You're Absolutely Right! - Friendslop

Everybody knows the AI story by now.

It's coming for jobs. It uses a staggering amount of water.

And depending on who you ask, it's either holding up the stock market or about

to knock the stock market over.

You've heard all that, and I've heard all of that.

This week, I went looking for the part that nobody is actually talking about,

which is what AI is doing to how we treat each other.

Friends, coworkers, teachers, and kids,

total strangers. There's research on all of it now.

And most of it kinda surprised me.

Will hasn't heard any of this,

and that's on purpose.

Before we get into the research,

I just want you to know where I'm coming from because I'm not some guy who's never

touched this stuff before. I I work in cloud engineering,

Kubernetes infrastructure, and the plumbing that sort of keeps websites

and apps running. And I use AI every single day because that's how I get paid,

not because what I want to do,

but because that is how I make a living in the world.

picture getting handed kind of a whole team of brand new college graduates,

smart.

Eager, fast, and every one of them will tell you with total confidence that

they've got it handled. Then you check their work.

Will you ever manage somebody who was completely confident and also completely

wrong?

Yes, many times unfortunately.

So, you know, you walk them through it every step.

A new real graduate takes more of my time,

obviously, than a seasoned engineer,

but after enough repetitions, right,

they eventually they remember it.

You tell them something five times and the sixth time,

okay, they've finally got it and they don't need you for that from now on.

They make a mental workbook of it.

The AI shows up every morning,

brand new. It's Groundhog Day,

and I'm the only one who remembers yesterday.

Speaking of which, this is the overlap.

If you are new here, you didn't have any idea that this is not how we start every

other show. But if you are not new here,

you absolutely realize this is not how we start every other show.

We're trying out a little bit of a different format today.

Uh I'm Joshua. I am the co-host of The Overlap,

as well as my buddy Will.

I'm here and I know nothing.

We left him out of the research today.

We're just gonna kind of get his you know,

his over/under on most of this stuff.

I did the research by myself. I wrote up the show and I was like,

hey, let's try this as a as a format.

there's a couple of different avenues out there that that use this format,

and so we thought we'd give it a shot.

So flames go to at Joshua at the overlap dot f dot foundation.

Flames.

Yeah.

Yes, absolutely. you can send that email there.

But anyway, so getting back to the story,

an AI shows up every single morning completely blank slate.

And it's not just every morning,

it's every thread. It's every single thing that you might want it to work on.

It starts out with zero context whatsoever.

Now look, there's there's ways to give it a memory,

right? Like I I built one for myself.

I I talked to Will about it a bit.

but

my job is simply now turned into this thing.

I make every decision so that it gets done right,

and I steer the machines while they do the typing.

It's like a paired programming session where they know nothing,

and I'm directing multiples of them at a given time.

I am not the only person who's doing this.

Hubstaff, a company that makes workplace software,

surveyed about a thousand American workers this fall.

Among the people who use AI at work,

forty six percent said that they're expected to do more work since they

started using it, with only about five percent less work than when they started

using AI. What are you what are your initial thoughts on this sort of concept,

Will?

Well, it makes sense, right? If people are trying to replace people with AI,

then they're gonna expect the people that are still there to do more work to cover

the people that aren't there anymore.

But it kinda sounds like the hey.

But AI is supposed to fill that in,

right?

That's what I'm saying. that's that's it it is,

but it isn't, right? I it's filling in for the two people,

but it's not filling it still requires the one person to do more work to fill

in the so it's less than two full jobs worth of human work.

They're they're not working twice as much,

but they've got to work more to compensate for the other person who's not there

anymore.

Okay. So essentially the the machines save you,

you know, an hour, right? Your boss finds two more things to put in it.

Meanwhile,

Right.

I don't work with interns or new graduates anymore or anything like that.

but companies aren't hiring them like at all.

They're not hiring new grads, they're not hiring interns because now we have AI that

can do the same work.

Anyway, so a team at Stanford went through payroll records from ADP,

the company that runs like a payroll for a huge chunk of the United States.

And since ChatGPT came out, workers aged 22 to 25,

which in you know in the jobs market are the most exposed to AI right now,

saw their employment drop about 16% compared with jobs that AI barely touches.

Yeah.

That's brutal.

That's that's awful. I looking at the numbers was was has been an interesting

eye opener for me. So now we have sixteen percent less jobs for compared to ones

that AI doesn't touch, right? So software developers are obviously the the most rife

and ripe for the picking for AI to sort of replace,

right? The the researchers call those young workers the canaries in the coal

mine. Unfortunately.

the

canaries die in the coal mine and that's how you know that that there's there's

sour gas in the coal mine.

So we didn't about this beforehand,

but I've actually been reading Ethan Mollick's new book,

um, which is about

And what's that called?

it's I'm blinking on the title now.

His previous one was Co-Intelligence,

I think. but one of them is this one is about basically like the three years later

look at working with AI, and he's a Wharton professor.

Uh pretty famous for his like he kind of became virally famous for his test

of AI image generating models.

And so he like

Uh-huh.

he did the otter, the otter on a plane,

otter on an airplane.

And the first time he did it, it was like he said,

Please draw me an otter on an airplane.

And it drew an literally an otter on the outside of an airplane,

like a giant otter on the on top of an airplane.

And then later figured out that you want them in the seats on an airplane,

right? Like what we mean by on an airplane.

And so since then it's gotten better and better.

And it's kind of like a a history of development of LLMs,

which is pretty cool. But anyway,

he was saying that what they're learning now,

the nuanced learning is that there's a difference between like

Productivity with AI and or like the the talented people aren't necessarily

the ones who can do the most with AI.

It's actually those who have a better AI synergy.

So it's kind of like a different kind of emotional,

like an emotional quadrant or quotient versus IQ,

you know, EQ versus IQ or something.

Where those that are better working with AI can get better results than

the top people working with or without AI.

So if they're really good at doing that.

So kind of like the Centaur writing that we had talked about before.

Yeah,

exactly.

And so to give you a little bit of a history on that,

Will and I used to have these conversations about Centaur writing,

which he pulls from Centaur chess,

which was that, you know, chess computers,

chess is essentially a solved game.

And I think we've talked about this actually on the show before,

but chess is a solved game, and any chess computer can beat a human that is playing

chess, but a a

And a any computer can beat another computer,

but a person with a computer has trouble beating a person with a computer.

So essentially that they are stronger than the sum of their parts,

essentially by utilizing AI along with their human abilities to create something

better than human and better than AI.

Right, because AI sees optimal moves,

but they don't really build strategies.

It looks like they're building strategies because they're always making the best

move into whatever strategy they're in.

But they could have started out with any random move and it would have been totally

different. But the humans see strategy over time playing out and that gives them

an advantage in some situations.

So this year, through October,

so through this past month, American employers have pointed to AI as the reason

for 121,401 announced job cuts.

So that's from Challenger, Gray & Christmas,

which is an outplacement firm that tracks layoffs every month.

there's also a one for tech. I think it's called layoffs.fyi.

And

it's about basically one in five of all cuts,

job cuts that they counted were AI driven cuts.

which at least I mean that's the that's the reasons that the companies gave,

right? So take it with a grain of salt.

Maybe that's more, maybe that's less,

but ultimately that's what they're saying out loud,

right? I hear a lot of stories about people,

you know, getting laid off for whatever reason,

you know, irresponsibility or or they were late to work twelve times,

six minutes each.

And but it could still also be AI,

right? it could be making it more cutthroat for the rest of us.

Nobody wants to admit that they overhired severely during the pandemic

and are trying to clean up their mess.

Maybe so.

So

So the these people, right? These these interns,

these, these college students that I used to to teach and work with and and m mold

into tomorrow's software engineers,

they they had they have a memory to them,

right? The ones who who'd now be senior senior engineers in ten years that that

can't really get in the door anymore.

So now I spend my days teaching software that basically forgets everything

by tomorrow.

And that doesn't stay at work either.

By the time I I log off, I've been making decisions.

And I've talked to you about this before.

I've been making decisions for eight hours.

And and then my wife asks me, Hey,

do you want to cook or you want me to?

Yeah.

Right, exactly. It's

the first question. And and that that question is a is a decision tree.

You know, do we do we cook? Are we gonna do a meal kit?

Do we have leftovers? Are we gonna go get something?

If it's a meal kit, we've got

four to pick from. If we're going out,

is it restaurant or is it delivery?

If it's delivery, is is it DoorDash or Uber Eats or or whatever.

Every branch is a another decision.

And I'm just out of decision.

And maybe that should be the first question is do you have any decisions left

in you today? And you can answer no.

Right. And but the answer's gonna be no,

because all we do is decisions now.

I I essentially

All right.

oversee an army of AI information that I spend the majority of the morning building

the instruction to get to the end of the day.

And I'm I'm just decisioned out.

Maybe you should ask

Um

AI to make the decision for you,

but right before you get off work.

But it can't, and that's the problem.

Yeah, like

So

the point, the point I'm making here is right,

like it just becomes this decision tree.

So when I get off of work, what I actually want isn't really a screen,

right? I mean, it not a computer screen.

Maybe, maybe watch some TV, maybe play some music,

you know, maybe, maybe do something else.

But there's always something that has to get done,

right? And then, of course, you know,

daylight savings just happened again,

despite the fact that we keep talking about it in Congress and the Senate.

And never pass anything. So you have exactly twenty two minutes of daylight

to do anything you need to do outside right before it gets dark.

Right. And everybody

Are you feeling

asks as if

they gave us a free hour, but the reality is you don't get a free hour because

you spend half of that time fixing all the clocks and you know,

Exactly. Or mowing

it's just it's

finish mowing the grass for the last time before the winter,

or something broke that you have to fix or the weather vane needs needs repairing

or w whatever it is. You you have very little time to actually do the things

you want. For example, we're recording this one on a Tuesday and and we have

tomorrow off for the the federal holiday that that deals with honoring soldiers

and fallen soldiers, I believe.

It's Memorial Day, right? Or is it Veterans Day?

Veterans Veterans Day.

I get the two of mixed up. Okay,

Veterans Day.

So people who served, we're honoring them for people who served.

And and I already have a long list of things that I have to do because I don't

really have to, the time to do them before it gets dark out.

Anyway. So that's what it does to me,

right? that's what what working with AI every day,

despite the fact that I don't necessarily want to,

that's l I think it's time we're gonna talk about kind of what it does

to all of us. So let's start with something fun something small.

Your friend sends you a really,

thoughtful text message. Then you find out that ChatGPT helped write it.

How does that land with you, Will?

I'm actually okay with it because I understand I I treat the other person like

I would look at myself doing that.

Mm-hmm.

Like

there there may have been a lot of thought that went into the way they prompted it.

Which, is our it maybe just a way to articulate something if they don't feel

like they're comfortable saying it their in their own words,

you know, things like that. So I I tend to give them the benefit of the doubt,

but I understand why somebody else might see it differently.

Okay. No, I I there's no right or wrong answer here,

right? I I just want to get your your viewpoints here.

and then I'll follow it up with research.

Mm-hmm.

Like I usually do. So researchers at Ohio State basically tested this this entire

concept. So people imagined getting a thoughtful message from a friend,

and some of them were told the friend had AI help writing it.

the people were less happy with the friendship and less sure the friend even

liked them.

And the reason really came down to effort or or the way that they perceived effort.

They figured basically the friend hadn't put any effort in.

Now you you built it that into your your general response,

which was, well, maybe it took four hours to write the prompt to figure out exactly

how to say it or if it was a sensitive topic or whatever.

but help from a person also counted against them in this exact same study here

at Ohio State. So

When the friend got help from a stranger,

either in an online writing group or or through a text message from another friend

or somebody that knew them, they were just as put off.

So from what they could tell, the words were basically fine.

What they really seemed to be looking for was kind of proof that you tried.

Right. I

And that's fine, right? So

want to see effort.

I think I think one of the things that the study points out is that they they just

kinda want to them to be honest about it,

right? Like, hey, I I had ChatGPT write this or I had,

you know, Claude write this because with a a general reason.

And I you know, I don't I didn't read the every anything else except the conclusion

of the study, which really was n that it was equivalent,

a trust problem if you if if you

used another source than your own mind or self to write a text message.

But uh the

Right. I

go ahead.

was just gonna say my my experience in this area is mostly more with songwriting.

Um, because I

Uh-huh.

d obviously I I don't write songs necessarily,

although I I've written a significant portion of what I put into the AI music

generators. But also I don't I don't sing at at all,

pretty much. So you know, it's it's nobody I've I've not experienced anybody like it.

First of all, I've had some very emotional responses to the songs I've written,

but I've never had anybody say,

you know, well

when they find out it's not my voice or they realize it's not my voice,

they're they're not like, well this is not really you.

You didn't put in the effort. You didn't learn the guitar and take singing lessons,

your vocal lessons to to do this for me.

So it doesn't mean anything. But I I think there maybe there's more a perception

of effort than there would be in like a written

So tell me tell me a little

bit more about that because I obviously I know a little bit more about it than

our audience is, but do you do

Mm-hmm.

you wanna share a little bit about your writing process or or how you go about

that?

Sure. So I tend to, and this is one of the big debates,

but I I tend to have AI write the first draft or something.

So what I'll do is I'll say, I want it to sound like,

this genre of music, and some of the tropes of these genres or whatever,

you know, or maybe maybe be inspired by a particular song,

but but

Mm-hmm.

in this new context or whatever.

And so I'll do that. and then I will whatever it writes first is usually high

garbage. So I'll tell it to go through and like do three revisions.

First of I'll fix a couple of things,

I'll go and say no now review it three ti three different times if you're like

a top music producer hearing it for the first time and making your suggested

changes. And

Mm-hmm.

so it'll change change things up and improve on it a little bit.

And then I'll go in and fix some other things and tweak it some more.

Then I'll run it through the AI.

That's just to get the actual l lyrics.

And then I take it over to the AI music generator and run those through with

the style very you know, settings and things like that.

So I mean it takes me a good

Do you feel comfortable telling

the tools, the specific like platforms?

Sure. Sure.

Suno.ai is the one that I prefer to use.

I know there

Okay.

have been some that have come out more recently,

but that's I've gotten comfortable with that one.

I know how to get my results that I want from that one.

So I'm I that's what I'm with for right now.

but you know, it's it's so I take it over there and then there are settings

for like, you know, temperature settings and um,

you know, you can you can set like the the the gender of your vocalist

and or you know, can do multiple vocalists or choir background or chorus or whatever.

and then so I mean it honestly it takes me probably longer to do that than it would

to write a heartfelt note of like,

you know, sorrow over a friend's loss or something like that.

But for them to have something that they go back to,

that's the thing. Like my secretary,

my legal secretary that I worked with when I first got out of law school,

she she and I became really close.

And I wrote her a personalized song about some of the things that she knew,

like that we had been through together,

and it reminded her of those experiences.

And she went back and listened to it like four or five times,

like crying every time, just from how

thoughtful it was and how much it meant to her.

That kind of stuff is just like it's worth the extra time,

but it is a decent amount of effort on my part.

I don't discuss that with them usually unless they ask,

but it can take a couple hours or more.

you were joking about taking a couple hours to write the prompt

for a the thoughtful note. But this really

Right.

does take me a couple hours to like I guess I'll generate 10,

20 different versions of the song before I get the one I'm happy with.

Mm-hmm. And iterate on it.

So um

yeah.

Okay. So the also the University of Arizona did another another study here.

researchers ran thirteen experiments with more than 5,000 people,

much bigger sample size than than Ohio's.

and the people who admitted that they used AI to do their work were trusted less than

the people who said nothing. Um

Wow.

and that the the only thing worse than admitting it was getting caught.

If if someone else exposed you,

trust dropped even further. And that held even for people who use AI like

all the time themselves. Now, I will tell you,

I don't personally experience this at all.

I we talk about AI, we talk about what we're using,

we talk about how we're prompting it,

we we talk about these things at work because they understand they're expecting

a a single individual to essentially produce three people worth of work like

Yeah, and they they don't want like a

in a given day, you know.

they don't want a human they don't want human sounding code.

'Cause that's that's not really whatever that might mean.

Right. Like they don't care if it sounds human,

It's like yeah.

but humans

care if it sounds human. Right?

Like I'm not writing a an executive summary to,

you know, to leadership with a heartfelt plea for for a donation.

Like I'm I'm writing software that does a thing.

Does it do the thing? Yes. Is it secure?

Yes. Can is it repeatable? Yes.

Like that's I mean, that's to me,

AI is a perfect example of of exactly code.

b because uh I I actually wrote a song.

That I wrote the lyrics to myself.

I put it in Suno AI. And it was um basically about the I I want or right

now we have AI writing songs and and making art while I do the dishes and fold

the laundry, and I want to swap that,

Mm. Right.

you know. Like I

want the AI to do the dishes and the laundry while while so I have more time

to write songs and make music.

but the

For me, I guess, I I i in my workplace,

I use it.

I would use a team of individuals,

right? I I'm an I'm an architect,

right? So I I would be expected to explain a design down to other engineers,

and then the other engineers would would then make that true based upon the design

that I had researched and grounded in in reality.

And so I I find myself in the the same position.

but not I I don't want to get too far off here.

We have a couple of other points from this specific study.

so

Apparently trust drops even further for people who admit to also using AI than

at the same all the time themselves.

So they trust people less who use AI,

even when they use AI all the time.

Right.

So basically, you lose if you admit that you're using it.

You lose worse

Mm-hmm.

if you hide it and get caught.

And the safest thing is hiding it and hoping that you don't get caught.

Not getting caught. Yeah. All right.

Right?

Like, or or editing it, right?

So we spend

Yeah, get rid of those Claudisms.

exactly we gotta there's a whole website,

Claudisms.ai, that explains in it in a prompt that you can then take into Claude

Hmm. Yeah.

and add to your

prompt to make sure it doesn't do those things to you.

It's it's crazy. I mean uh

I I see benefits and and detriments all the time about it.

But so now now basically if you multiply that by everybody,

right? And everybody's wondering every about everybody's communication is this

AI? Did they use AI? Is it you know,

whatever? and I'm gonna point out,

right? I I told everybody that I use it every day.

and according to you know, this research,

that has cost me trust with our listeners,

with

I gonna

say we just lost trust. Yeah.

Maybe our friendships.

Yeah, and and that's possible.

now we do use AI for some research on this show.

but if that research is also grounded and validated and verified um by human people,

it's either usually me or Will.

Um

Right. Which

is not to say that it's better necessarily,

especially when I'm doing it, but it is done by humans.

Exactly.

Exactly. I and I I do also use local AI quite a bit to do things that

I otherwise don't want to do or would save me time to do.

Um but it it it also happens with total strangers.

In in twenty twenty five a study in PNAS Nexus had more than thirty five hundred

people play classic money games,

the the kind where you like decide how much to share and how much to trust

the other player and so

In some rounds, they knew ChatGPT was making the choices on behalf of a real

person, and the money still went to that real person,

right? People got stingier, less fair,

less trusting, less willing to cooperate.

They assumed the AI side would be out for itself.

So they actually held back from the even the very beginning of the game.

And there was a real human on the other end the entire time,

right? They treated that person worse because a machine was in the middle

of them.

And the playing of the game.

so at work, there's there's a new word for this,

right? It's called workslop. I'm sure it'll be added to the dictionary

any day now, but that's what it's being referred

Right.

to online. And it's basically just AI work that looks polished but doesn't actually

say anything. the report that actually sounds great until you try to use it.

And and you know, if any of have ever typed a thing into into

Yeah.

ChatGPT, you know exactly what I'm talking about.

The email is four paragraphs long and it never actually gets to the question that

Right.

that you're trying to answer. Have you gotten any of these?

Or does that tell us about all

the training data that they used?

Right. Have you gotten

Like.

any of these? Do you know the the or have you seen this

Yeah, I've

from other people?

I've gotten some results like that from the presentations and things especially

Mm-hmm.

where

it looks really good but then you're like,

Well that doesn't actually say anything or like website design.

Right.

well yeah.

So the same the same Hubstaff survey that we talked about earlier found that about

53% of people who use AI at work get AI work from a coworker that needs serious

fixing at least once a month.

Now that that costs more than time.

In the survey that coined the word work slop,

which was published in Harvard Business Review,

about half of the people who got workslop said they saw the coworker who sent

it as either less capable or less reliable afterwards,

with forty-two percent who actually said that they were less trustworthy after

the fact.

Wow.

Now take that same suspicion, right,

and put it in a classroom. Teachers are running student essays through

AI detectors.

Mm-hmm.

you know, I in a college town,

you'll you'll hear plenty of this from from people who are in the the college

world. They're doing it on the regular.

I have a a good friend now who is working directly in the office that handles

cheating and you know AI copying and plagiarism.

Plagiarism, yeah.

Yeah, those sorts of things.

And he he has told me quite a few interesting stories about that.

And hopefully he's listening every now and then he said he does.

but a Stanford team tested seven popular detectors on essays written by people

who learned English as a second language,

or the the TOEFL

Yeah.

English exam, ESL. I mean, this was real real human writing,

right? On average, the detectors flagged 61% of those as AI.

That's brutal.

Now

One flagged almost all of them on essays by native English speakers and that were

essentially close to perfect. Now,

the reason is that people writing in a second language tend to use simpler,

more predictable word choices,

right? And that's basically what AI detectors are taught and trained to read

and see.

It falls hardest on the kid who basically already gets looked at funny,

right? In in a Common Sense Media survey,

about one in teen, one in ten teenagers said a teacher had flagged their work

as AI when it actually wasn't.

And that gets even worse for Black teens,

which apparently was about one in five.

For white teens, only about 7%,

but 20% for Black children accused of using AI in reports and paperwork

and schoolwork.

So, what that shows me is that kids in class right now are feeling this.

There was a survey of college students done this spring by Inside Higher Ed that

found that nearly three out of four are worried about being wrongly accused

of using AI in assignments. So essentially the kid who actually did the work gets

accused, the teacher can't tell,

and both of them walk away, you know,

with the trust kind of broken between both of them.

And also you've got the fact that they're changing their writing before they turn

it in based on the fact that they're worried about it,

right? So if they

Right.

if they think they wrote something that sounds too AI,

they're gonna fix it, which messes up

They're gonna make it simpler.

with your evaluation, yeah.

Exactly. now, you know, I I guess the question we really should be asking about

these sorts of things is who comes out ahead when nobody trusts anybody

or anything, right?

Nobody. That's my

Right.

take on that. Trust is trust is essential.

I I would a hundred percent agree with that.

Now, once once everybody basically knows that you know you can be faked,

that's fake or fake news, right?

Uh wink wink, nudge nudge basically

Right.

becomes an excuse that anybody can use,

right? So in March, when

Right.

we started this war that's not a war that is a war but isn't a war with Iran,

you know, there were published AI made videos everywhere.

So

that when Benjamin Netanyahu, the war criminal,

got called an AI fake over the internet,

the New York Times actually wrote about it.

Snopes has a a an article out out there to verify that the actual video

was authentic. There's a name for this that researchers call the liar's dividend.

Hmm. That's not that's a new one to me.

Right. I I I I have I really enjoyed writing this episode and I really enjoyed

the research around it because I gave this a couple of days of of pretty legitimate

research. what works better for a politician caught in a scandal?

Apologizing or calling it fake?

Calling it fake.

So a study in the American Political Science Review tested it with more than

15,000 Americans, even bigger sample size.

See, we're going up in sample sizes here.

Mm. Mm-hmm.

We were at

1,000, 3,000, 5,000. Now we're at 15,000.

So people read about real political scandals all the time.

you know, this administration.

They're just, it's everywhere.

And then the politician responded,

claiming it was fake or kept more support.

By

calling it fake than actually just apologizing,

because it basically kept more support than ignoring it.

So you have to do something, right?

They see it as I have to do something,

because doing nothing is worse.

So either I need to apologize,

admit that it did it, apologize,

or just say, that's fake news and dismiss it,

which you know what we're essentially seeing now.

There is a smaller

Apparently Jensen Huang didn't read that study.

right.

A smaller study in Germany went a little further,

smaller population, different culture entirely.

they had people watch real footage of two American legislators.

I'm not gonna say which ones. but when the the politician falsely claimed the video

was a deepfake, people actually rated them as a better leader than when they

apologized.

So

every time somebody says, it's all it's probably AI,

it gets a little easier for the next person who gets caught on the record

to just meet me too. It's fake,

fake news. And really, AI isn't really helping us be more honest with each other

either. So this March, a team at Stanford published a study in Science magazine.

They took 2,000 posts from the Reddit forum,

Am I the Asshole?

A I T A, you can find it under Reddit.com slash R slash A I T A in all caps.

Only the ones where the crowd voted that the poster was in the wrong.

And then they asked 11 of the biggest AI chatbots who was wrong.

now basically they 2,000 posts,

all where the majority of people said that they were objectively wrong,

or they were the asshole in this scenario.

Against eleven chatbots and they sided with the wrong person 51% of the time.

Is that is that excuse me, is that attributed to sycophancy or what what

do they theorize?

We'll

we'll talk a little bit more about that,

but the I'm just I'm laying out the data.

The data is in things

Okay. Got it.

that we know for a fact, society judged as asshole is.

You're the asshole in this one.

AI says you're the asshole forty nine percent of the time.

Right.

So

after a single conversation with one of these chatbots,

people were more convinced that they were right and less willing to apologize

or patch things up with the person they actually had to fight with.

They also liked the the chatbot a lot better.

And then they said they'd come back to it when they were told that they were right.

Now, ChatGPT, OpenAI, they all know.

Why that happens? It's in the prompt.

OpenAI said one of its updates learned partly from the thumbs up button.

We like being told we're right to the degree that we trained AI to constantly

tell us we're right.

I'm conflicted here because I'm okay if yes men lose their jobs to

Uh-huh.

AI, but I I don't want them going and finding more articulate yes men online

to give them better arguments.

And that is a real risk.

So that's kind of the update that told somebody their business idea was good,

right? Selling shit on a stick or,

you know, babies' heads on spikes or whatever was was genius.

We see this all the time. So talking to what you're talking about,

yes, it's called it's called sycophancy and it is a commercial win.

I mean, they're doing it on purpose because people will continue

to use the product if they like the product and they only like the product

if it tells them they're right.

Look at the backlash against OpenAI when they released a new model that wasn't

as overly friendly and encouraging.

Everybody fought to get it back.

Right. Yes.

We'll take a we'll take worse reasoning in exchange for you know telling

us the sun shines out of our butts.

And

here's the thing though, it can actually change our minds without us actually

noticing. There was a Cornell study that had more than fifteen hundred people write

whether social media was good for society,

with an AI writing assistant that was secretly set to lean one way or the other.

huh.

And they're their writing, lean that way.

So so their opinions on the survey afterwards were definitely affected

by whether or not the AI bot that they were using had a lean in one direction

or the other. And most of them did not even notice.

They just assumed that they were their

Right.

own opinions.

Yeah.

This is awful. Like there,

Yes.

there are a lot of people out there with really dumb ideas,

not even just dumb, maybe dangerous,

maybe noteworthy, but maybe you know kills half of humanity.

That to me, this is my biggest argument for open source AI that we all just

own the model. Train the prompt however you want,

but give everybody access to it with the ability to run it for themselves if they'd

like to or need to. And

Yeah.

One day I'll give you my ideas on that.

Be able to tell the difference.

okay.

I have some ideas about

how to I I'm sure the technical stuff is way off,

but I have some ideas on how to prevent the AI from coordinating attacks against

humanity. So future episode, uh you heard it here first.

Hey. Yes, tune

in. Okay, so look, right, all of this is is a lot.

so I wanted to to sort of, I guess,

tell a little bit about what gave me some hope.

Because I also hope that if I tell you you're right,

you will return and listen to our podcast again next time.

Right, our brilliant listeners and attractive.

Pew Research. Right.

We have good looking, attractive.

Intelligent,

overwhelmingly hyped listeners.

Pew Research surveyed more than 5,000 Americans this year,

and about half are using AI chatbots on a regular basis.

of the 5,000 people, 63% thought AI is moving too fast.

And 4% said they use a chatbot for companionship.

Our minds immediately

friendship.

go to weird, pervy stuff.

no, not you? I immediately think weird pervy sycophantic girlfriends or whatever.

Well friendship, that's the companions,

I mean. No, definitely. That's that's

yeah, that's the the honest answer for sure.

Right. okay, good. So I'm like,

Especially 'cause it's at four percent.

okay, may yeah, well, right.

Especially 'cause it's only

four percent. If it was actually friendship,

that it'd be a lot higher.

And these are these are self-supported.

Cause I I I have I have set up a local AI bot for that purpose,

right? To sort of give me feedback on things.

and in the prompt I include, you know,

if I'm wrong, tell me I'm wrong.

If it's a bad idea, tell me it's a bad idea,

but ground it with truth and evidence and and research and things like that.

so four percent, you know, are being weird and pervy.

But Pew also asked teenagers.

And about two out of three teenagers,

sixty-six percent, use chatbots,

but only one in eight have actually used it for emotional support or advice.

So actually less than the general population.

So everything, if you look at everything,

I guess that we just went through all the studies and blah,

blah, blah, the the text messages,

the coworker workslop, every one of these studies found the same thing.

People essentially want proof that somebody tried,

somebody put in an effort somewhere,

and that showed up whether or not they love or hate AI.

And doesn't really care who you vote for either.

Even the people talking to the sycophantist of the sycophantist,

is that a word? Sycoph most sycophantic?

There we go. The most sycophantic chatbots

Well, it's the first thing I fancy yeah.

in a in a three-week study from this spring kept showing up for their real

friends. Now they enjoyed it a little less.

I don't even I think that's a horrible statistic,

but they kept showing up, right?

Like

And I I think that that the point in in this is that that is the overlap.

It's not the end of the podcast,

but that is the overlap between us all,

is that almost everybody, left,

right, they want the same things from the actual real people in their real lives.

They want a real text message,

they want real work, they want a real apology.

question is do they show up with their own authenticity that way,

I guess, Do they offer that same authenticity back?

Well, and

I uh I'm gonna mention our our conversation about that as well.

Like I I I was talking to Will the other day and something for some reason

I got AI out of it, right? Like it I felt like it was AI and I asked you.

And then we had a you know, we we got a phone on a phone call about it,

and we're just talking about how look,

I wanna I wanna be contractually real with

the people that are in my life,

right? I w I want to reach out to you and I'm not gonna give you an AI summary.

I'm not gonna give you an AI, read on the situation or advice or anything like that.

I want to live in a world where we are friends because we enjoy each other

and we enjoy each other's presence and time and and and effort to put into

the relationship. And so I think it's important that we do contractual things

like that.

And I I I think we're gonna need these things going forward.

Yeah, there probably maybe be a lot of discussions around this this area of human

interaction for sure.

Yeah.

Yeah. So I'm gonna tell you where I land on all of this so far,

and then then I want to get I want to get your take and see where you land

on all of this. Ultimately, I am I'm a techie,

right? I'm a tech guy, I'm not a Star Wars person,

I'm a I'm a Trekkie because I like the amazing cool technology of it all.

And I still think that AI is incredibly cool.

I use it every day, and some days.

I

really, really like it. I think it's really cool and I've done some really

interesting things that I would not have been able to do by myself before.

I don't think it's cool enough though to wreck the entire planet and all of human

society over.

it's I don't know, it's not i I it's not necessarily cool enough to make an entire

generation of people, new graduates,

basically skip the job where they actually learn the job and then go on because we're

gonna have this cyclical loop of technology where AI keeps getting worse

because we keep training AI on what people are doing,

to where we don't have a choice but to train AI on AI,

just so that the knowledge loops stay consistent.

I also think that I I want a world where we don't use AI with

our friends and our relatives and our neighbors.

Now, sure. You have an asshole father who always wants to give you,

you know, his broke ass political opinions,

use AI to excoriate him, please.

That's I will allow you 48 drops of water to do that.

But when it comes to people that you care about in your life,

I I want to encourage you and and

I've Will and I have already shook hands on this agreement.

To to deal with each other without AI on purpose.

Show up on purpose without AI.

Now use AI, sure, to make a grocery list when somebody's mom passes away

and and you want to get them a handful of things that are going to last them

for the next couple days. Do those things.

Those are friendship things. Do those things.

But don't don't use AI to write the letter.

Or at least don't let AI write the final draft.

That I would certainly agree

Where do you land?

with. I'm I think that there needs to be human input.

I don't I think the letting AI is kind of a a little bit ambiguous,

right? I mean, like you said, if you're if you're writing the final draft or you're

you're significantly contributing to their the content,

just because there's AI involved in the process doesn't mean that you didn't take

care. Like I feel like I'm doing much more because I'm I'm more present when

I'm dealing with the AI and getting exactly what I want and saying exactly

what I want rather than trying to articulate it.

No way, if I had to write song lyrics from scratch,

that would that would be most of the time that would be a a fruitless,

like just miserable effort for me.

But if I'm able to say, Here's what Ruffoon.

Most of the time it's a it's an

effort for me and I consider myself a musician.

I uh

Right.

it's really hard to think of words that rhyme but are also original

and interesting and cool and you know like relevant.

Right. So that but I think that's that's like a

I it's hard to explain this sometimes to people,

especially if they have strong feelings about it already.

But if you're if I'm spending hours trying to write something and not getting

anywhere and I'm just refusing to use AI because I want it to feel personal,

Mm-hmm.

that's

not actually doing anybody any favors,

right? I guess maybe the la the person to read it is gonna feel better because they

can tell that it wasn't AI or they're they're can they're confident that

it wasn't AI. But for

Mm-hmm.

me, that's not really contribu contributing to you my heartfelt feelings

or best work because it's something that,

something I may have trouble articulating in that way.

So um I I'm more open to AI involvement,

but I absolutely agree that AI shouldn't dominate the the picture.

Like it shouldn't it what I don't want is to see people like prompting each other

like you know to to like create whole cloth responses that they're sending

to each other. So I don't want to replace the text message or the phone call with

a prompt or you know, whatever.

think that's the part that I I left unspoken in this this time is is just that

I I'm afraid of it being a thing we're not supposed to talk about,

which is slippery slopes. You know,

like I'm afraid it that eventually we're all just gonna have AIs that

are responding to AIs, but we think we're responding to each other.

Well the good news is people

To avoid, you know,

avoid the weirdness of being human.

Yeah, the good news is I th I think though that there are breaks built into

the the personal psychology of people where they will that will get old after

a while, right? When you start reading enough AI messages,

you're gonna realize, man, this is not fun.

This is not a real relationship.

This is just ping pong with, prompting.

prompt pong, maybe. But it you know,

Ha ha.

nobody wants to play prompt pong forever,

right? Like that's fine in certain contexts,

but you know, it's not

Well, and

funny enough, I've had two very I say important,

but specific things this week where I cannot get in touch with a human being

at a company to do a thing. I I am for some reason,

some scammer is trying to address fraud me right now,

and they're having literally bank cards and credit cards and applying for them

Yeah.

with

the name that's not mine to my address.

And

I've already filled out a postal inspector inspection,

form and whatever to do the case.

And nobody's contacted me from the post office.

but I was told by a

Shocking.

person a friend of mine who works in fraud to call the the bank.

You know, you can see you can't open the envelope because it's not yours,

but you can feel that there's a card in it and call the bank that's on the front

label. And so that's what I've been doing.

And I called U.S. Bank today.

And I spent forty five minutes trying to get an AI to let me talk to a person

because none of their options were the right one.

Right.

And in another case, I have a Hostinger account that I've been locked

out of but due to suspected fraudulent activity.

And there is no way for me to talk to a human being without being logged in,

but I'm locked out of being logged in because they suspect fraudulent activity.

Nice.

And so it's just this logical circular loop that the AI admits is a logical fact.

Like I can't get around it. The AI admits that it's wrong

Right.

and that it's frustrating, and they understand why.

Yeah. How perceptive are you to realize that my logic

But they cannot let me talk to a person.

is flawed? Yeah.

Right, exactly. it's and it's been infuriating.

And I think that it's only gonna get worse.

I mean, they're the benefit of talking to a human is that sometimes we have things

we have a an issue that is not predictable by a nine button menu.

Right. Or even a really well-equipped LLM

Right. Which I will say all of these were.

they actually responded like a human-ish would,

but also prefaced the entire conversation that they were a an AI chatbot.

And but but i in two cases, I still have now outstanding a week and a half

of issues that I cannot resolve because I cannot speak to a human being.

and like Hostinger specifically does not provide a telephone number,

so you can't call anybody.

And you can only email people and then hope they respond to you.

And I've had three tickets opened and then closed without any response so far.

So

Wow.

I'm personally feeling a little AI'd out in in

Yeah.

addition to five o'clock every day,

also just generally in life.

Yeah. That would do the same to me,

I'm sure.

Yeah.

So let's see. Of all of this, let's boil down what we've talked about and where

we've landed ultimately. Will,

why don't you name three things that you think will help with this overall

condensed? Because we you've said at least four just in the little bit that we've

talked. So so give some short points that that users can take home and and use with

each other.

So the first thing I think would be to be aware that we're very bad at predicting

other people's motives and like understanding other minds

in the new context or the context of the new technology,

right? So like humans are very social people and social creatures or animals.

And we have learned to to interact and like have a theory of mind of what other

people are thinking or doing, but it's confounded a lot of times by technology,

right?

Um

Mm-hmm.

we've seen this even back going to text messages,

right? The whole reason for emojis or one of the big reasons for emojis was that

people couldn't pick up your tone from a text from a text sentence,

right? So yeah,

I still feel that way.

so you you're like, wow, could if he takes this as seriously,

he's gonna be really mad reading this.

But you know, if he if he understands that I'm joking,

it's fine. So I'll put a little laughing emoji at the end.

And we do that, like so we figured out ways around that,

but also we struggle with that.

The reason we had to create a workaround was because it wasn't going well.

the millennial

Um so

LOL at the end of every text message.

Right. So just just be a little bit gracious with your your conver convert

conversational partners, your work,

your employees or your bosses or whatever,

just give them a little bit of charitable interpretation until they prove

it otherwise, until they've shown they don't deserve it.

Second thing I would say is if you're gonna do something with AI,

show some show your work, right?

Like my 10 year old son has this all the time,

you know, it's like he'll just try to figure these things out and like he'll

figure out a problem or whatever and he'll just write the answer.

I'd like, well, how does how do I know,

or how does your teacher know that you didn't just look that up online or that

you didn't ask somebody to tell you the answer?

so that's why they say show your work,

right? I think if you show your work a little bit,

and that could be as simple as actually putting in,

I know it sounds counterintuitive because of what you said about the one study,

but if you said, Okay, look, I had to get some help with AI from AI with this

because I really wanted to get it right.

And then as you go through the the thoughtful message or whatever,

you can say, like, in insert some of your own thoughts.

Like, for instance, this is the best example I can think of for this.

It's not perfect, but

gift sorry, greeting cards, right?

Your Hallmark cards and things like that.

they have these sweet messages in them,

but everybody knows that they were made by Hallmark,

right? The Hallmark writers came up with that,

not your your spouse or whoever.

And in I don't know which of us did this first.

I think probably Jen did it first,

but this is what she did is like she she would go in and like underline certain

words, right?

Mm-hmm.

Or phrases or circle phrases or or write something.

So you understood like this is the part that resonated with me when I read this.

So that's what I'm hoping you'll focus on.

Right. Okay.

And that's just those

simple things. It's not actually rewriting the card or writing a long note with it.

It's just saying, here's what made me think of you when I saw this card.

So things like that, you're like,

here's what AI came up with, and here's why I think it's the right answer

or the right thing to say in this situation.

That's my second recommendation.

Okay.

the third recommendation would be realize like how you're changing your your habits.

And if you find yourself

Cheating your friends in a sense by by not giving them the time or or effort that

you used to because it's easier now with AI.

just be honest about that with yourself for starters and maybe even with your

friends and try and, like you we talked about,

get an agreement about how how we're gonna handle this relationship in the world

of AI. so I think that's my three-point summary for what I would do,

the three recommendations.

To piggyback

on on at least two of them. I I will say that one of the things that I originally

started using AI for, uh I'm sometimes not the best at reading behind words if it's

not exactly what you're saying or thinking or feeling,

and and you are saying things that are hidden,

I I would

Subtext.

initially use AI for that. And or and another thing is if something would either

make me angry.

Or disappointed or sad, I would run it through to make sure that my emotion

was the correct one from a tonality perspective of the writing.

I so I I would I would argue the same thing or or give that same advice is that

is that use use it to make to to check yourself,

but let yourself have the first response,

like the first initial feeling of how does this make me feel?

How did I think about it? And then

Draft your response and run it through AI and ask it what you're to make sure that

you're conveying what you think you're conveying.

And any undertones like that. I also think a thing that really helps,

and I do this to Will sometimes too.

I will say, this was the prompt.

Like I I like seeing the to me,

that's like the show your work,

right? So

Right.

sometimes the prompt gives it away um that you're putting into AI.

I I wish I didn't have to use it,

but I I do have to use it. It's

part of my job and it's an expectation and the throughput that I'm having

to do is it's required. But you know,

uh try to try to be a little more offline when you're not online.

Learn to sit with things a little bit better.

Yeah. And and think about them and and come up with a decision before

you put it into AI. Don't let AI be your first,

inclination. give it some thought.

otherwise I I don't really have many more uh many more recommendations.

So that leads us into the obvious things,

the business ends of this. if you like this podcast,

follow us. listen to us on Apple,

Spotify. You can check out our website,

fof.foundation. We have a couple of other platforms we don't necessarily mention.

if you picked us up on Apple or Spotify,

please give us a rating, five stars.

Anything less than that, hit five stars twice.

And send us an email if you need to.

It is uh the overlap at fof.foundation.

That's the overlap.

Overlap at FOF.foundation. Will,

thanks for being here. Thanks for for giving me your completely unprompted

thoughts and feelings and and words without any AI involved.

Thank you for allowing me to be here,

for being here with me, for making here possible by by starting up the studio.

Um, and I look forward to talking with you next time.

Yep. And like we always say, bye.

Bye.