April 29, 2025 • 59 mins
Ted McRae - Certified Prompt Specialist will be joining Matt and Dan to overview strategies accounting professionals can utilize in their practice to incorporate AI.
QB Power Hour is a free, biweekly webinar series for accountants, ProAdvisors, CPAs, bookkeepers and QuickBooks consultants presented by Dan DeLong and Matthew Fulton who are very passionate about the industry, QuickBooks and apps that integrate with QuickBooks.
Earn CPE through Earmark: https://bit.ly/QBPHCPE
Watch or listen to all of the QB Power Hours at https://www.qbpowerhour.com/blog
Register for upcoming webinars at https://www.qbpowerhour.com/
00:00 Introduction and Welcome
01:06 AI in the Accounting Industry
06:06 Housekeeping and Session Details
15:45 Understanding Large Language Models (LLMs)
24:10 Practical Applications of AI in Accounting
32:46 The Cost of Private AI Environments
33:10 Handling Personal Identifiable Information
33:44 AI in Business: Practical Examples
34:21 Different Types of AI
35:24 Generative AI: Creating Content
37:17 Machine Learning: Continuous Improvement
38:23 The Future of AI: Risks and Opportunities
43:33 AI Hallucinations: Understanding the Concept
48:59 Effective Prompting Techniques
51:42 Upcoming AI Cohort and Final Thoughts
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Dan DeLong (00:32):
Welcome everybody to another QB Power Hour today.
Of course, clockwork.
As we go live, my, my machine,my other machine restart and me
if hear if you hear anything,they tear apart the RV park that
(00:53):
we're, so there's Bobcat heavymachinery behind me, so I
apologize for that becausethat's, I can't move, a
construction job.
Unfortunately, I don't have thatmuch full inmate in the park
here.
But Matthew, how are you today,
Matthew Fulton (01:12):
Dan?
I'm doing good.
But first real quickly, luckilyAI has done such a great job of
doing noise reduction.
We don't hear any of that stuff.
Oh, cool.
It's good seeing you recentlyout in Florida and I'm excited
for this conversation today,Ted.
I'm glad you're here with us.
We were pre-talk a little bitearlier and I think this is
(01:33):
gonna be a, fun episode forsure.
Dan DeLong (01:37):
Yeah, we you guys.
We always, this particular topicof AI in general has always has
a lot of unknowns and, I don'tknow, maybe fear, scared of the
machines type of thing.
(01:57):
And that's, what we wanna try tounpack here today because as Ted
was saying right before here,it's pretty amazing how little,
the adoption of AI in theaccounting industry has has
occurred.
Ted what, were you thinkingabout that?
Ted McRae (02:20):
I was just, for a.
Industry that has so muchtechnology.
I can't think of any industrybesides, actually, I can't think
of any industry that has so manydedicated apps and technology
and all this stuff.
(02:40):
Holy smokes there.
I, there's so much technology tolearn.
I cannot believe that, thatsomething like this is that what
people are afraid to, get into?
And it's so underutilized, evenif a lot of people get into it
and start using it, it's sounderutilized for what it can
(03:02):
do.
It just amazes me.
And I think a lot of it,obviously we all have our, real
lives that we go into, but itmight just be fear and that's
and I can tell a story laterabout my fear doing it and what
stopped me.
But I.
That's all we were talking aboutreally is just how underutilized
it is, I believe, in thisindustry.
Dan DeLong (03:24):
Yeah.
And the subtitle of thisparticular episode is because
the second best time to start istoday, like that when's the best
day to plant a tree was 20 yearsago, but when's the second best
time it's today?
So this is really what we'refocus on here today is about a
way to, to jump in to the to theend of the AI pool, without
(03:52):
losing your float, my name.
Matthew Fulton (04:13):
Hello everybody.
Matthew Fulton here with ParkwayBusiness Solutions and the
Facebook group QB CommunityLive.
Really, again you're talkingabout going backwards and
planting things, and I just, I'mgonna use a quick second to say,
I just imagine this AI image ofBill and Ted going into their
phone booth, going backwards intime and planting trees for all
(04:33):
of us would've been great, butyeah, that's me.
C Community Live ParkwayBusiness Solutions.
Dan DeLong (04:39):
Alright, and join me again, not with Maker code, but
but as a certified engineerspecialist today.
Ted McRae (04:48):
What did you call me yesterday?
I wanted, yeah, go ahead.
I wanted
Dan DeLong (04:50):
to call you a
Ted McRae (04:51):
ologist.
I'm a ologist, not aproctologist.
Thank you.
No, yeah.
And I'm a Did you find
Dan DeLong (05:00):
The the latex gloves or No?
Ted McRae (05:02):
No I forgot the latex gloves.
They're actually, they were alltoo small for my hands, no, and
I'm a self-proclaimed propellerhead.
Anything, technology, app-wise,everything.
I just completely love anythingthat's gonna build efficiencies,
and that's really how I've triedto build my career is how do we
(05:23):
build efficiencies for peoplethat, especially folks that are
trying to do this wholevalue-based pricing, right?
So efficiency is key in ourindustry.
The faster you can do somethingthe better it is, as long as
it's accurate, right?
Because speed is important, butspeed without accuracy is
horrible.
So how can we be more efficientbut still contain and keep that
(05:47):
accuracy?
Dan DeLong (05:49):
Yeah, there was one of the values that Intuit was, I
can't remember what the valuewas but it was work fast and and
they ended up changing thatvalue to work smart, work fast.
Because you can do dumb thingsfast.
04.29.25_-_behind_on_ai_a (06:08):
Yeah.
Dan DeLong (06:10):
And that's how AI is fitting into this picture.
It the speed of which thistechnology is coming in and
really disrupting a lot ofthings.
I think that's part of the fear,I think that of people feeling
(06:32):
like it's gonna leave thembehind, with, with ai.
So let's, before we get intomeat, let's do a little
housekeeping details about theQB Power Hour.
It's every other Tuesday at noonEastern.
It's eligible for CPE Maker Hub.
Used to be our CPE sponsor.
So appreciate what you did therebefore us Ted, but now we're,
(06:56):
using earmark for that.
So after the session, about fivedays later, there will be a.
Course to take over a earmark toget the CPE credit.
And I'm sure Ted is appreciativethat he doesn't have to do all
that.
It was
Ted McRae (07:11):
a Dan, it was a good transition period.
Yeah, it was a good, it was goodhelping you out in the
transition period, but yes, itwas a very a lot of work to
create 88 CPE certificates andmaking sure that everyone was
on, was there on time and stuff.
Exactly.
Dan DeLong (07:31):
We do have an earmarked channel for QB Power
Hour.
You can always check the, linkout there.
If you have specific questionsabout the topic that we're being
talking here today, please putthem in the q and a.
You should see a little q and abutton at the bottom of the
webinar.
That helps us keep track ofthose questions because as all
(07:52):
of the AI bots that are joiningus are announcing that in the
comments.
Those comments just seem to goby and scroll by really quickly.
So appreciate if you havespecific questions, please put
them in the q and a so we canmake sure, we can either address
(08:12):
them live or follow up.
And but if you have just generalcomments, please put them in the
chat.
And then we also have the linksfor the handouts, which I was
about to socialize, but again,my computer restarted.
Hopefully we've got those in thesocial media, live stream as
well.
We also do have a QB power hourstore.
(08:34):
So you two can have a pink hatlike me.
I need to order one all.
So we're gonna start off withour first polling question as a
setting the stage and settingthe, The foundation of what
we're gonna talk about here as aself-assessment.
What is your comfort level withai?
Anything from novice where itscares me to, I'm an AI ninja
(08:59):
fully embracing this technologyand incorporating the advanced
features.
And Ted can you talk about yourjourney with, ai?
Like where how you, became aologist.
Ted McRae (09:12):
A ologist?
Yeah.
So I started, you know what thefunny thing is, so I was hearing
about this AI thing, and thiswas way back when, a few years
ago, when AI actually was onlyup until, I think it was
September of 2021.
So anything new we were in 2022or the late ends of 2021 when I
(09:35):
started using it.
So anything new I.
It wouldn't, it had no idea.
You couldn't ask any who won theWorld Series or what are the a
You can't do anything like that.
And so when I first started out,and probably there's a lot of
people like novice, it scaresme.
A lot of people or myself inparticular thought there was a
(09:59):
specific like magic to askingthe prompts.
So I searched all over theinternet when I first started,
using chat GPT for chat, GPTprompts what are the, how do I
do prompts, how do I do this?
And what I soon found out aftersearching for weeks and weeks is
(10:20):
that there are no speci anyonecan give you, oh, here are 10
prompts that you should use,everyone should use in
accounting.
But realistically I.
That's not really what youshould be doing in chat GPT.
And there's really no way,there's no specific here's
exactly how you should do it.
(10:41):
There's a framework.
And so that's what I didn'treally realize until I became a,
ologist, is that it's dirtywork.
No, I'm kidding.
It there, it's not about
Dan DeLong (10:57):
digital.
Ted McRae (10:58):
Yeah, it's digital work.
It's really about where youwanna end up what you, want.
And you gotta think of it as ifI were to ask Matthew Hey, can
you help me create a webinarthat I wanna do next month?
I wouldn't just tell Matt that'cause he would be like, yeah,
(11:19):
sure here, and he would justtell me everything that he
thinks that I should do.
But if, I tell him, Hey Matt,I'm doing a webinar next month
for a specific type of personand it's at this conference.
Here's all the research on theconference.
And then I do even more thingslike that.
Now Matt has a framework towork, or Matthew has a framework
(11:41):
to work with where we can createawesome content for that.
But if I don't give him all thatbasic stuff, like having a
conversation with him, he'snever gonna know.
So that's really how I startedout with it way, way back when.
And I've kept up with it and Ihate to say it, but I use it
(12:01):
probably 50, 60 times a day.
Dan DeLong (12:06):
Yeah.
It's now it's it's part of your,almost your everyday thing.
The whole idea of the.
Dyslexia where, you know, if youdon't have your phone to do
basic math skills, you justYeah.
Purge that out.
But I've got a calculator on myphone.
So we're gonna be doing Ted andI are gonna be doing a four part
(12:29):
weekly cohort series over aschool bookkeeping where we're
gonna be talking and unpacking alot of these things from the Mr.
Rogers level of ai.
And we'll, we maybe go up toSesame Street or maybe electric
company if we're, lucky.
Matthew Fulton (12:48):
I'm sorry, Mr.
Rogers is going to, the ologistis gonna be the first episode of
this.
Yeah,
Dan DeLong (12:53):
the French.
He'll be happy taking off morethan his shoes for that.
But what we're gonna be doingthere is we're gonna be talking,
and today we're gonna be talkingabout AI and large language
model statistics.
We'll just really keep this, asa fireside chat type of thing
where we'll talk about whatwe're, gonna be unpacking in the
(13:16):
cohort series which will becalled prompts for practices,
crafting effective prompt foraccountants where to go to
register.
And then we'll have a q and asession about that.
So let's let's talk about, oh,Noah, did you share the results
of the first poll?
Because I thought that wasreally oh.
(13:38):
Did I Just, so it looks likeabout 60, almost 70% are in the
novice, the basic level of, ofthe com comfortability with with
ai.
So let's talk a little bitabout, the, impact of, AI in, in
small business as far as anunderstanding of how businesses
(14:01):
are using AI and the largelanguage models.
I really like this this slidethat you created here, Ted,
about.
'cause that's part of the thingfor me was what is clawed, what
is all of these other, theGemini or Bard, I think they
called it originally and nowit's got a new name to it.
(14:23):
It's tr trying to keep all theplayers straight.
But talk a little bit about thisthis Ted
Ted McRae (14:32):
Yeah, so there's a few things here that really
strike me as, especially in thisindustry, chat, EPT is like the
Intuit in cases of this whole ailarge language models.
They were pretty much the first,the Kleenex, right?
(14:57):
Yeah.
They're the Kleenex.
But the thing is they are, in myopinion the, most reliable, I
never stray.
Personally for me, I never strayfrom chat GPT except for
perplexity.
And so there's different typesof AI that people need to to,
(15:21):
actually keep in mind.
There's, AI that can generatespeech for you.
11, 11 labs, which I can give ita, I can give it a script and I
can either choose my, the audioperson or I can upload snippets
of myself and create audio.
There's like Gemini, there'sMicrosoft.
(15:43):
So it all depends on what you'redoing it for.
So like Google Gemini, if youput it into the Google suite of
tools works perfect if that'swhat you're gonna use it for.
You can also attach chat, GPT,to, to Google as well.
But it, so it all really dependson what you're using it for,
but.
(16:04):
The thing is really quickly hereis that some of all these other
ones while you might like to use'em and they're fine, they may
not have the same technology aslike just a regular old chat
GPT.
So it really matters what youwanna use it for.
Matthew Fulton (16:18):
Ted, can I just real quickly have you step back
just a little bit and can yougive us a quick definition of
what a large language model is?
What is an LLM and why is thatimportant in what we're talking
about?
Ted McRae (16:31):
Yeah so a large language model takes context
from a person, right?
And then will spit out what itan answer a real person would.
So it takes your question, spitsit out and, contextualizes it.
(16:54):
Like, a per, like you would in aregular conversation.
Large language models are hugedata sets.
And I could really talk aboutlike the prompts, what it does.
But Matthew, what it really doesis it takes your sentence and
then it breaks it out intolittle chunks.
All right.
So it's like for instance, thebill is due on the on May 1st.
(17:21):
So we'll take that and itactually chunks that out, the
bill and then it says is due,and then it has the date.
So that, so it separates allthat stuff out.
Yeah.
So it understands.
Then puts it back together so itcan understand what you're
talking about, the date,actually referring to a date in
(17:41):
the future that this document.
So almost like Matthew, do youremember when we were, not you
and I together, but when youwere in grade school and you had
to cut a sentence apart and findout where, what is the
adjective, what is the noun,what is the verb?
And all of that stuff to, toshow the context of a sentence.
So large language models do theexact same thing, but in a more
(18:04):
granular process.
So they, they chunk that out andright now I have old man brain,
so I can't remember the name ofthe, of what it's called.
Oh, tokens.
They chunk it out into tokens.
Okay.
And then each token, theyanalyze that token.
And so what large languagemodels have, it's just a huge
data set, and now it goes and itreturns it to you like a normal
(18:26):
person would.
Matthew Fulton (18:27):
So another way to kinda say it'cause Grace
asked a good question earlier ofwhat exactly can AI do in the ac
in the accounting to assist inour work?
That's obviously a hugequestion, but a quick answer
way.
I feel when we're talking aboutLLMs large language models,
while we the, word language isin there, it's, it is, it's
(18:48):
truly large data sets.
So shameless plug here, maker'sHub has taken the ability to,
when you scan in a whole bunchof different documents in
different formats, there's thislarge database to try to
understand where valid orimportant information is on a
sheet Yep.
Can extract it and thenunderstand the words that are on
there to place it appropriatelyinto the right fields for your
(19:10):
accounting platform.
Same thing, where there's a lotof companies working on like
check images now, trying to beable to pull that type of stuff.
So the more data you have, themore it can reference to try to
give you a better result.
Is the intent of it all, is thata good way to.
Summarize it.
Yeah.
Ted McRae (19:27):
Yeah.
And like the really cool thingthat you can do with like chat
GPT now and all the other onesis, and this is what I did, I
created an accountant a ai, soit's just a, all he is his, name
is Milo.
And Milo knows a lot aboutaccounting because every little
(19:49):
thing that I see on LinkedIn,like people will do infographics
and stuff like that, I just copyand paste it into Milo.
So like, how to analyze thespread, how to analyze a balance
sheet versus a cash flowsummary, how to ba versus a, a p
(20:09):
and l and all of these otherthings, income statement.
And so I just upload as much asinformation on accounting that I
could find into Milo.
I also have him look at ddifferent websites.
Then now I can do things in a,in accounting that maybe we
couldn't do before, but I don'tuse that as the true source.
(20:32):
So people will be like, I don'ttrust it.
You shouldn't, that's the point.
But it helps you get to whereyou want to be faster.
Dan DeLong (20:41):
No there's a lot of words here about the statistics
usage and adoption.
Any, anything you wannaencapsulate, take away of all
this?
Of course you can download theslides and take a closer look at
some of these statistics,
Ted McRae (20:56):
Ted.
Yeah.
I think that what people need torealize on this is that the
businesses don't think of it as,it's gonna take my job away.
Think of it, especially like Isaid, being the propeller head
that I am, what can we in thisindustry use it for?
(21:21):
And we all know that there is a,there is a, gap between like how
many hours we can work and howmuch we, especially if you're if
you ba bill by the hour, you canonly bill so many hours.
But if you're, but if you'revalue-based, the faster we can
(21:43):
do things, the more hours we canbill.
So if I can take a, if I cantake a balance sheet and ha have
the first pass of analyzing it,go through a AI before I look at
it, it's almost like Dan, likewhat you say, having an intern
look through my stuff.
I wouldn't just send my internsreport to my client, but the
(22:06):
intern's gonna look in and lookfor anomalies for me and things
that are like red flags.
Then I can review them with thedata, with the mindset of an
accountant to put my spin on it.
And then I could even put it ina and say, okay, chat GPT create
the the summary report for myclient for this.
(22:27):
But that doesn't mean I'm gonnasend that straight to my client.
I still need to review.
Now I take that summary reportand build my report off of it.
Yeah, but how much time doesthat save us?
It's astronomical when you useit as just the first set of eyes
and create, instead of staringat a blank piece of paper, I
always tell Chad PT to writedown a summary of my ideas.
(22:52):
And then I don't start with ablank piece of paper.
I just save myself three hoursprobably.
And then the other thing, 49% ofcompanies are currently using it
and 30% intend to use it in thefuture.
So for all of you, I was amazedat how many people said it
scares me, is what'd you say?
60% of people said they're inthe first two buckets.
04.29.25_-_behind_on_ai_a (23:11):
Yeah.
Ted McRae (23:12):
We're like five years almost into chat.
GPT artificial intelligenceright now in large language
models.
It's time, like you said, Dan,it's time to start thinking
about it.
it's,
Dan DeLong (23:25):
It's the new disruptor of, the day.
You look back at all thedifferent things that dis
disrupted the accountingindustry.
Cloud accounting is the thelatest up until this point,
Excel, right?
You can go back to all of thesethings where you think about,
(23:46):
I'm sorry.
Oh my God, it's gonna change.
It's gonna change the it's gonnatake my job, right?
But it's not, it's just anothertool that can assist and to your
point, provide you more leverageto do the things that you wanna
do.
Faster, right?
Think about
Ted McRae (24:04):
it.
They're in Dan, they talk about'EM as ERA technology eras,
right?
Yeah.
So we had the, really, the firstera was the www era where we
connected everyone online.
And that was a game changer forall of us.
And then there was the, thenthere was the cloud era where we
actually started sharing datawith each other.
Then there was the mobile erawhere now we're connected to our
(24:28):
mobile phones, and now we're inthe AI first era.
And so this is going to, it'snot going away, and this is
where it's really gonna be.
Yeah.
Thank you Matthew.
You're right.
Dan DeLong (24:41):
Thank.
So this one what, accountantsare, using AI for, right?
You've got the basic level whereit's creating some content.
So starting you off with thatthree hours of savings, like you
were just talking about, insteadof staring at a blank document.
Get it to start with that roughdraft and then the intermediate
(25:03):
level.
So you wanna talk us throughthis these, different levels?
Ted McRae (25:08):
Yeah.
So if you think about it at thebasic level, and it seems like
probably 70% I don't mean to, ifit's less, that's fine, but it
seems like 70% of the people onhere are at the very basic level
using it to or look at an email.
Is it right?
(25:28):
I want to create an email forthis and that, or I need to
create a blog.
Which people are probably notusing it the right way, even if
they're using it in the basiclevel.
Now the, intermediate level,we're allowing large language
models or AI chat, GPT, whateveryou wanted to.
Whatever you want to call it.
(25:49):
We're allowing it to take it up.
'cause we're starting to trustit more.
So we're allowing it to take usto a different step where,
alright, I want to create maybeDema, generate a demand letter,
but I'm still not using it toits full potential.
Or I'm generating a letter to aclient that I need to fire or
(26:11):
I'm maybe creating small easySOPs, little things like that.
So that's more of anintermediate level.
And then maybe you're going inand if you know what A GPT is,
we're gonna talk about itextensively in in our school of
bookkeeping training that we'redoing, but creating GPTs or
using it to ask questions.
(26:34):
About a product.
How do I create a formula, anExcel formula that sums all of
this really cool stuff and itjust spews it out for us and
shows us how to do it?
That's pretty intermediate,right?
Or how do I create a button inHTML?
That's pretty, easy.
But you can say, how do I createa button that does, that, goes
(26:55):
to this specific link with thesecolors, and when I hover over
it, it changes colors and whenmy mouse clicks on it, it
shakes.
So you could have it do morethan just create a button in h
TM l you can have it do insingle thing and create the
webpage for you that the buttongoes on.
So, intermediate is just reallybasic.
(27:19):
A tiny bit above the basic.
We're getting into knowing howto create somewhat crafty
prompts.
The advanced level we've createdMilo, like I have.
Who is a, an accountant in mylittle chat, GPT world.
And Milo knows a lot aboutaccounting, on a pretty awesome
(27:43):
high level because I start, Ifeed in every single news thing
that I get about it.
I have him look at websites allsorts of stuff.
So anything I want to ask Miloor analyze or Milo, what, you
know what it's, today is the endof April.
You're as an accountant.
What are you doing today?
(28:04):
And Milo will tell me what theaverage accountant is doing.
So now me as a marketer knowswhat to actually talk to the
accountants about.
So now we're getting more at anadvanced level, right?
Milo can analyze reports.
I can have Milo generate demandletters.
I can have Milo actuallygenerate them based on a level
(28:26):
of friendliness, aggressiveness,and how late it is.
Milo, I can say Milo generateddemand letter from my client
with an aggressiveness of seven'cause he already knows of one
to out of seven, a friendlinessof two, and the, it was due on
this date.
(28:46):
And here's, and I upload theinvoice and he'll generate a
demand letter that is more thanjust you need to pay this is
like the third or fourth time.
And then you can also do somepretty advanced.
Or semi advanced coding at thatlevel where maybe you're even,
(29:09):
you're doing more than J you'reextracting databases and you're
doing really cool things likethat because you can ha you can
do JavaScripts, you can do allsorts of stuff within within
chat GPT.
And in fact, my daughter's a asoftware engineer and they are
not allowed, they have a programthat scans their they use
(29:34):
artificial intelligence to scantheir backend code for minute
errors.
They're not allowed to use largelanguage models to create code,
but they use it to scan the codefor errors.
And then at the ninja levelyou're really, you're using it
to.
(29:54):
Think about it this way, if youcould create a GPT and we, like
I said, we'll talk about it, butA GPT is memorized information
that you put in a little in a,little area of chat.
GPT.
Think about this.
If I have a GPT where I canupload all of my SOPs, I can
upload all of my company, likehiring information and all that
(30:15):
stuff.
So when I have a new employeecome on board that has the
questions on standard operatingprocedures, they can just go
into my GPT that I did.
And I said, so if I, if we havea client that refuses to give us
their bank account informationfor QuickBooks Online, what do I
(30:36):
do?
And it just spits out all thatinformation that employees just
saved.
A a lot of time having to go andask people around the office.
What they should do in specificareas of the business.
So now I'm like, all right, howdo I do X, Y, and z?
And the chat GPT just comes outand tells me how to do it based
(30:59):
on my company.
Dan DeLong (31:02):
Yeah.
Good question here in thecomments how secure and I think
that was part of what you werealluding to, where you don't
want what you are training it onto be training everybody else.
Which gives that specialness, Iguess to Milo over sharing it
(31:25):
with, with everyone in the
Ted McRae (31:29):
world, right?
Yep.
Yep.
'cause Milo is basically anextension of what Ted wants.
So we, if we each had our ownMilos, it would have, its, it
would have its owncharacteristic of the company
that we work for.
I can have Milo be Ja, Jasonstats.
And react to me like Jason Statsand I actually have a Jason Stat
(31:51):
bot if I ever want to create avideo that's halfway funny I put
it through my Jason Stats scriptbuilder and it builds that
script out.
So I can have I can have chatGPT review, YouTube videos or,
(32:11):
things like that.
I actually had chat.
GPTI have a Ron Baker chat, GPTfor value-based pricing, believe
it or not.
So it's, there's no end to whatyou can do.
Dan DeLong (32:24):
And how does someone keep that from being leaked?
Is be like how do you,
Ted McRae (32:34):
your chat, GPTs, you can either share publicly, or
keep them private to yourself.
Now if somebody gets it I reallydon't care because I don't have
a, I don't have any, I don't putsocial security numbers in there
and, stuff like that, right?
Which I still I still wouldnever do, I wouldn't upload a
(32:54):
report and say, analyze all ofmy employees, give me a social
security list alpha ornumerically from smallest to
large.
I wouldn't do anything likethat.
I would think of chat.
GPT.
If you were in a doctor's officeand had to follow HIPAA rules,
would you put, what would youput into a computer system
outside the doctor's office?
(33:15):
I.
Dan DeLong (33:17):
It's a good good rule of thumb.
Matthew Fulton (33:19):
There are certain service providers out
there that allow you to spinthese up into your own closed
environment, but it becomes likethe cost of it is exponentially
higher.
So that would be something you'dresearch.
Once you've really developedsomething and you've got a tool
(33:39):
if you really needed thatprivate information there.
The key thing really is personalidentifiable information.
You don't want to specific, youdon't wanna have that in there.
You don't want the liability ofthat would be the key, I'd say.
Ted McRae (33:52):
Yeah, I would, I I.
I would think of it, Matthew,just like any other thing, are
you gonna, are you gonna putlike your employee, or not your
employees, but all your client'sdata up in an unsecured Google
Drive that maybe has a sharedlink to somebody that you don't
like, a deep folder that mightbe shared with somebody.
(34:14):
You wouldn't do it.
And I wouldn't do it on hereeither.
Agree.
But the thing is realisticallywe shouldn't even be using it
for that.
What can we analyze withpeople's personal data?
Like a company, I can understanda company, alright, I'm
uploading a balance sheet for acompany.
Okay.
It's just nu it's really justnumbers, right?
(34:34):
It's not like I can take thatnow and infiltrate the company
and pretend like I'm the CEO andaccess his credit card
information.
For the most part.
That's not what you're uploadinganyway.
Matthew Fulton (34:48):
Good point.
Dan DeLong (34:51):
When I asked AI.
How many different types of AIare there?
It spit out, I think 16different ones, of how AI is
used.
So we want to just talk aboutthe different types of ai.
So we've got traditionalgenerative and, machine
learning.
(35:12):
So Ted, how would you, qualifyor quantify the different type
of AI technologies?
Ted McRae (35:21):
There's traditional, which is like the chat GPTs and
the copilots and things of thatnature.
They're just like let's talkabout it this way.
Traditional AI was like, freechat, GPT.
There's, they have limitedscope.
It doesn't learn.
So that's just like whattraditional back in the day was.
(35:43):
Now, if we talk about, and thenyou can talk about traditional
AI is almost like chat GPT rightnow.
But again, without thesubscription, right?
Because it's not gonna learnfrom what you do.
Then there's generative ai.
And generative AI is, I use itall the time.
I can you and I just do thiswith chat.
(36:05):
CPTI need an icon for, so myMatthew, you know the arch out
on the Channel Islands, right?
That, yep.
The little, okay.
So I wanted to create an iconfor the arch, okay.
For Ventura.
And so I just uploaded a pictureand I said, here's the company I
(36:29):
want to generate this icon for.
And I want the theme to becoastal, so use coastal colors.
And so it generated a logo, andI shouldn't say icon, a logo.
I.
For that.
So I uploaded just a regularreal picture.
So generative AI createscontent, creates images it can
(36:54):
create all all sorts of stufffor, you.
So generative, you give it anidea and it's really creating
like an image for, you or I playthe guitar.
I can have it create a song forme based on an, accountant.
I actually, one, one time I wasdoing a a webinar for AI and I
(37:18):
had it create a song foraccountants around April 15th.
About how, about like how crappytheir clients are when they get
right around that time askingfor things or not giving them
the right paperwork that theywant and stuff like that.
And it wrote out the song, itgo, gave me all the chords.
(37:40):
It told me how to strum it andeverything.
And I sang that song and I sangthat and played that song on one
of my webinars.
Dan DeLong (37:49):
And then the machine learning and then machine
learning.
Yeah,
Ted McRae (37:53):
data.
Data.
It's like my milo, right?
My Milo learns.
I have to tell it to learn.
But my milo also learns when Ijust type things into him as
well.
He just keeps learning.
I can hard chord, hard codelearning into Milo, or as I'm
(38:14):
typing, Milo learns as well.
Dan DeLong (38:16):
So this would be just to put it in a QuickBooks,
speak.
This would be like the bankfeeds as they are trying to
homogenize all thesetransactions.
And it will start to recommend.
The things that you wannacategorize it to, or something
in like Maker Tub where you putin a, purchase order or a
(38:38):
document and it's different forone over the other.
Where it starts to learn thesales
Ted McRae (38:45):
tax.
Sales tax should go to the salestax chart of account.
And not just not just to cost ofgoods sold or whatever you wanna
put it.
To put it to.
Yeah.
Dan DeLong (38:56):
Which I think this leads to a question that was
posted in here.
What is the possibility of AIbasically becoming self-aware
very or, becoming corruptedbased on the things that it
learns?
Matthew Fulton (39:11):
Skynet is real.
Yeah.
Ted McRae (39:14):
Yeah.
You know what's the funny thingis I was listening to a podcast
and they were talking about.
These guys that are doing thelarge learning models, didn't
they watch all the movies wewatched growing?
I know, right?
Come on.
Literally out whole lot ofmovies saying this is a bad
(39:37):
idea.
Yeah, and guess what?
They know it too.
They're like, yeah, we know thatthis is a possibility, but
they're, but obviously if it'swe're in a race, we're in an AI
race right now, is it going tobe the United States?
Is it gonna be China?
Who's it gonna be?
That is, is the AI dominantcountry and it really matters.
(40:00):
'cause ai, if you think aboutai, what we do just in
accounting, think about whatwe're gonna use AI for in
defense.
In the future, what we're gonnause AI for in all sorts of like
negotiations and stuff likethat.
So
Matthew Fulton (40:16):
yesterday I was listening to a podcast as well.
And then they were specificallytalking about ai.
They had mentioned that as anexample, having AI research, AI
itself and how to do certainthings.
This kinda leads us towards thedifferent levels or layers of
where it could become a problem.
(40:36):
But currently.
AI can complete a human taskthat's, say, that's like a, one
hour task.
It can complete it faster with a50% accuracy.
And the imp it is improving anddoubling in time, like how much
it can do every four months.
So when they talk aboutartificial intelligence, they
also talk about human artificialintelligence.
(41:01):
In other words, artificialintelligence becoming as
intelligent as humans insteadof, I'm not gonna get the term
right, but like superintelligent the, next level
above.
And they compared it to atoddler versus an adult.
So a toddler trying to teach anadult something is not gonna
happen very easily.
(41:21):
The adult can even placate thetoddler and make'em, oh, you're
so cute.
You're teaching us suchwonderful things.
I.
So we have to be, the concern iswhen we go beyond human
artificial intelligence, are wegonna actually understand what
it's truly capable of or not?
And is it gonna go Skynet?
(41:44):
Nobody really knows, but wegotta, I don't think, I don't
think
Ted McRae (41:46):
we'll have to, I don't think in our lifetime
we're gonna have to worry aboutit.
But it's, the fastest it'sactually increasing in knowledge
is amazing.
I'm amazed every time I log in,what I can do with chat GPT
every fricking time I log in andI do something.
(42:08):
Not because it's actuallychanged, but because I just
thought of a different way ofusing it and I'm like, holy
smokes that.
And then like, when you can dothings like copy and paste
images and it can read it.
Then store that information.
That's why I said, every time Isee anything about accounting,
infographic, whatever, as longas I, as long as it comes from a
source that I trust, I just copyand paste it into Milo and have
(42:31):
Milo learn it.
Matthew Fulton (42:33):
Do you ever talk to yours?
Little tip I like to use is onmy chat, GPT, you can hit the
side and ask it.
Can you explain how AI can beused in accounting today?
AI (42:45):
Absolutely.
AI can be a game changer inaccounting.
It can automate repetitive taskslike data entry, invoice
processing, and reconciliations,which saves time and reduces
human error.
It can also analyze largevolumes of data quickly, helping
identify trends, anomalies, andpotential risks.
AI powered tools can assist inforecasting, budgeting, and
(43:08):
financial planning by providingmore accurate predictions based
on historical data.
Matthew Fulton (43:14):
Thank you.
Plus, sorry to interrupt you,but can you tell me, does that
mean AI accountants are gonnatake over the world someday in
the future?
AI (43:22):
No worries.
It's unlikely that AIaccountants will take over the
world.
While AI can handle many routineand data.
Ted McRae (43:29):
Yeah, it it's pretty, of course they would say
Dan DeLong (43:33):
no.
Ted McRae (43:35):
Hey so Linda Russell, how you doing?
I, Hey,
Matthew Fulton (43:39):
Linda.
Yeah.
Ted McRae (43:40):
Talked to you a few times before.
So try using perplexity.ai.
So Linda says, I'm using chat,GPT as my search engine, pro tip
perplexity.ai.
It was designed to be an AIsearch engine, and we'll give
you better results than chat.
(44:01):
GPT, just ex FYI.
Dan DeLong (44:05):
Yeah.
And, I want to talk about thiswhole concept of AI
hallucinations, which was reallylike the catalyst of this, talk
that we're having, this coursethat we're creating because, we,
we've all seen like the news ofhey, AI took the bar exam.
(44:27):
Or AI can't, couldn't do math orcouldn't do simple math.
Where AI hallucinates the truth.
Sometimes like it could havejust hallucinated that response
that, you said when you, askedit, is it gonna take over the
world?
Maybe that's what you think thatyou want to hear.
(44:51):
And this whole idea of ofhallucinations isn't necessarily
a flaw but more of a feature.
Ted can you talk a little bitabout the, concept of AI
hallucinating and just makingstuff up.
Ted McRae (45:09):
Yeah.
What ai hallucinating nowadaysis more of, I don't know if
you've ever heard of, pep Cap.
Problem exists between keyboardand chair.
So honestly, AI hallucinationsprior to 4.0, chat GPT was
(45:34):
because it just didn't have thedata right?
And it just tried to make upwhatever it is.
Now, realistically, it's becauseyou like what I said to you
guys?
Hey, let's before we got onhere, I was telling Dan and
Matthew, if I told them, Hey,let's all create a blog and we
all typed into chat GPT create ablog about accounting and chat
(46:01):
GPT, it would might, if we justsaid that it might give us all
the exact same, roughly the sameinformation.
And it might not even know whatwe're talking about.
Is it talking about like accountchat, GPT taking over
accounting?
Is it talk?
So AI hallucinations are a notso well crafted prompt now,
(46:29):
except for like it used to be.
It used to be like you would askchat GPT how many Rs are in
strawberry?
And it would tell you two, itdidn't matter how many times you
told it.
There were three, it would tellyou two Mine at least now, and I
can only go off of mine, saysthat there are three.
So it is starting to learn, butgo.
(46:49):
But it is the, in most cases nowit's the user.
Dan DeLong (46:54):
Yeah.
And it's this George Costanza Ghere about it's not you, it's
me.
It's if you're not getting theoutputs that you want.
You really need to change theinput that you're giving it.
Matthew Fulton (47:09):
I'm gonna disagree in one area
specifically, which is whenyou're creating an image and you
want words in there.
04.29.25_-_behind_on_ai_adop (47:14):
Oh yeah.
Matthew Fulton (47:15):
Like you can type out exactly what you want
it to say and it will stillalways spell it wrong every
time.
Ted McRae (47:25):
I never, I always, but Matthew, you can correct the
chat, g you can correct it tocreate the actual words in the
image.
If you tell it to focus on thewords in the image.
Matthew Fulton (47:40):
I years sounds like it's got a better
education.
Mine, I think kindergarten orsomething.
So maybe I don't know how tospell it.
Could be.
So it's probably my prompting.
Dan DeLong (47:49):
Yeah.
Like I, I saw a post wheresomebody had searched Google
just did a Google search for BSin the computer.
And Gemini, the AI overview thatcame up actually said.
There is, there are bees in thecomputer that have been there
since the year 2000.
(48:10):
And, of course that's the,poster child for, okay, this AI
doesn't work at all because hereit's saying that there are bees
inside of your computer.
04.29.25_-_behind_on_ai_ad (48:23):
Like honeybees.
Dan DeLong (48:25):
Yeah, like real bee, like buzzy bees.
You can google it, you cangoogle it now.
And the AI overview does not saythat.
04.29.25_-_behind_on_ai_a (48:31):
Okay.
Dan DeLong (48:31):
But, there what it was doing is it was searching
the internet and it found thiscompany that created a blog
article about such, things.
And it was an it, an fool joke,right?
This one things where.
(48:52):
It found that article andtherefore provided you that as
truth.
And that's something that AIcan't do yet, right?
Like you cannot measure thevalue of truth, even though it
found an answer for you.
Yeah, there were bees in thecomputer.
But in this case it was totallya joke and it can't figure out
(49:15):
sarcasm or truth or things yetuntil you, unless you tell it.
Now, apparently somebody toldthe AI overview that is not a
true thing and it doesn't prooffer you that as true.
That's funny.
So let's talk a little bit aboutsome prompt examples, right?
(49:36):
Where they're ineffective andmaybe effective, right?
So in this example, make areport about accounting, how
that is an ineffective prompt.
Ted McRae (49:49):
Yeah.
So make a, this is almost likethe, what we talked about the
blog, right?
We all, it's very vague.
You have no idea who am I makingthe report for?
What, is the report?
Am I talking about accounting oram I talking about numbers?
From from something that youwant to give to me.
(50:09):
So you're there, it lacks anycontext whatsoever.
And can I just share somethingreally quickly?
I wanted to show you guyssomething.
Matthew, this will show you,allow multiple, I'm just gonna
replace the current share.
Sure.
Is that okay?
Yep.
04.29.25_-_behind_on_ai_adopt (50:27):
I stop sharing.
Ted McRae (50:28):
So look at this.
I said can you create an imageof the bookkeeping department
with QuickBooks online on thecomputer?
And look at what it created forme.
So accounting, a bookkeepingdepartment actually spelt it ex.
So look at how good it's gettingnow.
QuickBooks, it's spelt wrong.
I could, redo it now and I couldsay you spelled QuickBooks wrong
(50:51):
up here, and, but it like small,like it says profit on loss, but
it's still pretty darn good now,right?
Holy smokes.
It's, really, crazy.
Go ahead and share again, guys.
I'm sorry.
Matthew Fulton (51:07):
That's where, so I'm sure I've, other people seen
these reels, right?
Where they'll be like, show us aunicorn with great, beautiful
wings and make it magical.
No more magical.
Make it extremely magical andthey just keep building on top
of it, on top of it, on top ofit, on top of it until it's just
absurd.
That's some fun stuff you to do.
Dan DeLong (51:26):
Yeah.
So when here we're talkingabout, being more effective in
your prompt, create a one pagesummary report on the latest
accounting trends for smallbusinesses focusing on
automation and AI tools.
So just by adding more.
Context to the input, you'regoing to get a, more specific,
(51:48):
answer from the from the output,right?
And then also
Ted McRae (51:53):
Dan, we were talking about basic and intermediate
levels.
That's more, that's anintermediate prompt right there,
right?
That's not even an advanceprompt because there's still,
it's contextually wrong stillbut it's better than the other
one, honestly.
Yeah.
And that's what we're gonna talkabout in our little prompt thing
(52:13):
as well.
Dan DeLong (52:14):
Yeah.
So in the month of July Ted andI are gonna be doing a four,
four week cohort where each weekwe'll break down four different
modules going from basic what isit to intermediate advance and
drafting prompts.
And each week we'll also havesome homework.
(52:35):
So because it's at the school.
We'll have some homework, someassignments.
There will be a cohort I can'tsay it, cohort only community
where you can, put in questionsand we can talk about them in a,
weekly follow up and q and asession.
We can do a group discussionabout the prompts that are being
(52:57):
created and the outputs.
And Ed will be there as thesubject matter expert and
theologist of the group to beable to really dissect and and
allow us to really unpack thatprompt and, why maybe it gave us
the the output that it is.
(53:20):
And I'm gonna try somethingreally silly and allow you to
pay for it what you want, right?
There is just a a small setupfee to make sure that it is,
that, that you're able to attendand we can keep the lights on,
but whatever you feel the valueof this sort of thing would be
is entirely up to you.
(53:40):
If you think it's just worth the$97, that's great.
That's fine to you.
If you think it's worth morethan that's fine as well.
And hopefully at the end of ityou'll find that it was worth
what you paid for it and we'llbe able to level set that.
So there's a link there to beable to do that.
So any any questions or conthoughts about what it is what
(54:05):
it is that we talked about heretoday?
Matthew, I know you weretraveling the globe, before all
of this.
And what do you in your journeywith adopting.
Your, thoughts about thistechnology.
Matthew Fulton (54:24):
I, it's been extremely helpful.
I would, I actually probably putmyself in between the beginner
and intermediate just becausethere are some more advanced
functions I've used by accident.
But it's not something where Iwas always using.
It's consistently every singleday.
I've definitely been using it alot more.
I like being able to talk to itlike I was just demonstrating
(54:46):
and ask questions because I canresearch different things while
I'm doing stuff around thehouse, have this conversation be
learning at the same time.
So I think that's prettypowerful.
But I do also wanna help spreadan important message for open
ai.
They've asked you to stop beingso polite to the overlords of AI
because it's costing'em a wholebunch of money.
Every time you run those extraprompts by saying thank you and
(55:07):
your response back to,
Ted McRae (55:09):
But Matthew, I always say please, just in case AI
takes That's
Matthew Fulton (55:13):
exactly right.
You can, you.
Ted McRae (55:17):
That way.
I, that way I know that whenthey take over, they're like,
Hey, that's the guy who alwayssaid, please let's not make him
a slave.
Let's make him like, run theslaves or whatever.
Matthew Fulton (55:28):
I love it.
Dan DeLong (55:29):
Yeah.
There's some good questions inthere.
When are we doing this?
So this is in July.
We'll be having the the weeklysessions for content on Monday
at noon Eastern.
There'll be for an hour, maybean hour and a half.
We're baking in some extra time'cause I'm sure there'll be a
lot to talk about.
(55:50):
And then the follow ups aregonna be on Thursday for an hour
at, 12, I'm sorry 2:00 PMEastern Time.
So whatever that is for for you.
And they will be recorded.
So if you can't make it for thatsession they'll be there for
you, to review.
And I'm sure we'll have an AIsummary created from all from
(56:15):
transcript because the courseand, a lot of the content was
created by ai.
So we've we've actuallyleveraged AI to create a course
on ai.
So we're teaching humans how touse AI with ai, which is,
Ted McRae (56:32):
and you know what Dan, there's one person that,
this is actually a pretty goodcomment says it might be
possible that AI could replacethe jobs, but for those who will
update themselves and adoptrelated technology, AI can't
replace it.
Yeah.
There's more likelihood that AIis gonna replace your job if you
don't use it.
(56:52):
Because you're gonna beobsolete.
You're people want you're notgonna be able to do, so if you
think about the competition inthe industry, you're not gonna
be, you're not gonna be able todo it as quick as a guy down the
street that actually has adoptedai.
So the more you can adopt thisstuff, the better it is for your
practice.
You might be obsolete if youdon't adopt it because somebody
(57:13):
else did, and they're gonna doit better.
Matthew Fulton (57:15):
Not, it's what Uber did to independent taxi
drivers.
Ted McRae (57:18):
Exactly.
Yeah.
They didn't update theirtechnology and stuff like that.
And now they're, who takes ataxi nowadays?
I only take a taxi if it's moreconvenient, one sitting right
there and I'm like, all right,I'll just take a taxi.
Yeah.
But then the taxi driver isalways arguing with me that, oh,
you don't have cash that you canpay.
Exactly.
(57:39):
Yeah.
You know what?
My wife let me have$150 for thistrip.
I'm not gonna spend it on.
Dan DeLong (57:47):
Yeah, the I think Joe Woodard said it best at the
at the last scaling new Heights,which if you're going, you won't
see me there.
'cause unfortunately I won't beable to go, but you'll see
Matthew.
But he said AI's not gonna takeyour job.
It's the people who adopt AIthat are gonna be taking your
(58:09):
job.
Yep.
So if you're not going to see, arobot, come into the space.
It's the person that is adoptingthe is bringing that, Milo with
them as part of their team.
Yeah, a hundred percent.
(58:32):
All right.
We appreciate you joining ushere today, Ted, and for those
of you that, that might beconsidering joining the cohort,
we look forward to continuingthis conversation over at School
Bookkeeping.
And we really appreciate you alljoining us here this week.
And I'm glad that theconstruction stopped, as as I
(58:57):
think there was some rain.
So they went for lunch, I think.
Nice.
So we will see you next time onthe QB Power Hour, which we'll,
be actually talking aboutkeyboard shortcuts.
That's what we thought we weregonna be talking about today,
but we, I didn't even know myschedule, you wanna check it out
(59:18):
on the QB Power Hour.
We have, the upcoming events andtopics there.
So we appreciate you joining ushere today, and we'll see you
next time on the QB Power.
Have a great day, everyone.
Welcome everybody to another QB Power Hour today.
Of course, clockwork.
As we go live, my, my machine,my other machine restart and me
if hear if you hear anything,they tear apart the RV park that
(00:53):
we're, so there's Bobcat heavymachinery behind me, so I
apologize for that becausethat's, I can't move, a
construction job.
Unfortunately, I don't have thatmuch full inmate in the park
here.
But Matthew, how are you today,
Matthew Fulton (01:12):
Dan?
I'm doing good.
But first real quickly, luckilyAI has done such a great job of
doing noise reduction.
We don't hear any of that stuff.
Oh, cool.
It's good seeing you recentlyout in Florida and I'm excited
for this conversation today,Ted.
I'm glad you're here with us.
We were pre-talk a little bitearlier and I think this is
(01:33):
gonna be a, fun episode forsure.
Dan DeLong (01:37):
Yeah, we you guys.
We always, this particular topicof AI in general has always has
a lot of unknowns and, I don'tknow, maybe fear, scared of the
machines type of thing.
(01:57):
And that's, what we wanna try tounpack here today because as Ted
was saying right before here,it's pretty amazing how little,
the adoption of AI in theaccounting industry has has
occurred.
Ted what, were you thinkingabout that?
Ted McRae (02:20):
I was just, for a.
Industry that has so muchtechnology.
I can't think of any industrybesides, actually, I can't think
of any industry that has so manydedicated apps and technology
and all this stuff.
(02:40):
Holy smokes there.
I, there's so much technology tolearn.
I cannot believe that, thatsomething like this is that what
people are afraid to, get into?
And it's so underutilized, evenif a lot of people get into it
and start using it, it's sounderutilized for what it can
(03:02):
do.
It just amazes me.
And I think a lot of it,obviously we all have our, real
lives that we go into, but itmight just be fear and that's
and I can tell a story laterabout my fear doing it and what
stopped me.
But I.
That's all we were talking aboutreally is just how underutilized
it is, I believe, in thisindustry.
Dan DeLong (03:24):
Yeah.
And the subtitle of thisparticular episode is because
the second best time to start istoday, like that when's the best
day to plant a tree was 20 yearsago, but when's the second best
time it's today?
So this is really what we'refocus on here today is about a
way to, to jump in to the to theend of the AI pool, without
(03:52):
losing your float, my name.
Matthew Fulton (04:13):
Hello everybody.
Matthew Fulton here with ParkwayBusiness Solutions and the
Facebook group QB CommunityLive.
Really, again you're talkingabout going backwards and
planting things, and I just, I'mgonna use a quick second to say,
I just imagine this AI image ofBill and Ted going into their
phone booth, going backwards intime and planting trees for all
(04:33):
of us would've been great, butyeah, that's me.
C Community Live ParkwayBusiness Solutions.
Dan DeLong (04:39):
Alright, and join me again, not with Maker code, but
but as a certified engineerspecialist today.
Ted McRae (04:48):
What did you call me yesterday?
I wanted, yeah, go ahead.
I wanted
Dan DeLong (04:50):
to call you a
Ted McRae (04:51):
ologist.
I'm a ologist, not aproctologist.
Thank you.
No, yeah.
And I'm a Did you find
Dan DeLong (05:00):
The the latex gloves or No?
Ted McRae (05:02):
No I forgot the latex gloves.
They're actually, they were alltoo small for my hands, no, and
I'm a self-proclaimed propellerhead.
Anything, technology, app-wise,everything.
I just completely love anythingthat's gonna build efficiencies,
and that's really how I've triedto build my career is how do we
(05:23):
build efficiencies for peoplethat, especially folks that are
trying to do this wholevalue-based pricing, right?
So efficiency is key in ourindustry.
The faster you can do somethingthe better it is, as long as
it's accurate, right?
Because speed is important, butspeed without accuracy is
horrible.
So how can we be more efficientbut still contain and keep that
(05:47):
accuracy?
Dan DeLong (05:49):
Yeah, there was one of the values that Intuit was, I
can't remember what the valuewas but it was work fast and and
they ended up changing thatvalue to work smart, work fast.
Because you can do dumb thingsfast.
04.29.25_-_behind_on_ai_a (06:08):
Yeah.
Dan DeLong (06:10):
And that's how AI is fitting into this picture.
It the speed of which thistechnology is coming in and
really disrupting a lot ofthings.
I think that's part of the fear,I think that of people feeling
(06:32):
like it's gonna leave thembehind, with, with ai.
So let's, before we get intomeat, let's do a little
housekeeping details about theQB Power Hour.
It's every other Tuesday at noonEastern.
It's eligible for CPE Maker Hub.
Used to be our CPE sponsor.
So appreciate what you did therebefore us Ted, but now we're,
(06:56):
using earmark for that.
So after the session, about fivedays later, there will be a.
Course to take over a earmark toget the CPE credit.
And I'm sure Ted is appreciativethat he doesn't have to do all
that.
It was
Ted McRae (07:11):
a Dan, it was a good transition period.
Yeah, it was a good, it was goodhelping you out in the
transition period, but yes, itwas a very a lot of work to
create 88 CPE certificates andmaking sure that everyone was
on, was there on time and stuff.
Exactly.
Dan DeLong (07:31):
We do have an earmarked channel for QB Power
Hour.
You can always check the, linkout there.
If you have specific questionsabout the topic that we're being
talking here today, please putthem in the q and a.
You should see a little q and abutton at the bottom of the
webinar.
That helps us keep track ofthose questions because as all
(07:52):
of the AI bots that are joiningus are announcing that in the
comments.
Those comments just seem to goby and scroll by really quickly.
So appreciate if you havespecific questions, please put
them in the q and a so we canmake sure, we can either address
(08:12):
them live or follow up.
And but if you have just generalcomments, please put them in the
chat.
And then we also have the linksfor the handouts, which I was
about to socialize, but again,my computer restarted.
Hopefully we've got those in thesocial media, live stream as
well.
We also do have a QB power hourstore.
(08:34):
So you two can have a pink hatlike me.
I need to order one all.
So we're gonna start off withour first polling question as a
setting the stage and settingthe, The foundation of what
we're gonna talk about here as aself-assessment.
What is your comfort level withai?
Anything from novice where itscares me to, I'm an AI ninja
(08:59):
fully embracing this technologyand incorporating the advanced
features.
And Ted can you talk about yourjourney with, ai?
Like where how you, became aologist.
Ted McRae (09:12):
A ologist?
Yeah.
So I started, you know what thefunny thing is, so I was hearing
about this AI thing, and thiswas way back when, a few years
ago, when AI actually was onlyup until, I think it was
September of 2021.
So anything new we were in 2022or the late ends of 2021 when I
(09:35):
started using it.
So anything new I.
It wouldn't, it had no idea.
You couldn't ask any who won theWorld Series or what are the a
You can't do anything like that.
And so when I first started out,and probably there's a lot of
people like novice, it scaresme.
A lot of people or myself inparticular thought there was a
(09:59):
specific like magic to askingthe prompts.
So I searched all over theinternet when I first started,
using chat GPT for chat, GPTprompts what are the, how do I
do prompts, how do I do this?
And what I soon found out aftersearching for weeks and weeks is
(10:20):
that there are no speci anyonecan give you, oh, here are 10
prompts that you should use,everyone should use in
accounting.
But realistically I.
That's not really what youshould be doing in chat GPT.
And there's really no way,there's no specific here's
exactly how you should do it.
(10:41):
There's a framework.
And so that's what I didn'treally realize until I became a,
ologist, is that it's dirtywork.
No, I'm kidding.
It there, it's not about
Dan DeLong (10:57):
digital.
Ted McRae (10:58):
Yeah, it's digital work.
It's really about where youwanna end up what you, want.
And you gotta think of it as ifI were to ask Matthew Hey, can
you help me create a webinarthat I wanna do next month?
I wouldn't just tell Matt that'cause he would be like, yeah,
(11:19):
sure here, and he would justtell me everything that he
thinks that I should do.
But if, I tell him, Hey Matt,I'm doing a webinar next month
for a specific type of personand it's at this conference.
Here's all the research on theconference.
And then I do even more thingslike that.
Now Matt has a framework towork, or Matthew has a framework
(11:41):
to work with where we can createawesome content for that.
But if I don't give him all thatbasic stuff, like having a
conversation with him, he'snever gonna know.
So that's really how I startedout with it way, way back when.
And I've kept up with it and Ihate to say it, but I use it
(12:01):
probably 50, 60 times a day.
Dan DeLong (12:06):
Yeah.
It's now it's it's part of your,almost your everyday thing.
The whole idea of the.
Dyslexia where, you know, if youdon't have your phone to do
basic math skills, you justYeah.
Purge that out.
But I've got a calculator on myphone.
So we're gonna be doing Ted andI are gonna be doing a four part
(12:29):
weekly cohort series over aschool bookkeeping where we're
gonna be talking and unpacking alot of these things from the Mr.
Rogers level of ai.
And we'll, we maybe go up toSesame Street or maybe electric
company if we're, lucky.
Matthew Fulton (12:48):
I'm sorry, Mr.
Rogers is going to, the ologistis gonna be the first episode of
this.
Yeah,
Dan DeLong (12:53):
the French.
He'll be happy taking off morethan his shoes for that.
But what we're gonna be doingthere is we're gonna be talking,
and today we're gonna be talkingabout AI and large language
model statistics.
We'll just really keep this, asa fireside chat type of thing
where we'll talk about whatwe're, gonna be unpacking in the
(13:16):
cohort series which will becalled prompts for practices,
crafting effective prompt foraccountants where to go to
register.
And then we'll have a q and asession about that.
So let's let's talk about, oh,Noah, did you share the results
of the first poll?
Because I thought that wasreally oh.
(13:38):
Did I Just, so it looks likeabout 60, almost 70% are in the
novice, the basic level of, ofthe com comfortability with with
ai.
So let's talk a little bitabout, the, impact of, AI in, in
small business as far as anunderstanding of how businesses
(14:01):
are using AI and the largelanguage models.
I really like this this slidethat you created here, Ted,
about.
'cause that's part of the thingfor me was what is clawed, what
is all of these other, theGemini or Bard, I think they
called it originally and nowit's got a new name to it.
(14:23):
It's tr trying to keep all theplayers straight.
But talk a little bit about thisthis Ted
Ted McRae (14:32):
Yeah, so there's a few things here that really
strike me as, especially in thisindustry, chat, EPT is like the
Intuit in cases of this whole ailarge language models.
They were pretty much the first,the Kleenex, right?
(14:57):
Yeah.
They're the Kleenex.
But the thing is they are, in myopinion the, most reliable, I
never stray.
Personally for me, I never strayfrom chat GPT except for
perplexity.
And so there's different typesof AI that people need to to,
(15:21):
actually keep in mind.
There's, AI that can generatespeech for you.
11, 11 labs, which I can give ita, I can give it a script and I
can either choose my, the audioperson or I can upload snippets
of myself and create audio.
There's like Gemini, there'sMicrosoft.
(15:43):
So it all depends on what you'redoing it for.
So like Google Gemini, if youput it into the Google suite of
tools works perfect if that'swhat you're gonna use it for.
You can also attach chat, GPT,to, to Google as well.
But it, so it all really dependson what you're using it for,
but.
(16:04):
The thing is really quickly hereis that some of all these other
ones while you might like to use'em and they're fine, they may
not have the same technology aslike just a regular old chat
GPT.
So it really matters what youwanna use it for.
Matthew Fulton (16:18):
Ted, can I just real quickly have you step back
just a little bit and can yougive us a quick definition of
what a large language model is?
What is an LLM and why is thatimportant in what we're talking
about?
Ted McRae (16:31):
Yeah so a large language model takes context
from a person, right?
And then will spit out what itan answer a real person would.
So it takes your question, spitsit out and, contextualizes it.
(16:54):
Like, a per, like you would in aregular conversation.
Large language models are hugedata sets.
And I could really talk aboutlike the prompts, what it does.
But Matthew, what it really doesis it takes your sentence and
then it breaks it out intolittle chunks.
All right.
So it's like for instance, thebill is due on the on May 1st.
(17:21):
So we'll take that and itactually chunks that out, the
bill and then it says is due,and then it has the date.
So that, so it separates allthat stuff out.
Yeah.
So it understands.
Then puts it back together so itcan understand what you're
talking about, the date,actually referring to a date in
(17:41):
the future that this document.
So almost like Matthew, do youremember when we were, not you
and I together, but when youwere in grade school and you had
to cut a sentence apart and findout where, what is the
adjective, what is the noun,what is the verb?
And all of that stuff to, toshow the context of a sentence.
So large language models do theexact same thing, but in a more
(18:04):
granular process.
So they, they chunk that out andright now I have old man brain,
so I can't remember the name ofthe, of what it's called.
Oh, tokens.
They chunk it out into tokens.
Okay.
And then each token, theyanalyze that token.
And so what large languagemodels have, it's just a huge
data set, and now it goes and itreturns it to you like a normal
(18:26):
person would.
Matthew Fulton (18:27):
So another way to kinda say it'cause Grace
asked a good question earlier ofwhat exactly can AI do in the ac
in the accounting to assist inour work?
That's obviously a hugequestion, but a quick answer
way.
I feel when we're talking aboutLLMs large language models,
while we the, word language isin there, it's, it is, it's
(18:48):
truly large data sets.
So shameless plug here, maker'sHub has taken the ability to,
when you scan in a whole bunchof different documents in
different formats, there's thislarge database to try to
understand where valid orimportant information is on a
sheet Yep.
Can extract it and thenunderstand the words that are on
there to place it appropriatelyinto the right fields for your
(19:10):
accounting platform.
Same thing, where there's a lotof companies working on like
check images now, trying to beable to pull that type of stuff.
So the more data you have, themore it can reference to try to
give you a better result.
Is the intent of it all, is thata good way to.
Summarize it.
Yeah.
Ted McRae (19:27):
Yeah.
And like the really cool thingthat you can do with like chat
GPT now and all the other onesis, and this is what I did, I
created an accountant a ai, soit's just a, all he is his, name
is Milo.
And Milo knows a lot aboutaccounting because every little
(19:49):
thing that I see on LinkedIn,like people will do infographics
and stuff like that, I just copyand paste it into Milo.
So like, how to analyze thespread, how to analyze a balance
sheet versus a cash flowsummary, how to ba versus a, a p
(20:09):
and l and all of these otherthings, income statement.
And so I just upload as much asinformation on accounting that I
could find into Milo.
I also have him look at ddifferent websites.
Then now I can do things in a,in accounting that maybe we
couldn't do before, but I don'tuse that as the true source.
(20:32):
So people will be like, I don'ttrust it.
You shouldn't, that's the point.
But it helps you get to whereyou want to be faster.
Dan DeLong (20:41):
No there's a lot of words here about the statistics
usage and adoption.
Any, anything you wannaencapsulate, take away of all
this?
Of course you can download theslides and take a closer look at
some of these statistics,
Ted McRae (20:56):
Ted.
Yeah.
I think that what people need torealize on this is that the
businesses don't think of it as,it's gonna take my job away.
Think of it, especially like Isaid, being the propeller head
that I am, what can we in thisindustry use it for?
(21:21):
And we all know that there is a,there is a, gap between like how
many hours we can work and howmuch we, especially if you're if
you ba bill by the hour, you canonly bill so many hours.
But if you're, but if you'revalue-based, the faster we can
(21:43):
do things, the more hours we canbill.
So if I can take a, if I cantake a balance sheet and ha have
the first pass of analyzing it,go through a AI before I look at
it, it's almost like Dan, likewhat you say, having an intern
look through my stuff.
I wouldn't just send my internsreport to my client, but the
(22:06):
intern's gonna look in and lookfor anomalies for me and things
that are like red flags.
Then I can review them with thedata, with the mindset of an
accountant to put my spin on it.
And then I could even put it ina and say, okay, chat GPT create
the the summary report for myclient for this.
(22:27):
But that doesn't mean I'm gonnasend that straight to my client.
I still need to review.
Now I take that summary reportand build my report off of it.
Yeah, but how much time doesthat save us?
It's astronomical when you useit as just the first set of eyes
and create, instead of staringat a blank piece of paper, I
always tell Chad PT to writedown a summary of my ideas.
(22:52):
And then I don't start with ablank piece of paper.
I just save myself three hoursprobably.
And then the other thing, 49% ofcompanies are currently using it
and 30% intend to use it in thefuture.
So for all of you, I was amazedat how many people said it
scares me, is what'd you say?
60% of people said they're inthe first two buckets.
04.29.25_-_behind_on_ai_a (23:11):
Yeah.
Ted McRae (23:12):
We're like five years almost into chat.
GPT artificial intelligenceright now in large language
models.
It's time, like you said, Dan,it's time to start thinking
about it.
it's,
Dan DeLong (23:25):
It's the new disruptor of, the day.
You look back at all thedifferent things that dis
disrupted the accountingindustry.
Cloud accounting is the thelatest up until this point,
Excel, right?
You can go back to all of thesethings where you think about,
(23:46):
I'm sorry.
Oh my God, it's gonna change.
It's gonna change the it's gonnatake my job, right?
But it's not, it's just anothertool that can assist and to your
point, provide you more leverageto do the things that you wanna
do.
Faster, right?
Think about
Ted McRae (24:04):
it.
They're in Dan, they talk about'EM as ERA technology eras,
right?
Yeah.
So we had the, really, the firstera was the www era where we
connected everyone online.
And that was a game changer forall of us.
And then there was the, thenthere was the cloud era where we
actually started sharing datawith each other.
Then there was the mobile erawhere now we're connected to our
(24:28):
mobile phones, and now we're inthe AI first era.
And so this is going to, it'snot going away, and this is
where it's really gonna be.
Yeah.
Thank you Matthew.
You're right.
Dan DeLong (24:41):
Thank.
So this one what, accountantsare, using AI for, right?
You've got the basic level whereit's creating some content.
So starting you off with thatthree hours of savings, like you
were just talking about, insteadof staring at a blank document.
Get it to start with that roughdraft and then the intermediate
(25:03):
level.
So you wanna talk us throughthis these, different levels?
Ted McRae (25:08):
Yeah.
So if you think about it at thebasic level, and it seems like
probably 70% I don't mean to, ifit's less, that's fine, but it
seems like 70% of the people onhere are at the very basic level
using it to or look at an email.
Is it right?
(25:28):
I want to create an email forthis and that, or I need to
create a blog.
Which people are probably notusing it the right way, even if
they're using it in the basiclevel.
Now the, intermediate level,we're allowing large language
models or AI chat, GPT, whateveryou wanted to.
Whatever you want to call it.
(25:49):
We're allowing it to take it up.
'cause we're starting to trustit more.
So we're allowing it to take usto a different step where,
alright, I want to create maybeDema, generate a demand letter,
but I'm still not using it toits full potential.
Or I'm generating a letter to aclient that I need to fire or
(26:11):
I'm maybe creating small easySOPs, little things like that.
So that's more of anintermediate level.
And then maybe you're going inand if you know what A GPT is,
we're gonna talk about itextensively in in our school of
bookkeeping training that we'redoing, but creating GPTs or
using it to ask questions.
(26:34):
About a product.
How do I create a formula, anExcel formula that sums all of
this really cool stuff and itjust spews it out for us and
shows us how to do it?
That's pretty intermediate,right?
Or how do I create a button inHTML?
That's pretty, easy.
But you can say, how do I createa button that does, that, goes
(26:55):
to this specific link with thesecolors, and when I hover over
it, it changes colors and whenmy mouse clicks on it, it
shakes.
So you could have it do morethan just create a button in h
TM l you can have it do insingle thing and create the
webpage for you that the buttongoes on.
So, intermediate is just reallybasic.
(27:19):
A tiny bit above the basic.
We're getting into knowing howto create somewhat crafty
prompts.
The advanced level we've createdMilo, like I have.
Who is a, an accountant in mylittle chat, GPT world.
And Milo knows a lot aboutaccounting, on a pretty awesome
(27:43):
high level because I start, Ifeed in every single news thing
that I get about it.
I have him look at websites allsorts of stuff.
So anything I want to ask Miloor analyze or Milo, what, you
know what it's, today is the endof April.
You're as an accountant.
What are you doing today?
(28:04):
And Milo will tell me what theaverage accountant is doing.
So now me as a marketer knowswhat to actually talk to the
accountants about.
So now we're getting more at anadvanced level, right?
Milo can analyze reports.
I can have Milo generate demandletters.
I can have Milo actuallygenerate them based on a level
(28:26):
of friendliness, aggressiveness,and how late it is.
Milo, I can say Milo generateddemand letter from my client
with an aggressiveness of seven'cause he already knows of one
to out of seven, a friendlinessof two, and the, it was due on
this date.
(28:46):
And here's, and I upload theinvoice and he'll generate a
demand letter that is more thanjust you need to pay this is
like the third or fourth time.
And then you can also do somepretty advanced.
Or semi advanced coding at thatlevel where maybe you're even,
(29:09):
you're doing more than J you'reextracting databases and you're
doing really cool things likethat because you can ha you can
do JavaScripts, you can do allsorts of stuff within within
chat GPT.
And in fact, my daughter's a asoftware engineer and they are
not allowed, they have a programthat scans their they use
(29:34):
artificial intelligence to scantheir backend code for minute
errors.
They're not allowed to use largelanguage models to create code,
but they use it to scan the codefor errors.
And then at the ninja levelyou're really, you're using it
to.
(29:54):
Think about it this way, if youcould create a GPT and we, like
I said, we'll talk about it, butA GPT is memorized information
that you put in a little in a,little area of chat.
GPT.
Think about this.
If I have a GPT where I canupload all of my SOPs, I can
upload all of my company, likehiring information and all that
(30:15):
stuff.
So when I have a new employeecome on board that has the
questions on standard operatingprocedures, they can just go
into my GPT that I did.
And I said, so if I, if we havea client that refuses to give us
their bank account informationfor QuickBooks Online, what do I
(30:36):
do?
And it just spits out all thatinformation that employees just
saved.
A a lot of time having to go andask people around the office.
What they should do in specificareas of the business.
So now I'm like, all right, howdo I do X, Y, and z?
And the chat GPT just comes outand tells me how to do it based
(30:59):
on my company.
Dan DeLong (31:02):
Yeah.
Good question here in thecomments how secure and I think
that was part of what you werealluding to, where you don't
want what you are training it onto be training everybody else.
Which gives that specialness, Iguess to Milo over sharing it
(31:25):
with, with everyone in the
Ted McRae (31:29):
world, right?
Yep.
Yep.
'cause Milo is basically anextension of what Ted wants.
So we, if we each had our ownMilos, it would have, its, it
would have its owncharacteristic of the company
that we work for.
I can have Milo be Ja, Jasonstats.
And react to me like Jason Statsand I actually have a Jason Stat
(31:51):
bot if I ever want to create avideo that's halfway funny I put
it through my Jason Stats scriptbuilder and it builds that
script out.
So I can have I can have chatGPT review, YouTube videos or,
(32:11):
things like that.
I actually had chat.
GPTI have a Ron Baker chat, GPTfor value-based pricing, believe
it or not.
So it's, there's no end to whatyou can do.
Dan DeLong (32:24):
And how does someone keep that from being leaked?
Is be like how do you,
Ted McRae (32:34):
your chat, GPTs, you can either share publicly, or
keep them private to yourself.
Now if somebody gets it I reallydon't care because I don't have
a, I don't have any, I don't putsocial security numbers in there
and, stuff like that, right?
Which I still I still wouldnever do, I wouldn't upload a
(32:54):
report and say, analyze all ofmy employees, give me a social
security list alpha ornumerically from smallest to
large.
I wouldn't do anything likethat.
I would think of chat.
GPT.
If you were in a doctor's officeand had to follow HIPAA rules,
would you put, what would youput into a computer system
outside the doctor's office?
(33:15):
I.
Dan DeLong (33:17):
It's a good good rule of thumb.
Matthew Fulton (33:19):
There are certain service providers out
there that allow you to spinthese up into your own closed
environment, but it becomes likethe cost of it is exponentially
higher.
So that would be something you'dresearch.
Once you've really developedsomething and you've got a tool
(33:39):
if you really needed thatprivate information there.
The key thing really is personalidentifiable information.
You don't want to specific, youdon't wanna have that in there.
You don't want the liability ofthat would be the key, I'd say.
Ted McRae (33:52):
Yeah, I would, I I.
I would think of it, Matthew,just like any other thing, are
you gonna, are you gonna putlike your employee, or not your
employees, but all your client'sdata up in an unsecured Google
Drive that maybe has a sharedlink to somebody that you don't
like, a deep folder that mightbe shared with somebody.
(34:14):
You wouldn't do it.
And I wouldn't do it on hereeither.
Agree.
But the thing is realisticallywe shouldn't even be using it
for that.
What can we analyze withpeople's personal data?
Like a company, I can understanda company, alright, I'm
uploading a balance sheet for acompany.
Okay.
It's just nu it's really justnumbers, right?
(34:34):
It's not like I can take thatnow and infiltrate the company
and pretend like I'm the CEO andaccess his credit card
information.
For the most part.
That's not what you're uploadinganyway.
Matthew Fulton (34:48):
Good point.
Dan DeLong (34:51):
When I asked AI.
How many different types of AIare there?
It spit out, I think 16different ones, of how AI is
used.
So we want to just talk aboutthe different types of ai.
So we've got traditionalgenerative and, machine
learning.
(35:12):
So Ted, how would you, qualifyor quantify the different type
of AI technologies?
Ted McRae (35:21):
There's traditional, which is like the chat GPTs and
the copilots and things of thatnature.
They're just like let's talkabout it this way.
Traditional AI was like, freechat, GPT.
There's, they have limitedscope.
It doesn't learn.
So that's just like whattraditional back in the day was.
(35:43):
Now, if we talk about, and thenyou can talk about traditional
AI is almost like chat GPT rightnow.
But again, without thesubscription, right?
Because it's not gonna learnfrom what you do.
Then there's generative ai.
And generative AI is, I use itall the time.
I can you and I just do thiswith chat.
(36:05):
CPTI need an icon for, so myMatthew, you know the arch out
on the Channel Islands, right?
That, yep.
The little, okay.
So I wanted to create an iconfor the arch, okay.
For Ventura.
And so I just uploaded a pictureand I said, here's the company I
(36:29):
want to generate this icon for.
And I want the theme to becoastal, so use coastal colors.
And so it generated a logo, andI shouldn't say icon, a logo.
I.
For that.
So I uploaded just a regularreal picture.
So generative AI createscontent, creates images it can
(36:54):
create all all sorts of stufffor, you.
So generative, you give it anidea and it's really creating
like an image for, you or I playthe guitar.
I can have it create a song forme based on an, accountant.
I actually, one, one time I wasdoing a a webinar for AI and I
(37:18):
had it create a song foraccountants around April 15th.
About how, about like how crappytheir clients are when they get
right around that time askingfor things or not giving them
the right paperwork that theywant and stuff like that.
And it wrote out the song, itgo, gave me all the chords.
(37:40):
It told me how to strum it andeverything.
And I sang that song and I sangthat and played that song on one
of my webinars.
Dan DeLong (37:49):
And then the machine learning and then machine
learning.
Yeah,
Ted McRae (37:53):
data.
Data.
It's like my milo, right?
My Milo learns.
I have to tell it to learn.
But my milo also learns when Ijust type things into him as
well.
He just keeps learning.
I can hard chord, hard codelearning into Milo, or as I'm
(38:14):
typing, Milo learns as well.
Dan DeLong (38:16):
So this would be just to put it in a QuickBooks,
speak.
This would be like the bankfeeds as they are trying to
homogenize all thesetransactions.
And it will start to recommend.
The things that you wannacategorize it to, or something
in like Maker Tub where you putin a, purchase order or a
(38:38):
document and it's different forone over the other.
Where it starts to learn thesales
Ted McRae (38:45):
tax.
Sales tax should go to the salestax chart of account.
And not just not just to cost ofgoods sold or whatever you wanna
put it.
To put it to.
Yeah.
Dan DeLong (38:56):
Which I think this leads to a question that was
posted in here.
What is the possibility of AIbasically becoming self-aware
very or, becoming corruptedbased on the things that it
learns?
Matthew Fulton (39:11):
Skynet is real.
Yeah.
Ted McRae (39:14):
Yeah.
You know what's the funny thingis I was listening to a podcast
and they were talking about.
These guys that are doing thelarge learning models, didn't
they watch all the movies wewatched growing?
I know, right?
Come on.
Literally out whole lot ofmovies saying this is a bad
(39:37):
idea.
Yeah, and guess what?
They know it too.
They're like, yeah, we know thatthis is a possibility, but
they're, but obviously if it'swe're in a race, we're in an AI
race right now, is it going tobe the United States?
Is it gonna be China?
Who's it gonna be?
That is, is the AI dominantcountry and it really matters.
(40:00):
'cause ai, if you think aboutai, what we do just in
accounting, think about whatwe're gonna use AI for in
defense.
In the future, what we're gonnause AI for in all sorts of like
negotiations and stuff likethat.
So
Matthew Fulton (40:16):
yesterday I was listening to a podcast as well.
And then they were specificallytalking about ai.
They had mentioned that as anexample, having AI research, AI
itself and how to do certainthings.
This kinda leads us towards thedifferent levels or layers of
where it could become a problem.
(40:36):
But currently.
AI can complete a human taskthat's, say, that's like a, one
hour task.
It can complete it faster with a50% accuracy.
And the imp it is improving anddoubling in time, like how much
it can do every four months.
So when they talk aboutartificial intelligence, they
also talk about human artificialintelligence.
(41:01):
In other words, artificialintelligence becoming as
intelligent as humans insteadof, I'm not gonna get the term
right, but like superintelligent the, next level
above.
And they compared it to atoddler versus an adult.
So a toddler trying to teach anadult something is not gonna
happen very easily.
(41:21):
The adult can even placate thetoddler and make'em, oh, you're
so cute.
You're teaching us suchwonderful things.
I.
So we have to be, the concern iswhen we go beyond human
artificial intelligence, are wegonna actually understand what
it's truly capable of or not?
And is it gonna go Skynet?
(41:44):
Nobody really knows, but wegotta, I don't think, I don't
think
Ted McRae (41:46):
we'll have to, I don't think in our lifetime
we're gonna have to worry aboutit.
But it's, the fastest it'sactually increasing in knowledge
is amazing.
I'm amazed every time I log in,what I can do with chat GPT
every fricking time I log in andI do something.
(42:08):
Not because it's actuallychanged, but because I just
thought of a different way ofusing it and I'm like, holy
smokes that.
And then like, when you can dothings like copy and paste
images and it can read it.
Then store that information.
That's why I said, every time Isee anything about accounting,
infographic, whatever, as longas I, as long as it comes from a
source that I trust, I just copyand paste it into Milo and have
(42:31):
Milo learn it.
Matthew Fulton (42:33):
Do you ever talk to yours?
Little tip I like to use is onmy chat, GPT, you can hit the
side and ask it.
Can you explain how AI can beused in accounting today?
AI (42:45):
Absolutely.
AI can be a game changer inaccounting.
It can automate repetitive taskslike data entry, invoice
processing, and reconciliations,which saves time and reduces
human error.
It can also analyze largevolumes of data quickly, helping
identify trends, anomalies, andpotential risks.
AI powered tools can assist inforecasting, budgeting, and
(43:08):
financial planning by providingmore accurate predictions based
on historical data.
Matthew Fulton (43:14):
Thank you.
Plus, sorry to interrupt you,but can you tell me, does that
mean AI accountants are gonnatake over the world someday in
the future?
AI (43:22):
No worries.
It's unlikely that AIaccountants will take over the
world.
While AI can handle many routineand data.
Ted McRae (43:29):
Yeah, it it's pretty, of course they would say
Dan DeLong (43:33):
no.
Ted McRae (43:35):
Hey so Linda Russell, how you doing?
I, Hey,
Matthew Fulton (43:39):
Linda.
Yeah.
Ted McRae (43:40):
Talked to you a few times before.
So try using perplexity.ai.
So Linda says, I'm using chat,GPT as my search engine, pro tip
perplexity.ai.
It was designed to be an AIsearch engine, and we'll give
you better results than chat.
(44:01):
GPT, just ex FYI.
Dan DeLong (44:05):
Yeah.
And, I want to talk about thiswhole concept of AI
hallucinations, which was reallylike the catalyst of this, talk
that we're having, this coursethat we're creating because, we,
we've all seen like the news ofhey, AI took the bar exam.
(44:27):
Or AI can't, couldn't do math orcouldn't do simple math.
Where AI hallucinates the truth.
Sometimes like it could havejust hallucinated that response
that, you said when you, askedit, is it gonna take over the
world?
Maybe that's what you think thatyou want to hear.
(44:51):
And this whole idea of ofhallucinations isn't necessarily
a flaw but more of a feature.
Ted can you talk a little bitabout the, concept of AI
hallucinating and just makingstuff up.
Ted McRae (45:09):
Yeah.
What ai hallucinating nowadaysis more of, I don't know if
you've ever heard of, pep Cap.
Problem exists between keyboardand chair.
So honestly, AI hallucinationsprior to 4.0, chat GPT was
(45:34):
because it just didn't have thedata right?
And it just tried to make upwhatever it is.
Now, realistically, it's becauseyou like what I said to you
guys?
Hey, let's before we got onhere, I was telling Dan and
Matthew, if I told them, Hey,let's all create a blog and we
all typed into chat GPT create ablog about accounting and chat
(46:01):
GPT, it would might, if we justsaid that it might give us all
the exact same, roughly the sameinformation.
And it might not even know whatwe're talking about.
Is it talking about like accountchat, GPT taking over
accounting?
Is it talk?
So AI hallucinations are a notso well crafted prompt now,
(46:29):
except for like it used to be.
It used to be like you would askchat GPT how many Rs are in
strawberry?
And it would tell you two, itdidn't matter how many times you
told it.
There were three, it would tellyou two Mine at least now, and I
can only go off of mine, saysthat there are three.
So it is starting to learn, butgo.
(46:49):
But it is the, in most cases nowit's the user.
Dan DeLong (46:54):
Yeah.
And it's this George Costanza Ghere about it's not you, it's
me.
It's if you're not getting theoutputs that you want.
You really need to change theinput that you're giving it.
Matthew Fulton (47:09):
I'm gonna disagree in one area
specifically, which is whenyou're creating an image and you
want words in there.
04.29.25_-_behind_on_ai_adop (47:14):
Oh yeah.
Matthew Fulton (47:15):
Like you can type out exactly what you want
it to say and it will stillalways spell it wrong every
time.
Ted McRae (47:25):
I never, I always, but Matthew, you can correct the
chat, g you can correct it tocreate the actual words in the
image.
If you tell it to focus on thewords in the image.
Matthew Fulton (47:40):
I years sounds like it's got a better
education.
Mine, I think kindergarten orsomething.
So maybe I don't know how tospell it.
Could be.
So it's probably my prompting.
Dan DeLong (47:49):
Yeah.
Like I, I saw a post wheresomebody had searched Google
just did a Google search for BSin the computer.
And Gemini, the AI overview thatcame up actually said.
There is, there are bees in thecomputer that have been there
since the year 2000.
(48:10):
And, of course that's the,poster child for, okay, this AI
doesn't work at all because hereit's saying that there are bees
inside of your computer.
04.29.25_-_behind_on_ai_ad (48:23):
Like honeybees.
Dan DeLong (48:25):
Yeah, like real bee, like buzzy bees.
You can google it, you cangoogle it now.
And the AI overview does not saythat.
04.29.25_-_behind_on_ai_a (48:31):
Okay.
Dan DeLong (48:31):
But, there what it was doing is it was searching
the internet and it found thiscompany that created a blog
article about such, things.
And it was an it, an fool joke,right?
This one things where.
(48:52):
It found that article andtherefore provided you that as
truth.
And that's something that AIcan't do yet, right?
Like you cannot measure thevalue of truth, even though it
found an answer for you.
Yeah, there were bees in thecomputer.
But in this case it was totallya joke and it can't figure out
(49:15):
sarcasm or truth or things yetuntil you, unless you tell it.
Now, apparently somebody toldthe AI overview that is not a
true thing and it doesn't prooffer you that as true.
That's funny.
So let's talk a little bit aboutsome prompt examples, right?
(49:36):
Where they're ineffective andmaybe effective, right?
So in this example, make areport about accounting, how
that is an ineffective prompt.
Ted McRae (49:49):
Yeah.
So make a, this is almost likethe, what we talked about the
blog, right?
We all, it's very vague.
You have no idea who am I makingthe report for?
What, is the report?
Am I talking about accounting oram I talking about numbers?
From from something that youwant to give to me.
(50:09):
So you're there, it lacks anycontext whatsoever.
And can I just share somethingreally quickly?
I wanted to show you guyssomething.
Matthew, this will show you,allow multiple, I'm just gonna
replace the current share.
Sure.
Is that okay?
Yep.
04.29.25_-_behind_on_ai_adopt (50:27):
I stop sharing.
Ted McRae (50:28):
So look at this.
I said can you create an imageof the bookkeeping department
with QuickBooks online on thecomputer?
And look at what it created forme.
So accounting, a bookkeepingdepartment actually spelt it ex.
So look at how good it's gettingnow.
QuickBooks, it's spelt wrong.
I could, redo it now and I couldsay you spelled QuickBooks wrong
(50:51):
up here, and, but it like small,like it says profit on loss, but
it's still pretty darn good now,right?
Holy smokes.
It's, really, crazy.
Go ahead and share again, guys.
I'm sorry.
Matthew Fulton (51:07):
That's where, so I'm sure I've, other people seen
these reels, right?
Where they'll be like, show us aunicorn with great, beautiful
wings and make it magical.
No more magical.
Make it extremely magical andthey just keep building on top
of it, on top of it, on top ofit, on top of it until it's just
absurd.
That's some fun stuff you to do.
Dan DeLong (51:26):
Yeah.
So when here we're talkingabout, being more effective in
your prompt, create a one pagesummary report on the latest
accounting trends for smallbusinesses focusing on
automation and AI tools.
So just by adding more.
Context to the input, you'regoing to get a, more specific,
(51:48):
answer from the from the output,right?
And then also
Ted McRae (51:53):
Dan, we were talking about basic and intermediate
levels.
That's more, that's anintermediate prompt right there,
right?
That's not even an advanceprompt because there's still,
it's contextually wrong stillbut it's better than the other
one, honestly.
Yeah.
And that's what we're gonna talkabout in our little prompt thing
(52:13):
as well.
Dan DeLong (52:14):
Yeah.
So in the month of July Ted andI are gonna be doing a four,
four week cohort where each weekwe'll break down four different
modules going from basic what isit to intermediate advance and
drafting prompts.
And each week we'll also havesome homework.
(52:35):
So because it's at the school.
We'll have some homework, someassignments.
There will be a cohort I can'tsay it, cohort only community
where you can, put in questionsand we can talk about them in a,
weekly follow up and q and asession.
We can do a group discussionabout the prompts that are being
(52:57):
created and the outputs.
And Ed will be there as thesubject matter expert and
theologist of the group to beable to really dissect and and
allow us to really unpack thatprompt and, why maybe it gave us
the the output that it is.
(53:20):
And I'm gonna try somethingreally silly and allow you to
pay for it what you want, right?
There is just a a small setupfee to make sure that it is,
that, that you're able to attendand we can keep the lights on,
but whatever you feel the valueof this sort of thing would be
is entirely up to you.
(53:40):
If you think it's just worth the$97, that's great.
That's fine to you.
If you think it's worth morethan that's fine as well.
And hopefully at the end of ityou'll find that it was worth
what you paid for it and we'llbe able to level set that.
So there's a link there to beable to do that.
So any any questions or conthoughts about what it is what
(54:05):
it is that we talked about heretoday?
Matthew, I know you weretraveling the globe, before all
of this.
And what do you in your journeywith adopting.
Your, thoughts about thistechnology.
Matthew Fulton (54:24):
I, it's been extremely helpful.
I would, I actually probably putmyself in between the beginner
and intermediate just becausethere are some more advanced
functions I've used by accident.
But it's not something where Iwas always using.
It's consistently every singleday.
I've definitely been using it alot more.
I like being able to talk to itlike I was just demonstrating
(54:46):
and ask questions because I canresearch different things while
I'm doing stuff around thehouse, have this conversation be
learning at the same time.
So I think that's prettypowerful.
But I do also wanna help spreadan important message for open
ai.
They've asked you to stop beingso polite to the overlords of AI
because it's costing'em a wholebunch of money.
Every time you run those extraprompts by saying thank you and
(55:07):
your response back to,
Ted McRae (55:09):
But Matthew, I always say please, just in case AI
takes That's
Matthew Fulton (55:13):
exactly right.
You can, you.
Ted McRae (55:17):
That way.
I, that way I know that whenthey take over, they're like,
Hey, that's the guy who alwayssaid, please let's not make him
a slave.
Let's make him like, run theslaves or whatever.
Matthew Fulton (55:28):
I love it.
Dan DeLong (55:29):
Yeah.
There's some good questions inthere.
When are we doing this?
So this is in July.
We'll be having the the weeklysessions for content on Monday
at noon Eastern.
There'll be for an hour, maybean hour and a half.
We're baking in some extra time'cause I'm sure there'll be a
lot to talk about.
(55:50):
And then the follow ups aregonna be on Thursday for an hour
at, 12, I'm sorry 2:00 PMEastern Time.
So whatever that is for for you.
And they will be recorded.
So if you can't make it for thatsession they'll be there for
you, to review.
And I'm sure we'll have an AIsummary created from all from
(56:15):
transcript because the courseand, a lot of the content was
created by ai.
So we've we've actuallyleveraged AI to create a course
on ai.
So we're teaching humans how touse AI with ai, which is,
Ted McRae (56:32):
and you know what Dan, there's one person that,
this is actually a pretty goodcomment says it might be
possible that AI could replacethe jobs, but for those who will
update themselves and adoptrelated technology, AI can't
replace it.
Yeah.
There's more likelihood that AIis gonna replace your job if you
don't use it.
(56:52):
Because you're gonna beobsolete.
You're people want you're notgonna be able to do, so if you
think about the competition inthe industry, you're not gonna
be, you're not gonna be able todo it as quick as a guy down the
street that actually has adoptedai.
So the more you can adopt thisstuff, the better it is for your
practice.
You might be obsolete if youdon't adopt it because somebody
(57:13):
else did, and they're gonna doit better.
Matthew Fulton (57:15):
Not, it's what Uber did to independent taxi
drivers.
Ted McRae (57:18):
Exactly.
Yeah.
They didn't update theirtechnology and stuff like that.
And now they're, who takes ataxi nowadays?
I only take a taxi if it's moreconvenient, one sitting right
there and I'm like, all right,I'll just take a taxi.
Yeah.
But then the taxi driver isalways arguing with me that, oh,
you don't have cash that you canpay.
Exactly.
(57:39):
Yeah.
You know what?
My wife let me have$150 for thistrip.
I'm not gonna spend it on.
Dan DeLong (57:47):
Yeah, the I think Joe Woodard said it best at the
at the last scaling new Heights,which if you're going, you won't
see me there.
'cause unfortunately I won't beable to go, but you'll see
Matthew.
But he said AI's not gonna takeyour job.
It's the people who adopt AIthat are gonna be taking your
(58:09):
job.
Yep.
So if you're not going to see, arobot, come into the space.
It's the person that is adoptingthe is bringing that, Milo with
them as part of their team.
Yeah, a hundred percent.
(58:32):
All right.
We appreciate you joining ushere today, Ted, and for those
of you that, that might beconsidering joining the cohort,
we look forward to continuingthis conversation over at School
Bookkeeping.
And we really appreciate you alljoining us here this week.
And I'm glad that theconstruction stopped, as as I
(58:57):
think there was some rain.
So they went for lunch, I think.
Nice.
So we will see you next time onthe QB Power Hour, which we'll,
be actually talking aboutkeyboard shortcuts.
That's what we thought we weregonna be talking about today,
but we, I didn't even know myschedule, you wanna check it out
(59:18):
on the QB Power Hour.
We have, the upcoming events andtopics there.
So we appreciate you joining ushere today, and we'll see you
next time on the QB Power.
Have a great day, everyone.
Popular Podcasts
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.
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
On Purpose with Jay Shetty
I’m Jay Shetty host of On Purpose the worlds #1 Mental Health podcast and I’m so grateful you found us. I started this podcast 5 years ago to invite you into conversations and workshops that are designed to help make you happier, healthier and more healed. I believe that when you (yes you) feel seen, heard and understood you’re able to deal with relationship struggles, work challenges and life’s ups and downs with more ease and grace. I interview experts, celebrities, thought leaders and athletes so that we can grow our mindset, build better habits and uncover a side of them we’ve never seen before. New episodes every Monday and Friday. Your support means the world to me and I don’t take it for granted — click the follow button and leave a review to help us spread the love with On Purpose. I can’t wait for you to listen to your first or 500th episode!