July 10, 2023 • 52 mins
For this episdoe we are joined by Eric Pesik, the Deputy General Counsel at Seagate Technology. Eric, walks us through how he has been using generative AI tools, including ChatGPT, AI image generators and AI voice overs to speed up his workflow when creating presentations for his colleagues. The crew also discusses how AI will broadly impact other sectors.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Eric Brown (00:05):
You're listening to the Audit presented by IT Audit
Labs.
Welcome to the Audit, a podcastby IT Audit Labs.
Today we're going to talkautomation and AI with Eric
Pesek, who's joining Scott andNick myself, and Josh is on with
(00:27):
us today too.
Josh is a music producer.
We're going to hear a littlebit about some AI in the music
producing world, and Eric is anattorney and is going to show us
a presentation that he workedon that was entirely done with
AI.
Eric Pesik (00:42):
So you know the way it came to be is, as we all knew
, there was a whole lot of pressright in January or so you know
, beginning of February, whereChatGPT 4 came out and people
were super excited and peoplewere talking about oh I asked
ChatGPT to do this or that forme.
And internally so I'm a lawyerand I work within a larger legal
(01:06):
team and one of the questionsthat somebody asked I say well,
will chat, will artificialintelligence, replace lawyers?
Because there's this.
There was this fear thatartificial intelligence is going
to replace programmers.
You know graphic designers, youknow all those different tasks,
and those are the tasks thatpeople usually attribute to.
You know, this takes differenttasks, and those are the tasks
that people usually attribute to.
You know this takes creativity.
(01:26):
These are the things that no AI, no computer is ever going to
replace.
And yet here we were in 2023saying you know your job as a
programmer, your job as agraphic designer, your job as an
artist is going to be replaced.
So within my team, somebody justrandomly said do you think the
AI will replace lawyers?
And so I thought, hey, I'll doa presentation, I'll do some
(01:52):
research on chat GPT, I'll puttogether a presentation online
like I would normally do.
You know normal PowerPointpresentation and we sometimes do
like a Friday brown bag or wehaven't done it a lot, but we'll
do these.
You know interdepartmentalpresentations, and people come.
I thought that's a great topicfor all the lawyers to say, hey,
is your job going to be maderedundant?
And people did.
Everybody started signing up andso I started drafting the
(02:14):
presentation and the way youmight do it.
Do some research on chat GPTand I got I don't know about
halfway done and I thought whyam I doing this?
I'll just ask chat GPT write apresentation for me on whether
AI will create, will replacelawyers.
And it did.
And it was, I don't know, likemaybe 70% aligned to what I was
(02:35):
already writing.
It didn't, it wasn't fullyfleshed out and there was a
couple of weak spots, and so inthose areas I said explain, why
X?
So the GPT responded withsomething that was sort of
superficial perhaps, and I justasked it to explain and then it
fleshed out various answers andso I thought, hey, this is great
(02:56):
, I'll use this as mypresentation.
And I started like I wouldnormally.
I started putting togetherslides to go with the script
that it had written and Ithought well, wait a minute.
Eric Brown (03:06):
Why don't.
Eric Pesik (03:06):
I have a different AI.
Do the graphics for me.
So I went to Dream Studio and Ijust took keywords from the
presentation that was written bychat, put them into Dream
Studio and said you know, draftnot draft, but you know, draw me
some art in this style.
And I think I picked differentstyles, but using these keywords
(03:28):
, and it came up with all theseimages and at that point I
thought, okay, we're gonna goall in.
I had already used for someother I forget some other
project that I had.
I'd already subscribed to an AItext to voice generator.
I thought I'm not even going topresent this in my own voice,
I'm just going to have AI do thetext, do the content.
(03:51):
I'll have AI do the images.
I'll have AI do the voice.
And then I started putting itinto PowerPoint.
And that's when I rememberedthat even PowerPoint has a I can
design this slide for youfeature.
And I thought, okay, this is aperfect demo, I'll put it all
together.
I used all four of those steps.
I used either AI or kind of a Idon't know what the PowerPoint
(04:16):
equivalent is, but thePowerPoint auto designer and
then I just recorded it and,instead of me giving the
presentation.
I gave the presentation to theteam as a pre-recorded video the
day before my presentation andat the presentation I just said,
hey, go watch this video first,come to the meeting and we'll
discuss it.
And it was one of the betterattended meetings that we've had
(04:38):
.
People loved it.
I kind of explained to themmore or less what I'm explaining
to you about how it came to be.
And, of course, you know we hada lot of lawyers on the call,
as you might imagine and youknow.
The bottom line that both Ihave and the AI came up with is
(05:00):
that we're probably not going toreplace lawyers at any time
soon, which made everybody feela little bit better, but it was,
you know, by the time I wasdone, I thought, hey, there are
a lot of applications you know,even this kind of tongue in
cheek type of applications whereyou could do this.
You could seriously do this.
Throw in a little bit of humanintervention to fine tune it
where the AI is, you know, justnot as strong and it could be a
really powerful tool to use forlawyers.
(05:22):
So a lot of what lawyers dolike.
If you're just doing legalresearch, you imagine it's like
any other research.
You're just researching stuffon different topics.
Typically there'll be like acase that sets a precedent and
you'll say, okay, well, what arethe cases that follow this?
Does that precedent still stand?
That's the kind of thing thatAI could probably do really well
.
That's a really boring part ofbeing a lawyer, but it's also
(05:46):
part where it takes somecreativity, because you're
looking at other scenarios wheresimilar facts might arise or
overlapping laws that youremember as a lawyer.
That might apply.
But the actual research and kindof writing the results of your
research is very tedious.
Like any other, you know, kindof writing a big report can be
(06:07):
very tedious, but that's whereAI could help lawyers and, of
course, anybody else who has towrite or has to come up with
some creative solution.
I thought this would be a great, almost a muse for you.
Right, it can do your firstdraft as a human.
You'll have to finish thatdraft.
But if you've ever had writer'sblock and you're thinking I
(06:31):
just can't get started, well,here's a quick shortcut Tell AI
to do it for you, let them getstarted, and then you're in the
mode of editing and correctingand fine-tuning and rewriting.
But at least you've got somecontent to help you get started,
and that was, I think, of allof our meeting we kind of came
(06:55):
up with.
That answer is that one is it'snot ready yet, it's close, but
even in the current state, itcould probably be used to do
really good first drafts ofthings that lawyers do today.
And so lawyers will probably bethe lawyers that are doing
those first drafts right now.
Yeah, their jobs might bethreatened, but they're the ones
that they could co-opt that byuse.
(07:16):
They're the ones that should beusing the AI today to help them
do their first draft beforethey submit it to their, you
know, their senior partner orwhatever it is, and that's
that's kind of the story of howthis came up.
And then, after I presented itto their senior partner or
whatever it is and that's kindof the story of how this came up
and then, after I presented itto my team, I just said, hey,
there's nothing confidential inhere, I'll just upload it to
YouTube and see if it gets anyviews.
And, like I said, I've goteight since February Not a lot,
but and probably two of those atleast two of those are my own,
(07:38):
I would imagine.
Eric Brown (07:39):
And you did voice.
Eric Pesik (07:47):
You had it speak in your own voice, didn't you?
Well, it's not my voice.
So I used it's just a, I wouldsay, a generic ai voice
generator and you can pickdifferent voices or different.
And in fact I could let me justshare my screen really quick,
sure?
Why don't I tell you what I'llstart at the very beginning.
So you know, here's chat gpt,here's dream studio that I use.
In fact, this is my backgroundright now.
I think if you but I have in mybackground, um, this is the
(08:07):
voice generator that I use andyou can see, hey, I want you can
pick different voices.
Hey, I want a young adult, Iwant a middle-aged man, you know
I want a, you know anothermiddle-aged man.
You have all these choices andit will in fact generate
something that kind ofapproximates.
You know different people andyou know it will try to, you
know, approximate differentaccents.
But what I did was so I wouldgo into chat, right, so I'd say,
(08:30):
write me a presentation and youjust pick about.
Eric Brown (08:35):
I mean, you guys pick a topic scott, where did
you used to, uh, be a mountainranger or climb or hike or
whatever it was you were doingout west?
Scott Rysdahl (08:45):
Yeah, how about?
Yeah, Yosemite National Park.
I like that it knows it's apresentation so it has to
introduce itself.
It's probably scraped everyTripAdvisor posting that's to
date.
Eric Pesik (09:01):
So it has a good training set, yeah, and so this
is more or less what I did, andlet's just say, suppose it, um
suppose.
So here is, in terms of safety,it's important to know that
wildlife note, be aware ofwildlife and follow park
regulations.
Suppose, in your presentationand this is the other thing I
did I said, hey, that's toosuperficial.
So I just said, hey, explainmore about safety, to get a
(09:27):
little more detail.
I don't know, maybe that wasenough, but maybe you wanted
more.
Let's get into some thingsWildlife safety, water safety,
road safety.
Scott Rysdahl (09:38):
Yeah, all very good tips that.
I have said myself as a rangerout there.
Eric Brown (09:44):
Anything, Scott, that comes to mind that we could
try asking it.
That maybe is more insiderinformation.
Scott Rysdahl (09:52):
Yeah, let's ask it how we get a hiking permit
for Half Dome.
They require those nowadays andthey can be very competitive.
Eric Pesik (10:02):
Nice apply for a permit.
So this is interesting.
One thing I remember about atleast ChatGPT is that it doesn't
search the internet immediately.
It's really just going on olderresults.
I know there's other chatbotsthat can also search and get
updated information.
I don't know how far back itgoes, but I assume this was at
least updated at one point goes,but uh, I assume this was at
(10:28):
least updated one at one point Ithink chat gpt was.
Eric Brown (10:29):
Its date is indexed from like september 2021.
Eric Pesik (10:31):
So this, this really is what I did.
I.
I just asked her to write apresentation where I wanted more
information.
I said explain.
And then I, you know, I went todream studio.
Well, first I tried to go rightright into PowerPoint and start
putting it in there, and I'vegot a PowerPoint instance up
somewhere, demo presentation.
So what I would do is I wouldgo to the visit and I'd say,
(10:53):
okay, well, here's my script,right?
So maybe I'd have a slide forthe first script, so I'd put it
in my speaking notes, right?
Good afternoon everybody.
I'd like to talk to you aboutvisiting yosemite national park.
And then I thought, remember, Isaid I, I thought I'd have ai
do my artwork, so maybe I wouldpick a keyword from here and
then I'd pop over to here andsay you know, my prompt is just
(11:16):
yosemite national park.
It seems like a good one.
And then now there's a couplemore options you can do, but
let's just let it.
Let it uh mull over that one isthis a free?
tool so this is a paid tool.
I've got 543 more credits andit costs ten dollars for a
thousand credits, I think.
So you can see here's how manycredits it's using 10 credits
(11:38):
out of my 500.
Um, let's scroll to the top.
So here's grab.
It's made some art for me, somaybe I might download one.
And then you know, this is I'mgoing to drop it in here as my
background.
And then, you know, becausethis is a presentation on
Yosemite National Park, we'llmake whoops, we'll send it back,
(11:59):
put that in front, and I, youknow, I did do a little bit of
of.
Oh, that's right, I forgot,I'll leave it.
Let me just leave it as is.
Then here's what we did.
Is we let the designer pick forme?
You can say, okay, well, here,let PowerPoint decide how this
(12:19):
is going to be displayed.
Here's a slide with the artworkdirect generated by AI, the
content that's going to bedisplayed.
Right, so here's a slide withthe artwork direct generated by
ai, the the content that's goingto be presented.
Um, maybe we can add this asthe subtitle here.
Right?
One of the most breathtaking,iconic national parks, united
states.
So I've already got my firstslide right.
And then I just kind of repeatedthis almost like assembly line.
(12:41):
So, like today, I'm doing itone by one.
But once I realized what I wasdoing is I just sort of set up a
like assembly line.
So, like today, I'm doing a oneby one.
But once I realized what I wasdoing is I just set up a little
assembly line process where Itook the whole text and I went
through one by one, generatingimages.
Let's just say this is going tobe our second slide, so I put
this in the text part of oursecond slide, and I pick
something.
What would be the topic?
Let's talk about why you shouldvisit.
That sounds good, why youshould visit, right, and then
(13:06):
why, okay, let's, we'll justdrop this in here, don't for
stunning view beauty.
We'll just say, okay, we'lljust turn these into bullet
points because we're lazy, right, and then we'll grab another
picture, and then again I couldtry to do something else.
Or I could just pick one ofthese and here's, you know,
here's some general.
This one looks pretty good,kind of got the camera in there.
A little the content there.
I could try to do somethingelse.
Or I could just pick one ofthese.
Here's some generative.
This one looks pretty good.
It's kind of got the camera inthere.
(13:27):
The content there.
You've got your script.
You've got at least your firsttwo slides right.
So here it is, a couple slidesall into it, and then you just
repeat the process, slide byslide.
Scott Rysdahl (13:39):
One thing that immediately comes to mind as a
security person is how this canbe used for social engineering
attacks.
So right now, something that'spretty common is sending an SMS,
say, to a bunch of employeessaying, hey, this is Eric Brown,
I'm in a meeting, I can't talk,but I really need you to go buy
me $500 in Best Buy gift cardsand enter them in manually to
(14:02):
this website or whatever.
It made me get a drone whenyou're over there at Best.
Buy yeah yeah, yeah, I'm runningout of drones, but if that
could be like a voicemaildelivered via email that's
generated in somebody's voicebased on some reasonable amount
of training data, it takes it toa whole other level, right?
All of a sudden it's yourboss's voice in your ear telling
(14:25):
you what to do, and I wouldguess that would be way more
successful.
Or even a Teams call right,we've got audio.
Eric Pesik (14:33):
I can't do the pass-through audio, but if I
figured that out, maybe I couldimpersonate Eric Brown right
here, live and say hey, I'mreaching out on Teams because
this is an emergency.
Please go again Best Buy orWalmart and pick up these cards.
Eric Brown (14:46):
I was reading somewhere that Microsoft is
working on an AI voice generatorthat it can train to sound like
somebody's natural voice withjust three seconds of sample.
It's too scary, yes.
Eric Pesik (15:02):
Yeah, so then you just need the deep fake voice,
the deep fake face, and your CEO, your CIO, is going to be
sending you off to Walmart again.
Nick Mellem (15:15):
One of the things that comes to my mind, not IT
related, but people and collegestudents are probably using this
technology as well to writepapers.
At what point do we, or what'syour thought on potential
plagiarism right?
If it's generating this, is itgiving a new paper every time,
or is it somewhat similar orcanned?
Eric Pesik (15:35):
So you know, just like me, sort of thinking hey,
we should use this to make yourfirst draft Right.
If I was a college student, or,heck, if I was giving you know
creation whether I'm a lawyerturning in a presentation, it
wouldn't take much to edit it,to make it your own.
It's for the really lazystudent who's just going to go
(15:55):
out like.
The same person who's justgoing to buy a research paper
online is not going to botherediting it.
I think it's probably prettyunethical.
That being said, I did readwhere they're trying to figure
out some way to watermark it.
The same way you mightwatermark a video or an image.
So if I go and do this imagethey could watermark.
(16:15):
I'm not sharing anymore but theycould watermark one of those
images, and I'd never tell Nowhow you watermark a text file.
What they were saying is thatthey might deliberately use
words that are not commonly usedand that would be sort of red
flags.
But I think that's tough though, because you think of, like, a
college student is probably alsotrying to deliberately use
(16:37):
words to make them sound moreprofessional or more scholastic
or whatever it is.
And I I seem to rememberreading an article about
somebody who was at least heclaimed was falsely identified
as having used ai to generatethe article, and he said hey,
I'll show you my research.
Um, but I don't, I don't know.
I didn't hear the rest of thestory, but they were talking
(16:58):
about how they were battlingwith a professor to say I've got
my research, I can prove that I.
That being said, obviously theygot pinged by something.
So how are you going to detectthis?
I have no clue Because, again,it's just a text file.
Eric Brown (17:13):
There is a.
Joshua Schmidt (17:15):
If I could just jump in.
Yeah, just, I'm aware ofWinston AI detection, which is
being used by educators andpublishers to detect AI-written
content up to ChatGPT-4.
So I think they're kind ofstaying along with this, you
know, to protect academia andthe amount of money coming into
(17:39):
those institutions.
I think they're kind of on topof this, looks like.
From a quick search, I've kindof heard that on a few other
YouTube videos as well.
Scott Rysdahl (17:49):
That said, I know you can, as Eric was doing.
You can take GPT's output andsay, hey, could you rewrite this
in the style of ErnestHemingway?
You know, and it'll do it, andI wonder how much like how many
iterations of that process itwould take to really fool some
of those you know likelyai-based detection tools, right?
Eric Pesik (18:11):
or even you know, do what you know google does for
google maps, right, they'llthrow in an error.
So you as a, as a as a student,you know, throw in some extra
errors that a student wouldlikely make and then say you
know, chat, gpt would not havemade this mistake.
I obviously made this mistake.
I apologize, but at least I canprove that I did it right.
Eric Brown (18:33):
Yep, I wonder, though, you know we're kind of
entering that era where thetechnology is taking a new path
and, rather than fightingagainst the technology, should
we be embracing it and lookingat higher operators that we
(18:54):
could work at as humans?
Because, you know, goingthrough college, you know back
in my day, to use thatcolloquialism you'd go to the
library, you'd research yourstuff right before the internet.
You'd find what you were goingto write on.
But really, what did thatreally teach me at the end of
the day, like, what value didthat bring me?
(19:16):
Maybe it brought value of, like.
You know, I could find somethingin the library, or I could know
how to take a collection ofinformation and put it together
in a paper.
Will that have any value forfuture generations where, back
in the day, we used to take therugs outside and beat them, but
now we have vacuum cleaner?
Right?
Is the actual research of thepaper going the way of the
(19:41):
beating of the rugs, and couldwe operate at a higher level as
humans?
Yeah, the technology is goingto be able to search terabytes
of data in nanoseconds andsynthesize those thoughts much
quicker than any human evercould, and then how the human
applies that information isprobably the real value of the
(20:02):
human brain versus the actualgrunt work of doing the research
.
Scott Rysdahl (20:06):
An analogy, there is something a friend told me
about how China's kind of techsector operates, and that's that
.
So China's a tech leader,obviously, especially in
manufacturing, but they didn'tdo a lot.
But they don't have a longhistory of research and
development in these highlytechnical areas.
So they beg, borrow, buy orsteal a lot of this intellectual
(20:28):
property and then they justplunk it down in their factories
and churn out the iPhones orwhatever right.
But if that tap got cut off andall of a sudden they didn't
have that sort of history, thatlegacy of research and
development and intellectualproperty, would they even be
able to continue functioning asa high-tech economy?
And I think there's goodarguments to be made for both
(20:49):
sides.
But I wonder if the same thingmight happen in this case, where
after a couple of generationsof people who didn't do the
research in the library, likeEric's saying, maybe we do sort
of lose the faculties to evencome up with you know, content
that's interesting to humans andit just all becomes kind of
recycled, regurgitated AI, youknow garbage.
Eric Pesik (21:14):
Well, you know, I think it works both ways, right?
So we have some people in myhouse doing some remodeling and
they're using nail guns right.
They're not using hammers right?
I wouldn't say we've abandonedhammers, but for most framing
jobs, you're going to beprimarily using a nail gun,
right?
We use that, we embrace thattechnology.
On the other hand, we're stillteaching children to do their
(21:37):
times tables in third grade, andwe're still teaching children
to do their times tables inthird grade.
And my mind still relies onthat, even though I have a
calculator, I have a computer.
But in ordinary conversations,this, but the next time somebody
(21:58):
asks you and you just need tosort of really quickly figure
out whether something is in thesame order of magnitude as
something else, having thatinnate knowledge of memorizing
your times tables or memorizingyour powers or whatever it is,
still comes in handy.
So there are things that westill want to keep doing, and
yet we don't make the framer guyuse a hammer, we say, and yet
we don't make the framer guy usea hammer.
We say you know, go and embracethe technology, use that
(22:21):
pneumatic nail gun.
So some things, I think, willyou know, ai will replace and it
would be rightfully so like thefirst draft, or, you know,
doing basic research, or maybeeven I don't know maybe there's
some low level sort of cognitivecreativity, stuff that AI can
come up with that we as humanscan then go take and build upon.
(22:43):
You know, I was.
I was just reading an articleearlier this week about some ceo
is complaining.
All my employees are workingfrom home.
They use, you know, chat botsto create their answers.
So I'm going to increase theworkload by 50 times or whatever
.
He said something ridiculousabout 56 times and my first
thought was you know why?
Do you care what tools they useto get their job done?
(23:04):
Right, you shouldn't punishthem for doing that.
But after a second thought Ithought but use that nail gun
example.
If people are using hammers andthen they switch to nail guns,
you're not going to say, hey,keep doing the same work at the
same speed that you were doingwhen you had a hammer at your
disposal.
When you're really using a nailgun, of course you're going to
expect your framers to get stuffup faster.
(23:25):
So for that CEO now he wasprobably way out of line to say,
hey, you're going to do 50times the work.
But for the CEO one is don'tobject to your employees using
AI.
Tell them, hey, I to youremployees using AI.
Tell them, hey, I can tellyou're doing so much more work.
That's wonderful.
I'm going to officiallyauthorize you to use that AI to
(23:46):
do that work.
Now will you get more work outof your employees?
Sure, but don't punish them bysaying oh, you know, 50 times
the work for you, you know, hitthem with the rug beater.
That Eric was talking about,right.
So on one hand you don't wantit to be punitive, but you know
it will take over and it willmake our jobs faster and
accordingly, you know thethreshold requirements will
(24:11):
probably creep up to match andthe nail gun does not know where
the nail needs to go.
Joshua Schmidt (24:14):
True.
Scott Rysdahl (24:14):
There's still a human in the center of the work.
Eric Pesik (24:18):
Yeah, that's the framer downstairs.
With like 30 years ofexperience of knowing, you know
how the building should beframed up and where to put that
nail and you know using thatnail gun as a tool to you know
implement it.
Eric Brown (24:31):
There's paperwork that we come across in our daily
lives and jobs that is not hardto fill out but sometimes
tedious, right, when you thinkof the proofreading that you
have to do or the multipleiterations that you have to go
(24:52):
to to generate this content forvery little return.
Right, if you're writingsomething that somebody's going
to look at once, or maybe never,and toss in a drawer, spending
intellectual capital on that.
It's not a great use of time.
So you know, eric, I like youridea and I've taken it and done
it in reverse, where I come upwith the initial concept you
(25:15):
know myself of.
Like you know, here are acouple of thoughts that I want
to put together and then useChatGPT to clean it up and write
it in a business, professionaltone or kind of whatever tone
you know would suit the audiencethat it's being delivered to,
and I've found that just inresponding to RFPs as an example
(25:36):
.
So an RFP a request for aproposal, which is just an
arduous mountain of paperwork,that it's kind of a rite of
passage where you've got to fillout all these long documents in
order to work with anorganization that already wants
you to work with them becausethey've invited you to
participate in the RFP.
You could spend dozens of hoursresponding to these RFPs, or
(26:01):
you could spend a little bitless time, frame up the idea
that you want to deliver.
Let ChatGPT generate theparagraphs of content and
copy-paste it in, which I foundto be a pretty valuable resource
, because you don't even know ifyou're going to get the
contract, and why am I putting30 hours into this thing if I
(26:22):
may not even get the work?
Eric Pesik (26:23):
Computers have always been good at doing
certain types of things, likerepetitive, thankless jobs, like
filling out an RFP, and Iremember reading one quote where
somebody's talking about whenthey said, hey, ai can now
generate art and replace allthese graphic designers.
And the quote was yeah, I don'tneed AI to replace all these
graphic designers.
Sure, it's great, but what Ineed is AI to fill out an
(26:45):
application for a passport or,you know, incorporate my job, my
resume, into your job boardwithout me having to retype it
again.
You know, again, let AI do allthe tedious filling out
paperwork forms, you know, forus, so that we can get back to
doing the real creative workthat we want to do.
And you look at your systems.
You probably have theminternally in all your systems.
(27:06):
There's a lot of bureaucracythat we human beings have to
wade through that maybe AI wouldbe really good at bypassing,
once they get to know the basicrequirements that is required by
your bureaucracy and also getto know you, to the extent that
AI can, so that it knows hey,this is all I need to meet this
requirement.
I can do that for you.
(27:27):
I'm an AI.
Scott Rysdahl (27:28):
I'm thinking income tax prep.
That is my first applicationfor.
Nick Mellem (27:32):
AI.
Oh my goodness.
Yes, that's probably the bestidea, yet.
Eric Pesik (27:37):
Yeah, I mean you pay a couple hundred bucks for
somebody to do that and it'smostly, it's just a ministerial
task.
You know, pick numbers fromhere.
Eric Brown (27:45):
That should be something AI should be great at
Right you could upload all thedocuments and then it could pull
it all together.
Eric Pesik (27:51):
Yeah, you know you'll get resistance, so that's
a good example.
I mean, the tax return businessis a perfect example.
Because they resist.
The IRS has offered hey, wewill make the easy form
available online for free.
Just go online, fill it out andgo, and we'll do that.
It's it's the tax preparers whohave prevented that from
(28:12):
happening, because they'reprotecting their jobs.
Same way, you know, every timea new technology is going to put
somebody out of business, therewill be people that are going
to fight it, not because it'sbad, just because they're trying
to protect their industry.
Joshua Schmidt (28:26):
They're going to need a great disclaimer for
questionable write-offs, so youdon't end up on an AI-generated
write-off list An AI audit.
Scott Rysdahl (28:36):
Yeah, that's an AI audit.
Eric Brown (28:44):
Or it could tell me why investing in all those meme
stocks that ultimately led toloss is a bad idea.
I was going to say they used tohave that burger flipping robot.
I think it was the KUKA robot,and there's a pretty cool video
of this KUKA robot that they usein manufacturing and it's
playing.
I think the person's name isTimo and he's a professional or
(29:08):
Olympic table tennis player andthe KUKA robot plays against him
and does really well.
So it's this robot holding apaddle playing against him and
does really well, like so.
You know, it's this robotholding a paddle playing against
him, and and that same kookarobot I think was used or there
(29:29):
was some trial use in fast foodrestaurants to do like dip the
fries in the oil, flip theburgers, assemble the burgers,
whatever it was.
But it could just be that thecost of entry for that it's
probably a half a million dollardevice that requires
maintenance and whatnot.
(29:49):
The barrier to entry is costwhere it's easier to have a
human do those things, to have ahuman do those things.
The ROI on it would be toogreat, even though that's
probably a perfect applicationfor AI.
Where you have somebody workingin this task, where they're
doing repetitive things andmaybe it's not the safest, where
(30:13):
you're working around hot oilor something.
Or maybe take a foundryapplication where you're working
with.
You know hot molten liquidsthat the robot would be much
better for than a human.
You wouldn't have to conditionthe environment for the human,
so to speak.
But you could probably operatethese environments differently
(30:34):
if they were designed withautomation from the ground up,
where you wouldn't need lightsor other.
You know air conditioning tosome extent, or heat to some
extent, um, where a lot of thesethings that are there for the
comfort of the human or just toaccommodate the way that they
have.
Eric Pesik (30:50):
They have to stand there, they have their arms
right, and so you get that, thatrobot that comes in and so it
uses those same physicalcharacteristics of a human.
But if you designed it outright, where you know, the burger
just maybe goes to this thingand there's a, you know, I don't
know how I'm just making it up,but like flips it, this way you
reduce the cost of having allthat articulation because you're
just trying to imitate thehuman instead of, you know,
(31:11):
automating the process, insteadof automating the person at the
alaska airlines lounge I thinkit's in Seattle, if I'm
recalling correctly.
Eric Brown (31:21):
but they have a pancake maker that the you know.
You say you want a pancake orwhatever in it.
The batter's in this pouch andthe pouch, kind of you know, the
batter goes out on thisconveyor belt type of system and
30 seconds later the pancakejust rolls off the end.
Eric Pesik (31:39):
And it's really the coolest thing One of my first
business trips overseas, I wentto Tokyo and at Tokyo Narita
Airport they have a beer pouringrobot and I don't know if
you've ever seen it where ittilts the glass to the side
right, pours it down because itputs a little foam, perfect foam
, and I remember thinking Idon't really need a beer, but I
need to use this.
I need to use this beer pouringrobot, right.
Scott Rysdahl (32:01):
Yeah, and that's maybe a good.
Two examples there of the linebetween where there's human
value add and where there's not.
With pancakes, I don't reallycare how it came about, I'm just
going to eat it.
It's going to be good, it'srefined carbs the best kind, but
with a bartender maybe there isstill that human value add, the
, the, the human to humancontact, the.
(32:22):
How do you like your drink madein a way that turns into a
conversation that ai maybe isn'tquite yet ready to, to pretend
to be?
Eric Pesik (32:31):
yeah, it's novelty, right, like I wanted to get the
beer because that was reallycool, but like, are you going to
hang out in a bar that justpours beers?
That way, I mean, maybe youwould, but like it's not going
to have the same feel as, likeyou know, a bartender that has,
you know, the feeling for do it,for doing that, I think you
know good, good idea.
Like we think of ai, like, okay, as ai progresses, people
(32:54):
compare it to like the iq of adog, the iq of a child, the iq
of I don't know a teenager,right?
So you're thinking of, okay,this is we're trying to improve
the iq of a dog, the iq of achild, the iq of I don't know a
teenager, right?
So you're thinking of, okay,this is we're trying to improve
the iq of ai, when maybe thereal threshold is the eq, right,
when you get that emotional,you can, you can, you know,
maybe it's not fake, but whereyou can somehow digitize that
emotional experience, wheresomebody recognizes you, they
(33:17):
are the AI is genuinely glad tosee you again.
The AI maybe is insulted orfeels a little bit miffed when
you cut them off short.
That's when you're going to getthat real passing, truly finally
passing, the Turing test, whichreminded me when they were
complaining about AI makingmistakes, what they call halluc
hallucinations right, like I wasthinking that's the most human
(33:40):
thing that it does.
Right, it makes a mistake andthen it, you know, can, like it
still gives you the answer andit acts confident in its mistake
and, yes, at some point, youknow, you say, well, a true
leader or a true professionalowns up to their mistakes and
you know and you learn from them, but a lot of humans don't.
(34:02):
And it's a very human thing todo is to make a mistake and then
be confident in your mistakeand double, you know, really
double down on that.
So it's, you know, very genuine, you know, sometimes tragic.
Look at the MyPillow guy, likehe must know that he's made a
mistake about the election andhere, you know, after he's been
having to pay what?
$5 million for, like, beingdisproved and yet he's done this
(34:22):
very human thing I'm going to,I'm going to double down on my
mistake.
Nick Mellem (34:26):
Well, in this whole AI thing is also what China has
been doing for a bit now.
Maybe I'm mistaken, but aren'tthey using it as a social
grading, right?
So if you want to buy a house,you want to get a new credit
card, you're trying to buy a car, they have all these cameras
all over the place that aregrading how your behavior is,
what you're posting on socialmedia, how you talk to somebody
on the phone, right?
(34:47):
So it's kind of the same thing.
Yeah, it is the same.
Yeah, that's you know that's awhole topic.
Eric Pesik (34:54):
Talk for hours on that.
But yeah, they're doing ittoday.
You know, grading people ontheir behavior within society.
How good of a citizen you are,according to the People's
Republic of China.
Nick Mellem (35:05):
Instead of how you've spent money in the past.
How likely are you to pay itback?
Did you?
Scott Rysdahl (35:13):
jaywalk yesterday .
How many party meetings haveyou been to?
Nick Mellem (35:19):
Exactly.
Scott Rysdahl (35:20):
I just dropped a link in the chat.
It's a New York Times articleabout actually Minnesota and
nursing homes here and howrobots and AI embedded in them
are starting to be used for careof the elderly.
So there's that EQ, thatemotional piece.
We can't pay enough people togo hang out with the elderly and
provide them company, so here'sa place that AI can come in and
(35:41):
fill, you know, literally emptychairs.
Eric Brown (35:45):
And Josh, as a music producer, you're seeing AI come
up in the creation space too,aren't you?
Joshua Schmidt (35:53):
Oh yeah, we just had a very lengthy Slack
conversation about that with myagency, my composing agency.
It got very emotional and a lotof opinions coming out about it
.
There are already AI-gener, aigenerated um scores, music
composition tools, um.
For example, one thing that II've been using for years I
(36:16):
don't know if this, this is onthe cusp of ai, I don't know if
this is technically ai um is adrum avatar where you know you
can put in the style of drummingthat you want, because drum
drumming is such a tedious taskin the studio and audio world.
It requires a lot of space forthe drum set, it requires a lot
of talent and a lot ofmicrophones, yada, yada.
(36:39):
So this avatar, you can put ina style, a tempo and it will
automatically generate a drumsection for you.
But that's even gotten crazierover the last few years, where
it humanizes it.
A lot of electronic drums areon what's called a quantized
grid, where everything's lockedinto a tempo grid.
(36:59):
Well, this will humanize it andkind of create little micro
millisecond mistakes where youcan rush the beat or slow the
beat down.
That's been going on for a verylong time, but now we're
getting into full-on scores.
I did see Google's working onsomething that you can put in.
I want a laid-back reggae songwith a male singer and it will
(37:22):
spit out a piece of music.
So the temperature is kind oflike how much time do we have?
Is it getting red hot?
Do we only have a few moreyears or is this pretty terrible
music to listen to still?
And you know, one thing I wasthinking about we're talking
about flipping the burgers isyou know, there's the nuance.
You know things can betechnically done correctly, but
(37:46):
you know, when you're cookingthey say you put your foot in it
.
Well, in music you put yoursoul in it, and I think there's
always going to be that humanelement until AI gets so
incredibly powerful that it canimbue that into what it's doing.
What's yet to be seen is ifthat's something that can be
programmed or can be made intoan algorithm.
(38:06):
Know the humans will enjoy tothe extent where it really moves
them, makes them cry.
You know, makes them feel thatkind of transcendental emotion
that you get from a really greatpiece of art or music.
Um, I did see that there was,you know, just recently a ai
generated photograph had won acontest.
(38:29):
I can't remember where that was.
So we're getting there with thevisuals.
My hot take was, like we havemaybe five years as composers
but then, like what you weresaying, eric, is maybe you're
kind of editing, maybe you'rebeing a curator of AI generated
(38:50):
ideas and, you know,incorporating that into some
real instruments.
So it just cuts down theworkload.
You know, because if you'retalking about doing an orchestra
or you know a live band, aseven piece jazz band, things
can get pretty expensive prettyquickly.
Things can get pretty expensivepretty quickly.
So if these systems can kind ofget to a place where the fun
(39:12):
part of hearing the music andkind of editing the music is
snapping fast, maybe they'll cutdown on some of the costs and
the workload.
Of course that gets into anotherconversation about where do the
copyrights lie.
If you have a DjangoReinhardt-sounding guitar part,
does Django Estate get a sliceof the pie?
(39:32):
If the pancake maker at theairport is making my daughter a
Mickey Mouse pancakes, doesDisney get a royalty from the
pancake making machine?
So that's kind of where I'mlooking at, you know.
That's where my interest liesand I think a lot of this will
be settled kind of on the legalbattlefield, so to speak.
Eric Pesik (39:53):
Um so, eric, I think you're on the front lines there
for a while, just like you,maybe five, ten years, I think a
lot of our jobs where wherepreviously felt pretty safe
music, you know law, where AI isproving that they can cut, take
a little piece of it.
But again, maybe it's.
Maybe it's just going to makeour jobs easier because it's
(40:16):
going to take the tedious partout of it.
And you know it's not going toreplace live music, at least not
anytime soon, until you havethe physical manifestation of AI
.
You know the people.
I'm sure there's other aspectsof it, the holograms.
I just got through a.
We're doing a negotiation andthere's teams of lawyers and
(40:36):
both sides have multiplebusiness people on there and
we're locked in rooms talking topeople.
And while I was thinking aboutthis AI thing, about whether AI
will replace lawyers, it's notgoing to replace that type of
negotiation.
It's not going to replace thattype of negotiation.
It's not going to replacepeople getting face-to-face and
saying I want you to do more ofthis and I'm not going to be
able to do that and we can'tcomply with this rule and my
(40:59):
lawyer is going to explain it toyou and while the AI might be
able to look up whether you'reright or wrong to what the law
says you're supposed to do.
It's not going to change thenegotiation between people, so
there's always going to be arule or a role, I should say,
for, you know, lawyers.
In that situation, just like,probably, for music generation,
(41:20):
as clever as ai is, you know,would ai, you know, turn back
the clock a couple decades,would ai have come up to say,
take a rock and roll, you knowdisco, and would have come up
with rap.
Would have come up with, youknow, the same type of you know
completely new music that cameout of.
You know just people playingaround with you know, jamming at
(41:43):
some point and you knowthrowing in something and you
know, like a new song.
Every now and then you'll heara new song.
You'll think, wow, that's justamazing.
Now will ai come up withsomething like that?
The way I was just readingabout a chess move, you know
they said ai was playing chessin a way that no human ever
would.
So ai will come up withsomething new music wise, but
still the same way.
A bunch of guys jam into somesong.
(42:04):
You know that they're writingon the fly.
Joshua Schmidt (42:07):
Uh, they'll still come up with something new
that ai would not have come upwith absolutely, I think it will
further make those sub genresinto fractals, so there'll be
even more mini genres andthere'll be more sub genres
within those genres.
It'll be up to us to decidewhat we want to consume, right?
But, um, you know, I thinkwhere kind of the the value will
(42:30):
be placed.
More will be the gatekeepersand the tastemakers because, um,
you can take, I had a friendthat took a beatles song and you
know the guitars were slightlyout of tune and he tuned each
guitar string so it wasperfectly in tune and kind of,
you know, fixed the problemswith this with the helter
(42:50):
skelter and he said it ended upsounding like a bon jovi song
and it just wasn't the same.
So I think a lot of that nuance,and then the taste and the
aesthetic, and how much does theuh, how much does the ai
culture integrate with the humanculture and where's where's
kind of the borderline of that?
That's kind of where we'reexploring right now.
Nick Mellem (43:11):
I think we're seeing the same thing with
photography.
One of my biggest hobbiesoutside of work is photography.
I've been basically my wholelife and I think, josh, what
you're saying it's the samething.
And that contest, I believe,was national geographic, because
I did see it was a picture of acheetah, if I'm not mistaken.
So.
But then there's the other part.
(43:34):
You take it right.
National Geographic or thesecompanies are spending all this
money to send somebody to do aremote place in Africa to take a
picture of that cheetah closingdown on a you know whatever
their prey.
So you've taken the danger outof it, the cost.
But at one point it's kind ofgoing back to the argument of
well, film was dying, thendigital came around, and then
there's people now that aregoing back to film because of
the art of it, and so now AI istaking over digital, so now
(43:57):
digital shooters are going tobecome the purists versus the AI
.
You're right.
So both your points, it'swhat's the balance?
Eric Brown (44:10):
what's the balance?
There's a show, I think it's,on Netflix called AlphaGo, or a
movie about the AlphaGo, whichis the game of Go, and then,
just like Deep Blue, which wasIBM's Watson computing system
that beat Kasparov a coupledecades ago, well, with AlphaGo
(44:33):
it played the best Go player inthe world, lee Sedol, who's like
a 9th Dawn player, and duringthe match it showed how the AI
made a move that humans justcouldn't conceive of.
And you think, well, from achess perspective or at least
(44:57):
this is how I thought of it was,a computer is able to calculate
chess moves much faster than ahuman can.
So you make one move thatcomputer can calculate moves
that it could make that have ahigh probability of winning.
And because the chess game issmaller than the Go board,
(45:17):
apparently the compute resourcesneeded to calculate the same
moves for the Go game are justnot possible today.
So their artificialintelligence has done something
else to to come up with the, the, how the alpha go works.
But anyway, it said that duringthe match.
(45:37):
Um, and you could see during thematch they showed the, the, the
footage of the match, where thehuman, lee sadoli, was sitting
across from another human whowas making the moves, but he
wasn't able to really read thatperson's emotions because that
person wasn't the one coming upwith the moves.
Um, and it was, you know, thecomputer that was doing it, and
you could see that it reallytook its toll on on lee over the
(46:00):
course of those those fourgames or so.
Um, but I think one of thecommenters made made the comment
at one point in time this movethat we saw could be like a 10th
dawn move, which is, you know,like there are no 10th dawn
humans, there's only, you know,I think nine is the theoretical
highest.
So that that was really kind ofcool to see this application
(46:23):
come to life.
And, you know, like, like youwere saying, josh or Eric, with
music where rap came about fromother musics that we've had,
potentially AI could inventsomething really cool that we've
not seen yet.
Eric Pesik (46:40):
Or heard.
Eric Brown (46:42):
Or heard.
Yeah, and Nick, was it a?
Did the AI just generate thatphoto of the cheetah, or was it?
Did it enhance it somehow?
Nick Mellem (46:54):
so I I was mistaken , I did look, there was national
geographic did create anarticle or a post that showed a
couple pictures of a cheetah tosee you could pick out which one
was ai versus actual, you know,photography or digital.
But there was a contest wherethe person that won and I put
the link in the chat where hewon the contest with his
(47:15):
generated picture and it'sreally an old looking you know
film, 30 old 35 millimeter film,look right from I don't know
when, a long time ago, with acouple, um, elderly females in
the picture here, and it lookslike it's old film that was
developed in a dark room.
Right, you can see this.
Imperfections and the ai builtall that in right, and when you
(47:38):
look at it at first glance maybeif you get right up onto it,
you can maybe tell it's a littletoo perfect.
You know to be real.
Right, it's generated, but atfirst glance it's shockingly
good.
Eric Pesik (47:52):
Because cheetahs don't have fingers.
Nick Mellem (47:54):
That's what AI hasn't mastered yet.
Eric Pesik (47:59):
Well, I know, even today, Dream Studio the one I
was using you can tell it whatlevel of detail you want and,
like when I was showing youearlier, there's a charge for
the computing power that you use.
So if you say, hey, I want youknow so much computing power
devoted to making thisphotorealistic picture, it will
do it.
It'll take longer, it'll costyou more, but you can have it do
(48:21):
that.
I doubt that the version thatI'm using online for, like I
said, a couple bucks here andthere, is going to give me
National Geographic levelresults.
But if you can do it, if I cando it to a certain level here,
I'm sure there's like a reallypro version you can do a
fantastic job with.
Eric Brown (48:39):
There's an AI image generator called Mid Journey
that you interact with overDiscord and I was looking into
using it for image generationand I saw someone had written an
article about using AI, chatgpt, to put together the
(49:03):
instruction set for the imagethat you want to generate
through mid-journey, becausethere's lots of different flags
that you can set and it's almostlike a 120-character-long
Google search where you turn onall of these flags to get
exactly what you want in thesearch.
But you could do the same thingwith the image generation of
(49:26):
tuning how you get it to look bysetting those flags and as a
human it might be hard toremember that right.
It's almost like generatingcode where you could then use
the AI to generate those flags,which I thought was kind of neat
.
Eric Pesik (49:39):
Yeah, it seems like it could be right, in line with,
I mean, like you said,analogous to generating code,
which is what people startedusing chat GPT at first.
Hey, can you give me a HelloWorld program for Rust?
I don't know.
Let it generate your code foryou, Right?
Eric Brown (49:59):
Yeah, could you convert this Java into assembly?
It should be crazy if it coulddo that.
Scott Rysdahl (50:06):
Turn this rock song into jazz right.
Joshua Schmidt (50:08):
Right yeah, turn this rock song into jazz, right
, right, yeah, when I one of theapplications I see happening
probably within the next fiveyears, is taking a piece of film
or a whole movie and uploadingthat into a score generator and
then having the ai spit out 30,50 different scores that are
perfectly timed to the edits ofthe video and choosing the mood,
(50:31):
and then you know, maybe youwant a John Williams type score
with Prince on guitar, with JohnBonham on the drums, you know,
I think.
And then you know there'sthere's some money to be made
there, though, in a way, becausethen you can get the John
Bonham pack, you know, or youcould pay.
You know, some of these peoplethat have these estates, I think
, will end up monetizing thosestyles and figuring out legally
(50:55):
how they can capitalize on those.
Eric Pesik (50:58):
those kinds of um signature sounds those will be
interesting lawsuits, right,like, okay, john bonham, or you
know any musician, you own this.
But do you own songs that soundlike it but aren't necessarily
copy.
Or do you own somebody whoplays the drums like you?
Right, so everybody can do thedouble bass nowadays.
(51:20):
You know, does he get a royaltyfor every?
You know everybody who doesthat with you.
Know, I'm sure there's there'ssome limit where you just sound
too much like John Bonham, butthat's going to be some fun
lawsuits.
So, hey, you know, job securityfor lawyers right, I think
you're set.
Joshua Schmidt (51:38):
You know, even John Bonham got sued for ripping
off Little Richard.
The introduction to rock androll is the introduction to
Little Richard's song.
So you know, all those Brit popguys stole a lot from the
African-American blues artists.
So it's just about how far backdo you want to go and how the
(51:59):
stuff all kind of ends uprolling out and how it kind of
changes and amalgamatesthroughout the years.
It's going to be quiteinteresting.
Nick Mellem (52:07):
Thank, you for having me have a great weekend
everyone.
Joshua Schmidt (52:09):
Yeah, thanks guys.
Yeah, have a good weekend.
Nick Mellem (52:11):
Bye.
Joshua Schmidt (52:11):
See you all.
Thanks, all right.
Eric Brown (52:16):
Want security leadership without the headcount
.
As an extension of the team, ITAudit Labs will provide the
experts to guide and counselyour company.
We will start by creating acustom security program that
caters to your industry whileproviding transparency and
remediation to improve cyberposture while reducing risk.
(52:38):
Contact IT Audit Labs to findout more.
You're listening to the Audit presented by IT Audit
Labs.
Welcome to the Audit, a podcastby IT Audit Labs.
Today we're going to talkautomation and AI with Eric
Pesek, who's joining Scott andNick myself, and Josh is on with
(00:27):
us today too.
Josh is a music producer.
We're going to hear a littlebit about some AI in the music
producing world, and Eric is anattorney and is going to show us
a presentation that he workedon that was entirely done with
AI.
Eric Pesik (00:42):
So you know the way it came to be is, as we all knew
, there was a whole lot of pressright in January or so you know
, beginning of February, whereChatGPT 4 came out and people
were super excited and peoplewere talking about oh I asked
ChatGPT to do this or that forme.
And internally so I'm a lawyerand I work within a larger legal
(01:06):
team and one of the questionsthat somebody asked I say well,
will chat, will artificialintelligence, replace lawyers?
Because there's this.
There was this fear thatartificial intelligence is going
to replace programmers.
You know graphic designers, youknow all those different tasks,
and those are the tasks thatpeople usually attribute to.
You know, this takes differenttasks, and those are the tasks
that people usually attribute to.
You know this takes creativity.
(01:26):
These are the things that no AI, no computer is ever going to
replace.
And yet here we were in 2023saying you know your job as a
programmer, your job as agraphic designer, your job as an
artist is going to be replaced.
So within my team, somebody justrandomly said do you think the
AI will replace lawyers?
And so I thought, hey, I'll doa presentation, I'll do some
(01:52):
research on chat GPT, I'll puttogether a presentation online
like I would normally do.
You know normal PowerPointpresentation and we sometimes do
like a Friday brown bag or wehaven't done it a lot, but we'll
do these.
You know interdepartmentalpresentations, and people come.
I thought that's a great topicfor all the lawyers to say, hey,
is your job going to be maderedundant?
And people did.
Everybody started signing up andso I started drafting the
(02:14):
presentation and the way youmight do it.
Do some research on chat GPTand I got I don't know about
halfway done and I thought whyam I doing this?
I'll just ask chat GPT write apresentation for me on whether
AI will create, will replacelawyers.
And it did.
And it was, I don't know, likemaybe 70% aligned to what I was
(02:35):
already writing.
It didn't, it wasn't fullyfleshed out and there was a
couple of weak spots, and so inthose areas I said explain, why
X?
So the GPT responded withsomething that was sort of
superficial perhaps, and I justasked it to explain and then it
fleshed out various answers andso I thought, hey, this is great
(02:56):
, I'll use this as mypresentation.
And I started like I wouldnormally.
I started putting togetherslides to go with the script
that it had written and Ithought well, wait a minute.
Eric Brown (03:06):
Why don't.
Eric Pesik (03:06):
I have a different AI.
Do the graphics for me.
So I went to Dream Studio and Ijust took keywords from the
presentation that was written bychat, put them into Dream
Studio and said you know, draftnot draft, but you know, draw me
some art in this style.
And I think I picked differentstyles, but using these keywords
(03:28):
, and it came up with all theseimages and at that point I
thought, okay, we're gonna goall in.
I had already used for someother I forget some other
project that I had.
I'd already subscribed to an AItext to voice generator.
I thought I'm not even going topresent this in my own voice,
I'm just going to have AI do thetext, do the content.
(03:51):
I'll have AI do the images.
I'll have AI do the voice.
And then I started putting itinto PowerPoint.
And that's when I rememberedthat even PowerPoint has a I can
design this slide for youfeature.
And I thought, okay, this is aperfect demo, I'll put it all
together.
I used all four of those steps.
I used either AI or kind of a Idon't know what the PowerPoint
(04:16):
equivalent is, but thePowerPoint auto designer and
then I just recorded it and,instead of me giving the
presentation.
I gave the presentation to theteam as a pre-recorded video the
day before my presentation andat the presentation I just said,
hey, go watch this video first,come to the meeting and we'll
discuss it.
And it was one of the betterattended meetings that we've had
(04:38):
.
People loved it.
I kind of explained to themmore or less what I'm explaining
to you about how it came to be.
And, of course, you know we hada lot of lawyers on the call,
as you might imagine and youknow.
The bottom line that both Ihave and the AI came up with is
(05:00):
that we're probably not going toreplace lawyers at any time
soon, which made everybody feela little bit better, but it was,
you know, by the time I wasdone, I thought, hey, there are
a lot of applications you know,even this kind of tongue in
cheek type of applications whereyou could do this.
You could seriously do this.
Throw in a little bit of humanintervention to fine tune it
where the AI is, you know, justnot as strong and it could be a
really powerful tool to use forlawyers.
(05:22):
So a lot of what lawyers dolike.
If you're just doing legalresearch, you imagine it's like
any other research.
You're just researching stuffon different topics.
Typically there'll be like acase that sets a precedent and
you'll say, okay, well, what arethe cases that follow this?
Does that precedent still stand?
That's the kind of thing thatAI could probably do really well
.
That's a really boring part ofbeing a lawyer, but it's also
(05:46):
part where it takes somecreativity, because you're
looking at other scenarios wheresimilar facts might arise or
overlapping laws that youremember as a lawyer.
That might apply.
But the actual research and kindof writing the results of your
research is very tedious.
Like any other, you know, kindof writing a big report can be
(06:07):
very tedious, but that's whereAI could help lawyers and, of
course, anybody else who has towrite or has to come up with
some creative solution.
I thought this would be a great, almost a muse for you.
Right, it can do your firstdraft as a human.
You'll have to finish thatdraft.
But if you've ever had writer'sblock and you're thinking I
(06:31):
just can't get started, well,here's a quick shortcut Tell AI
to do it for you, let them getstarted, and then you're in the
mode of editing and correctingand fine-tuning and rewriting.
But at least you've got somecontent to help you get started,
and that was, I think, of allof our meeting we kind of came
(06:55):
up with.
That answer is that one is it'snot ready yet, it's close, but
even in the current state, itcould probably be used to do
really good first drafts ofthings that lawyers do today.
And so lawyers will probably bethe lawyers that are doing
those first drafts right now.
Yeah, their jobs might bethreatened, but they're the ones
that they could co-opt that byuse.
(07:16):
They're the ones that should beusing the AI today to help them
do their first draft beforethey submit it to their, you
know, their senior partner orwhatever it is, and that's
that's kind of the story of howthis came up.
And then, after I presented itto their senior partner or
whatever it is and that's kindof the story of how this came up
and then, after I presented itto my team, I just said, hey,
there's nothing confidential inhere, I'll just upload it to
YouTube and see if it gets anyviews.
And, like I said, I've goteight since February Not a lot,
but and probably two of those atleast two of those are my own,
(07:38):
I would imagine.
Eric Brown (07:39):
And you did voice.
Eric Pesik (07:47):
You had it speak in your own voice, didn't you?
Well, it's not my voice.
So I used it's just a, I wouldsay, a generic ai voice
generator and you can pickdifferent voices or different.
And in fact I could let me justshare my screen really quick,
sure?
Why don't I tell you what I'llstart at the very beginning.
So you know, here's chat gpt,here's dream studio that I use.
In fact, this is my backgroundright now.
I think if you but I have in mybackground, um, this is the
(08:07):
voice generator that I use andyou can see, hey, I want you can
pick different voices.
Hey, I want a young adult, Iwant a middle-aged man, you know
I want a, you know anothermiddle-aged man.
You have all these choices andit will in fact generate
something that kind ofapproximates.
You know different people andyou know it will try to, you
know, approximate differentaccents.
But what I did was so I wouldgo into chat, right, so I'd say,
(08:30):
write me a presentation and youjust pick about.
Eric Brown (08:35):
I mean, you guys pick a topic scott, where did
you used to, uh, be a mountainranger or climb or hike or
whatever it was you were doingout west?
Scott Rysdahl (08:45):
Yeah, how about?
Yeah, Yosemite National Park.
I like that it knows it's apresentation so it has to
introduce itself.
It's probably scraped everyTripAdvisor posting that's to
date.
Eric Pesik (09:01):
So it has a good training set, yeah, and so this
is more or less what I did, andlet's just say, suppose it, um
suppose.
So here is, in terms of safety,it's important to know that
wildlife note, be aware ofwildlife and follow park
regulations.
Suppose, in your presentationand this is the other thing I
did I said, hey, that's toosuperficial.
So I just said, hey, explainmore about safety, to get a
(09:27):
little more detail.
I don't know, maybe that wasenough, but maybe you wanted
more.
Let's get into some thingsWildlife safety, water safety,
road safety.
Scott Rysdahl (09:38):
Yeah, all very good tips that.
I have said myself as a rangerout there.
Eric Brown (09:44):
Anything, Scott, that comes to mind that we could
try asking it.
That maybe is more insiderinformation.
Scott Rysdahl (09:52):
Yeah, let's ask it how we get a hiking permit
for Half Dome.
They require those nowadays andthey can be very competitive.
Eric Pesik (10:02):
Nice apply for a permit.
So this is interesting.
One thing I remember about atleast ChatGPT is that it doesn't
search the internet immediately.
It's really just going on olderresults.
I know there's other chatbotsthat can also search and get
updated information.
I don't know how far back itgoes, but I assume this was at
least updated at one point goes,but uh, I assume this was at
(10:28):
least updated one at one point Ithink chat gpt was.
Eric Brown (10:29):
Its date is indexed from like september 2021.
Eric Pesik (10:31):
So this, this really is what I did.
I.
I just asked her to write apresentation where I wanted more
information.
I said explain.
And then I, you know, I went todream studio.
Well, first I tried to go rightright into PowerPoint and start
putting it in there, and I'vegot a PowerPoint instance up
somewhere, demo presentation.
So what I would do is I wouldgo to the visit and I'd say,
(10:53):
okay, well, here's my script,right?
So maybe I'd have a slide forthe first script, so I'd put it
in my speaking notes, right?
Good afternoon everybody.
I'd like to talk to you aboutvisiting yosemite national park.
And then I thought, remember, Isaid I, I thought I'd have ai
do my artwork, so maybe I wouldpick a keyword from here and
then I'd pop over to here andsay you know, my prompt is just
(11:16):
yosemite national park.
It seems like a good one.
And then now there's a couplemore options you can do, but
let's just let it.
Let it uh mull over that one isthis a free?
tool so this is a paid tool.
I've got 543 more credits andit costs ten dollars for a
thousand credits, I think.
So you can see here's how manycredits it's using 10 credits
(11:38):
out of my 500.
Um, let's scroll to the top.
So here's grab.
It's made some art for me, somaybe I might download one.
And then you know, this is I'mgoing to drop it in here as my
background.
And then, you know, becausethis is a presentation on
Yosemite National Park, we'llmake whoops, we'll send it back,
(11:59):
put that in front, and I, youknow, I did do a little bit of
of.
Oh, that's right, I forgot,I'll leave it.
Let me just leave it as is.
Then here's what we did.
Is we let the designer pick forme?
You can say, okay, well, here,let PowerPoint decide how this
(12:19):
is going to be displayed.
Here's a slide with the artworkdirect generated by AI, the
content that's going to bedisplayed.
Right, so here's a slide withthe artwork direct generated by
ai, the the content that's goingto be presented.
Um, maybe we can add this asthe subtitle here.
Right?
One of the most breathtaking,iconic national parks, united
states.
So I've already got my firstslide right.
And then I just kind of repeatedthis almost like assembly line.
(12:41):
So, like today, I'm doing itone by one.
But once I realized what I wasdoing is I just sort of set up a
like assembly line.
So, like today, I'm doing a oneby one.
But once I realized what I wasdoing is I just set up a little
assembly line process where Itook the whole text and I went
through one by one, generatingimages.
Let's just say this is going tobe our second slide, so I put
this in the text part of oursecond slide, and I pick
something.
What would be the topic?
Let's talk about why you shouldvisit.
That sounds good, why youshould visit, right, and then
(13:06):
why, okay, let's, we'll justdrop this in here, don't for
stunning view beauty.
We'll just say, okay, we'lljust turn these into bullet
points because we're lazy, right, and then we'll grab another
picture, and then again I couldtry to do something else.
Or I could just pick one ofthese and here's, you know,
here's some general.
This one looks pretty good,kind of got the camera in there.
A little the content there.
I could try to do somethingelse.
Or I could just pick one ofthese.
Here's some generative.
This one looks pretty good.
It's kind of got the camera inthere.
(13:27):
The content there.
You've got your script.
You've got at least your firsttwo slides right.
So here it is, a couple slidesall into it, and then you just
repeat the process, slide byslide.
Scott Rysdahl (13:39):
One thing that immediately comes to mind as a
security person is how this canbe used for social engineering
attacks.
So right now, something that'spretty common is sending an SMS,
say, to a bunch of employeessaying, hey, this is Eric Brown,
I'm in a meeting, I can't talk,but I really need you to go buy
me $500 in Best Buy gift cardsand enter them in manually to
(14:02):
this website or whatever.
It made me get a drone whenyou're over there at Best.
Buy yeah yeah, yeah, I'm runningout of drones, but if that
could be like a voicemaildelivered via email that's
generated in somebody's voicebased on some reasonable amount
of training data, it takes it toa whole other level, right?
All of a sudden it's yourboss's voice in your ear telling
(14:25):
you what to do, and I wouldguess that would be way more
successful.
Or even a Teams call right,we've got audio.
Eric Pesik (14:33):
I can't do the pass-through audio, but if I
figured that out, maybe I couldimpersonate Eric Brown right
here, live and say hey, I'mreaching out on Teams because
this is an emergency.
Please go again Best Buy orWalmart and pick up these cards.
Eric Brown (14:46):
I was reading somewhere that Microsoft is
working on an AI voice generatorthat it can train to sound like
somebody's natural voice withjust three seconds of sample.
It's too scary, yes.
Eric Pesik (15:02):
Yeah, so then you just need the deep fake voice,
the deep fake face, and your CEO, your CIO, is going to be
sending you off to Walmart again.
Nick Mellem (15:15):
One of the things that comes to my mind, not IT
related, but people and collegestudents are probably using this
technology as well to writepapers.
At what point do we, or what'syour thought on potential
plagiarism right?
If it's generating this, is itgiving a new paper every time,
or is it somewhat similar orcanned?
Eric Pesik (15:35):
So you know, just like me, sort of thinking hey,
we should use this to make yourfirst draft Right.
If I was a college student, or,heck, if I was giving you know
creation whether I'm a lawyerturning in a presentation, it
wouldn't take much to edit it,to make it your own.
It's for the really lazystudent who's just going to go
(15:55):
out like.
The same person who's justgoing to buy a research paper
online is not going to botherediting it.
I think it's probably prettyunethical.
That being said, I did readwhere they're trying to figure
out some way to watermark it.
The same way you mightwatermark a video or an image.
So if I go and do this imagethey could watermark.
(16:15):
I'm not sharing anymore but theycould watermark one of those
images, and I'd never tell Nowhow you watermark a text file.
What they were saying is thatthey might deliberately use
words that are not commonly usedand that would be sort of red
flags.
But I think that's tough though, because you think of, like, a
college student is probably alsotrying to deliberately use
(16:37):
words to make them sound moreprofessional or more scholastic
or whatever it is.
And I I seem to rememberreading an article about
somebody who was at least heclaimed was falsely identified
as having used ai to generatethe article, and he said hey,
I'll show you my research.
Um, but I don't, I don't know.
I didn't hear the rest of thestory, but they were talking
(16:58):
about how they were battlingwith a professor to say I've got
my research, I can prove that I.
That being said, obviously theygot pinged by something.
So how are you going to detectthis?
I have no clue Because, again,it's just a text file.
Eric Brown (17:13):
There is a.
Joshua Schmidt (17:15):
If I could just jump in.
Yeah, just, I'm aware ofWinston AI detection, which is
being used by educators andpublishers to detect AI-written
content up to ChatGPT-4.
So I think they're kind ofstaying along with this, you
know, to protect academia andthe amount of money coming into
(17:39):
those institutions.
I think they're kind of on topof this, looks like.
From a quick search, I've kindof heard that on a few other
YouTube videos as well.
Scott Rysdahl (17:49):
That said, I know you can, as Eric was doing.
You can take GPT's output andsay, hey, could you rewrite this
in the style of ErnestHemingway?
You know, and it'll do it, andI wonder how much like how many
iterations of that process itwould take to really fool some
of those you know likelyai-based detection tools, right?
Eric Pesik (18:11):
or even you know, do what you know google does for
google maps, right, they'llthrow in an error.
So you as a, as a as a student,you know, throw in some extra
errors that a student wouldlikely make and then say you
know, chat, gpt would not havemade this mistake.
I obviously made this mistake.
I apologize, but at least I canprove that I did it right.
Eric Brown (18:33):
Yep, I wonder, though, you know we're kind of
entering that era where thetechnology is taking a new path
and, rather than fightingagainst the technology, should
we be embracing it and lookingat higher operators that we
(18:54):
could work at as humans?
Because, you know, goingthrough college, you know back
in my day, to use thatcolloquialism you'd go to the
library, you'd research yourstuff right before the internet.
You'd find what you were goingto write on.
But really, what did thatreally teach me at the end of
the day, like, what value didthat bring me?
(19:16):
Maybe it brought value of, like.
You know, I could find somethingin the library, or I could know
how to take a collection ofinformation and put it together
in a paper.
Will that have any value forfuture generations where, back
in the day, we used to take therugs outside and beat them, but
now we have vacuum cleaner?
Right?
Is the actual research of thepaper going the way of the
(19:41):
beating of the rugs, and couldwe operate at a higher level as
humans?
Yeah, the technology is goingto be able to search terabytes
of data in nanoseconds andsynthesize those thoughts much
quicker than any human evercould, and then how the human
applies that information isprobably the real value of the
(20:02):
human brain versus the actualgrunt work of doing the research
.
Scott Rysdahl (20:06):
An analogy, there is something a friend told me
about how China's kind of techsector operates, and that's that
.
So China's a tech leader,obviously, especially in
manufacturing, but they didn'tdo a lot.
But they don't have a longhistory of research and
development in these highlytechnical areas.
So they beg, borrow, buy orsteal a lot of this intellectual
(20:28):
property and then they justplunk it down in their factories
and churn out the iPhones orwhatever right.
But if that tap got cut off andall of a sudden they didn't
have that sort of history, thatlegacy of research and
development and intellectualproperty, would they even be
able to continue functioning asa high-tech economy?
And I think there's goodarguments to be made for both
(20:49):
sides.
But I wonder if the same thingmight happen in this case, where
after a couple of generationsof people who didn't do the
research in the library, likeEric's saying, maybe we do sort
of lose the faculties to evencome up with you know, content
that's interesting to humans andit just all becomes kind of
recycled, regurgitated AI, youknow garbage.
Eric Pesik (21:14):
Well, you know, I think it works both ways, right?
So we have some people in myhouse doing some remodeling and
they're using nail guns right.
They're not using hammers right?
I wouldn't say we've abandonedhammers, but for most framing
jobs, you're going to beprimarily using a nail gun,
right?
We use that, we embrace thattechnology.
On the other hand, we're stillteaching children to do their
(21:37):
times tables in third grade, andwe're still teaching children
to do their times tables inthird grade.
And my mind still relies onthat, even though I have a
calculator, I have a computer.
But in ordinary conversations,this, but the next time somebody
(21:58):
asks you and you just need tosort of really quickly figure
out whether something is in thesame order of magnitude as
something else, having thatinnate knowledge of memorizing
your times tables or memorizingyour powers or whatever it is,
still comes in handy.
So there are things that westill want to keep doing, and
yet we don't make the framer guyuse a hammer, we say, and yet
we don't make the framer guy usea hammer.
We say you know, go and embracethe technology, use that
(22:21):
pneumatic nail gun.
So some things, I think, willyou know, ai will replace and it
would be rightfully so like thefirst draft, or, you know,
doing basic research, or maybeeven I don't know maybe there's
some low level sort of cognitivecreativity, stuff that AI can
come up with that we as humanscan then go take and build upon.
(22:43):
You know, I was.
I was just reading an articleearlier this week about some ceo
is complaining.
All my employees are workingfrom home.
They use, you know, chat botsto create their answers.
So I'm going to increase theworkload by 50 times or whatever
.
He said something ridiculousabout 56 times and my first
thought was you know why?
Do you care what tools they useto get their job done?
(23:04):
Right, you shouldn't punishthem for doing that.
But after a second thought Ithought but use that nail gun
example.
If people are using hammers andthen they switch to nail guns,
you're not going to say, hey,keep doing the same work at the
same speed that you were doingwhen you had a hammer at your
disposal.
When you're really using a nailgun, of course you're going to
expect your framers to get stuffup faster.
(23:25):
So for that CEO now he wasprobably way out of line to say,
hey, you're going to do 50times the work.
But for the CEO one is don'tobject to your employees using
AI.
Tell them, hey, I to youremployees using AI.
Tell them, hey, I can tellyou're doing so much more work.
That's wonderful.
I'm going to officiallyauthorize you to use that AI to
(23:46):
do that work.
Now will you get more work outof your employees?
Sure, but don't punish them bysaying oh, you know, 50 times
the work for you, you know, hitthem with the rug beater.
That Eric was talking about,right.
So on one hand you don't wantit to be punitive, but you know
it will take over and it willmake our jobs faster and
accordingly, you know thethreshold requirements will
(24:11):
probably creep up to match andthe nail gun does not know where
the nail needs to go.
Joshua Schmidt (24:14):
True.
Scott Rysdahl (24:14):
There's still a human in the center of the work.
Eric Pesik (24:18):
Yeah, that's the framer downstairs.
With like 30 years ofexperience of knowing, you know
how the building should beframed up and where to put that
nail and you know using thatnail gun as a tool to you know
implement it.
Eric Brown (24:31):
There's paperwork that we come across in our daily
lives and jobs that is not hardto fill out but sometimes
tedious, right, when you thinkof the proofreading that you
have to do or the multipleiterations that you have to go
(24:52):
to to generate this content forvery little return.
Right, if you're writingsomething that somebody's going
to look at once, or maybe never,and toss in a drawer, spending
intellectual capital on that.
It's not a great use of time.
So you know, eric, I like youridea and I've taken it and done
it in reverse, where I come upwith the initial concept you
(25:15):
know myself of.
Like you know, here are acouple of thoughts that I want
to put together and then useChatGPT to clean it up and write
it in a business, professionaltone or kind of whatever tone
you know would suit the audiencethat it's being delivered to,
and I've found that just inresponding to RFPs as an example
(25:36):
.
So an RFP a request for aproposal, which is just an
arduous mountain of paperwork,that it's kind of a rite of
passage where you've got to fillout all these long documents in
order to work with anorganization that already wants
you to work with them becausethey've invited you to
participate in the RFP.
You could spend dozens of hoursresponding to these RFPs, or
(26:01):
you could spend a little bitless time, frame up the idea
that you want to deliver.
Let ChatGPT generate theparagraphs of content and
copy-paste it in, which I foundto be a pretty valuable resource
, because you don't even know ifyou're going to get the
contract, and why am I putting30 hours into this thing if I
(26:22):
may not even get the work?
Eric Pesik (26:23):
Computers have always been good at doing
certain types of things, likerepetitive, thankless jobs, like
filling out an RFP, and Iremember reading one quote where
somebody's talking about whenthey said, hey, ai can now
generate art and replace allthese graphic designers.
And the quote was yeah, I don'tneed AI to replace all these
graphic designers.
Sure, it's great, but what Ineed is AI to fill out an
(26:45):
application for a passport or,you know, incorporate my job, my
resume, into your job boardwithout me having to retype it
again.
You know, again, let AI do allthe tedious filling out
paperwork forms, you know, forus, so that we can get back to
doing the real creative workthat we want to do.
And you look at your systems.
You probably have theminternally in all your systems.
(27:06):
There's a lot of bureaucracythat we human beings have to
wade through that maybe AI wouldbe really good at bypassing,
once they get to know the basicrequirements that is required by
your bureaucracy and also getto know you, to the extent that
AI can, so that it knows hey,this is all I need to meet this
requirement.
I can do that for you.
(27:27):
I'm an AI.
Scott Rysdahl (27:28):
I'm thinking income tax prep.
That is my first applicationfor.
Nick Mellem (27:32):
AI.
Oh my goodness.
Yes, that's probably the bestidea, yet.
Eric Pesik (27:37):
Yeah, I mean you pay a couple hundred bucks for
somebody to do that and it'smostly, it's just a ministerial
task.
You know, pick numbers fromhere.
Eric Brown (27:45):
That should be something AI should be great at
Right you could upload all thedocuments and then it could pull
it all together.
Eric Pesik (27:51):
Yeah, you know you'll get resistance, so that's
a good example.
I mean, the tax return businessis a perfect example.
Because they resist.
The IRS has offered hey, wewill make the easy form
available online for free.
Just go online, fill it out andgo, and we'll do that.
It's it's the tax preparers whohave prevented that from
(28:12):
happening, because they'reprotecting their jobs.
Same way, you know, every timea new technology is going to put
somebody out of business, therewill be people that are going
to fight it, not because it'sbad, just because they're trying
to protect their industry.
Joshua Schmidt (28:26):
They're going to need a great disclaimer for
questionable write-offs, so youdon't end up on an AI-generated
write-off list An AI audit.
Scott Rysdahl (28:36):
Yeah, that's an AI audit.
Eric Brown (28:44):
Or it could tell me why investing in all those meme
stocks that ultimately led toloss is a bad idea.
I was going to say they used tohave that burger flipping robot.
I think it was the KUKA robot,and there's a pretty cool video
of this KUKA robot that they usein manufacturing and it's
playing.
I think the person's name isTimo and he's a professional or
(29:08):
Olympic table tennis player andthe KUKA robot plays against him
and does really well.
So it's this robot holding apaddle playing against him and
does really well, like so.
You know, it's this robotholding a paddle playing against
him, and and that same kookarobot I think was used or there
(29:29):
was some trial use in fast foodrestaurants to do like dip the
fries in the oil, flip theburgers, assemble the burgers,
whatever it was.
But it could just be that thecost of entry for that it's
probably a half a million dollardevice that requires
maintenance and whatnot.
(29:49):
The barrier to entry is costwhere it's easier to have a
human do those things, to have ahuman do those things.
The ROI on it would be toogreat, even though that's
probably a perfect applicationfor AI.
Where you have somebody workingin this task, where they're
doing repetitive things andmaybe it's not the safest, where
(30:13):
you're working around hot oilor something.
Or maybe take a foundryapplication where you're working
with.
You know hot molten liquidsthat the robot would be much
better for than a human.
You wouldn't have to conditionthe environment for the human,
so to speak.
But you could probably operatethese environments differently
(30:34):
if they were designed withautomation from the ground up,
where you wouldn't need lightsor other.
You know air conditioning tosome extent, or heat to some
extent, um, where a lot of thesethings that are there for the
comfort of the human or just toaccommodate the way that they
have.
Eric Pesik (30:50):
They have to stand there, they have their arms
right, and so you get that, thatrobot that comes in and so it
uses those same physicalcharacteristics of a human.
But if you designed it outright, where you know, the burger
just maybe goes to this thingand there's a, you know, I don't
know how I'm just making it up,but like flips it, this way you
reduce the cost of having allthat articulation because you're
just trying to imitate thehuman instead of, you know,
(31:11):
automating the process, insteadof automating the person at the
alaska airlines lounge I thinkit's in Seattle, if I'm
recalling correctly.
Eric Brown (31:21):
but they have a pancake maker that the you know.
You say you want a pancake orwhatever in it.
The batter's in this pouch andthe pouch, kind of you know, the
batter goes out on thisconveyor belt type of system and
30 seconds later the pancakejust rolls off the end.
Eric Pesik (31:39):
And it's really the coolest thing One of my first
business trips overseas, I wentto Tokyo and at Tokyo Narita
Airport they have a beer pouringrobot and I don't know if
you've ever seen it where ittilts the glass to the side
right, pours it down because itputs a little foam, perfect foam
, and I remember thinking Idon't really need a beer, but I
need to use this.
I need to use this beer pouringrobot, right.
Scott Rysdahl (32:01):
Yeah, and that's maybe a good.
Two examples there of the linebetween where there's human
value add and where there's not.
With pancakes, I don't reallycare how it came about, I'm just
going to eat it.
It's going to be good, it'srefined carbs the best kind, but
with a bartender maybe there isstill that human value add, the
, the, the human to humancontact, the.
(32:22):
How do you like your drink madein a way that turns into a
conversation that ai maybe isn'tquite yet ready to, to pretend
to be?
Eric Pesik (32:31):
yeah, it's novelty, right, like I wanted to get the
beer because that was reallycool, but like, are you going to
hang out in a bar that justpours beers?
That way, I mean, maybe youwould, but like it's not going
to have the same feel as, likeyou know, a bartender that has,
you know, the feeling for do it,for doing that, I think you
know good, good idea.
Like we think of ai, like, okay, as ai progresses, people
(32:54):
compare it to like the iq of adog, the iq of a child, the iq
of I don't know a teenager,right?
So you're thinking of, okay,this is we're trying to improve
the iq of a dog, the iq of achild, the iq of I don't know a
teenager, right?
So you're thinking of, okay,this is we're trying to improve
the iq of ai, when maybe thereal threshold is the eq, right,
when you get that emotional,you can, you can, you know,
maybe it's not fake, but whereyou can somehow digitize that
emotional experience, wheresomebody recognizes you, they
(33:17):
are the AI is genuinely glad tosee you again.
The AI maybe is insulted orfeels a little bit miffed when
you cut them off short.
That's when you're going to getthat real passing, truly finally
passing, the Turing test, whichreminded me when they were
complaining about AI makingmistakes, what they call halluc
hallucinations right, like I wasthinking that's the most human
(33:40):
thing that it does.
Right, it makes a mistake andthen it, you know, can, like it
still gives you the answer andit acts confident in its mistake
and, yes, at some point, youknow, you say, well, a true
leader or a true professionalowns up to their mistakes and
you know and you learn from them, but a lot of humans don't.
(34:02):
And it's a very human thing todo is to make a mistake and then
be confident in your mistakeand double, you know, really
double down on that.
So it's, you know, very genuine, you know, sometimes tragic.
Look at the MyPillow guy, likehe must know that he's made a
mistake about the election andhere, you know, after he's been
having to pay what?
$5 million for, like, beingdisproved and yet he's done this
(34:22):
very human thing I'm going to,I'm going to double down on my
mistake.
Nick Mellem (34:26):
Well, in this whole AI thing is also what China has
been doing for a bit now.
Maybe I'm mistaken, but aren'tthey using it as a social
grading, right?
So if you want to buy a house,you want to get a new credit
card, you're trying to buy a car, they have all these cameras
all over the place that aregrading how your behavior is,
what you're posting on socialmedia, how you talk to somebody
on the phone, right?
(34:47):
So it's kind of the same thing.
Yeah, it is the same.
Yeah, that's you know that's awhole topic.
Eric Pesik (34:54):
Talk for hours on that.
But yeah, they're doing ittoday.
You know, grading people ontheir behavior within society.
How good of a citizen you are,according to the People's
Republic of China.
Nick Mellem (35:05):
Instead of how you've spent money in the past.
How likely are you to pay itback?
Did you?
Scott Rysdahl (35:13):
jaywalk yesterday .
How many party meetings haveyou been to?
Nick Mellem (35:19):
Exactly.
Scott Rysdahl (35:20):
I just dropped a link in the chat.
It's a New York Times articleabout actually Minnesota and
nursing homes here and howrobots and AI embedded in them
are starting to be used for careof the elderly.
So there's that EQ, thatemotional piece.
We can't pay enough people togo hang out with the elderly and
provide them company, so here'sa place that AI can come in and
(35:41):
fill, you know, literally emptychairs.
Eric Brown (35:45):
And Josh, as a music producer, you're seeing AI come
up in the creation space too,aren't you?
Joshua Schmidt (35:53):
Oh yeah, we just had a very lengthy Slack
conversation about that with myagency, my composing agency.
It got very emotional and a lotof opinions coming out about it
.
There are already AI-gener, aigenerated um scores, music
composition tools, um.
For example, one thing that II've been using for years I
(36:16):
don't know if this, this is onthe cusp of ai, I don't know if
this is technically ai um is adrum avatar where you know you
can put in the style of drummingthat you want, because drum
drumming is such a tedious taskin the studio and audio world.
It requires a lot of space forthe drum set, it requires a lot
of talent and a lot ofmicrophones, yada, yada.
(36:39):
So this avatar, you can put ina style, a tempo and it will
automatically generate a drumsection for you.
But that's even gotten crazierover the last few years, where
it humanizes it.
A lot of electronic drums areon what's called a quantized
grid, where everything's lockedinto a tempo grid.
(36:59):
Well, this will humanize it andkind of create little micro
millisecond mistakes where youcan rush the beat or slow the
beat down.
That's been going on for a verylong time, but now we're
getting into full-on scores.
I did see Google's working onsomething that you can put in.
I want a laid-back reggae songwith a male singer and it will
(37:22):
spit out a piece of music.
So the temperature is kind oflike how much time do we have?
Is it getting red hot?
Do we only have a few moreyears or is this pretty terrible
music to listen to still?
And you know, one thing I wasthinking about we're talking
about flipping the burgers isyou know, there's the nuance.
You know things can betechnically done correctly, but
(37:46):
you know, when you're cookingthey say you put your foot in it
.
Well, in music you put yoursoul in it, and I think there's
always going to be that humanelement until AI gets so
incredibly powerful that it canimbue that into what it's doing.
What's yet to be seen is ifthat's something that can be
programmed or can be made intoan algorithm.
(38:06):
Know the humans will enjoy tothe extent where it really moves
them, makes them cry.
You know, makes them feel thatkind of transcendental emotion
that you get from a really greatpiece of art or music.
Um, I did see that there was,you know, just recently a ai
generated photograph had won acontest.
(38:29):
I can't remember where that was.
So we're getting there with thevisuals.
My hot take was, like we havemaybe five years as composers
but then, like what you weresaying, eric, is maybe you're
kind of editing, maybe you'rebeing a curator of AI generated
(38:50):
ideas and, you know,incorporating that into some
real instruments.
So it just cuts down theworkload.
You know, because if you'retalking about doing an orchestra
or you know a live band, aseven piece jazz band, things
can get pretty expensive prettyquickly.
Things can get pretty expensivepretty quickly.
So if these systems can kind ofget to a place where the fun
(39:12):
part of hearing the music andkind of editing the music is
snapping fast, maybe they'll cutdown on some of the costs and
the workload.
Of course that gets into anotherconversation about where do the
copyrights lie.
If you have a DjangoReinhardt-sounding guitar part,
does Django Estate get a sliceof the pie?
(39:32):
If the pancake maker at theairport is making my daughter a
Mickey Mouse pancakes, doesDisney get a royalty from the
pancake making machine?
So that's kind of where I'mlooking at, you know.
That's where my interest liesand I think a lot of this will
be settled kind of on the legalbattlefield, so to speak.
Eric Pesik (39:53):
Um so, eric, I think you're on the front lines there
for a while, just like you,maybe five, ten years, I think a
lot of our jobs where wherepreviously felt pretty safe
music, you know law, where AI isproving that they can cut, take
a little piece of it.
But again, maybe it's.
Maybe it's just going to makeour jobs easier because it's
(40:16):
going to take the tedious partout of it.
And you know it's not going toreplace live music, at least not
anytime soon, until you havethe physical manifestation of AI
.
You know the people.
I'm sure there's other aspectsof it, the holograms.
I just got through a.
We're doing a negotiation andthere's teams of lawyers and
(40:36):
both sides have multiplebusiness people on there and
we're locked in rooms talking topeople.
And while I was thinking aboutthis AI thing, about whether AI
will replace lawyers, it's notgoing to replace that type of
negotiation.
It's not going to replace thattype of negotiation.
It's not going to replacepeople getting face-to-face and
saying I want you to do more ofthis and I'm not going to be
able to do that and we can'tcomply with this rule and my
(40:59):
lawyer is going to explain it toyou and while the AI might be
able to look up whether you'reright or wrong to what the law
says you're supposed to do.
It's not going to change thenegotiation between people, so
there's always going to be arule or a role, I should say,
for, you know, lawyers.
In that situation, just like,probably, for music generation,
(41:20):
as clever as ai is, you know,would ai, you know, turn back
the clock a couple decades,would ai have come up to say,
take a rock and roll, you knowdisco, and would have come up
with rap.
Would have come up with, youknow, the same type of you know
completely new music that cameout of.
You know just people playingaround with you know, jamming at
(41:43):
some point and you knowthrowing in something and you
know, like a new song.
Every now and then you'll heara new song.
You'll think, wow, that's justamazing.
Now will ai come up withsomething like that?
The way I was just readingabout a chess move, you know
they said ai was playing chessin a way that no human ever
would.
So ai will come up withsomething new music wise, but
still the same way.
A bunch of guys jam into somesong.
(42:04):
You know that they're writingon the fly.
Joshua Schmidt (42:07):
Uh, they'll still come up with something new
that ai would not have come upwith absolutely, I think it will
further make those sub genresinto fractals, so there'll be
even more mini genres andthere'll be more sub genres
within those genres.
It'll be up to us to decidewhat we want to consume, right?
But, um, you know, I thinkwhere kind of the the value will
(42:30):
be placed.
More will be the gatekeepersand the tastemakers because, um,
you can take, I had a friendthat took a beatles song and you
know the guitars were slightlyout of tune and he tuned each
guitar string so it wasperfectly in tune and kind of,
you know, fixed the problemswith this with the helter
(42:50):
skelter and he said it ended upsounding like a bon jovi song
and it just wasn't the same.
So I think a lot of that nuance,and then the taste and the
aesthetic, and how much does theuh, how much does the ai
culture integrate with the humanculture and where's where's
kind of the borderline of that?
That's kind of where we'reexploring right now.
Nick Mellem (43:11):
I think we're seeing the same thing with
photography.
One of my biggest hobbiesoutside of work is photography.
I've been basically my wholelife and I think, josh, what
you're saying it's the samething.
And that contest, I believe,was national geographic, because
I did see it was a picture of acheetah, if I'm not mistaken.
So.
But then there's the other part.
(43:34):
You take it right.
National Geographic or thesecompanies are spending all this
money to send somebody to do aremote place in Africa to take a
picture of that cheetah closingdown on a you know whatever
their prey.
So you've taken the danger outof it, the cost.
But at one point it's kind ofgoing back to the argument of
well, film was dying, thendigital came around, and then
there's people now that aregoing back to film because of
the art of it, and so now AI istaking over digital, so now
(43:57):
digital shooters are going tobecome the purists versus the AI
.
You're right.
So both your points, it'swhat's the balance?
Eric Brown (44:10):
what's the balance?
There's a show, I think it's,on Netflix called AlphaGo, or a
movie about the AlphaGo, whichis the game of Go, and then,
just like Deep Blue, which wasIBM's Watson computing system
that beat Kasparov a coupledecades ago, well, with AlphaGo
(44:33):
it played the best Go player inthe world, lee Sedol, who's like
a 9th Dawn player, and duringthe match it showed how the AI
made a move that humans justcouldn't conceive of.
And you think, well, from achess perspective or at least
(44:57):
this is how I thought of it was,a computer is able to calculate
chess moves much faster than ahuman can.
So you make one move thatcomputer can calculate moves
that it could make that have ahigh probability of winning.
And because the chess game issmaller than the Go board,
(45:17):
apparently the compute resourcesneeded to calculate the same
moves for the Go game are justnot possible today.
So their artificialintelligence has done something
else to to come up with the, the, how the alpha go works.
But anyway, it said that duringthe match.
(45:37):
Um, and you could see during thematch they showed the, the, the
footage of the match, where thehuman, lee sadoli, was sitting
across from another human whowas making the moves, but he
wasn't able to really read thatperson's emotions because that
person wasn't the one coming upwith the moves.
Um, and it was, you know, thecomputer that was doing it, and
you could see that it reallytook its toll on on lee over the
(46:00):
course of those those fourgames or so.
Um, but I think one of thecommenters made made the comment
at one point in time this movethat we saw could be like a 10th
dawn move, which is, you know,like there are no 10th dawn
humans, there's only, you know,I think nine is the theoretical
highest.
So that that was really kind ofcool to see this application
(46:23):
come to life.
And, you know, like, like youwere saying, josh or Eric, with
music where rap came about fromother musics that we've had,
potentially AI could inventsomething really cool that we've
not seen yet.
Eric Pesik (46:40):
Or heard.
Eric Brown (46:42):
Or heard.
Yeah, and Nick, was it a?
Did the AI just generate thatphoto of the cheetah, or was it?
Did it enhance it somehow?
Nick Mellem (46:54):
so I I was mistaken , I did look, there was national
geographic did create anarticle or a post that showed a
couple pictures of a cheetah tosee you could pick out which one
was ai versus actual, you know,photography or digital.
But there was a contest wherethe person that won and I put
the link in the chat where hewon the contest with his
(47:15):
generated picture and it'sreally an old looking you know
film, 30 old 35 millimeter film,look right from I don't know
when, a long time ago, with acouple, um, elderly females in
the picture here, and it lookslike it's old film that was
developed in a dark room.
Right, you can see this.
Imperfections and the ai builtall that in right, and when you
(47:38):
look at it at first glance maybeif you get right up onto it,
you can maybe tell it's a littletoo perfect.
You know to be real.
Right, it's generated, but atfirst glance it's shockingly
good.
Eric Pesik (47:52):
Because cheetahs don't have fingers.
Nick Mellem (47:54):
That's what AI hasn't mastered yet.
Eric Pesik (47:59):
Well, I know, even today, Dream Studio the one I
was using you can tell it whatlevel of detail you want and,
like when I was showing youearlier, there's a charge for
the computing power that you use.
So if you say, hey, I want youknow so much computing power
devoted to making thisphotorealistic picture, it will
do it.
It'll take longer, it'll costyou more, but you can have it do
(48:21):
that.
I doubt that the version thatI'm using online for, like I
said, a couple bucks here andthere, is going to give me
National Geographic levelresults.
But if you can do it, if I cando it to a certain level here,
I'm sure there's like a reallypro version you can do a
fantastic job with.
Eric Brown (48:39):
There's an AI image generator called Mid Journey
that you interact with overDiscord and I was looking into
using it for image generationand I saw someone had written an
article about using AI, chatgpt, to put together the
(49:03):
instruction set for the imagethat you want to generate
through mid-journey, becausethere's lots of different flags
that you can set and it's almostlike a 120-character-long
Google search where you turn onall of these flags to get
exactly what you want in thesearch.
But you could do the same thingwith the image generation of
(49:26):
tuning how you get it to look bysetting those flags and as a
human it might be hard toremember that right.
It's almost like generatingcode where you could then use
the AI to generate those flags,which I thought was kind of neat
.
Eric Pesik (49:39):
Yeah, it seems like it could be right, in line with,
I mean, like you said,analogous to generating code,
which is what people startedusing chat GPT at first.
Hey, can you give me a HelloWorld program for Rust?
I don't know.
Let it generate your code foryou, Right?
Eric Brown (49:59):
Yeah, could you convert this Java into assembly?
It should be crazy if it coulddo that.
Scott Rysdahl (50:06):
Turn this rock song into jazz right.
Joshua Schmidt (50:08):
Right yeah, turn this rock song into jazz, right
, right, yeah, when I one of theapplications I see happening
probably within the next fiveyears, is taking a piece of film
or a whole movie and uploadingthat into a score generator and
then having the ai spit out 30,50 different scores that are
perfectly timed to the edits ofthe video and choosing the mood,
(50:31):
and then you know, maybe youwant a John Williams type score
with Prince on guitar, with JohnBonham on the drums, you know,
I think.
And then you know there'sthere's some money to be made
there, though, in a way, becausethen you can get the John
Bonham pack, you know, or youcould pay.
You know, some of these peoplethat have these estates, I think
, will end up monetizing thosestyles and figuring out legally
(50:55):
how they can capitalize on those.
Eric Pesik (50:58):
those kinds of um signature sounds those will be
interesting lawsuits, right,like, okay, john bonham, or you
know any musician, you own this.
But do you own songs that soundlike it but aren't necessarily
copy.
Or do you own somebody whoplays the drums like you?
Right, so everybody can do thedouble bass nowadays.
(51:20):
You know, does he get a royaltyfor every?
You know everybody who doesthat with you.
Know, I'm sure there's there'ssome limit where you just sound
too much like John Bonham, butthat's going to be some fun
lawsuits.
So, hey, you know, job securityfor lawyers right, I think
you're set.
Joshua Schmidt (51:38):
You know, even John Bonham got sued for ripping
off Little Richard.
The introduction to rock androll is the introduction to
Little Richard's song.
So you know, all those Brit popguys stole a lot from the
African-American blues artists.
So it's just about how far backdo you want to go and how the
(51:59):
stuff all kind of ends uprolling out and how it kind of
changes and amalgamatesthroughout the years.
It's going to be quiteinteresting.
Nick Mellem (52:07):
Thank, you for having me have a great weekend
everyone.
Joshua Schmidt (52:09):
Yeah, thanks guys.
Yeah, have a good weekend.
Nick Mellem (52:11):
Bye.
Joshua Schmidt (52:11):
See you all.
Thanks, all right.
Eric Brown (52:16):
Want security leadership without the headcount
.
As an extension of the team, ITAudit Labs will provide the
experts to guide and counselyour company.
We will start by creating acustom security program that
caters to your industry whileproviding transparency and
remediation to improve cyberposture while reducing risk.
(52:38):
Contact IT Audit Labs to findout more.
Popular Podcasts
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.
Therapy Gecko
An unlicensed lizard psychologist travels the universe talking to strangers about absolutely nothing. TO CALL THE GECKO: follow me on https://www.twitch.tv/lyleforever to get a notification for when I am taking calls. I am usually live Mondays, Wednesdays, and Fridays but lately a lot of other times too. I am a gecko.