that,
but you might want to take acrack at it yourself.
So you might want to log on toChatGBT, you might want to
describe the offering and youmight want to ask it to write an
email for you and, truth betold, it will probably do a
(19:36):
pretty good job of that.
It will write a compelling,interesting, accurate piece for
an email for your website, for amarketing blast.
Whatever you're doing with itbased on those criteria, the
problem is you don't want theguy down the street that selling
and competing, offering to dothe same thing for you and have
(19:56):
the same language in there.
Now again, I'm not saying thatin every case this language is
going to come out the same, butwhen you're using the same
machine learning model or Ishould say the same large
language model in this case, togenerate similar content,
chances are those results aregoing to be pretty close to each
other.
Now you can put your own slanton it.
You can ask to write it likeyou're a 14-year-old skater or a
(20:18):
25-year-old gamer or a pirateliving in the 1800s, like
whatever you choose to slant toput on it.
You can get it to write in sortof a different voice or a
different method, but chancesare you're just going to ask it
to write as a businessprofessional or not.
Ask it all and it will implythat you sorry, it will infer,
rather, that you think it shouldbe written as a business
(20:40):
professional.
And that's the danger of a lotof this stuff.
And leaving it what we callunshaperone, leaving it alone,
is wrought for opportunity forthings to go horribly wrong
Again.
You don't want your website tohave the same content.
I remember early on when wewere starting to sell managed
services, there were a bunch ofmarketing firms that were
selling, offering kind ofpre-canned articles that you
(21:03):
could put on your website, orthey were writing content for
your website.
And this is years and years ago.
They've gotten way better.
But if you had at the time, ifyou had asked one of these
companies to generate contentfor your website and then you
had Googled it, you would findother managed service providers
that had the same exact contenton their site.
Or maybe you guys were usingthe same Microsoft articles or
(21:24):
the same HP articles or whoeverwas putting out content for the
SMB, for IT services, and we'rekind of at risk of that
happening again.
So the moral of the storyreally is I think it's kind of
fun and interesting and helpsyou understand sort of how this
technology works, where it camefrom.
But really the moral of thestory is that, no matter what
(21:44):
the use case, never let this AIwork alone, never let it work in
a vacuum, always have itchaperoned by some sort of
professional right.
So whether you're asking it towrite code for you or write a
bunch of scripts for you forsome sort of automation great,
because it will absolutelyshortcut the work and it will
(22:05):
get done much faster.
But you're absolutely out ofyour mind if you roll that stuff
out without having someone thatunderstands it, checking it and
rechecking it and running it ina sandbox and doing all the
sort of due diligence you shoulddo before that stuff gets
deployed.
By the same token, don't ask itto write an email, copy and
paste that email into MailChimpand send it out.
(22:26):
Read it, edit it, make it yourown.
And if you could start on thirdbase, so to speak, with the
amount of content you needwritten or the content you want,
in this case, that's stillsaving you a tremendous amount
of time.
Don't assume it's going to doall your work for you, or, quite
frankly, don't ask it to do allyour work for you, and that, I
think, is something to remember.
(22:47):
Just about AI in generalgenerative AI specifically, but
AI in general.
We like to say that AI will notreplace people.
People that use AI will replacepeople, and this is a perfect
example of that at play.
Here's this technology playaround with it, see what's out
there, use it to generatecontent, but never let it work
alone in a vacuum.
(23:08):
That would be my takeaway ofthe day, my big advice, my big
lesson learned that, and don'tever draft a starting
quarterback from the DallasCowboys for your fantasy
football team.
Two lessons of the day.
Thanks again for joining metoday.
My name is David Tan.
This has been the Crush Bank AIfor MSPs podcast.
Check me out on LinkedIn, checkus out at CrushBankcom and
(23:31):
enjoy the rest of your week.
Hello everyone and welcome to the Crush Bank AI for
MSPs podcast, a podcast wherewe talk about everything related
to artificial intelligence,machine learning, managed
services and really anythingelse that I think about that I
think you might find interestingthat I feel like talking about.
My name is David Tan, I'm yourhost and I'm really excited
(00:22):
about today's episode because Ithink it's going to be a lot of
fun.
We're going to talk about whatis arguably the hottest
technology on the planet rightnow generative AI.
But that's not it.
We're going to talk abouteverything from financial
reporting to high schoolfootball to fantasy football and
a bunch of things in between,and if you have no idea how
those things can be connected,strap in, because this is going
(00:44):
to be a bit of a ride.
We're going to have a littlebit of story time.
I'm going to talk about how Igot interested in AI.
I'm going to talk about thepower of generative AI a bit.
I'm going to talk about some ofthe dangers of generative AI.
But, to get started, the ideabehind this podcast is not to
talk down to my audience.
I fully expect that most of youwill know what generative AI is
(01:08):
, understand the technology,have a pretty good idea of what
I'm talking about, so I'm reallygoing to try and skip the
basics where possible.
That being said, I also thinkthere's a really good
possibility the only person thatlistens to these episodes is
going to be my mom.
She doesn't really know whatgenerative AI is, so, for my
mom's sake and anyone else,that's not technical.
(01:28):
Let's start with a little bit ofbackground.
Generative AI is, as it sounds,a type of artificial
intelligence where largelanguage models, or large models
, I should say, in general, areused to create original content.
Now, that content can beanything from text to images, to
(01:49):
computer code and a lot ofthings in between.
Obviously, generative AI hitthe mainstream back in about
November of 2022, when OpenAIreleased their ChatGPT product,
gave people the ability to logon for free and chat with a
computer bot that could doanything from answer questions,
(02:12):
create original content, writestories, write narratives,
create outlines, build recipesyou name it.
You can do it with ChatGPT, butbefore ChatGPT, openai was
getting a little bit ofpublicity around their Dolly
product.
Dolly is also a generative AIengine.
(02:34):
The difference there is agraphic generative engine.
What that means is you candescribe a picture and Dolly
will spit out what that picturewould look like.
So, in other words, you couldtell Dolly to create you a
picture of an Eskimo riding amotorcycle through Times Square
on New Year's Eve and it wouldcome back with something that
probably doesn't accuratelyrepresent that, but at least in
(02:54):
name represents that.
In other words, it will haveall of those elements.
It will be recognizable as whatyou asked for.
When OpenAI released ChatGPT,like I said, it got incredibly
popular, incredibly mainstream,and everyone started using it
and since then probably the yearalmost a year since not quite a
(03:15):
lot of other companies havecome out with similar
technologies that people arestarting to leverage in all
aspects of their business.
I mentioned, you can use it tocreate code.
There's a lot of low codeplatforms built on generative AI
.
People are using it to createcontent.
Like I said, people are usingit to create images, slideshow,
decks, things like that.
(03:36):
You name it.
There are really powerfulmodels that can be used to
create that.
And again, I'm also not going toget too deeply technical on
some of these episodes.
There may be some where I am,but this one's a little bit more
about having fun, a little bitmore story time, so I thought it
might be fun to take a stepback and give you guys a little
bit of a background of why Ifirst ever got interested in
(03:56):
artificial intelligence andwhere my fascination with some
of this technology came from.
So I'm going to go back,flashback to 1994.
My business partner, evan, whoyou will hear on a bunch of
these episodes, and I had juststarted our IT company.
We were two snot-nosed kidsliving at our parents' homes, no
(04:16):
expenses, really no idea whatwe wanted to do.
But we got a great opportunityand we started an IT company.
That's a long story for anotherday.
Maybe it'll come up on anotherepisode.
That's not important.
What's important was we reallydidn't have much to do.
We had a couple of customers,we were trying to build a
business, but again, thispredated the internet in a lot
(04:37):
of ways That'll become relevantin just a minute.
It predated what we know of asmanaged services.
It certainly predated the cloud.
We got our start back then,just to give you an idea from a
time frame standpoint grippingout Novel Network and installing
Windows NT for businesses.
I remember vividly having tosell our clients on email so not
(04:59):
to pay to myself as someonethat's a million years old, but
you get the idea of when thiswas.
So one of our friends had abrilliant idea of starting a
fantasy football league.
We thought it was great becausewe love football.
Listen, I'm not going to say wewere like the guys sitting in
the back of a diner that createda rotisserie baseball back in
the late 70s or early 80s Ithink it was but we were fairly
(05:22):
early on in the fantasy footballdays and again, we're talking a
time before mainstream internetas you think of it today, no
worldwide web as you know it.
So all of this stuff was donemanually.
So basically there was 12 of usin the league, a bunch of guys
that all grew up together, wentto high school together, great
friends, 12 of us in the leagueand Evan and I were the
commissioners.
We ran the league.
(05:42):
We were responsible essentiallyfor managing the lineups and
the scoring and basically theway it would work back then was
on Sunday mornings, everyone inthe league would call me, leave
a message on my answeringmachine and leave their lineup,
and then Sunday night I woulddownload the box scores from
copy server, prodigy, whateverdial up service I was using, and
(06:05):
we would manually or withspreadsheets we'd compile the
scores and then Tuesday morningafter the Monday night football
game we would fax the finalscores out to all of our friends
.
And again, like I said, wedidn't have much business to do
at the time, so we certainly hada bunch of spare time.
So I also took it upon myselfto start writing newsletters
(06:26):
Really, quite frankly, it was away to make fun of each other.
But it was also good fun.
We called it the PigskinChronicles and it was a great
time.
I would write articles aboutthe individual games I would
write about.
I'd have some sort of aninterest piece, like talk about
a made-up story about someone'slife.
It was fun.
It sounds a little bit childishand probably nonsensical to you
(06:48):
now, but trust me, at the timeit was a good time.
So that was 1994.
Flash forward to about 1999,2000.
The internet as we know itstarted to come more into shape.
Obviouslycom was all the rageback then and a bunch of
websites to run fantasy footballleagues started to sprout up.
We chose CBS Sportsline forsome reason.
(07:11):
I don't remember if it was theonly one or if we liked it, if
it was free, if it was thecheapest, don't remember, but
it's what we chose.
We still use it to this day andyou could do all of everything
online.
You could do it electronically.
I didn't have to manuallycalculate scores, I didn't have
to listen to my answeringmachine to get lineups.
(07:32):
Everyone could handle itthemselves.
They could create their ownlineups.
And what would happen was, astime went on, cbs would send out
an email every Tuesday morningwith the results of the games,
the final scores, but also a bitof a narrative.
Now, at this point our businesshad started to grow so I didn't
(07:52):
have time to write the pigs inchronicles anymore, and I can
assure you CBS Sportsline, whohad probably hundreds of
thousands, if not millions, ofleagues by this point in the
early 2000s, certainly didn'thave staff writers that were
writing fantasy box scorestories about the junkies that
was my team playing thesledgehammers that was Evan's
(08:13):
team.
But every Tuesday we would geta newsletter from CBS that had a
fairly well-written narrativeabout the outcome of the game
and it would talk about coachingmistakes and sitting and
starting the wrong player andbasically just describe the
games and the outcomes of theweek, and then they would come
out with another one which is apreview for the next week, and I
was fascinated to find out howthey did it.
(08:34):
So I dug into it and it turnsout they were using a very early
form of generative AI.
So basically they fed detailsabout the scores of the game,
the rosters, the possiblestarters and sit and players
that could be set free agents,things like that.
They'd feed all that into themachine I'm oversimplifying it,
(08:55):
obviously and the machine wouldspit out a narrative that looked
like an article about theparticular game and, like I said
, you couldn't necessarily tellthat it was written by a
computer.
It was interesting.
As time went on you started tosense a little bit that was
written by a computer, but stillvery cool.
And, like I said, I was superfascinated with the technology
(09:17):
and I found out where it camefrom.
So a few years prior to that,reuters, who had to cover
financial reporting news, foundthemselves in a bit of a dilemma
.
So if you think about publiccompanies, they report earnings
quarterly, obviously, and bigcompanies like Apple, google,
microsoft, you know, caterpillar, you name it the large
(09:41):
companies have teams of analyststhat listen in on earnings
calls and they dig through thefilings and they look inside the
P&Ls and the balance sheet andthey write long narrative
stories about the performance ofthese companies.
That's all well and good forthese large companies, but if
you look at a small company thatmaybe trades over the counter,
(10:02):
they're still responsible forreporting their financials every
quarter and they're stillinterested to the stockholders
to read about those reportings.
So what Reuters did was theybasically fed the same type of
financial reporting data into alanguage model.
I'm not going to call it alarge language model because it
very much predated what we havetoday.
So they would feed these into alanguage model and they would
(10:26):
write a financial news storyabout some over the counter
stock.
The company reported earnings.
Again.
The stock was probably tradingat 12 cents.
So I have no idea who wasreading it probably the same
people that are going to listento my podcast but it was a great
solution for Reuters and thatkind of grew into, like I said,
(10:47):
the technology that was used byfantasy football and that was
really the precursor of what wesee as generative AI now and
again.
It fascinated me.
I've always been fascinatedwith technology, emerging
technology specifically.
I wouldn't say that I sat downand started to try and figure
out business uses of it.
It was the type of thing thatwas kind of in the back of my
(11:07):
mind.
I've told that story many times.
It's like I said, it's lit thespark around AI specifically for
me and when we started CrushBank initially back in 2017, one
of the reasons I was sofascinated with AI was kind of
where that came from.
So I'm going to fast forward nowand this is going to get much
more relevant in a minute.
So thank you for indulging meand learning a little bit of my
(11:29):
background.
I'm going to fast forward nowto a newspaper article from the
Columbus Dispatch.
It's a newspaper in Columbus,ohio, a date line August 18,
2023.
It is a article about a highschool football team game
between two rival high schoolsand I'm going to read it.
(11:49):
So please indulge me.
I'm not going to read theentire thing, but so indulge me
for a minute.
I'm going to read a little bitof it to you.
So the headline is WestervilleNorth escapes Westerville
Central in thin wind in Ohio,high School football action, and
the byline of that is lead AIL-E-D-E-A-I.
Obviously, lead AI.
(12:10):
Well, not, obviously, but leadAI is a service that newspapers,
news stations, newsrooms ingeneral any news reporting
service can subscribe to thatwill essentially quote, unquote,
replace journalists and writeAI articles for you Fairly well
known product.
I know a lot of newsrooms useit, so this is not just not
(12:35):
meant to be critical of them.
It's more of a conversationaround generative AI and some of
the things we should bethinking of when we embrace this
technology.
So that headline again Westescapes Westerville Central and
thin wind in Ohio not what yourtypical sports journalist would
write, obviously right.
I grew up reading Mike Lupica inthe New York Post, right.
(12:58):
I used to wake up early everymorning to get the post to read
Mike Lupica's article about theRanger game from the night
before.
When I was in college I was inMichigan and Mitch Albom was the
journalist of Note in Michiganand he wrote stories about the
fab five or Michigan's footballgate teams or things like that,
and we couldn't wait to readthem.
Anything Mitch Albom wrote wasgold.
(13:19):
This is not the type ofheadline that someone like a
Mitch Albom or Mike Lupica wouldwrite.
Obviously, journalism ingeneral has changed
significantly in the last 30 to35 years, but you can tell
reading that that it is not,that it is written by AI and to
the Columbus Dispatches creditthey.
The byline is the AI productthat they use to write this
(13:41):
article.
I'm gonna read the first coupleparagraphs, so please indulge me
for another few seconds.
The Westerville North Warriorsdefeated the Westerville Central
War Hawks 2112 in an Ohio highschool football game on Friday.
Westerville North edgedWesterville Central at 2112 in a
close encounter of the athletickind at Westerville North High
(14:02):
on August 18th In Ohio footballaction.
Those are the first twoparagraphs.
The first paragraph of a newsarticle, if you don't know, is
called the lead LED and the ideaof it is to grasp your
attention and make youinterested in the article.
Both of those paragraphs wouldqualify as a lead.
So in other words they're bothmeant to be first paragraph
(14:26):
Pieces or parts of a newspaperarticle.
So that's the first strikethere.
The second strike is it justsounds very awkward, right?
If you read it again.
It's not what a ambitiousJournalist, what journalists,
would write, especially when andI'm extrapolating out now here
but especially when you considerthat someone covering high
school football in Ohio Probablyhas dreams of covering the
(14:47):
Browns or the Bengals or evenOhio State, not Westerville
North versus Westerville Central.
But you need to start somewhereand you would assume they would
put their best foot forward.
So we'll kind of leave all ofthat for a minute.
I want to focus on one sentencethere.
I read it to you.
Hopefully you caught it,hopefully you picked up on it
when I was talking about it, butif not, just in case, I'm gonna
(15:09):
read it again Westerville Northedged Westerville Central 2112
in a close encounter of theathletic kind at Westerville
North High.
That sentence Jumps right outof the page at you.
That is absolutely notsomething that a sportscaster
would write.
Our sports journalist wouldwrite.
I should say again it's a I,it's meant to sound like a human
(15:31):
, some ones and zeros gotcrossed someplace.
So it thinks that that is aCompelling sentence in a sports
article.
It is not.
But let's go a step further.
So I find that sentence veryinteresting.
So I want you to play along athome with me now and I want you
to open up Google and I want youto type a close Encounter of
(15:51):
the athletic kind into yoursearch bar and as soon as you do
that, I want you to switch tothe news tab.
So, in other words, you'regetting news articles and you
will notice dozens and dozens ofarticles that come up and if
you click on any one of them,let's see.
I'm gonna go back.
I'm gonna go to the fourth pagehere I'm gonna go back to let's
(16:17):
see how about January 10th of2023?
I'm gonna click on the articleand I'm gonna read it for you.
Raymond Lincoln would edgeverdant North Mac 41 34 in a
close encounter of the athletickind on January 10th in Illinois
boys high school basketball.
(16:39):
And every one of them is thesame.
You can find that sentence noexaggeration in 50 articles
without the blink of an eye, andprobably even way more than
that.
And again, which is actuallypretty funny, if you were to go
to lead a eyes website, whatthey tout is original and
creative content that soundslike it was written by a real
(17:01):
person.
I'm paraphrasing there.
The original and creative partis what strikes out.
That's funny to me.
Now, again, I think that lastarticle I picked a random was
from like the Tennessee and orsomething, and the other one I
read you, like I mentioned, wasfrom the Columbus dispatch.
So certainly neither of themare gonna lose any readers over
relying on a generative AIengine that uses the same silly
(17:24):
sounding sentence over and overagain.
But we need to take a step backand we need to think about this
and how it affects usspecifically.
Again.
Managed services, it companiesin general, but really any
enterprise, any business that'srelying on this technology.
So people often ask me myopinion on generative AI and the
uses of it, and open AIspecifically, and some other
(17:47):
vendors and again, I'm very highon the technology.
I think the promise of it isincredible.
In a lot of ways, it's notready for prime time.
The biggest thing and we'll talkabout this probably on future
episodes, the biggest thing isthat it's not required to be
accurate.
So, in other words, if you ask,if you go to open AI or chat,
you can ask it to writesomething.
(18:08):
It'll write something thatsounds real and sounds realistic
but doesn't necessarily meanit's true and I have a bunch of
great stories that we'll shareabout that at some other time.
So that's the biggest problemin it.
So one of the things I oftenencourage people to use it for
is to create content fromsomething that you know is real.
So, in other words, feeding ininformation about the final
(18:30):
score of a high school footballgame and asking it to generate a
notepad article about it.
In that scenario, under thoseconstraints, it will generally
create an article that isaccurate and truthful and, again
, it will use the technology,will use the language that it
finds appropriate, that thinksit's interesting or that it that
is usable.
In that case, the reason I bringthis up and the reason I talk
(18:52):
about it is because one of themost common uses that I
recommend for things like thisis around sales and marketing
pieces.
So, in other words, you're asmall upstart MSP and you need
to do a marketing blitz or anemail campaign and you want to
get a bunch of new clients tosign up for some new managed
security plan that you're aboutto launch, but you just don't
have the expertise on writingmarketing pieces around managed
(19:15):
services or managed securityservices in this case.
Now you can go out and you canhire someone, or you can
contract someone, or you can useone of the multitude of great
partners there are on thechannel to do stuff like that,
but you might want to take acrack at it yourself.
So you might want to log on toChatGBT, you might want to
describe the offering and youmight want to ask it to write an
email for you and, truth betold, it will probably do a
(19:36):
pretty good job of that.
It will write a compelling,interesting, accurate piece for
an email for your website, for amarketing blast.
Whatever you're doing with itbased on those criteria, the
problem is you don't want theguy down the street that selling
and competing, offering to dothe same thing for you and have
(19:56):
the same language in there.
Now again, I'm not saying thatin every case this language is
going to come out the same, butwhen you're using the same
machine learning model or Ishould say the same large
language model in this case, togenerate similar content,
chances are those results aregoing to be pretty close to each
other.
Now you can put your own slanton it.
You can ask to write it likeyou're a 14-year-old skater or a
(20:18):
25-year-old gamer or a pirateliving in the 1800s, like
whatever you choose to slant toput on it.
You can get it to write in sortof a different voice or a
different method, but chancesare you're just going to ask it
to write as a businessprofessional or not.
Ask it all and it will implythat you sorry, it will infer,
rather, that you think it shouldbe written as a business
(20:40):
professional.
And that's the danger of a lotof this stuff.
And leaving it what we callunshaperone, leaving it alone,
is wrought for opportunity forthings to go horribly wrong
Again.
You don't want your website tohave the same content.
I remember early on when wewere starting to sell managed
services, there were a bunch ofmarketing firms that were
selling, offering kind ofpre-canned articles that you
(21:03):
could put on your website, orthey were writing content for
your website.
And this is years and years ago.
They've gotten way better.
But if you had at the time, ifyou had asked one of these
companies to generate contentfor your website and then you
had Googled it, you would findother managed service providers
that had the same exact contenton their site.
Or maybe you guys were usingthe same Microsoft articles or
(21:24):
the same HP articles or whoeverwas putting out content for the
SMB, for IT services, and we'rekind of at risk of that
happening again.
So the moral of the storyreally is I think it's kind of
fun and interesting and helpsyou understand sort of how this
technology works, where it camefrom.
But really the moral of thestory is that, no matter what
(21:44):
the use case, never let this AIwork alone, never let it work in
a vacuum, always have itchaperoned by some sort of
professional right.
So whether you're asking it towrite code for you or write a
bunch of scripts for you forsome sort of automation great,
because it will absolutelyshortcut the work and it will
(22:05):
get done much faster.
But you're absolutely out ofyour mind if you roll that stuff
out without having someone thatunderstands it, checking it and
rechecking it and running it ina sandbox and doing all the
sort of due diligence you shoulddo before that stuff gets
deployed.
By the same token, don't ask itto write an email, copy and
paste that email into MailChimpand send it out.
(22:26):
Read it, edit it, make it yourown.
And if you could start on thirdbase, so to speak, with the
amount of content you needwritten or the content you want,
in this case, that's stillsaving you a tremendous amount
of time.
Don't assume it's going to doall your work for you, or, quite
frankly, don't ask it to do allyour work for you, and that, I
think, is something to remember.
(22:47):
Just about AI in generalgenerative AI specifically, but
AI in general.
We like to say that AI will notreplace people.
People that use AI will replacepeople, and this is a perfect
example of that at play.
Here's this technology playaround with it, see what's out
there, use it to generatecontent, but never let it work
alone in a vacuum.
(23:08):
That would be my takeaway ofthe day, my big advice, my big
lesson learned that, and don'tever draft a starting
quarterback from the DallasCowboys for your fantasy
football team.
Two lessons of the day.
Thanks again for joining metoday.
My name is David Tan.
This has been the Crush Bank AIfor MSPs podcast.
Check me out on LinkedIn, checkus out at CrushBankcom and
(23:31):
enjoy the rest of your week.
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.