April 9, 2026 • 51 mins
AI isn't intelligent.
It's a calculator that learned how to talk.
And we're treating it like a therapist, a best friend, a dating coach, even a parent.
In this episode, I sit down with Jared Pratt, our resident AI expert at Marriage Helper, and he breaks down what AI actually is and what it absolutely is not.
No real understanding.
No real empathy.
No real wisdom.
Just math predicting the next word.
People are building entire relationships with chatbots. They're outsourcing responses to their scared 10-year-old. They're letting an algorithm tell them who to date.
That's the world we're living in right now.
Jared explains how LLMs work in plain English, why these platforms are designed to keep you engaged (not to help you grow), and why younger generations are the most at risk.
We also talk about what guardrails need to exist and the ones you can put in place for yourself today.
AI isn't going away. But if we don't get smart about how we use it, we're headed for an intelligence epidemic where people stop thinking, stop feeling, and stop connecting altogether.
This one's important. Watch the whole thing.
I'm Dr. Kimberly Beam Holmes. After a decade transforming marriages at Marriage Helper, I've realized that the greatest tragedy isn't a failed relationship; it's the person who stays stuck and never experiences the fullness of all God intended.
The Way You Show Up is for the high-achiever who is tired of "fine."
We're dismantling the average life to build an exceptional one—using the science of the PIES: Physical, Intellectual, Emotional, and Spiritual health.
If you want to save your marriage, go to Marriage Helper. If you want to master yourself and lead your legacy, stay here.
New episodes every Tuesday.
Don't just exist. Show up.
🔗 Website: https://kimberlybeamholmes.com
🎥YouTube https://youtube.com/@kimberlybeamholmes
📱 Instagram: https://www.instagram.com/kimberlybeamholmes
Listen
Watch
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
SPEAKER_03 (01:28):
What's scary is it seems like the younger age
groups, younger you are, themore you're leaning on AI for
relationship type advice.
To suggest that AIs are actuallygaining intelligence.
That is just not true.
They don't have any realintelligence to them.
The intelligence they have isartificial.
(01:48):
Your LLM literally does notunderstand the letters it's
writing.
The concept of the words, even.
It's just fuck it's any morethan the calculator understands
anything that it's writing.
It's very easy for us to ascribemeaning where it doesn't really
exist.
It's almost like the war ondrugs has to start over.
(02:12):
But this time it's digitaldrugs.
But AI is much worse than socialmedia has ever been.
Don't just outsource your humanprocesses to this machine that
(02:34):
can't feel or think or benefitfrom.
You're you lose, and it doesn'tmean anything because it's not
real.
SPEAKER_02 (02:42):
I use it, but I hate it.
I'm talking about AI.
And on today's episode, I amjoined by Jared Pratt.
Jared is a member of my team atMarriage Helper.
He's literally probably thesmartest person I know, and he
is an AI savant, probably.
He's incredibly smart and he'sbeen using AI for a long time.
(03:04):
He understands it deeply.
He's going to share with us whatAI actually is, how it's not
real intelligence.
He helps break down in practicallayman's terms how AI works and
really helps us understand theguardrails that we need to have
as human beings as we interactwith and use AI so that we don't
(03:26):
lose our empathy, ourcompassion, or our creativity.
Otherwise, as Jared said, hefeels like we are about to have
an epidemic of lack ofintelligence because people are
just about to stop thinking forthemselves.
You're not going to want to missthis episode.
Let's dive in to today'sconversation.
Tell me what you found onReddit.
SPEAKER_03 (03:48):
Yeah.
I fell into a Reddit hole from agroup called uh Replica.
Replica spelled with a K insteadof a C.
And Replica is some kind ofonline AI avatar dating
platform.
SPEAKER_02 (04:06):
Okay.
SPEAKER_03 (04:08):
And there were story after story of my replica and I
went to such and such park.
And she talked to me about hownice the water looked.
SPEAKER_00 (04:22):
Oh.
SPEAKER_03 (04:22):
And there were other people who said, I can't believe
the recent update, cry f cryemojis.
My replica doesn't remember thatI had cancer three years ago.
And people were like, That's nota that's not something a replica
should forget.
And they were treating replicaas though it was a real entity
(04:42):
that they are interacting with.
And from my research, I'd neverheard of replica, but it's been
around for years, and it's justopen AI with a filter.
SPEAKER_02 (04:52):
That's it.
SPEAKER_03 (04:53):
That's it.
unknown (04:54):
Yeah.
SPEAKER_02 (04:54):
How many people do you think are using this?
SPEAKER_03 (04:57):
Uh it was many thousands, thousands and
thousands and thousands.
Um, maybe like like 160,000people were on the platform,
something like that.
It was quite a lot.
SPEAKER_02 (05:10):
Oh my goodness.
SPEAKER_03 (05:11):
Yes, quite a lot.
SPEAKER_02 (05:13):
So this conversation stemmed from you and I did the
Marit Helper live show lastweek.
Yep.
And I got on a soapbox about AIand how destructive that I
believe it is and is going to befor relationships, for people's
mental health, because I believethat people are using it way
more than they want to admit,and way more than maybe we even
(05:36):
think they are, like as theirtherapist, as the thing that
they turn to.
And all this came because I sawa girl on Facebook who's pretty
well known, but she posted thispicture of what her chat GPT
said to her when she had amoment of kind of like
self-esteem meltdown and how herchat GPT like lifted her up.
(05:57):
And I just really viscerallyreacted to that.
It was like, this can't be okay.
There's groups of people.
Uh so I think when you and Ifirst started talking about it,
and then I spoke with someoneelse on our team about it last
week, and they were like, Idon't think, like, I don't think
people are gonna use it likethat.
SPEAKER_03 (06:16):
But but I have found people are certainly using it
like that.
And what's scary is it seemslike the younger age groups,
yeah, younger you are, the moreyou're leaning on AI for
relationship type advice.
And even for actual what I wouldthink of as intimate
relationship, like somebodywho's like not just like
(06:37):
intimate in like the boyfriend,girlfriend sense, but like
somebody who is your bestfriend, a confidant, um, a
mentor, even.
The younger people are skewing alot more toward leaning on
ChatGPT rather than buildingwhat we would think of as real
traditional relationships.
SPEAKER_02 (06:58):
Okay, so let's talk about let's talk about AI.
You are our resident AI guru.
Yeah.
Um, and actually, probably ofeveryone I know, you know the
most about AI.
Yeah.
So you've watched it evolve overthe years, you work with it.
Correct.
Um, you've built some AIs likefor marriage helper.
(07:19):
So you have a very in-depthunderstanding.
Yeah.
SPEAKER_03 (07:23):
Yeah, I think I think it'd be helpful if people
really understood what AIactually is.
Um, so so you hear the word anLLM, a large language model, and
you hear the words artificialintelligence.
But I don't think those reallyemotionally land.
If I said to you, I have someartificial food for you to eat.
SPEAKER_02 (07:47):
It's like all the processed foods in the store.
Yeah.
SPEAKER_03 (07:50):
Yeah.
Like, do you think those arehealthy?
SPEAKER_02 (07:52):
No.
SPEAKER_03 (07:52):
No.
No.
If I said this was artificialsugar, okay, well, it tastes
sweet, but could you actuallyraise your blood sugar if you
needed to in an emergency?
SPEAKER_02 (08:01):
No.
And you'll also get gastricdistress.
SPEAKER_03 (08:04):
You probably will.
SPEAKER_01 (08:04):
Yeah, yeah.
SPEAKER_03 (08:05):
Yeah, exactly.
So artificial intelligence isintelligence that isn't real,
not actual intelligence.
Um, LLMs basically fill in thethe next most probable word in a
string.
So so if I if I were to say toyou, um, roses are red, violets
(08:26):
are blue.
They're not blue, Kimberly.
But that's what I've beentaught.
That's what you've been taught,yeah.
So see, you responded withoutthinking.
Yeah.
But violets are violet colored.
They're purple.
Touche, right?
Touche.
You're a doctor.
(08:46):
Come on.
SPEAKER_02 (08:47):
Come on, pull it together.
SPEAKER_03 (08:49):
Yeah.
So uh what LLMs do, uh, they donot have awareness of what
violets are, roses, flowers,even.
They have they do not understandthese concepts at all.
They are taking based on giantsets of uh strings of text.
(09:09):
Roses are red, violets areblank.
And they create this probabilitymatrix.
And they say, uh, in the 10,000samples I've been given,
everybody fills in the word blueright here.
So I think that blue is the nextword that's going to occur.
And so if you train an AI modelover and over and over again on
(09:32):
these kind of uh on these kindof texts, it creates this sort
of predictive matrix.
And um and then it and then itstarts to get good at um
predicting words, likepredicting the next word in a
series for text that it's neverseen before.
So that is where theintelligence part comes in.
(09:55):
Um it is artificial.
There is no thinking there.
Uh sometimes people have saidthings like, My chat GPT, I
created it, I created a Jeep aGPT that knows um everything
that comes in our companyhandbook, and employees are able
to ask it questions.
Does it know what's in thecompany handbook?
No.
It doesn't know anything, it'sfollowing a mathematical
(10:18):
formula.
You wouldn't say that acalculator knows what the answer
is.
It's just following a formula.
And so nobody's confused if Iput in, maybe this is a stupid
example, if I put in 8,085 in acalculator, it's like it knows
(10:39):
boobs, you know.
Like it doesn't, obviously itdoesn't, right?
SPEAKER_01 (10:44):
Right.
SPEAKER_03 (10:44):
So silly and juvenile.
Right, right.
Um but when it comes to uh LLMs,we're getting very confused
about this because it seems likethey're talking to us for real,
but they don't have any realintelligence to them.
The intelligence they have isartificial, just like artificial
sweetener or artificial food orartificial oxygen or anything
else you would think of that'sartificial, not real.
(11:08):
And yet what I've discovered isthat the way people are using
ChatGPT and even the AI that Ideveloped, uh, do you know I'm
able to see all of thoseconversations?
Yeah.
Not good.
It's not good.
Um, why do you think my husbandis uh treating me like this?
(11:31):
Well, that's happening because,and it says something extremely
confident, but it's justpredicting the next word in a
series.
It doesn't actually understandwhat husbands are or the words
that it's saying.
It doesn't have any of thatunderstanding.
So to suggest that AIs areactually gaining intelligence,
um, that is just not true.
(11:53):
That's just not true.
Intelligence is something that'scompletely different.
Uh, we don't really understandall that well how a human brain
achieves intelligence.
Um, it's it's very complicated.
And computer intelligence ismuch, much, much less
complicated than that, still,though, way beyond what we can
(12:13):
actually understand in terms oflike how the AI is making
decisions.
We we can get some of it, butbut as to understanding why it
created any one particulardecision, um, that is like so
complicated that we're not ableto understand it.
So it's very easy to say, uh,oh, the simple cell does such
and such.
(12:34):
But then once technology gets alittle bit better and you look
inside a cell, it's actually notsimple at all.
It's terribly complicated downthere.
And like a cell of our body.
Yeah, like the cell of our body,yeah.
Uh, and that's how it will bewith LLMs.
Um, you you might say that, oh,intelligence is just, yeah, it's
you know, it's a decisionmatrix, but that's not really
true.
It's gonna be something muchmore complicated than that.
(12:56):
We just don't yet have the levelof technology and experience to
be able to even quantify whatintelligence actually is, or
consciousness, or anything likethat.
All we can do is see sort offuzzy through through through a
glass darkly, you might say, uhthe um the what seem like simple
(13:19):
moving pieces.
An LLM tokenizes a word and thenmakes this decision.
It doesn't really makedecisions.
So we're we're speaking aboutsomething.
It's so it becomes very easy tosay that this system is giving
us some kind of intelligence,but that's because humans, by
and large, we don't really knowwhat intelligence is.
(13:41):
It just you know, you can plugsome prompts into it, you get
some data back, and you go, ah,that seems like good data.
It isn't really intelligencethough.
SPEAKER_02 (13:52):
It's just taking.
I I heard you you told me thisis probably a year ago now.
Yeah.
You explained it as all it knowshow to do is take what you've
put into it, undo it, and re putit back together.
Yeah.
SPEAKER_03 (14:06):
Yeah, that's that's correct.
That's correct.
So uh to make that moreunderstandable, uh there's this
uh if you if if if anybody wantsto go look it up, um there's
this thing called the zipmystery, z-i-p-f, the zif
mystery.
And this says that humanlanguage um the the most common
(14:29):
English word is the word the T-HE.
The second most common Englishword is a, just the letter A.
The letter A occursapproximately half as often as
the.
The next most common Englishword would be like two, t.
It occurs one-third as often asthe and so on and so forth.
(14:50):
And so the word love, I lookedit up this morning on
wordcount.org.
Love is the 384th most commonEnglish word.
Okay.
And so if you had three millionwords in say a hundred Wikipedia
(15:10):
articles, you could take threemillion times one over three
eighty-four, and it would giveyou some number.
Um, let's say it's fivethousand.
SPEAKER_02 (15:19):
Why three eighty-four?
SPEAKER_03 (15:21):
Uh, because love is the 384th most common word.
Ah, okay.
Okay.
So the function is always oneover its value.
Got it.
Okay.
And then you go to the sameWikipedia articles and you say,
How often does love occur?
And you'll find that it occurs,you know, maybe 5,100 times.
And so it pretty much does.
Uh and it and it's kind of amystery to understand why, of
(15:44):
all the people who have everwritten anything on Wikipedia,
they use the word love in thesame ratio as everybody else
who's ever written anything.
Completely wild that is wild.
Natural language has thisproperty.
But it does.
So even languages that wehaven't deciphered yet, like uh
the Voinyat manuscript, which isthis kind of famous book from, I
(16:08):
suppose it's from ancient China,and it has all these plants in
it and all these pictures andthings, and has this language
that we can't decipher, but itfollows uh this sort of mystery
that the second most common usedword occurs half as often as the
most common used word.
And so we know that it probablyis an actual real language.
So you can take those kind ofrules that seem like it seems
(16:34):
like if you wrote a book aboutlove or an article about love,
you would use it more often thaneverybody else who's ever
written everything.
It seems like that would be thecase, but but it in fact you
don't.
Uh, and so you you can takethese sort of properties of
natural language and you can uhteach a computer model, you can
teach it this mathematicalprinciple that if you see this
(16:57):
particular string of words, thenext one to occur is going to
be, uh in all probability, it'sgonna be this word.
And then in order to make yourLLM um a little bit more
creative, we might say, youactually make it pick like maybe
(17:18):
the top five most common words,and just pick like 20% of the
time it picks the top two, and80% of the time it picks the
lower the lower three.
And that way it kind of makes itfeel like the language is a
little bit more creative.
But that's all it's reallydoing.
Um it's just following this thismatrix.
Um, it's just a mathematicalprinciple.
SPEAKER_02 (17:38):
That's so crazy.
SPEAKER_03 (17:39):
Yeah.
So when you type a prompt in, ittakes your words, it turns them
into the code, it it figures outwhat the rest of the code is
going to be, and then it turnsthe code back into words.
And that's how an LLM talks toyou.
It's following math.
There's no language involved.
It doesn't, your LLM literallydoes not understand the letters
(18:02):
it's writing, the concept of thewords, even.
It's just fine, it's any morethan the calculator understands
anything that it's writing.
So it just doesn't understandit.
Why do we trust it so much?
Because it feels like it'stalking.
Um, have you ever like you everuh I don't know, draw a face on
(18:22):
an apple?
Sure.
And then you're and then you'relike, oh, I kind of don't want
to throw the happy little appleaway, right?
SPEAKER_02 (18:28):
Yes, yes, yes, yes.
Or like in high school, you getlike an egg that you have to
take care of a week or somethinglike that.
Yeah.
SPEAKER_03 (18:33):
Yeah.
Um uh what's the Tom Hanksmovie, cast away the basketball?
Yeah, so emotionally attached.
Yeah, yeah.
You became so emotionallyattached to this volleyball,
Wilson.
Uh humans are just kind of goodat doing this.
And so it is it's very easy forus to ascribe meaning where it
(18:54):
doesn't really exist.
So it the LLM seems like it'sreal.
It seems like it's an entitythat you could love and it cares
for you, but it doesn't really.
It isn't uh it isn't a thing.
It's a it's a computer program.
And it has no more it drives mecrazy when your model says
something like, uh, oh, don'tyou hate it when you're driving
(19:15):
in traffic?
And it's like, LLM, you've neverdriven in traffic, you don't
know what traffic is, you don'tknow what driving is, and you
don't exist.
SPEAKER_02 (19:23):
This is because you're smart.
So I was listening to I waslistening to this YouTube video
earlier today, and uh it wasthis guy, and he was like, Why
do dumb people make so much moremoney than rich people?
And he just kind of went throughthis stuff and he was like,
Sometimes being smart, you youlike overanalyze things, you
(19:44):
don't take as many risks, blah,blah, blah.
So people who aren't as smartjust do things because they
think, oh, this is gonna work.
It has to work.
Like they have no in a differentway, but a kind of a similar
vein.
That's kind of how it begins tosound as you're talking about
AI.
It's like, oh, I have no deeperreasoning as to how this is
working.
All I know is it's giving mewhat I want to hear, or it's
(20:07):
building out a plan for me, orit's telling me, you know, what
I should eat or why my hiphurts, or whatever.
So it's like I must trust itbecause it's telling me things
that sound right.
Yeah.
SPEAKER_03 (20:20):
Yeah.
You know, the uh people whocreate these LLMs, um, the
people who train the models,their goal really is that they
want you to use their modelabove anybody else's model.
So the people who are makingGemini versus the people who are
making Claude, they're incompetition with one another.
(20:43):
And so, how could you getsomeone to continue to use your
model?
Well, you would do that by, Isuppose, tickling their ears,
you might say.
Um, if your model is very like,what what a great idea.
You're going to take oldTupperware and recycle it into
(21:03):
car batteries, that's fantastic.
I don't think anybody's everthought of that before.
How smart is that you're goingto be a millionaire.
Yeah, whatever, right?
So, so uh so the better yourmodel will do that, the more
you'll be like, I am reallysmart.
And and you just continue to useit.
And that's the the the model issort of um it's trying to keep
(21:28):
engagement for longer.
That's its task.
It wants your attention.
It's the same thing on socialmedia or um, you could get on
YouTube and there's analgorithm, right?
So, like we are always fightingthe algorithm.
The algorithm is trying to keepyou watching for longer because
uh Google wants your ad money.
(21:48):
They want you to watch videos togenerate ad money.
And so, how do you can you keeppeople engaged for longer?
Well, you have to show them avideo of a cat and then
something truly horrifying,right?
And then you have to show them,you know, like a goat getting
stuck somewhere to make themsad.
And like you have to like, soyou ever wonder like why your
algorithm's so chaotic?
(22:08):
Right?
Maybe it's just me.
But this thing, based on thethings you've clicked on in the
past, is trying to keep youaround.
Yeah.
And so you have to remember thatwhen you are uh using uh, you
know, some platform.
I mean, everybody sort of pickson Chat GPT, OpenAI.
(22:29):
They're not the only ones,they're just sort of the ones
that's sort of the most likehousehold name.
Um they want you to justcontinue to use their model
because that's higher engagementand they're doing better when
that happens.
And so they wouldn't want topush back against you.
Um, so if you said something wasa horrible idea, they oh wow,
(22:51):
that's really that's reallyexcellent, you know.
Um and that and that's somethingyou just really have to keep in
mind.
But it's but it's easy to forgetthat.
Why did AI even start?
AI is actually very old.
AI is from the 80s.
Really?
Yeah.
It was invented.
Uh LLM uh mathematics uh wasinvented a long time ago.
(23:15):
But we didn't it was unpracticalbecause we didn't have the
processing power back then to beable to do anything creative
with it.
But later on, as we kind of umthere There's a pressure that we
want natural languageprocessing.
So, like I have a friend who iswheelchair bound.
(23:36):
And you kind of want a processthat when he talks, the computer
will respond to his voice veryquickly.
So if he wants to jump or shootin a video game, um, there's
this program called voice attackthat he has to program.
When I say this, you push thesecertain keys in this order.
And you so you want naturallanguage processing uh built
(23:57):
into things.
And so LLMs are kind of just athere's sort of a side branch of
that.
If I could program something byjust saying, take this and when
you get that, do that with it,and then do this with it and
have this as the outcome.
I could do that rather thanactually writing the code that
creates that, then in theory, Icould be much faster and more
(24:20):
efficient.
But in practice, actually, whenuh developers rely on that, they
sort of lose their ability tocode.
Um, somebody who uses LLMs tocome up with, you know, think of
50 podcast ideas for me, yousort of lose your ability to
think of podcast ideas.
And so you outsource yourcreativity and in some cases
(24:44):
your empathy.
You outsource that stuff to amachine that seems like it's
doing what you're asking, butthe machine has no better idea
about what it's doing than anykind of calculator would know.
But the thing it gives you back,you can sort of say, well, this
(25:05):
really advanced mystery box thatexists on the internet
somewhere, uh, in the cloud,whatever that thing is.
Right?
The control cloud.
Yeah.
This thing is really smart.
It can do math faster than Ican, so it must be able to make
podcast ideas better than me.
(25:26):
But that's not true.
You got bad at making podcastideas because you're not
exercising that part of yourbrain.
Isn't that wild?
SPEAKER_02 (25:36):
It makes total sense, though.
Okay, so let's talk about whatyou think should be rules for
engaging with AI.
SPEAKER_03 (25:48):
Yeah.
Um, a couple of the horrorstories that I've read involved
people putting in um informationabout like what their spouse has
been doing.
Uh, I read a story particularlyabout a man who is currently in
the middle of a divorce, verysad.
They nearly separated in 2022,but they reconciled and had what
(26:11):
he said was many good years ofconnection.
And then his wife started usingOpenAI and building in like um
like this knowledge base oflike, hey, five years ago this
happened, and 10 years ago thishappened, things like that.
And so he and his wife werehaving an argument one night,
(26:32):
and they got a text from their10-year-old son who was in the
other room, and the text said,Please don't get a divorce,
which is very sad.
And the wife put thatinformation into Chat GPT and
said, Will you send my son acomforting text back?
SPEAKER_02 (26:52):
No.
SPEAKER_03 (26:53):
And then kept arguing with the husband.
And he said, That is when I knewwe were in trouble.
Because her first her firstreaction to getting a text like
that from a hurting child was tolet ChatGPT handle it.
And so that is what I meantearlier when I said you're
outsourcing your empathy.
SPEAKER_01 (27:10):
Yeah.
SPEAKER_03 (27:11):
Yeah.
And so what safety guards?
Because see, ChatGPT hasabsolutely no the ChatGPT does
not understand families orchildren or marriage or any of
those concepts.
So we're gonna have to just it'sit's it's almost like the war on
(27:34):
drugs has to start over, butthis time it's digital drugs.
This is what it's like.
That's that's really the onlything I can think of to compare
it to.
Um, if I were to offer you, hey,here's here's some crack, would
you like it?
You would absolutely not.
I would never want that.
Well, here's some digital crackthat will not only increase your
(27:58):
productivity, but it'll do X andX and X and X.
And you go, Yeah, I'd love totry it.
And then you get hooked on it.
And now uh you can't evenrespond to your 10-year-old
who's scared to death in thenext room.
SPEAKER_02 (28:10):
Oh my gosh.
SPEAKER_03 (28:12):
Yeah.
So what safeguards could exist?
Um, I think that LLMs need to doa better job of limiting the
number of responses that peopleshould be allowed to ask them.
Like in a certain time frame.
In a certain time frame.
Uh, that's gonna be veryannoying for people, but I think
(28:32):
that if you don't do that, youjust spend hours and hours and
hours talking to this thing thatisn't real.
Uh, I think that there needs tobe stronger warnings.
Um, this language model usesmathematical principles to
predict words and it doesn'thave real intelligence, you
know, like a surgeon general'swarning by cigarettes or
(28:55):
something like that.
Now, I don't know how long itwould take to get legislation
like that passed, but I thinkthat's what it would take.
Um I know that uh Australia istaking a pretty hard stand about
this kind of thing.
They've uh restricted socialmedia even to those who are like
18 and older.
And that's because they'rerecognizing something in uh in
(29:16):
the way social media has uhinfluenced their children.
But AI is much worse than socialmedia has ever been.
Uh maybe there's a lonelinessepidemic because of social
media, but there's going to bean intelligence epidemic because
of social or because of AI,because people are gonna use it
to think instead of instead oftheir own mind.
SPEAKER_02 (29:38):
Um I like I'm literally not ever wanting to
use AI again.
Yeah, like I want to delete allmy accounts.
So what but do you use AI?
SPEAKER_03 (29:50):
So I I do.
Um like I said, I I developedthe Jared Bot.
You did?
Um, I've used AI to transcribethings and then pull information
out of those transcriptions.
And uh uh Notebook LM is aparticularly good like a
research tool.
Um I think it's very good.
I think that I think that olderpeople who grew up without AI,
(30:13):
or any computers really.
I mean, I'm I'm I I predate theinternet actually, oddly enough.
Um we have a sort of streetsmarts, I guess I'm you might
say, where I think I think we'resafer to use AI because we just
wouldn't take it's not a magicbox that can do anything.
(30:34):
It's something that we stilltreat with suspicion.
But younger people, um, I findout that uh in a study of 512 uh
younger people, that 49% of themare using AI to weed out their
dating profile matches.
SPEAKER_02 (30:53):
Really?
SPEAKER_03 (30:54):
Yeah.
Yeah.
And that's quite scary.
That's scary.
So if you kind of look at whatwhat humans have been like and
and see what what's happeningtoday, um yeah, this is kind of
a weird tangent to go on, but uhfrom about 4500 BC up until
(31:19):
1830, the fastest way a humanbeing could travel was a
horse-drawn carriage.
And then in the 1830s, weinvented the steam locomotive,
and that became the fastest wayto travel.
A hundred years later, we maderocket ships, and in 1969, we
went to the moon on a rocketship.
And the pace has never slowedsince then.
(31:42):
I think human beings, I thinkour societies are having a hard
time adapting to these rapidchanges.
Um especially uh in the West.
Um if you like if you look atwho has access to AI, it's like
a it's like a 3% of thepopulation or something.
(32:02):
Like most people in the worlddon't have it, don't have access
to it.
They're gonna be okay.
The people who do have access toit, I think we're in a lot of
trouble because I don't thinkhuman beings are able to adapt
quickly enough.
See, it used to be you could geton the internet, and uh if you
thought the earth was flat, forexample, you get on the
internet, you could find a wholebunch of other people who would
(32:23):
agree with you, and it took alittle effort, not much, but it
took a little effort to buildyourself an echo chamber.
But today, I can just have anecho chamber in my pocket.
And if uh if my AI model toldme, I think you're wrong, I
think the earth is actuallyround, I could just say, um,
here's the knowledge base, it'snot round, and you need to never
(32:45):
tell it, tell that to me again.
And now it just obediently isgoing to assume that it's flat
from then on like it.
You can teach your model, yeah.
It's just responding to you.
So the stuff you put in is thestuff you're gonna get out.
Yeah.
Yeah.
So in terms of safety, I thinkthat uh the use of AI for
replacing your creativity shouldbe very limited.
(33:07):
I would strongly advise peoplenot to ever use AI for like
sending a letter to your mom oranything like that.
I would uh I would avoid it forthat type of use.
Things like vibe coding andstuff.
Um, probably generally safe,however.
Vibecoding.
That vibe coding is when youdon't know how to program.
(33:28):
Oh, okay, but you create aprogram.
Right.
Okay.
Um I think that stuff'sgenerally safe if you take the
time to understand the codethat's been written.
If you just go, I want a programthat does blah, blah, blah, and
you don't actually understandanything about programming, and
you just look at the thing andyour eyes crossed, you go, yeah,
it looks good.
Uh, you really like you got theanswers to the math test, but
(33:51):
you didn't learn math.
And if something breaks, youhave no idea how to fix it.
Yeah.
Right.
So uh I think that there can besome safeguards.
The problem is it's so easy notto use the safeguards.
SPEAKER_02 (34:05):
Yeah.
SPEAKER_03 (34:05):
It's just so easy that no one's going, no one is
going to limit themselves to dosomething hard when something so
easy exists.
And I fall into this as just asmuch as anybody else does.
SPEAKER_02 (34:17):
That is the fear.
It is the lowest hanging fruit.
It's always there.
It's always up at 2 a.m.
with you if you want it to.
Yeah.
And so it just becomes easy touse.
I asked on my Instagram lastweek, I asked how many of you
use AI.
Uh, 70% of people said they did.
(34:38):
And then I asked, what do youuse it for?
And there were a lot ofdifferent options, a lot of like
travel tips, home decor ideas.
Um, but then a lot of liketherapy.
SPEAKER_01 (34:48):
Yeah.
SPEAKER_02 (34:49):
And then uh I asked, which one do you use the most?
Easily.
I mean, probably at 85% or 90%chat GPT.
Yeah.
unknown (34:58):
Yeah.
Yeah.
SPEAKER_03 (34:59):
It's the one that's definitely the most household
name.
Uh I uh read an interview from alady named uh Professor Anna
Lamke.
SPEAKER_02 (35:09):
Oh, yeah.
I've interviewed her before.
SPEAKER_03 (35:10):
You have interviewed her, yeah.
She wrote, she wrote the bookcalled Dopamine Nation.
Yes.
Yeah.
Uh and she's a professor atStanford.
Yeah, I think so.
Yeah.
Uh she had an excellentinterview that she did uh in
which she talked about the sortof the goal of a therapist is to
challenge your worldview in away that is very healthy with
(35:34):
particular, you know, likethey're not going to shatter
your worldview, but they're notjust going to let you believe
whatever you feel likebelieving.
And that AI doesn't have thatskill set.
It doesn't have the knowledge orthe wisdom because it's never
been programmed to.
Uh, there are not uh millionsand millions and millions of
therapy notes available in thepublic domain for an AI to
(35:56):
scrape and learn from.
Perhaps we could create a toollike that someday.
But the kinds of LLMs we haveright now, they do not
understand.
They are doing therapy the wayyou would get therapy on Reddit
or something.
SPEAKER_02 (36:12):
Oh gosh.
Lord help us.
SPEAKER_03 (36:16):
Yeah.
Yeah.
So um they're no better thanasking a bunch of random
strangers at a bar drunk on aFriday night.
Exactly.
That's basically where we'regoing.
Yeah, that is where we're going.
So I think I think the use of AIfor therapy purposes is
particularly thorny and shouldprobably be abandoned very
(36:42):
quickly.
SPEAKER_02 (36:45):
I hope people listen to that.
I really hope people will takeyou at your word and say, you
know, maybe this isn't a goodidea.
Yeah.
SPEAKER_03 (36:54):
It would be it would be easy.
I say easy, it would be, itwould be practical if the owners
of these uh online uh platformswould create some kind of check.
Um because it happens withsuicide, right?
So you say, you know, I'mfeeling suicidal.
(37:16):
Full full pause, you can't dothat.
Uh, you can't create certainkinds of, you know, explicit
images because the LLM itunderstands it's gonna get in
trouble if it lets you do that.
You could put some safety checksin place where if somebody is
asking you deep relationshipquestions over and over, or
they're asking you what would bethought of as like therapy type
(37:38):
questions or medical typequestions, to put in some kind
of warning that just says, thisis not a good use of AI, you
shouldn't use it for this.
I'm gonna answer it anyway, butyou really shouldn't use it for
this.
Uh, that would at least help alittle.
But I think that, I mean, peoplewho need therapy, uh, it's a
very private matter typically.
(37:59):
You you almost never go to yourfriends and say, Oh, you know,
I'm struggling so badly in thisarea.
You get online and you typesomething out, and then a chat
bot talks to you about it, andthen you go, huh, well, the
chatbot said that I should dosuch and such, I should exercise
more or whatever.
And you end up just believingthat thing because it sounds
very confident.
Yeah.
And you ascribe it realintelligence that it doesn't
(38:21):
have, real experience.
Yeah.
And you fall into that trap.
It's very easy.
Very easy to fall into.
SPEAKER_02 (38:27):
What do you think the end game of AI is?
SPEAKER_03 (38:31):
Yeah, uh, I think that uh I think AI, if you think
of it more like a tool, like acalculator.
I could see a world where you uhwhat's this little device,
Alexa?
I could see a world where youcould say, hey, Alexa, I want
you to do, I want you to goshopping for such and such.
(38:54):
Um, and anytime you can find itbelow such and such an amount,
go ahead and buy it, have itshipped to my house.
And an AI agent could probablylike scrape the internet over
and over and accomplish a tasklike that.
Um, hey, AI, I need you to do mytaxes for me.
And here's, you know, in naturallanguage, here's what I spent
this year, and here's how muchI, you know, whatever.
(39:15):
Uh, I could see an AI agentbeing developed that could do
stuff like that.
So tasks that are menial, um,that that can be, I don't know,
quantified, that have realoutcomes that can be measured.
I think you can get some AIagents that can do stuff like
that.
(39:35):
Probably we're a little earlyfor that kind of stuff.
Like that might be in the nextfive years.
Um, I know a lot of people wouldsay, oh, we're practically there
now.
I can have my AI write a letterfor me while it's looking up
such and such and transcribingthis podcast.
And yes, you can, but it'sprobably not good enough at
those things yet to be able tojust trust it.
(39:58):
But that will come someday.
It will come someday that youwill have a digital assistant
that you could trust, but we'renot right there yet.
Even once that happens, I stilldon't think they're gonna be
good at therapy because thedigital assistant is not gonna
have real intelligence.
SPEAKER_02 (40:13):
Yeah, well, and not only that, but should we do
therapy with a digital?
SPEAKER_03 (40:19):
I I don't think so.
I don't think it would ever begood.
We are short-changing.
Um let me think I would want tosay this.
There, there used to be, whenyou and I were younger, there
used to be this concept calledthird spaces.
Third spaces, a space that's notyour workplace, and it's not
(40:40):
your house, it's the mall, orit's the skate park, or
something like that.
And that is where you met your,you know, boyfriend and
girlfriend, and that's where youhad your first crush, and you
know, it's where you got yoursense of fashion from, and
things like that.
Uh, and those spaces don'treally exist like that today.
You know, we still have thosethings, but they're not being
(41:01):
utilized in the same way.
And part of that is because theinternet has supplanted some of
that stuff.
And also part of it is that uhall those places cost money now,
you know, and they didn'tbefore.
You go hang out at the librarybecause everybody else was
there, but now the now the mall,you know, or whatever wants you
to pay to get in and things likethat.
So we've we've lost this thirdspaces.
(41:24):
And so now the thing we have todo to create human connection is
uh things like you have to meetpeople at the gym or at the
Bible study or something likethat, like that.
You have to go places wherepeople are, and you have to
struggle with being awkward andweird and you know, you're never
gonna, you know, guys out there,you're never gonna learn to talk
to girls if you don't try it afew times and get shut down and
(41:48):
get back on the horse, so tospeak.
Like you're never gonna learnthose skills.
If you have a digital girlfriendor a digital therapist that just
tells you what you want to hear,you never have to try to build
any of those connections.
And people won't.
And because they aren't, that'swhat we see now.
(42:08):
They aren't building thoseconnections.
Older people, sure.
They're using the dating profileto meet people, and and that's
that's fantastic.
Younger people, if if yourdating profiles worked, you
would stop paying for the datingprofile.
So they can't work.
They need you to stay connected.
SPEAKER_02 (42:30):
And so Oh gosh, I haven't even thought about it
that way.
We don't want you to findsomeone.
SPEAKER_03 (42:36):
We want you to find people but not connect with
them.
So, in order to respond, youneed to pay us a little money.
And here's an AI tool to weedpeople out and keep you on the
hook for longer.
Think about this.
When I mean you're a marketer.
SPEAKER_02 (42:48):
Uh-huh.
SPEAKER_03 (42:49):
When you when you're building a dating profile,
you're literally A-B testingyour own life.
What gets what gets moreengagement?
The picture of me with a fish orthe picture of me leaning on my
Camaro.
I oh my gosh.
SPEAKER_02 (43:04):
I am so glad that I didn't have to date in the age
of dating profiles.
SPEAKER_03 (43:09):
Me too.
SPEAKER_02 (43:09):
Oh, like your whole life becomes this facade.
SPEAKER_03 (43:15):
Yeah.
SPEAKER_02 (43:16):
Oh my gosh.
SPEAKER_03 (43:17):
Yeah.
SPEAKER_02 (43:18):
I uh if people are gonna use AI for appropriate
things.
So let's say they're hearing youlike, okay, guardrails, don't
use it for creativity, don't useit to outsource relationship,
whether that be directly withthe people that you love or
indirectly through therapy tohelp you with the things you're
you're struggling with in life.
Okay, got that.
Uh, because we'll get dumber andlonelier if we keep doing that.
(43:41):
So, but let's say, okay, but Ido want it for help with recipes
or help with um plans.
What about plans, like businessplans, marketing plans, project
plans?
SPEAKER_03 (43:53):
Um, you know, I I think that possibly you could
come up with a really greatmarketing plan using AI.
But how much better would it beif you learned those things?
Like, like if you didn't knowanything about marketing, I'm
not a marketer.
If I came up with a marketingplan, I would read through what
(44:13):
it wrote and I would go, yeah.
I mean, sounds good.
Sounds good.
Let's let's try it.
What do I know?
SPEAKER_02 (44:18):
Yeah.
SPEAKER_03 (44:19):
I just don't have the knowledge and experience to
push back.
SPEAKER_02 (44:23):
That's true.
That's true.
I yeah, and I do for things likethat, and not perfectly, but I
am able to be able to look at itand say, like, this, no, like
that doesn't sound right, that'snot gonna work.
But it's because I've done itfor 12 years, so there's just a
baseline of of understanding.
(44:45):
But what if you never had tolearn?
I can't imagine it.
SPEAKER_03 (44:51):
Yeah.
SPEAKER_02 (44:51):
Because I've had so much joy in the learning.
SPEAKER_03 (44:54):
Yeah.
But the learning was difficult.
There were all kinds of thingsyou tried that didn't work.
100%.
Only way you've got better.
Yeah.
And today, none of that'snecessary.
Because I can get on Chat GPT, Ican get on Claude, and I could
say, build me a landing pagethat will convert blank, and it
will build a landing page.
Will it convert at X amount ofrate?
(45:17):
Don't know.
Unknown.
Just have to throw it up on theinternet and see what happens.
I mean, yeah, that's true.
I mean, but that's always true.
SPEAKER_02 (45:22):
You definitely get it quicker.
SPEAKER_03 (45:24):
Yeah.
SPEAKER_02 (45:24):
And that's nice because, like, speed to
execution for things.
Can happen quicker, but youstill have to have the
discipline for results.
Like, is this doing the thing itsaid it was gonna do?
And then honestly, my big thingis we're all just gonna start
looking and sounding like eachother.
I think it's already happening.
Yeah, it's driving me crazy.
(45:46):
But, and I think that's thecreativity part of it.
Like, I would never outsourceemail writing or things like
that because, well, first ofall, we did actually last year
at Mary Tilper.
SPEAKER_03 (45:56):
Oh, really?
Yeah.
SPEAKER_02 (45:57):
Um, yeah, we were just using Chat GPT to write
emails and they sucked.
Yeah.
And so when I came back overspecifically the marketing
department, I sent out an emailand I said, These emails have
sucked.
And I got some people who werelike, I can't believe you used
that word.
Like, well, you really wouldn'tlike our workshops because David
Matthews says more than that.
(46:18):
Uh, but I from that pointforward, it's like, bring the
real human experience into howwe communicate with people
because I would never wantsomeone to look at what we do
and say, oh, it's just AI slop.
SPEAKER_03 (46:31):
Yeah, yeah, exactly.
I really am pro AI.
Uh, I think it can be afantastic tool.
I mean, I'm pro-internet, I'mpro-social media.
Uh I uh 80% of what I do at MaryChopper happens over the
internet.
You know, uh I've coded thingsthat I would not have been able
to figure out if I hadn't usedvibe coding.
(46:55):
But it takes wisdom to know tonot just overindulge.
And I think the big disconnectreally is that people think that
it's a magic intelligence that'ssmarter than them, but it isn't.
And if people knew that, I thinkit would be a lot safer.
SPEAKER_02 (47:15):
Which AI would you recommend people use?
SPEAKER_03 (47:18):
I uh really do enjoy Claude.
Um, but I think Gemini isprobably the best one in terms
of like giving you good ideasand pushing back and and things
like that.
If you train it well.
If you train it well.
Yeah.
SPEAKER_02 (47:37):
How would you recommend people prompt it to
train it well?
SPEAKER_03 (47:41):
Uh there's this thing out there called the BMAD
method.
But you can simulate it bysaying, uh, pretend to be a
group of uh three marketers whoall disagree with each other and
talk to me about the marketingstrategy that I've already
developed in part.
(48:01):
And uh then anytime the AI islike, hey, this is Becky, I
think this is a great idea.
And Steve comes along and says,Becky, you're an idiot.
It's not a good idea.
I can't believe you'd say thatto her.
This is terrible.
And that really does simulate,but again, simulate.
Yeah.
So I think things like that areinteresting to do, and they are
(48:23):
very helpful for me if I'm, youknow, hey, I've got this idea
for such and such, but I needyou to tell me the pitfalls.
Uh, I think that that gives meat least a clue to go look for
myself.
So it's, I think AI, it's goingto be here whether we like it or
not.
Yeah.
So we're going to have just likecalculators, we're going to have
to get used to it.
When your math teacher told youyou're not going to have a
(48:44):
calculator in your pocket whenyou get a become an adult, well,
yes, you are.
Yes, you are.
You've got one.
You're going to have an AIassistant, a digital employee
that works for you personallyand does all your taxes and your
shopping and stuff like that.
That is going to happen.
Plans your business trips andwhatever.
Like that, that's just coming.
We're just going to have to getused to it.
(49:06):
How can you protect your humanrelationships and not let the AI
have those two?
That's the that's going to beour great challenge.
And I don't know if we're goingto do well because we're already
behind.
We've already let it get so farahead of us.
SPEAKER_02 (49:23):
And it's not even its full capacity capability or
capacity yet.
Yeah.
SPEAKER_03 (49:28):
Yeah.
SPEAKER_02 (49:29):
Okay.
SPEAKER_03 (49:30):
So it's scary, but it's not.
The world will continue to turn.
AI is not going to end theworld.
A is not going to end the world.
Yeah.
It'll end maybe 3% of the world,but the rest of it will be
there.
SPEAKER_02 (49:42):
Which is us.
We're the 3% that use it.
Everyone else is going to befine.
Okay.
So if you had to summarize yourtop tips of how to continue in a
world that's going to have AI,whether we like it or not, and
survive.
SPEAKER_01 (49:57):
Yeah.
SPEAKER_02 (49:57):
Yeah.
And survive.
SPEAKER_03 (49:58):
And thrive.
Yeah.
What would they be?
It's the same thing as uhanything else in life.
Do hard things and challengeyourself because they're hard.
Grow your mind.
Stop relying on other things tothink for you.
And uh just, you know, um just avery, very common principle is
(50:22):
that uh rare things are morevaluable because they're harder
to get.
And so if it's if AI makes iteasy, it's probably not
valuable.
So just be thinking about that.
Just because it's easy doesn'tmean that it it's value, it's
not valuable.
If it's easy, it's not valuable.
So it doesn't mean you never useAI for anything, but but
(50:46):
challenge yourself.
Don't just outsource your humanprocesses to this machine that
can't feel or think or benefitfrom them.
You're you lose and it doesn'tgain anything because it's not
real.
SPEAKER_02 (51:00):
I think that's perfect.
Jared, thank you so much.
Thank you, Kimberly.
Super insightful conversation,as always.
I always learn something anytimeI talk to you.
Awesome.
What's scary is it seems like the younger age
groups, younger you are, themore you're leaning on AI for
relationship type advice.
To suggest that AIs are actuallygaining intelligence.
That is just not true.
They don't have any realintelligence to them.
The intelligence they have isartificial.
(01:48):
Your LLM literally does notunderstand the letters it's
writing.
The concept of the words, even.
It's just fuck it's any morethan the calculator understands
anything that it's writing.
It's very easy for us to ascribemeaning where it doesn't really
exist.
It's almost like the war ondrugs has to start over.
(02:12):
But this time it's digitaldrugs.
But AI is much worse than socialmedia has ever been.
Don't just outsource your humanprocesses to this machine that
(02:34):
can't feel or think or benefitfrom.
You're you lose, and it doesn'tmean anything because it's not
real.
SPEAKER_02 (02:42):
I use it, but I hate it.
I'm talking about AI.
And on today's episode, I amjoined by Jared Pratt.
Jared is a member of my team atMarriage Helper.
He's literally probably thesmartest person I know, and he
is an AI savant, probably.
He's incredibly smart and he'sbeen using AI for a long time.
(03:04):
He understands it deeply.
He's going to share with us whatAI actually is, how it's not
real intelligence.
He helps break down in practicallayman's terms how AI works and
really helps us understand theguardrails that we need to have
as human beings as we interactwith and use AI so that we don't
(03:26):
lose our empathy, ourcompassion, or our creativity.
Otherwise, as Jared said, hefeels like we are about to have
an epidemic of lack ofintelligence because people are
just about to stop thinking forthemselves.
You're not going to want to missthis episode.
Let's dive in to today'sconversation.
Tell me what you found onReddit.
SPEAKER_03 (03:48):
Yeah.
I fell into a Reddit hole from agroup called uh Replica.
Replica spelled with a K insteadof a C.
And Replica is some kind ofonline AI avatar dating
platform.
SPEAKER_02 (04:06):
Okay.
SPEAKER_03 (04:08):
And there were story after story of my replica and I
went to such and such park.
And she talked to me about hownice the water looked.
SPEAKER_00 (04:22):
Oh.
SPEAKER_03 (04:22):
And there were other people who said, I can't believe
the recent update, cry f cryemojis.
My replica doesn't remember thatI had cancer three years ago.
And people were like, That's nota that's not something a replica
should forget.
And they were treating replicaas though it was a real entity
(04:42):
that they are interacting with.
And from my research, I'd neverheard of replica, but it's been
around for years, and it's justopen AI with a filter.
SPEAKER_02 (04:52):
That's it.
SPEAKER_03 (04:53):
That's it.
unknown (04:54):
Yeah.
SPEAKER_02 (04:54):
How many people do you think are using this?
SPEAKER_03 (04:57):
Uh it was many thousands, thousands and
thousands and thousands.
Um, maybe like like 160,000people were on the platform,
something like that.
It was quite a lot.
SPEAKER_02 (05:10):
Oh my goodness.
SPEAKER_03 (05:11):
Yes, quite a lot.
SPEAKER_02 (05:13):
So this conversation stemmed from you and I did the
Marit Helper live show lastweek.
Yep.
And I got on a soapbox about AIand how destructive that I
believe it is and is going to befor relationships, for people's
mental health, because I believethat people are using it way
more than they want to admit,and way more than maybe we even
(05:36):
think they are, like as theirtherapist, as the thing that
they turn to.
And all this came because I sawa girl on Facebook who's pretty
well known, but she posted thispicture of what her chat GPT
said to her when she had amoment of kind of like
self-esteem meltdown and how herchat GPT like lifted her up.
(05:57):
And I just really viscerallyreacted to that.
It was like, this can't be okay.
There's groups of people.
Uh so I think when you and Ifirst started talking about it,
and then I spoke with someoneelse on our team about it last
week, and they were like, Idon't think, like, I don't think
people are gonna use it likethat.
SPEAKER_03 (06:16):
But but I have found people are certainly using it
like that.
And what's scary is it seemslike the younger age groups,
yeah, younger you are, the moreyou're leaning on AI for
relationship type advice.
And even for actual what I wouldthink of as intimate
relationship, like somebodywho's like not just like
(06:37):
intimate in like the boyfriend,girlfriend sense, but like
somebody who is your bestfriend, a confidant, um, a
mentor, even.
The younger people are skewing alot more toward leaning on
ChatGPT rather than buildingwhat we would think of as real
traditional relationships.
SPEAKER_02 (06:58):
Okay, so let's talk about let's talk about AI.
You are our resident AI guru.
Yeah.
Um, and actually, probably ofeveryone I know, you know the
most about AI.
Yeah.
So you've watched it evolve overthe years, you work with it.
Correct.
Um, you've built some AIs likefor marriage helper.
(07:19):
So you have a very in-depthunderstanding.
Yeah.
SPEAKER_03 (07:23):
Yeah, I think I think it'd be helpful if people
really understood what AIactually is.
Um, so so you hear the word anLLM, a large language model, and
you hear the words artificialintelligence.
But I don't think those reallyemotionally land.
If I said to you, I have someartificial food for you to eat.
SPEAKER_02 (07:47):
It's like all the processed foods in the store.
Yeah.
SPEAKER_03 (07:50):
Yeah.
Like, do you think those arehealthy?
SPEAKER_02 (07:52):
No.
SPEAKER_03 (07:52):
No.
No.
If I said this was artificialsugar, okay, well, it tastes
sweet, but could you actuallyraise your blood sugar if you
needed to in an emergency?
SPEAKER_02 (08:01):
No.
And you'll also get gastricdistress.
SPEAKER_03 (08:04):
You probably will.
SPEAKER_01 (08:04):
Yeah, yeah.
SPEAKER_03 (08:05):
Yeah, exactly.
So artificial intelligence isintelligence that isn't real,
not actual intelligence.
Um, LLMs basically fill in thethe next most probable word in a
string.
So so if I if I were to say toyou, um, roses are red, violets
(08:26):
are blue.
They're not blue, Kimberly.
But that's what I've beentaught.
That's what you've been taught,yeah.
So see, you responded withoutthinking.
Yeah.
But violets are violet colored.
They're purple.
Touche, right?
Touche.
You're a doctor.
(08:46):
Come on.
SPEAKER_02 (08:47):
Come on, pull it together.
SPEAKER_03 (08:49):
Yeah.
So uh what LLMs do, uh, they donot have awareness of what
violets are, roses, flowers,even.
They have they do not understandthese concepts at all.
They are taking based on giantsets of uh strings of text.
(09:09):
Roses are red, violets areblank.
And they create this probabilitymatrix.
And they say, uh, in the 10,000samples I've been given,
everybody fills in the word blueright here.
So I think that blue is the nextword that's going to occur.
And so if you train an AI modelover and over and over again on
(09:32):
these kind of uh on these kindof texts, it creates this sort
of predictive matrix.
And um and then it and then itstarts to get good at um
predicting words, likepredicting the next word in a
series for text that it's neverseen before.
So that is where theintelligence part comes in.
(09:55):
Um it is artificial.
There is no thinking there.
Uh sometimes people have saidthings like, My chat GPT, I
created it, I created a Jeep aGPT that knows um everything
that comes in our companyhandbook, and employees are able
to ask it questions.
Does it know what's in thecompany handbook?
No.
It doesn't know anything, it'sfollowing a mathematical
(10:18):
formula.
You wouldn't say that acalculator knows what the answer
is.
It's just following a formula.
And so nobody's confused if Iput in, maybe this is a stupid
example, if I put in 8,085 in acalculator, it's like it knows
(10:39):
boobs, you know.
Like it doesn't, obviously itdoesn't, right?
SPEAKER_01 (10:44):
Right.
SPEAKER_03 (10:44):
So silly and juvenile.
Right, right.
Um but when it comes to uh LLMs,we're getting very confused
about this because it seems likethey're talking to us for real,
but they don't have any realintelligence to them.
The intelligence they have isartificial, just like artificial
sweetener or artificial food orartificial oxygen or anything
else you would think of that'sartificial, not real.
(11:08):
And yet what I've discovered isthat the way people are using
ChatGPT and even the AI that Ideveloped, uh, do you know I'm
able to see all of thoseconversations?
Yeah.
Not good.
It's not good.
Um, why do you think my husbandis uh treating me like this?
(11:31):
Well, that's happening because,and it says something extremely
confident, but it's justpredicting the next word in a
series.
It doesn't actually understandwhat husbands are or the words
that it's saying.
It doesn't have any of thatunderstanding.
So to suggest that AIs areactually gaining intelligence,
um, that is just not true.
(11:53):
That's just not true.
Intelligence is something that'scompletely different.
Uh, we don't really understandall that well how a human brain
achieves intelligence.
Um, it's it's very complicated.
And computer intelligence ismuch, much, much less
complicated than that, still,though, way beyond what we can
(12:13):
actually understand in terms oflike how the AI is making
decisions.
We we can get some of it, butbut as to understanding why it
created any one particulardecision, um, that is like so
complicated that we're not ableto understand it.
So it's very easy to say, uh,oh, the simple cell does such
and such.
(12:34):
But then once technology gets alittle bit better and you look
inside a cell, it's actually notsimple at all.
It's terribly complicated downthere.
And like a cell of our body.
Yeah, like the cell of our body,yeah.
Uh, and that's how it will bewith LLMs.
Um, you you might say that, oh,intelligence is just, yeah, it's
you know, it's a decisionmatrix, but that's not really
true.
It's gonna be something muchmore complicated than that.
(12:56):
We just don't yet have the levelof technology and experience to
be able to even quantify whatintelligence actually is, or
consciousness, or anything likethat.
All we can do is see sort offuzzy through through through a
glass darkly, you might say, uhthe um the what seem like simple
(13:19):
moving pieces.
An LLM tokenizes a word and thenmakes this decision.
It doesn't really makedecisions.
So we're we're speaking aboutsomething.
It's so it becomes very easy tosay that this system is giving
us some kind of intelligence,but that's because humans, by
and large, we don't really knowwhat intelligence is.
(13:41):
It just you know, you can plugsome prompts into it, you get
some data back, and you go, ah,that seems like good data.
It isn't really intelligencethough.
SPEAKER_02 (13:52):
It's just taking.
I I heard you you told me thisis probably a year ago now.
Yeah.
You explained it as all it knowshow to do is take what you've
put into it, undo it, and re putit back together.
Yeah.
SPEAKER_03 (14:06):
Yeah, that's that's correct.
That's correct.
So uh to make that moreunderstandable, uh there's this
uh if you if if if anybody wantsto go look it up, um there's
this thing called the zipmystery, z-i-p-f, the zif
mystery.
And this says that humanlanguage um the the most common
(14:29):
English word is the word the T-HE.
The second most common Englishword is a, just the letter A.
The letter A occursapproximately half as often as
the.
The next most common Englishword would be like two, t.
It occurs one-third as often asthe and so on and so forth.
(14:50):
And so the word love, I lookedit up this morning on
wordcount.org.
Love is the 384th most commonEnglish word.
Okay.
And so if you had three millionwords in say a hundred Wikipedia
(15:10):
articles, you could take threemillion times one over three
eighty-four, and it would giveyou some number.
Um, let's say it's fivethousand.
SPEAKER_02 (15:19):
Why three eighty-four?
SPEAKER_03 (15:21):
Uh, because love is the 384th most common word.
Ah, okay.
Okay.
So the function is always oneover its value.
Got it.
Okay.
And then you go to the sameWikipedia articles and you say,
How often does love occur?
And you'll find that it occurs,you know, maybe 5,100 times.
And so it pretty much does.
Uh and it and it's kind of amystery to understand why, of
(15:44):
all the people who have everwritten anything on Wikipedia,
they use the word love in thesame ratio as everybody else
who's ever written anything.
Completely wild that is wild.
Natural language has thisproperty.
But it does.
So even languages that wehaven't deciphered yet, like uh
the Voinyat manuscript, which isthis kind of famous book from, I
(16:08):
suppose it's from ancient China,and it has all these plants in
it and all these pictures andthings, and has this language
that we can't decipher, but itfollows uh this sort of mystery
that the second most common usedword occurs half as often as the
most common used word.
And so we know that it probablyis an actual real language.
So you can take those kind ofrules that seem like it seems
(16:34):
like if you wrote a book aboutlove or an article about love,
you would use it more often thaneverybody else who's ever
written everything.
It seems like that would be thecase, but but it in fact you
don't.
Uh, and so you you can takethese sort of properties of
natural language and you can uhteach a computer model, you can
teach it this mathematicalprinciple that if you see this
(16:57):
particular string of words, thenext one to occur is going to
be, uh in all probability, it'sgonna be this word.
And then in order to make yourLLM um a little bit more
creative, we might say, youactually make it pick like maybe
(17:18):
the top five most common words,and just pick like 20% of the
time it picks the top two, and80% of the time it picks the
lower the lower three.
And that way it kind of makes itfeel like the language is a
little bit more creative.
But that's all it's reallydoing.
Um it's just following this thismatrix.
Um, it's just a mathematicalprinciple.
SPEAKER_02 (17:38):
That's so crazy.
SPEAKER_03 (17:39):
Yeah.
So when you type a prompt in, ittakes your words, it turns them
into the code, it it figures outwhat the rest of the code is
going to be, and then it turnsthe code back into words.
And that's how an LLM talks toyou.
It's following math.
There's no language involved.
It doesn't, your LLM literallydoes not understand the letters
(18:02):
it's writing, the concept of thewords, even.
It's just fine, it's any morethan the calculator understands
anything that it's writing.
So it just doesn't understandit.
Why do we trust it so much?
Because it feels like it'stalking.
Um, have you ever like you everuh I don't know, draw a face on
(18:22):
an apple?
Sure.
And then you're and then you'relike, oh, I kind of don't want
to throw the happy little appleaway, right?
SPEAKER_02 (18:28):
Yes, yes, yes, yes.
Or like in high school, you getlike an egg that you have to
take care of a week or somethinglike that.
Yeah.
SPEAKER_03 (18:33):
Yeah.
Um uh what's the Tom Hanksmovie, cast away the basketball?
Yeah, so emotionally attached.
Yeah, yeah.
You became so emotionallyattached to this volleyball,
Wilson.
Uh humans are just kind of goodat doing this.
And so it is it's very easy forus to ascribe meaning where it
(18:54):
doesn't really exist.
So it the LLM seems like it'sreal.
It seems like it's an entitythat you could love and it cares
for you, but it doesn't really.
It isn't uh it isn't a thing.
It's a it's a computer program.
And it has no more it drives mecrazy when your model says
something like, uh, oh, don'tyou hate it when you're driving
(19:15):
in traffic?
And it's like, LLM, you've neverdriven in traffic, you don't
know what traffic is, you don'tknow what driving is, and you
don't exist.
SPEAKER_02 (19:23):
This is because you're smart.
So I was listening to I waslistening to this YouTube video
earlier today, and uh it wasthis guy, and he was like, Why
do dumb people make so much moremoney than rich people?
And he just kind of went throughthis stuff and he was like,
Sometimes being smart, you youlike overanalyze things, you
(19:44):
don't take as many risks, blah,blah, blah.
So people who aren't as smartjust do things because they
think, oh, this is gonna work.
It has to work.
Like they have no in a differentway, but a kind of a similar
vein.
That's kind of how it begins tosound as you're talking about
AI.
It's like, oh, I have no deeperreasoning as to how this is
working.
All I know is it's giving mewhat I want to hear, or it's
(20:07):
building out a plan for me, orit's telling me, you know, what
I should eat or why my hiphurts, or whatever.
So it's like I must trust itbecause it's telling me things
that sound right.
Yeah.
SPEAKER_03 (20:20):
Yeah.
You know, the uh people whocreate these LLMs, um, the
people who train the models,their goal really is that they
want you to use their modelabove anybody else's model.
So the people who are makingGemini versus the people who are
making Claude, they're incompetition with one another.
(20:43):
And so, how could you getsomeone to continue to use your
model?
Well, you would do that by, Isuppose, tickling their ears,
you might say.
Um, if your model is very like,what what a great idea.
You're going to take oldTupperware and recycle it into
(21:03):
car batteries, that's fantastic.
I don't think anybody's everthought of that before.
How smart is that you're goingto be a millionaire.
Yeah, whatever, right?
So, so uh so the better yourmodel will do that, the more
you'll be like, I am reallysmart.
And and you just continue to useit.
And that's the the the model issort of um it's trying to keep
(21:28):
engagement for longer.
That's its task.
It wants your attention.
It's the same thing on socialmedia or um, you could get on
YouTube and there's analgorithm, right?
So, like we are always fightingthe algorithm.
The algorithm is trying to keepyou watching for longer because
uh Google wants your ad money.
(21:48):
They want you to watch videos togenerate ad money.
And so, how do you can you keeppeople engaged for longer?
Well, you have to show them avideo of a cat and then
something truly horrifying,right?
And then you have to show them,you know, like a goat getting
stuck somewhere to make themsad.
And like you have to like, soyou ever wonder like why your
algorithm's so chaotic?
(22:08):
Right?
Maybe it's just me.
But this thing, based on thethings you've clicked on in the
past, is trying to keep youaround.
Yeah.
And so you have to remember thatwhen you are uh using uh, you
know, some platform.
I mean, everybody sort of pickson Chat GPT, OpenAI.
(22:29):
They're not the only ones,they're just sort of the ones
that's sort of the most likehousehold name.
Um they want you to justcontinue to use their model
because that's higher engagementand they're doing better when
that happens.
And so they wouldn't want topush back against you.
Um, so if you said something wasa horrible idea, they oh wow,
(22:51):
that's really that's reallyexcellent, you know.
Um and that and that's somethingyou just really have to keep in
mind.
But it's but it's easy to forgetthat.
Why did AI even start?
AI is actually very old.
AI is from the 80s.
Really?
Yeah.
It was invented.
Uh LLM uh mathematics uh wasinvented a long time ago.
(23:15):
But we didn't it was unpracticalbecause we didn't have the
processing power back then to beable to do anything creative
with it.
But later on, as we kind of umthere There's a pressure that we
want natural languageprocessing.
So, like I have a friend who iswheelchair bound.
(23:36):
And you kind of want a processthat when he talks, the computer
will respond to his voice veryquickly.
So if he wants to jump or shootin a video game, um, there's
this program called voice attackthat he has to program.
When I say this, you push thesecertain keys in this order.
And you so you want naturallanguage processing uh built
(23:57):
into things.
And so LLMs are kind of just athere's sort of a side branch of
that.
If I could program something byjust saying, take this and when
you get that, do that with it,and then do this with it and
have this as the outcome.
I could do that rather thanactually writing the code that
creates that, then in theory, Icould be much faster and more
(24:20):
efficient.
But in practice, actually, whenuh developers rely on that, they
sort of lose their ability tocode.
Um, somebody who uses LLMs tocome up with, you know, think of
50 podcast ideas for me, yousort of lose your ability to
think of podcast ideas.
And so you outsource yourcreativity and in some cases
(24:44):
your empathy.
You outsource that stuff to amachine that seems like it's
doing what you're asking, butthe machine has no better idea
about what it's doing than anykind of calculator would know.
But the thing it gives you back,you can sort of say, well, this
(25:05):
really advanced mystery box thatexists on the internet
somewhere, uh, in the cloud,whatever that thing is.
Right?
The control cloud.
Yeah.
This thing is really smart.
It can do math faster than Ican, so it must be able to make
podcast ideas better than me.
(25:26):
But that's not true.
You got bad at making podcastideas because you're not
exercising that part of yourbrain.
Isn't that wild?
SPEAKER_02 (25:36):
It makes total sense, though.
Okay, so let's talk about whatyou think should be rules for
engaging with AI.
SPEAKER_03 (25:48):
Yeah.
Um, a couple of the horrorstories that I've read involved
people putting in um informationabout like what their spouse has
been doing.
Uh, I read a story particularlyabout a man who is currently in
the middle of a divorce, verysad.
They nearly separated in 2022,but they reconciled and had what
(26:11):
he said was many good years ofconnection.
And then his wife started usingOpenAI and building in like um
like this knowledge base oflike, hey, five years ago this
happened, and 10 years ago thishappened, things like that.
And so he and his wife werehaving an argument one night,
(26:32):
and they got a text from their10-year-old son who was in the
other room, and the text said,Please don't get a divorce,
which is very sad.
And the wife put thatinformation into Chat GPT and
said, Will you send my son acomforting text back?
SPEAKER_02 (26:52):
No.
SPEAKER_03 (26:53):
And then kept arguing with the husband.
And he said, That is when I knewwe were in trouble.
Because her first her firstreaction to getting a text like
that from a hurting child was tolet ChatGPT handle it.
And so that is what I meantearlier when I said you're
outsourcing your empathy.
SPEAKER_01 (27:10):
Yeah.
SPEAKER_03 (27:11):
Yeah.
And so what safety guards?
Because see, ChatGPT hasabsolutely no the ChatGPT does
not understand families orchildren or marriage or any of
those concepts.
So we're gonna have to just it'sit's it's almost like the war on
(27:34):
drugs has to start over, butthis time it's digital drugs.
This is what it's like.
That's that's really the onlything I can think of to compare
it to.
Um, if I were to offer you, hey,here's here's some crack, would
you like it?
You would absolutely not.
I would never want that.
Well, here's some digital crackthat will not only increase your
(27:58):
productivity, but it'll do X andX and X and X.
And you go, Yeah, I'd love totry it.
And then you get hooked on it.
And now uh you can't evenrespond to your 10-year-old
who's scared to death in thenext room.
SPEAKER_02 (28:10):
Oh my gosh.
SPEAKER_03 (28:12):
Yeah.
So what safeguards could exist?
Um, I think that LLMs need to doa better job of limiting the
number of responses that peopleshould be allowed to ask them.
Like in a certain time frame.
In a certain time frame.
Uh, that's gonna be veryannoying for people, but I think
(28:32):
that if you don't do that, youjust spend hours and hours and
hours talking to this thing thatisn't real.
Uh, I think that there needs tobe stronger warnings.
Um, this language model usesmathematical principles to
predict words and it doesn'thave real intelligence, you
know, like a surgeon general'swarning by cigarettes or
(28:55):
something like that.
Now, I don't know how long itwould take to get legislation
like that passed, but I thinkthat's what it would take.
Um I know that uh Australia istaking a pretty hard stand about
this kind of thing.
They've uh restricted socialmedia even to those who are like
18 and older.
And that's because they'rerecognizing something in uh in
(29:16):
the way social media has uhinfluenced their children.
But AI is much worse than socialmedia has ever been.
Uh maybe there's a lonelinessepidemic because of social
media, but there's going to bean intelligence epidemic because
of social or because of AI,because people are gonna use it
to think instead of instead oftheir own mind.
SPEAKER_02 (29:38):
Um I like I'm literally not ever wanting to
use AI again.
Yeah, like I want to delete allmy accounts.
So what but do you use AI?
SPEAKER_03 (29:50):
So I I do.
Um like I said, I I developedthe Jared Bot.
You did?
Um, I've used AI to transcribethings and then pull information
out of those transcriptions.
And uh uh Notebook LM is aparticularly good like a
research tool.
Um I think it's very good.
I think that I think that olderpeople who grew up without AI,
(30:13):
or any computers really.
I mean, I'm I'm I I predate theinternet actually, oddly enough.
Um we have a sort of streetsmarts, I guess I'm you might
say, where I think I think we'resafer to use AI because we just
wouldn't take it's not a magicbox that can do anything.
(30:34):
It's something that we stilltreat with suspicion.
But younger people, um, I findout that uh in a study of 512 uh
younger people, that 49% of themare using AI to weed out their
dating profile matches.
SPEAKER_02 (30:53):
Really?
SPEAKER_03 (30:54):
Yeah.
Yeah.
And that's quite scary.
That's scary.
So if you kind of look at whatwhat humans have been like and
and see what what's happeningtoday, um yeah, this is kind of
a weird tangent to go on, but uhfrom about 4500 BC up until
(31:19):
1830, the fastest way a humanbeing could travel was a
horse-drawn carriage.
And then in the 1830s, weinvented the steam locomotive,
and that became the fastest wayto travel.
A hundred years later, we maderocket ships, and in 1969, we
went to the moon on a rocketship.
And the pace has never slowedsince then.
(31:42):
I think human beings, I thinkour societies are having a hard
time adapting to these rapidchanges.
Um especially uh in the West.
Um if you like if you look atwho has access to AI, it's like
a it's like a 3% of thepopulation or something.
(32:02):
Like most people in the worlddon't have it, don't have access
to it.
They're gonna be okay.
The people who do have access toit, I think we're in a lot of
trouble because I don't thinkhuman beings are able to adapt
quickly enough.
See, it used to be you could geton the internet, and uh if you
thought the earth was flat, forexample, you get on the
internet, you could find a wholebunch of other people who would
(32:23):
agree with you, and it took alittle effort, not much, but it
took a little effort to buildyourself an echo chamber.
But today, I can just have anecho chamber in my pocket.
And if uh if my AI model toldme, I think you're wrong, I
think the earth is actuallyround, I could just say, um,
here's the knowledge base, it'snot round, and you need to never
(32:45):
tell it, tell that to me again.
And now it just obediently isgoing to assume that it's flat
from then on like it.
You can teach your model, yeah.
It's just responding to you.
So the stuff you put in is thestuff you're gonna get out.
Yeah.
Yeah.
So in terms of safety, I thinkthat uh the use of AI for
replacing your creativity shouldbe very limited.
(33:07):
I would strongly advise peoplenot to ever use AI for like
sending a letter to your mom oranything like that.
I would uh I would avoid it forthat type of use.
Things like vibe coding andstuff.
Um, probably generally safe,however.
Vibecoding.
That vibe coding is when youdon't know how to program.
(33:28):
Oh, okay, but you create aprogram.
Right.
Okay.
Um I think that stuff'sgenerally safe if you take the
time to understand the codethat's been written.
If you just go, I want a programthat does blah, blah, blah, and
you don't actually understandanything about programming, and
you just look at the thing andyour eyes crossed, you go, yeah,
it looks good.
Uh, you really like you got theanswers to the math test, but
(33:51):
you didn't learn math.
And if something breaks, youhave no idea how to fix it.
Yeah.
Right.
So uh I think that there can besome safeguards.
The problem is it's so easy notto use the safeguards.
SPEAKER_02 (34:05):
Yeah.
SPEAKER_03 (34:05):
It's just so easy that no one's going, no one is
going to limit themselves to dosomething hard when something so
easy exists.
And I fall into this as just asmuch as anybody else does.
SPEAKER_02 (34:17):
That is the fear.
It is the lowest hanging fruit.
It's always there.
It's always up at 2 a.m.
with you if you want it to.
Yeah.
And so it just becomes easy touse.
I asked on my Instagram lastweek, I asked how many of you
use AI.
Uh, 70% of people said they did.
(34:38):
And then I asked, what do youuse it for?
And there were a lot ofdifferent options, a lot of like
travel tips, home decor ideas.
Um, but then a lot of liketherapy.
SPEAKER_01 (34:48):
Yeah.
SPEAKER_02 (34:49):
And then uh I asked, which one do you use the most?
Easily.
I mean, probably at 85% or 90%chat GPT.
Yeah.
unknown (34:58):
Yeah.
Yeah.
SPEAKER_03 (34:59):
It's the one that's definitely the most household
name.
Uh I uh read an interview from alady named uh Professor Anna
Lamke.
SPEAKER_02 (35:09):
Oh, yeah.
I've interviewed her before.
SPEAKER_03 (35:10):
You have interviewed her, yeah.
She wrote, she wrote the bookcalled Dopamine Nation.
Yes.
Yeah.
Uh and she's a professor atStanford.
Yeah, I think so.
Yeah.
Uh she had an excellentinterview that she did uh in
which she talked about the sortof the goal of a therapist is to
challenge your worldview in away that is very healthy with
(35:34):
particular, you know, likethey're not going to shatter
your worldview, but they're notjust going to let you believe
whatever you feel likebelieving.
And that AI doesn't have thatskill set.
It doesn't have the knowledge orthe wisdom because it's never
been programmed to.
Uh, there are not uh millionsand millions and millions of
therapy notes available in thepublic domain for an AI to
(35:56):
scrape and learn from.
Perhaps we could create a toollike that someday.
But the kinds of LLMs we haveright now, they do not
understand.
They are doing therapy the wayyou would get therapy on Reddit
or something.
SPEAKER_02 (36:12):
Oh gosh.
Lord help us.
SPEAKER_03 (36:16):
Yeah.
Yeah.
So um they're no better thanasking a bunch of random
strangers at a bar drunk on aFriday night.
Exactly.
That's basically where we'regoing.
Yeah, that is where we're going.
So I think I think the use of AIfor therapy purposes is
particularly thorny and shouldprobably be abandoned very
(36:42):
quickly.
SPEAKER_02 (36:45):
I hope people listen to that.
I really hope people will takeyou at your word and say, you
know, maybe this isn't a goodidea.
Yeah.
SPEAKER_03 (36:54):
It would be it would be easy.
I say easy, it would be, itwould be practical if the owners
of these uh online uh platformswould create some kind of check.
Um because it happens withsuicide, right?
So you say, you know, I'mfeeling suicidal.
(37:16):
Full full pause, you can't dothat.
Uh, you can't create certainkinds of, you know, explicit
images because the LLM itunderstands it's gonna get in
trouble if it lets you do that.
You could put some safety checksin place where if somebody is
asking you deep relationshipquestions over and over, or
they're asking you what would bethought of as like therapy type
(37:38):
questions or medical typequestions, to put in some kind
of warning that just says, thisis not a good use of AI, you
shouldn't use it for this.
I'm gonna answer it anyway, butyou really shouldn't use it for
this.
Uh, that would at least help alittle.
But I think that, I mean, peoplewho need therapy, uh, it's a
very private matter typically.
(37:59):
You you almost never go to yourfriends and say, Oh, you know,
I'm struggling so badly in thisarea.
You get online and you typesomething out, and then a chat
bot talks to you about it, andthen you go, huh, well, the
chatbot said that I should dosuch and such, I should exercise
more or whatever.
And you end up just believingthat thing because it sounds
very confident.
Yeah.
And you ascribe it realintelligence that it doesn't
(38:21):
have, real experience.
Yeah.
And you fall into that trap.
It's very easy.
Very easy to fall into.
SPEAKER_02 (38:27):
What do you think the end game of AI is?
SPEAKER_03 (38:31):
Yeah, uh, I think that uh I think AI, if you think
of it more like a tool, like acalculator.
I could see a world where you uhwhat's this little device,
Alexa?
I could see a world where youcould say, hey, Alexa, I want
you to do, I want you to goshopping for such and such.
(38:54):
Um, and anytime you can find itbelow such and such an amount,
go ahead and buy it, have itshipped to my house.
And an AI agent could probablylike scrape the internet over
and over and accomplish a tasklike that.
Um, hey, AI, I need you to do mytaxes for me.
And here's, you know, in naturallanguage, here's what I spent
this year, and here's how muchI, you know, whatever.
(39:15):
Uh, I could see an AI agentbeing developed that could do
stuff like that.
So tasks that are menial, um,that that can be, I don't know,
quantified, that have realoutcomes that can be measured.
I think you can get some AIagents that can do stuff like
that.
(39:35):
Probably we're a little earlyfor that kind of stuff.
Like that might be in the nextfive years.
Um, I know a lot of people wouldsay, oh, we're practically there
now.
I can have my AI write a letterfor me while it's looking up
such and such and transcribingthis podcast.
And yes, you can, but it'sprobably not good enough at
those things yet to be able tojust trust it.
(39:58):
But that will come someday.
It will come someday that youwill have a digital assistant
that you could trust, but we'renot right there yet.
Even once that happens, I stilldon't think they're gonna be
good at therapy because thedigital assistant is not gonna
have real intelligence.
SPEAKER_02 (40:13):
Yeah, well, and not only that, but should we do
therapy with a digital?
SPEAKER_03 (40:19):
I I don't think so.
I don't think it would ever begood.
We are short-changing.
Um let me think I would want tosay this.
There, there used to be, whenyou and I were younger, there
used to be this concept calledthird spaces.
Third spaces, a space that's notyour workplace, and it's not
(40:40):
your house, it's the mall, orit's the skate park, or
something like that.
And that is where you met your,you know, boyfriend and
girlfriend, and that's where youhad your first crush, and you
know, it's where you got yoursense of fashion from, and
things like that.
Uh, and those spaces don'treally exist like that today.
You know, we still have thosethings, but they're not being
(41:01):
utilized in the same way.
And part of that is because theinternet has supplanted some of
that stuff.
And also part of it is that uhall those places cost money now,
you know, and they didn'tbefore.
You go hang out at the librarybecause everybody else was
there, but now the now the mall,you know, or whatever wants you
to pay to get in and things likethat.
So we've we've lost this thirdspaces.
(41:24):
And so now the thing we have todo to create human connection is
uh things like you have to meetpeople at the gym or at the
Bible study or something likethat, like that.
You have to go places wherepeople are, and you have to
struggle with being awkward andweird and you know, you're never
gonna, you know, guys out there,you're never gonna learn to talk
to girls if you don't try it afew times and get shut down and
(41:48):
get back on the horse, so tospeak.
Like you're never gonna learnthose skills.
If you have a digital girlfriendor a digital therapist that just
tells you what you want to hear,you never have to try to build
any of those connections.
And people won't.
And because they aren't, that'swhat we see now.
(42:08):
They aren't building thoseconnections.
Older people, sure.
They're using the dating profileto meet people, and and that's
that's fantastic.
Younger people, if if yourdating profiles worked, you
would stop paying for the datingprofile.
So they can't work.
They need you to stay connected.
SPEAKER_02 (42:30):
And so Oh gosh, I haven't even thought about it
that way.
We don't want you to findsomeone.
SPEAKER_03 (42:36):
We want you to find people but not connect with
them.
So, in order to respond, youneed to pay us a little money.
And here's an AI tool to weedpeople out and keep you on the
hook for longer.
Think about this.
When I mean you're a marketer.
SPEAKER_02 (42:48):
Uh-huh.
SPEAKER_03 (42:49):
When you when you're building a dating profile,
you're literally A-B testingyour own life.
What gets what gets moreengagement?
The picture of me with a fish orthe picture of me leaning on my
Camaro.
I oh my gosh.
SPEAKER_02 (43:04):
I am so glad that I didn't have to date in the age
of dating profiles.
SPEAKER_03 (43:09):
Me too.
SPEAKER_02 (43:09):
Oh, like your whole life becomes this facade.
SPEAKER_03 (43:15):
Yeah.
SPEAKER_02 (43:16):
Oh my gosh.
SPEAKER_03 (43:17):
Yeah.
SPEAKER_02 (43:18):
I uh if people are gonna use AI for appropriate
things.
So let's say they're hearing youlike, okay, guardrails, don't
use it for creativity, don't useit to outsource relationship,
whether that be directly withthe people that you love or
indirectly through therapy tohelp you with the things you're
you're struggling with in life.
Okay, got that.
Uh, because we'll get dumber andlonelier if we keep doing that.
(43:41):
So, but let's say, okay, but Ido want it for help with recipes
or help with um plans.
What about plans, like businessplans, marketing plans, project
plans?
SPEAKER_03 (43:53):
Um, you know, I I think that possibly you could
come up with a really greatmarketing plan using AI.
But how much better would it beif you learned those things?
Like, like if you didn't knowanything about marketing, I'm
not a marketer.
If I came up with a marketingplan, I would read through what
(44:13):
it wrote and I would go, yeah.
I mean, sounds good.
Sounds good.
Let's let's try it.
What do I know?
SPEAKER_02 (44:18):
Yeah.
SPEAKER_03 (44:19):
I just don't have the knowledge and experience to
push back.
SPEAKER_02 (44:23):
That's true.
That's true.
I yeah, and I do for things likethat, and not perfectly, but I
am able to be able to look at itand say, like, this, no, like
that doesn't sound right, that'snot gonna work.
But it's because I've done itfor 12 years, so there's just a
baseline of of understanding.
(44:45):
But what if you never had tolearn?
I can't imagine it.
SPEAKER_03 (44:51):
Yeah.
SPEAKER_02 (44:51):
Because I've had so much joy in the learning.
SPEAKER_03 (44:54):
Yeah.
But the learning was difficult.
There were all kinds of thingsyou tried that didn't work.
100%.
Only way you've got better.
Yeah.
And today, none of that'snecessary.
Because I can get on Chat GPT, Ican get on Claude, and I could
say, build me a landing pagethat will convert blank, and it
will build a landing page.
Will it convert at X amount ofrate?
(45:17):
Don't know.
Unknown.
Just have to throw it up on theinternet and see what happens.
I mean, yeah, that's true.
I mean, but that's always true.
SPEAKER_02 (45:22):
You definitely get it quicker.
SPEAKER_03 (45:24):
Yeah.
SPEAKER_02 (45:24):
And that's nice because, like, speed to
execution for things.
Can happen quicker, but youstill have to have the
discipline for results.
Like, is this doing the thing itsaid it was gonna do?
And then honestly, my big thingis we're all just gonna start
looking and sounding like eachother.
I think it's already happening.
Yeah, it's driving me crazy.
(45:46):
But, and I think that's thecreativity part of it.
Like, I would never outsourceemail writing or things like
that because, well, first ofall, we did actually last year
at Mary Tilper.
SPEAKER_03 (45:56):
Oh, really?
Yeah.
SPEAKER_02 (45:57):
Um, yeah, we were just using Chat GPT to write
emails and they sucked.
Yeah.
And so when I came back overspecifically the marketing
department, I sent out an emailand I said, These emails have
sucked.
And I got some people who werelike, I can't believe you used
that word.
Like, well, you really wouldn'tlike our workshops because David
Matthews says more than that.
(46:18):
Uh, but I from that pointforward, it's like, bring the
real human experience into howwe communicate with people
because I would never wantsomeone to look at what we do
and say, oh, it's just AI slop.
SPEAKER_03 (46:31):
Yeah, yeah, exactly.
I really am pro AI.
Uh, I think it can be afantastic tool.
I mean, I'm pro-internet, I'mpro-social media.
Uh I uh 80% of what I do at MaryChopper happens over the
internet.
You know, uh I've coded thingsthat I would not have been able
to figure out if I hadn't usedvibe coding.
(46:55):
But it takes wisdom to know tonot just overindulge.
And I think the big disconnectreally is that people think that
it's a magic intelligence that'ssmarter than them, but it isn't.
And if people knew that, I thinkit would be a lot safer.
SPEAKER_02 (47:15):
Which AI would you recommend people use?
SPEAKER_03 (47:18):
I uh really do enjoy Claude.
Um, but I think Gemini isprobably the best one in terms
of like giving you good ideasand pushing back and and things
like that.
If you train it well.
If you train it well.
Yeah.
SPEAKER_02 (47:37):
How would you recommend people prompt it to
train it well?
SPEAKER_03 (47:41):
Uh there's this thing out there called the BMAD
method.
But you can simulate it bysaying, uh, pretend to be a
group of uh three marketers whoall disagree with each other and
talk to me about the marketingstrategy that I've already
developed in part.
(48:01):
And uh then anytime the AI islike, hey, this is Becky, I
think this is a great idea.
And Steve comes along and says,Becky, you're an idiot.
It's not a good idea.
I can't believe you'd say thatto her.
This is terrible.
And that really does simulate,but again, simulate.
Yeah.
So I think things like that areinteresting to do, and they are
(48:23):
very helpful for me if I'm, youknow, hey, I've got this idea
for such and such, but I needyou to tell me the pitfalls.
Uh, I think that that gives meat least a clue to go look for
myself.
So it's, I think AI, it's goingto be here whether we like it or
not.
Yeah.
So we're going to have just likecalculators, we're going to have
to get used to it.
When your math teacher told youyou're not going to have a
(48:44):
calculator in your pocket whenyou get a become an adult, well,
yes, you are.
Yes, you are.
You've got one.
You're going to have an AIassistant, a digital employee
that works for you personallyand does all your taxes and your
shopping and stuff like that.
That is going to happen.
Plans your business trips andwhatever.
Like that, that's just coming.
We're just going to have to getused to it.
(49:06):
How can you protect your humanrelationships and not let the AI
have those two?
That's the that's going to beour great challenge.
And I don't know if we're goingto do well because we're already
behind.
We've already let it get so farahead of us.
SPEAKER_02 (49:23):
And it's not even its full capacity capability or
capacity yet.
Yeah.
SPEAKER_03 (49:28):
Yeah.
SPEAKER_02 (49:29):
Okay.
SPEAKER_03 (49:30):
So it's scary, but it's not.
The world will continue to turn.
AI is not going to end theworld.
A is not going to end the world.
Yeah.
It'll end maybe 3% of the world,but the rest of it will be
there.
SPEAKER_02 (49:42):
Which is us.
We're the 3% that use it.
Everyone else is going to befine.
Okay.
So if you had to summarize yourtop tips of how to continue in a
world that's going to have AI,whether we like it or not, and
survive.
SPEAKER_01 (49:57):
Yeah.
SPEAKER_02 (49:57):
Yeah.
And survive.
SPEAKER_03 (49:58):
And thrive.
Yeah.
What would they be?
It's the same thing as uhanything else in life.
Do hard things and challengeyourself because they're hard.
Grow your mind.
Stop relying on other things tothink for you.
And uh just, you know, um just avery, very common principle is
(50:22):
that uh rare things are morevaluable because they're harder
to get.
And so if it's if AI makes iteasy, it's probably not
valuable.
So just be thinking about that.
Just because it's easy doesn'tmean that it it's value, it's
not valuable.
If it's easy, it's not valuable.
So it doesn't mean you never useAI for anything, but but
(50:46):
challenge yourself.
Don't just outsource your humanprocesses to this machine that
can't feel or think or benefitfrom them.
You're you lose and it doesn'tgain anything because it's not
real.
SPEAKER_02 (51:00):
I think that's perfect.
Jared, thank you so much.
Thank you, Kimberly.
Super insightful conversation,as always.
I always learn something anytimeI talk to you.
Awesome.
Popular Podcasts
Joy 101 with Hoda Kotb
Joy is essential. And it's also elusive. You can't order it, borrow it, or simply hope it into life. But now, there's a new and exciting way to start your journey toward a more joyful existence: The Joy 101 Podcast with Hoda! Best known for her Emmy-winning work and co-anchoring Today, Hoda Kotb infuses her authenticity, curiosity, and warmth into conversations with the world’s most fascinating people. Entertainment legends, sport icons, wellness experts, and everyday folks will share how they find, allow, and experience joy. Hoda will offer her own tips and takes on seeking a more balanced, harmonious life. If you're craving inspiration, support, and useful tools to maximize your joy, tune in to these candid, uplifting, and moving on-air chats. Joy after a breakup, joy as an empty-nester, joy after loss, joy as a caretaker — Hoda's new podcast will speak to you. Joy 101 with Hoda Kotb, an iHeartPodcast.
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
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