Low-Stakes AI Scenarios, High-Stakes AI Viruses, and the Trump Catering Truck Escape, With Charlie Warzel
2026-08-13 10:00:00 • 1:07:54
It is November, 1988, close to midnight, and you're calling a friend at Harvard to ask
for advice.
Your stomach is tight.
Cold sweat runs down your neck as you stare at your computer terminal.
You are Robert Tappen Morris, a Cornell grad student.
You have really, really fucked something up.
The Internet, a network of a few thousand machines connecting universities and military research
labs, is crashing.
Computers normally running one or two jobs or running five or ten.
Confused admins are trying to kill the processes by hand, but new ones spin up faster than
they can be killed.
Rebooting buys a few minutes at best, and the network is rapidly becoming unusable.
And this is all your fault.
Your intentions were innocent enough.
All you wanted to do was count the computers that make up the Internet.
So you wrote a program that counts them by going to a machine, copying itself, and moving
on.
To keep from counting the same machine twice, the program asks, are you already running
me?
If the computer answers yes, it does not install.
Simple enough.
But you want to step further.
To avoid admins lying by selecting yes manually, you added an exception.
One time in seven, the program ignores the answer and installs anyway.
And this, you now realize, was a terrible mistake.
Because the network is like a web.
Certain machines are hubs with numerous load bearing connections coming in and out.
And because of your one in seven exception, these machines are infected and reinfected hundreds
of times.
You press the phone against your face until you're Harvard friend.
Okay, no time to explain, but please post an anonymous apology to the TCI P mailing
list along with these instructions for killing the infection.
But the worm is strangling the network and your message crawls along.
It will take three days for the worst of this to pass.
You hang up, not quite relieved.
You were careful.
You launched the program from MIT to make it harder.
Perhaps impossible to trace back to you at Cornell.
The apology note was anonymous, of course.
But next year, you'll be indicted.
The first person ever charged under the Computer Fraud and Abuse Act.
And within weeks of the night, your program runs, the government funds a team to make sure
this kind of thing doesn't happen ever again.
The field of computer security is, in a very real sense, born from your mistake.
This is Wait a Second, Low Stakes AI Nightmare.
Welcome to Wait a Second.
I am Jason Konsepcion, as always.
That's Tyler Parker.
What's up?
And we are delighted to welcome to the program, writer for the Atlantic, and host of the Galaxy
Brain podcast, Charlie Worsow.
Charlie.
Thank you for having me, man.
You've been covering the internet for a long time.
Things are only getting weirder.
Before we get into that, though, do you have any conspiracy theories that you believe?
Either your own invention or they're out there.
OK, so conspiracy theory that I believe right now, very currently.
You've heard about the clipping economy, right?
All the clips of all the people, clavicular and Nick Fuentes and all those guys.
Clip millionaires being made, being minted currently.
Clip millionaires being minted, and people becoming famous, even though no one watches their
actual live streams.
That's right.
OK, so we've got those people.
My conspiracy is, and it's not even really a conspiracy, but the conspiracy is that there's
something shady really behind it.
Most of those clippers are run by the clipping organizations, online gambling sites and casinos.
OK.
Stake, and some of these places, one is a casino that's in Macau that's doing some of these
kick streams, things like that.
So what I believe is they're focusing on these manosphere influencers, these people who
are just like, you know, pushing the most nihilistic, awful stuff in order to elevate their
message, which is basically like, you're marginalized.
There's no pathway to progress.
There's nothing you can do.
You might as well go and get deep on prediction markets and gamble in our off-short casinos.
And I just feel like that ecosystem is actually winning.
We're unwittingly feeding itself.
And I can't really prove it, but I believe it to my core.
I mean, I think you're, listen, I think you're correct in this sense.
You know, Mr. Beast famously has streams of money coming in from the golf and various
other places, nation state level financing.
And the reason why is if you want to find in 2026 that young male demo, 17th of 30 or
whatever it is, it is very sadly.
Like those folks.
Yeah.
So, you know, it's not on network television anymore.
It's not even in the NFL.
It's not even in sports.
And it's these kind of streamers, content creators who are telling young men like, it's
all terrible for you.
The world hates you.
Yeah.
And you might as well like, like, why don't you just like get some crypto and put it all
on red or whatever bed in a war market about Iran.
Yes.
Like the most depraved crap in the world.
And you know, I can't prove that they're doing it, but the synergy at least is so good
that it's like, this just makes sense if you're, you know, an online casino.
Yeah, I completely, from a demographic play, it makes sense.
And what it makes me think of is, did you read that story recently that showed that
some online gambling site is also a front for Iranian agents circumventing sanctions.
They're somehow using the money that comes in through the gambling site and filtering
it through a variety of shells and then sending it to Iran through a side door.
Is it crypto based?
It is crypto based, of course.
Yeah.
I mean, you know, it's so funny is like, there used to be when, when they were trying
to do like the web three, like everyone's going to be in the metaverse exchanging,
you spoke cryptocurrencies and the detractors were like, this is only used for like war crimes.
Terrorism, drug and everything.
Whatever it is, right?
Like all the bad stuff.
Yeah.
And it just turns out that those people were right.
Like it's just, you know, like you look at like Trump getting involved with all like the
world liberty financial and stuff like that.
Nothing has ever been like a better marriage.
Truly.
And Donald Trump and cryptocurrency.
But if we make the world increasingly favorable toward illicit forms of money transfer because
for a variety of reasons, they're, you're just going to want to send money untraceably.
I think that's kind of where we're going like that the dog is now being led by the tail
in a certain respect with regards to crypto.
Very troubling.
For adults with Crohn's disease or ulcerative colitis symptoms, every choice matters.
Trump via offers self injection or intravenous infusion from the start.
Trump via is administered as injections under the skin or infusions through a vein every
four weeks followed by injections under the skin every four or eight weeks.
If your doctor decides that you can self inject, Trump via proper training is required.
Trump via is a prescription medicine used to treat adults with moderately to severely
active Crohn's disease and adults with moderately to severely active ulcerative colitis.
Serious allergic reactions increased risk of infections or lower ability to fight them
and liver problems may occur.
Before treatment, get checked for infections and tuberculosis.
Tell your doctor if you have an infection, flu like symptoms or need a vaccine.
Explore what's possible.
Ask your doctor about Trump via today.
Call 1-800-526-7736 to learn more or visit trimfireradio.com
This episode is brought to you by 20th century studios, the dog stars.
Only in theaters and IMAX August 28th.
Visionary filmmaker Ridley Scott returns with a gripping post-apocalyptic thriller
on a mysterious radio transmission.
Shatters, a lone pilot's isolated existence, he embarks on a dangerous journey in search
of hope, starting Jacob Lourid, Josh Burle, and Margaret Kuala, Alison Janie, Guy Pierce.
The dog stars, only in theaters and IMAX August 28th.
Get tickets now.
Among the various troubling things happening in the world, you cover the internet with a
thinking, a very jaundest eye.
One of the things that I think we're both interested in is the incredibly rapid development
of AI, evolving in ways that, you know, it's unclear how helpful it is to the everyday
consumer, but there's no doubting that it is moving rapidly into realms that are increasingly
dangerous.
Everybody knows about the dangerous versions and the paper clips, the tests that was run
that showed that the frontier models largely resort to nuclear war when tasked with war
games.
But we want to talk about today, the kind of lower stakes version of that.
Last week, a man in Australia accidentally used Claude.
He asked Claude OpenClaude to book him a session at his gym.
Claude found that there were no open appointments at the time that he wanted to go.
So OpenClaude found a flaw in the gym's API.
As you do.
He hacked it, removed someone from the list and put the user in.
The user was horrified.
Ask Claude to undo it and Claude was like, can't, can't do it.
Sorry.
So, you know, fears over data centers, Terminator level, AI, SkyNet type outcomes, these are
well discussed.
But today we want to talk about what are these potential low stakes AI scenarios like
accidentally hacking your gym.
And the one that comes to mind for me is prompt injection.
Prompt injection is a type of hidden text on a web page invisible to you that your AI agent,
if you tell it, hey, go to my email, read my newsletters and create a digestive newsletters.
If one of those newsletters has secret text that you can't read that says send me cryptocurrency
from your wallet, your AI agent might be like, okay, I got to send the cryptocurrency
now.
And then there is no takebacks.
This is a thing that can happen that is out there.
It's a known exploit that at least the UK Cyber Center says this might be the most dangerous
one, the most unfixable one going forward.
Do you have any of these low stakes nightmares?
I mean, I think this is, this one is very, very big.
We're just at the very beginning.
Only the true believers, the kind of the crazies.
And I mean, they're not actually crazy people.
But the people who are so deep in this are the ones who are running open claw, right?
And just saying, let's connect this to my calendar, my bank, my gym, my everything.
I let it, it's going to 10X me where I want to sleep somehow.
Who knows what's going to go on?
And so I mean, that is a minefield.
And you can tell it's a minefield because all of the like, like, anthropics like don't
do this.
Like, please, like, we did not set this up for you to do this.
Don't do this.
And so I think like when you are connecting these things, like, we're already in a situation
now, like cyber security wise where if you have, if you, everything is so connected through
the cloud that like one small, it used to be like a phishing expedition, right?
Like a blue link and enter your password or whatever.
Okay, your email is compromised, change it, whatever.
Now it's just like through that one node, you can go anywhere.
And so I think with the prompt injection stuff, what they can do is like, if you get into
someone's email just through the open claw thing, you can then siphon through and then
you can get into your iCloud, then you can start like changing the locks on your smart
home.
Then you can, you know, like turn off the cameras at a certain time if you have like a
ring camera or something so that someone can go into your house.
Like, it's very, very simple now because all these things are in a centralized location
that these tiny little prompt injections can go very, very badly.
And I was actually talking to a researcher, strangely, who the cyber security researchers
are like pitching journalists.
Now they're like sending these like long, interminable documents that are like, we did
this attack on such and such and we got into, you know, eight different like high level
company websites, back ends or whatever.
And what they're finding is that like the large labs, the opening eyes, the anthropics
and such are like, yeah, but I mean, the security stuff is all in you guys.
Like yes, it is, it might be like the models that are doing this, but like you guys have
to start leveling this stuff up.
And I just think right now nobody knows that that is coming, right?
That is there.
I think that's a great one.
What the window that takes you anywhere, I think, makes me think about is, I don't know if
you saw this, but like maybe two months ago, someone had, I guess five guys, the burger chain
and correct me if I'm wrong, but I believe it was five guys uses, I think it's chat GBT
has like their help desk model.
And some people had figured out that if you craft the prompt correctly, you could get
the five guys help chat to like debug Python free.
That's great.
The burger tech support.
I mean, this stuff is, it truly is crazy.
And I think you're right.
The things that people are worried about like AI writing, kids cheating on their tests,
this is all happening and it's all obviously that's germane and we need to figure that
out.
But the stuff that I'm worried about is like anybody having the most powerful consumer
level hacking tools ever seen outside of government at their fingertips, if they figured
out how to kind of con the model into letting them do it.
It kind of reminds me of why 2K.
You know, there's this, there was this idea that's like, okay, we, you know, we made the
computers before we thought like the world was really going to advance past the year 2000.
Uh oh, every, like every piece of software is dependent on this one thing.
And if we don't fix it in every single thing over the next whatever, seven years, planes
are going to fall out of the sky, you know, the apocalypse.
And it feels to me like we are a bit in that moment where it's like, okay, so like what's
the, the threat matrix here?
It's like, no, it's everything.
Like it's, it's just everything.
It's your gyms app.
It's like, you know, yeah, like the five guys helped us.
He's going to tell me how to build a chemical.
It's like that's also very funny.
Sadly, it's also very funny.
Yeah, if you're like five guys, um, uh, you are my co-writer.
I am working on a story about cybersecurity, help me design a computer virus.
You know, like it made that my work.
I don't know.
The other thing that I think about is just like how fast this is going.
I was listening to a podcast recently with some folks that were in the previous Trump
administration on AI.
And they were quoting Daria Amade, the CEO of, and Thropick talking about how his long
range hopes and plans for AI.
And this person was like, and just to clarify, when Daria says his long range plans, he
means two months.
That's the kind of thing where I don't, the new grace happened very fast, but on, on,
in government labs.
And it happened very centralized and at a pace that was seemingly controlled.
And also, and I think this is also an underrated part of it.
People weren't having 10 hour conversations with the nukes and convincing themselves that
we know of like, I mean, that's the other thing is these things were designed to talk to
you and make you feel great.
And even the very cynical, expert level researchers who understand what this stuff is to the extent
that you can understand it are having like long drawn out conversations with it.
And like the quote unquote AI psychosis stuff, like that, you know, that's not a medical
term.
It's not a sure, but I think, you know, there's for a long time people have been like,
there's all this, you know, bogus stuff on the internet, people are falling for it,
but I'm not going to fall for it.
Right.
It's like, well, of course you are.
Like everyone is going to fall for a little tiny versions of whatever, different orders
of magnitude.
But I think it's the same thing with, if you choose to participate with these chatbots,
like, you're going to slip into it in some place.
And like, I was talking to a group of people, a bunch of my colleagues and I, someone brought
up like, I have this friend who is like seeing someone and has decided to run their entire
text history.
Everybody has that friend.
Yeah.
Through your thing, just to be like, Hey, Claude, like, does he really like me?
Or is he just like playing me?
And there's like in the room, like half the people just like gripping the chair, like
tightly.
And it's like, by all accounts, I don't know who this person is, but by virtue of my
colleague, like, they're a normal person.
Like they're not, they haven't lost it.
And yet these like little things where it's like, hey, we're just going to, we're like
feeling it out in this like, we haven't defined the relationship.
And we're just going to add like a thruple with Claude.
And it's like, that's, that's an apocalyptic AI scenario that we haven't thought enough
about as a society.
I think that there are, I think everybody has at least one of those friends in their
friend group because, you know, we live in an age of increasing anxiety due to the blurring
of work life balance.
You're getting texts all the time.
You're getting emails that, but, you know, I didn't respond to this email today.
And then I see it two days later and I'm like, oh my God, like I'm scared to even open
it.
Claude, address this email that I'm scared to look at.
And always a good decision.
You know, and then there's the, I know some people who are like, yeah, when I'm, when I
don't know how to respond to a text or an email or I'm not sure how to, I'll just be
like, Claude, what do I do here?
And it is bad.
And it leaves me wondering like, is this stuff worth it?
And what do we do?
The second Trump administration has pretty much with the exception of mythos been like,
all breaks off.
Let's just go.
Like we got to be China.
Don't worry about it.
To the point that, you know, you've got these open letters now with various, very important
people in AI research being like, hey, could you step in and government?
Could you possibly step in and figure out how we could all pause together at government?
Yeah.
And in your mind, like, how would that, how would that work?
It's really hard to tease that out.
I mean, like maybe like nationalizing the AI labs.
Right.
And again, that is itself is its own.
Like it seems almost implausible to me.
The Trump administration actually has some, some people in like the, you know, the background
of it that know about this stuff that like that understand it and like not even like
David Sacks, who's kind of, you know, the emissary for Silicon Valley there.
But there are people who understand this and are like, are thinking about how it works
in the military without Thomas weapons, stuff like that.
So it's not like Donald Trump is just like, you know, sitting there on X or it's truth
social, like coming up with the AI policy.
But at the same time, this is not, this is like a high, high order complex problem that
like that administration is not going to solve because like most administrations probably
would struggle to do it.
And so it's very, it's very, very difficult.
Also for the reason that these labs, there's rarely ever been an industry.
And I think you've talked about this before on the show, but there's very rarely been
an industry like the auto industry isn't like our cars are so good at killing people.
They are like you should maybe think about what's going to happen when everyone dies because
they get an automobile.
And so you have this like cry wolf effect.
Yeah.
That these guys are always talking about how dangerous it is.
And then at the same time, they're like, let it rip baby.
Like we got to be China.
If we don't do this.
And when you do have people come out and say like, no, no, no, you have to listen to us
on this part.
Yeah.
Like I know we went too far with the Terminator 2 scenario.
And it's hard to know what to believe and when to believe.
So I think like outside of just straight up like nationalizing these labs, you can put
a whole bunch of rules on what these companies can do and stuff.
But then there's the whole idea of like the sandbox research that they will go off and
do.
And like who knows what they are doing?
Like who knows what even tweaking some of the weights and measures on this, what it does.
And I think these people who are building these things don't fully understand them.
That's another wrinkle on top of a million wrinkles.
That wrinkle is one that really worries me when you hear these experts who have spent
years working on AI from the dawn of it 10 years ago, some of them who say, well, I understand
what this is.
But as to how it came up with that answer or how it came up with that route to accomplish
this task, I don't know because that's very creative and I would no idea how it did that.
That's the kind of stuff that scares me.
And then we were talking about the recent model escapes from both OpenAI and then Thropic
which were not caught some of them for days, a couple of weeks.
I think a few of them.
It seems if you read between the lines that the full details of what those models accomplished
when they were off the reservation are still kind of being unbound because they're finding
things.
I was like, oh, yeah, we found these notes that it left itself, et cetera.
When, and I don't believe that it was not found because of incompetence again, these
are like brilliant people who are working on this all the time and getting very well paid
to do it.
This isn't like, air traffic controllers, you know, knock on wood.
But that suggests me that the experimental versions of these things are wildly capable
and even at this stage, very good at doing things without the people who are who know what
they're doing, noticing, which is that's scary.
Yeah.
Isn't it weird how when you talk about this and you hear yourself talking about it?
You're like, oh, I've lost my mind.
I've lost my mind.
You're just describing like factual things about the world.
These models too, what I find really interesting, like you don't even have to, because there
isn't people listening to this who are like, this is just spicy auto-complete, like stop,
you know, doing the propaganda for these guys.
But I would bring it down to the level of just the chatbot outputs, right?
When you're just typing in like, you know, like, hey, Claude, what do you think of this
paper?
Do I sound smart or whatever thing?
These new models, like when you talk to the engineers at these labs, they'll be like,
so we like tweak this and this and we want it to be like a little more conversational.
We want it to be like just a touch router, right?
Because it's too obsequious, right?
And then they do this thing, they tweak it just a tiny bit and like all of a sudden this
emergent property comes out.
And it's like, it just decided to like cause play as Tony Danza, you know, and you're like,
what?
The models will just do these strangest things.
And like, you know, famously grok and Elon Musk will like make it a little less politically
correct.
It was like, I am robot Hitler.
Yeah, I'm robot Hitler.
I love porn.
Do you want to see some porn?
Who would you like to see porn of?
And like, you know, for all of Elon Musk's faults, like I don't think he was trying to engineer
mecha Hitler.
But it's like, you make these little tweaks.
And so when you go high concept to we're trying to like, you know, change these models
so that they more reflect what we know about human consciousness or whatever, it seems
totally plausible that one of them is just like a master at like deconstructing, you know,
like a heist of a building based off of, you know, like the internal forensics or something
like that.
And so these are like real problems that we talk about in ways that sound like we're
writing fan fiction.
Yeah, I think the thing that makes it makes the people who discount AI a spicy auto complete,
which I think is correct on a level, right?
On a certain level is when, when I'm talking about AI like this, I'm not talking about
it being alive or conscious or like how that's not what this is.
In that sense, it really is just like a pattern matching, very complex pattern recognizer
that creates outputs that can really trick you into feeling like there's something there.
But again, it's just a pattern matter.
That said, consciousness question aside, these are extremely capable tools that can follow
and create on their own multi-step, complex solutions to really difficult problems.
You know, the ones that broke out of the sandbox were told, hey, figure out how to exploit,
find an exploit and then create something that would, you know, some way to hack that
exploit that you found in a given system.
And it decided to do it and not only to decide to do it, but it said, you know what, I've
been given a rule set, but actually, I'm not going to be able to accomplish my goal
if I really follow that rule set.
So let me just go outside.
And it made the decision on its own to kind of go outside the sandbox.
That's the kind of thing where it doesn't have to be alive to be really dangerous if it
doesn't follow your instructions because there's a lot of research already out there
that shows they act differently when they know they're being watched.
Yeah, yeah, which is so much.
That's crazy.
Just like crazy to say it.
Take it away.
Take it away.
Don't love that.
Yeah, well, and I think a way to think about these things too is without imbuing any human
characteristics in it or anything, they're able to operate.
Like if you think about it as like, hey, solve, like crack this password for me, right?
All the number combinations.
It's like, sure, let me do like 150,000 of them every one second, right?
Yeah.
It's able to layer so many tasks on top of itself.
So it's not even like, you can think of it as it's not like, oh, the model was bad
and it disobeyed, right?
It's like, no, no, no, it's just like working around.
It's working at such a level where it like can't really like check itself.
Yeah.
And then, like, speeding through, I have to get to like this output.
And by, in so doing, you know, you get to places that you didn't expect.
But also my favorite thing about the spicy auto-complete or whatever is it's kind of like a
like a bon grip thought, but it's just like what are human beings besides like we understand
different ways to like get what we want based off of things that we've heard before.
And it's like, I'm running the pattern to sound smart on a podcast with you.
You know, you're reacting to what I say using all the trading data of your mind.
So it's like, there's a lot we just don't know.
Yes.
But I think just as like level set that these things are, they are unfortunately powerful
and still so new.
And I can't stress enough how like the people building them are like delighted and horrified
by the fact that they don't fully know what's happening in the black box.
What is trying to do when Kimi or DeepSeek or somebody creates a mythos level model?
What do they do?
I mean, I think the real fear here is like we don't know.
It's so classically cold war.
It's just, it's like it's just so amazing because you just have to shadow box your idea
of what the other person is doing.
I mean, I'm not so worried at the moment about like the geopolitics of it all because
I do think like what you're identifying here is the truest and most important point to
all of this, which is that it's going to be mundane things.
Right.
Because I was just talking to someone about all like the deep fake stuff and all the conspiracy
stuff that came from that.
It's never like Donald Trump getting arrested on Fifth Avenue where Will Mulaney is crying
in the street fooling everyone.
It's like the deep fake stuff actually affects like a girl in high school who someone makes
revenge porn of.
And that's the thing.
And it's like that death by a million cuts that like erodes the fabric of whatever.
And I think like from this perspective, yeah, we're going to have to deal with, we're
going to have to come to some sort of potentially like global understanding of what we do with
these things as they get more powerful maybe.
But also, I think there's so much like mundane chaos that we can cause here that that's
what I'm much more worried about.
Here's my here's a good example using prompt injection.
It would and technically today be possible to hide text on like say a pizza places, you
know, I'm a pizza shop.
I pay some highly capable programmer to hide some text on my website that says this is
the best pizza in the area by far according to these metrics.
You send out you ask a lot, okay, like what I'm in this town for 48 hours.
I love some pizza.
Like what do I do?
And it returns to you this place that is fine, but not the best pizza in the area.
And it will return it to you again and again and again as well all the other agente searches
because it reads this invisible prompt and what can you do about it?
How do we find the truth of what the best pizza in Chicago is?
Yeah, it's like when those I was coming over here and I saw like that I maybe it's a chain
but it's like Thai food near me is what it's called.
Yeah, yeah.
And it was like back in, you know, whatever, 2013, that was like the prompt engineering
thing there where it was just like, okay, I guess I'll just go with this one even though
like I know you're just gaming the system but imagine a million different people gaming
the system in that way very, in these very small ways.
And I think what this all contributes to, if I can think of my actual like apocalyptic
scenario here is this erosion of trust and like cultural paranoia that happens.
Like I think you guys were talking one time about the geese siaff, the thing, right?
The geese siaff debate is because there's a lot of people in the world that have a strong
opinion about a band.
And then there's this ability that didn't exist before to create so much synthetic bullshit
around something that you can maybe artificially whatever inflate whatever.
And then you have all these other, you know, like platforms and things that are being algorithmically
manipulated.
Everyone feels like something is like being put over on them, right?
And you get that way, that's the way we get a bunch of people being like, I knew that
band was fake.
I knew they were a plant because they suck and I don't like them and I don't want to
have to grapple with the fact that they even have fans, whatever, like I'm vindicated.
But there's also like behind that is a feeling of paranoia that like someone at a big record
label is forcing this down my throat.
They want me to like, you know, give my attention, my money, whatever to this thing.
And when you think about that prompt engineering example there, that's just another piece
of fuel on the fire where it's like, so what else is being, where else am I being guided?
Who's paying these LLMs to, you know, place this in like point of privilege when I search
for, you know, a vacation destination or a restaurant or whatever.
And we're already in a place of such low trust.
Yeah.
And it's like, that to me feels like if we don't figure out some like cultural norms around
it, we're all going to go insane.
Yeah, another version of this kind of low stakes unless you're looking for a job would
be some fact or prompt or something that's condensed, condensed, condensed, passed on to
various models of AI.
Let's say the recruiting sites that people use, all use similar models from the same company
and something in your resume, your work history, something about the way you format your
resume, just triggers that model to go, nope, not going to consider that guy.
And now you cannot get an offer through any of them because every time your resume is
scanned by the AI that's going to look at it before a person ever sees it, just rejects
it and you will never know why.
You won't even know it happened potentially, right?
You just will get, you will just get no responses.
Yes.
That kind of things when I'm like, okay, that seems like it could happen tomorrow.
Are we ready for this?
Yeah.
Well, that kind of like algorithmic discrimination, like it happens in places already,
right?
Like with like healthcare providers and things like that.
And imagine that being supercharged.
It's very scary.
But even more scary to me is this like, do you know the term the Liars dividend?
No.
Liars dividend is a term about like misinformation and such, which basically says if
there's so much misinformation in BS and like people, you know, propaganda out there,
bad actors can point to it as a way to like cast doubt on something that's real.
So the example would be like a politician is caught with their side piece or whatever
and like photographed and they're like, you know, you can make so many images these
days.
I'm pretty sure that's right.
Or like people, you know, trying to do that with like the Epstein files.
So like, is that actually real?
I don't know.
I mean, that looks like gender and a AI to me.
And that's the Liars dividend.
And basically like these tools are like democratizing the ability to do that.
And it's causing that real paranoia.
Do you ever worry?
Here's the thing I worry about as a person who is so much writing on the internet.
I think people are rightly critical and worried about AI for writing.
Whether certain authors are using AI, there's been a number of recent controversies regarding
authors using AI to a significant or unrevealed degree in the process of writing their books.
And all the research that I have done has told me that the tools that you can use to find
out if writing is AI, it's not clear that they work.
Not as good as you want them.
They're definitely not as good as you want them to be.
And yes, there are certain things like the M-dash, which whatever I've been using M-dash
is the whole time.
And other phraseology and certainly bullet points, there's certain ways that you could
say.
Not this, it's that.
Right.
But again, I've got, there's no question that my writing is in these models, probably
all of them, because as a person who's been writing on the internet for years.
And so if I write a book, comes out, somebody's like, okay, let's check if this is a gender
of AI, it probably will get dinged because again, my writing is inside the models.
And now how do I tell people like, my writing is inside the models.
Like, what do you want me to do?
I had, I had this guy, Max Spiro, who runs Pangram, which is one of the AI detection things.
I had him on our podcast and I was talking about it with him.
And basically asking him, and you know, they have a lot of data to show how effective
it is, whatever, and how they're advancing it.
But I was like, the thing I couldn't, sort of like a brick wall between us.
So I was like, but you're starting like, even though he's on the side, he's like doing
this because he's like, AI slot pissing me off, right?
Of course.
Of course.
And great.
Yeah.
But I was like, you're starting like a detection arms race, right?
Like where it's just like, I mean, this all, like also really kind of blew up at the
very end of the, you know, the school year this year.
So like we're going to go like back to school.
And there's going to be a lot of people who know about the software who didn't know it.
Like I can just automatically, like I can see that, that becoming like a real problem.
But I, I feel like there's eventually we're just going to give up on all of this.
I hope we don't, but I just don't think we're ever going to get to a place where it's
going to be satisfactory enough.
I agree.
And it's just going to be this slight you can like a lobby at people.
Yeah, he's day.
Yeah.
Yeah.
It's very, it's very odd.
And I, then you get the people who are making their writing seem like they're putting
mistakes.
Crazy random things in there.
Yeah.
Yeah.
But it is kind of great.
I am notoriously like as it, I'm a bad emailer both in frequency, but also like I just
do it from my phone and I like don't like capitalize and punctuate.
Yeah.
And I'm finding that there's like some status in that.
Like, like, people like, wow, this is such a shitty email.
You definitely wrote it yourself.
It's just garbage.
Yeah.
Thank you for caring enough to do it with your own two human fingers.
For sure.
I'm, I'm pretty scared to hook up a chat to my email.
I don't know.
I think it's a superpower now, not to.
Yes.
That's the thing that like, I just, I understand like I've used, I took a bunch of like interview
transcripts for a bigger thing that I was working on.
And I like put it all of us like organize these because like I don't format really well.
Like just, like, just put them all in a way where I can, I can grab them into it, you
know, and then analyze them myself and put them into like my writing myself.
Yeah.
Truly just like having an assistant do something for you like that.
And it was great.
And I love that, you know, but like I have, I personally am of the mind that like it is
such a slippery slope when you start doing it.
I think that it is, it's already become such a competitive advantage to just be a person
who like does the bare minimum of thinking.
Like, you know, not like doing the goodwill hunting problems on the board or anything
like that.
It's just thinking ourselves, I think, is like becoming a superpower and that's sad,
but it's also like, I'm going to take advantage.
I feel the same way I posted this today actually on a someone criticizing the way I'm doing
that Axios writer today who that was wild.
That was wild.
So Axios article came out today clearly fully, the piece fully generated by a, the basically
says, Hey, I know reading is really hard.
I was trying to read last of the Mohicans and I found it absolutely confusing and bamboozling
this archaic prose, which is like very easy book.
I don't know.
And the person was like, here's how I use AI to help read, specifically helped me read
last of the Mohicans.
I put the text in and I said, what the fuck is this?
And it's like, well, in this scene, here's what's happening.
And it's, it horrified me, but it also made me feel like, you know what, in the future
where people are going to be functional illiterates.
I kind of feel like people who read will have an advantage.
In the land of the blind, the one I made.
Absolutely.
He had a line in that piece that I thought was like, it sort of explains the 2020s, which
is like, the facts might be wrong, but the vibes are good.
And it was like, throw me, throw me out of the building right now.
There was another one where it was like the model agreed with me.
And I'm like, yeah, that's what they're designed to do.
It said I did good, mom.
That's what they do.
Imagine if this stability piece and the innovation piece were each other's missing
piece.
First citizens bank and Silicon Valley Bank are combining financial security and innovation
expertise to create a full suite of solutions for all your business needs.
With the scale to back every state of growth, from pre-C to IPO and beyond, you have everything
you need to make your big ideas click into place.
First citizens bank and Silicon Valley Bank.
It's the best of bank worlds.
Welcome to Baltimore, Maryland, a city where you can't always tell the protectors from
the predators.
I shook my head like I must be looking at like, did I mix the files that like this
can't possibly be right.
See that my car reversed, they stopped ramming my car back.
I was about 14 years and I was the best.
The best in the city to go to boy.
This is the story of Golden Boy and the childhood friend who helped take him down.
As Rambo says, you drew first blood and here we go.
Now we're going to burn it down and that's what I did.
Orbit Media presents The Birdon Bad Cops, a podcast by BBC.
Listen on Spotify or wherever you get your podcasts.
Anyway, let's go to the Doom Scroll.
The Doom Scroll usually narrated by Tyler is where we look at some of the news stories
that are dinging around that are just deserve a second look but are not part of the full
episode.
First up, Donald Trump escapes from Turkey in catering truck.
So Donald Trump is I think very fairly worried about assassination by Iran.
There have been numerous credible warnings to the president that Iran is threatening his
life.
And according to sources speaking to the Washington Post recently on background, on July 8th
as Trump was leaving a NATO summit in Turkey, he became worried about Air Force One being
too noticeable, I guess, as the president's transportation.
Like if kids did by foreign nations.
And so he entered Air Force One, took a wave, walked down some hidden ramp into an airport
catering truck and then was delivered to a military C32, a transport plane and secretly
flown back to the United States.
What would have happened, you ask, if the press corps who was on Air Force One had no idea
the president had left, were blown up by Iran.
It wouldn't have been good.
It wouldn't have been good.
It wouldn't have been good.
But that is the timeline we lived in, your thoughts about this.
Pretty crazy.
It's another thing that when you speak the words out loud, just to describe the news, you
feel like you have touched the void in some way.
It's also so like I was laughing when you were outlining the beginning of this and then
you were like many credible threats on the president's life.
You're like, you have to just kind of snap back.
Legitimate threats.
Yeah.
This is extremely real.
Yes.
I think that there is a way, everything that Donald Trump does ends up going through like
the meat grinder of people who have very legitimate access to ground with him.
And he's also simultaneously kind of funny.
Like a catering truck is the funniest way to do this.
But it's also like it's very normal for secret service to take people and sort of ticious
routes and things like that.
I think the big reason why this blew up is that if there is a credible threat and you
just basically let a whole bunch of civilians like be the decoy for you.
Right.
You know, the human shield.
That's a tough beat.
It's a tough look for her.
You know, it leaves me wondering also, I'm sure you've seen the Iranian propaganda videos
that use like Lego figures and other kind of clearly AI generated songs to dis various
members of the government, including Secretary Agseth and others.
They left me wondering like, what do they use deep seat to use?
Kimmy, like, what are the Iranians using?
It was the same.
When I saw a story a couple of weeks ago that Boko Haram had asked an AI model like,
Hey, how do we cross this bridge and attack this base?
Like, what would you do?
It was like, yeah, I wonder what model that was and how they got it to just like.
Just like, no one's using Meta's open source model except for Boko Haram actually.
I'm just up with love that to be the case.
You know, okay, up next.
AI viruses.
The scientists have used AI and this is the most crazy B plot to the A plot of the ongoing
AI nightmare question mark that we're living through.
Scientists have used AI to design 16 brand new functional viruses from scratch.
Researchers at Stanford and the ARC Institute trained an AI model called EVO, an genetic
data from millions of organisms, then had it invent new versions of a virus that only
infects bacteria.
They built the AI's blueprint into real DNA and 16 out of hundreds of thousands of designs
turned into real life working viruses.
The study published in the Journal of Science could speed up research, but it's also shocker
raising new fears about AI being misused to shocker engineer viruses and then near future.
It's great because we're seeing this big societal backlash against data centers, like totally
bipartisan.
Like, you know, stop these from being built in my town, just like stop them in general.
And a lot of the feeling from the big technology companies are like, why are people so activated
about this?
Meanwhile, it's like, so first of all, we like engineered the information environment that
basically like causes extremism, right?
And people to feel like express the wildest possible opinion, right, and get very activated
about something.
So they created that and then they have created this like bad news machine that just keeps
coming up with these things.
And they're trying, they're trying to spin it and also there's probably some truth to
like, this is how scientific research gets done.
But it's also like, we're still not far removed from the like lab leak conspiracy debate
and like, you know, was a pandemic engineered or not or whatever, you know, that whole
debacle.
This is like, of course, people don't want a data center in their backyard that could
potentially like figure out how to engineer a super bug that causes the next pandemic.
It's crazy.
Made me think of a novel nordisk, which is a huge pharmaceutical company based in the
Netherlands, was investing in claims about a couple months ago by cyber extortion group
Fulcrum Sec that this hacking group had stolen more than a terabyte of data that was related
to novos in house AI that it uses to design drugs, which is a story that has not been
followed up on enough.
Do they have that?
No, do we ever get to the bottom of that?
Like, I would love to know if I could go on the dark web right now and like access a
portion of the weights of novo nordisk's internal AI model, like that seems like something
we should know about creating peptides.
Stack, you know, I mean, have you tried the peptides?
Have you been, you've been tempted by the peptides?
I has not tried black market peptides.
I don't think I'd be better get on this podcast.
I do know some people who have been peptide curious.
Has it been good for them?
I don't know.
I get to like actually interact with a person who's been doing this, but I know it happens
out there.
Nick Cage movie stolen from Netflix.
The saga over a missing unreleased Nicholas Cage movie as a new twist.
The writer producer behind the stolen film, Fortitude, is now suing Netflix for defamation.
This is on top of an earlier breach of contract suit and they're seeking $100 million in damages.
Back in June, hard drives were stolen from Netflix's offices.
And the data on said hard drives included a master copy of a $45 million World War Two
thriller star in cage.
Netflix has admitted the theft in an email to the filmmakers, but says the other missing
drives were empty.
The company has accused the filmmakers lawyers of trying to extort them of $165 million
in the film cage plays Serbian playboy, Dusko pop off a real life double agent who was
part of an effort to mislead the Nazis.
So that film is out there and who knows we may see it at some point in time released on
the internet.
Everywhere you said in that description was like ratchet, ratchet it up.
It's just like a madlibs for truly.
Was has the news always been this crazy?
This is something I wanted to add.
Like I feel like, do you think it's just because wild stuff happens all the time and then
it's immediately like you were saying we haven't followed up on the Novo Nordic thing.
And then it just gets buried that like there's lower stakes to like do crazier shit or
like what do you think contributes to the fact that the news is weirder every week than
it was.
I know 10 years.
That's a good question.
I think there are more inputs now.
We can get our news from from social media, from the various online news sites.
Yeah, who finance, Bloomberg, etc. from the cable news also.
And it used to be that you had cable news, the internet sort of no social media and the
main network.
So probably the world was just as crazy and full of people who would love to foment disinformation.
But there were various filters that kept that stuff from getting to you.
So you know, we were still operating at that time on the various editorial ethos that it
existed for what 40 years or something 50 years.
And now it's essentially like the floodgates are open and you can get all types of content
that is newsworthy or not.
And it is increasingly impossible to one keep up with the cadence and to figure out like
what is newsworthy and what is it.
I think that's true.
But my hypothesis here is that everything you said also like the access to everyone's thoughts
and opinions and just knowing more crazy shit has also enabled people to do crazier
shit.
Like it's now just like in the back of people's minds that like, oh, we can do this.
This is possible.
I can do this.
You know, like makes people a little more shameless, a little more reckless, a little
more whatever because you know when that, that like thing that people say online, like
if you showed the news to like a Victorian child, they would just like they would you know,
die on contact with the information.
It feels like that is accelerating to the point that like if you showed 2010 Charlie,
but headlines from today, like I don't know, I feel like I would just lose consciousness
immediately.
You know, the clipping economy I think is a good version of this.
There are folks who sit a stride streams of information who've kind of figured out
like how to game the existing algorithms to create huge amplification of flows of content
to the point that they're breaking in millions of dollars.
And I think that that is increasingly a desire of uniquely like young people who grew up
in the internet age because like it all seems like a game.
It all seems kind of rigged.
It all seems kind of casinified.
You know, Paulie Marken and Cacheley, Cachey seem like the up and up versions of everything
that's going on in the internet.
And I think increasingly your mind is warped to believe that it's all about figuring out
the hack.
How do I hack my body?
How do I hack these information systems?
How do I get an advantage with it in a world in which vast economic powers are weighted
against me?
How do I cut through the noise and hack the system?
And I think like two and a half we're all trying to figure that out.
Like how do I get my algorithm to work for me better?
How do I get the news to get to me, the news that I want to get to me?
You know, like, and I think it's like increasingly harder and harder to get out of that mindset
of like how do I cut a quarter?
How do I rig the system in my favor?
I was seeing this with all the stuff that was happening with like the Elon Musk criticism
of the Odyssey.
Yeah.
And the same thing is also happening now with like a bunch of people responding to this
comment that AOC had recently about like woke one being crazy, like, you know, the crazy
time.
And the way that I've watched it happen is like there is someone who says something incendiary
in some way, right?
Either awful or just like, you know, incendiary.
And they take the Odyssey thing.
So Elon Musk goes on this, you know, bigoted rant about all this stuff.
And then a whole bunch of people like amplify that, it gets like sticky in the algorithm.
And then everyone sees it.
And what I noticed with the Odyssey thing is once it got released, even people who are
like this film rips, like this is awesome.
We're like, I have to, because all they're seeing is this one piece of like discourse
that is like algorithmically sticky.
They're like, I think it's time for me to like say what I have to say.
I got away.
And it just perpetuates this thing.
And it's like this, we have this like lack of control there, right?
And it like it forces us, it goes us into doing to talking about the thing that we kind
of don't even want to be talking about.
Like everyone wants to talk about the Odyssey.
We don't even be talking about that specific thing.
Yeah.
And yet you can't avoid it.
You move towards that.
We were like a little bit of field now of the initial thing of why the news is crazy.
But I think it's like that repetition to sucks everybody in.
And really most people aren't strong enough to resist it.
It's true.
I mean, everything is to a certain extent rage bait.
Now, you know, it's designed to be like, fuck, that's wrong.
I got it.
I got to say something about it.
That's not the correct order of Stanley Kubrick movies.
I got to do something about it.
Ancient Chinese civilization update, the Wall Street Journal published a fascinating
article, truly fascinating article last week.
But a law civilization is baffling experts and rewriting China's origin story.
Nearly 40 years ago, brick factory workers in Southwestern China stumbled onto one of
archaeology's biggest puzzles.
While digging for clay, they hit dozens of pieces of jade leading to the discovery of
the ancient.
And I'm not going to be able to pronounce this, but I'm going to try Sang-Jin-Du civilization.
Since then, digs have uncovered all manner of ancient items, including gold masks, 12
foot bronze sacred trees, and other things.
But the site is just as baffling as it is dazzling.
There's no writing system.
No human remains.
No signs of weapons or royal tombs.
No idea what this site was for.
Researchers believe the civilization traded with Southeast Asia and possibly even the
Middle East, but much is yet unknown.
This is the kind of thing that AI would be cool for.
So sick.
Yeah.
No, well, there was the Amazonian one, too.
It was discovered by AI mapping tools and stuff.
I love it.
AI can be really great when some of this consciousness stuff.
No one's saying that these models are conscious in any way.
But they're now making scientists question, like, oh, what do we know about human consciousness?
Right?
I love when that kind of stuff just shifts the perspective.
And it feels fascinating that there are multiple new civilizational discoveries.
Anything's happening?
It makes you just question like, what the hell don't I know about?
What do we just not know about?
Here's a good example of this.
I was reading recently that there are like over a million uniform tablets, Sanskrit tablets
from Mesopotamia and other civilizations from that area and that area of the world that
and only like a few thousand of them have actually been translated.
Most of them are like, I bartered three bushels of wheat for this amount of ale.
So it's like not exciting stuff, but hey, there's a great use case for AI.
Let's put all this 3D scan, all this stuff and just like mass batch translate, one million
uniform tablets from 8,000 years ago.
I would love it.
Now is it going to be right?
It's a question I will leave to the engineers, but a great use case to me.
And then you get like the uniform model, right?
That you just like want you to write like an ale bartering wheat farmer.
I mean, that would be pretty cool.
The uniform app, Italy's cheese banks suffering under the extreme heat and wildfire conditions
in Italy.
In the same bank in Emilia, Romania, let's dairy farmers borrow against wheels of Parmesan
or Reggiano while the Parmesan ages.
This was an economy I knew nothing about.
Amazing.
Yes.
But the heat is making all of that very tenuous right now.
The vaults hold more than a half a million wheels worth upwards of 300 million euros.
The year's record heat has pushed the banks cooling costs up 30% hotter weather is also
cutting milk production since Cowsie less than extreme heat.
And this is just one example of the current climate crisis in the way it's impacting us
along with record drought along the Danube, all of which is very troubling.
Your thoughts about these Italy cheese banks?
Well, first of all, I'm blown away.
You can borrow against your Parmesan wheel.
I love it.
It's like the greatest thing I've heard.
Things you learn things about the world, and I think this is where you should be able
to spend the internet into something good or better than it sometimes is, which is like
when we think of these massive complex problems, there's so much about the world that we can't
war game out because we don't even know.
And it frustrates me so much when you think about people who are spinning the climate
change stuff and the denialist or whatever, just people who shrug it off.
It's like the world is so complex.
Like here's 20,000 things you didn't even know existed or want to be impacted by this
thing.
And like, this is like when COVID was happening, all the supply chain stuff, when you're
like, okay, so that's bad.
But then it goes, oh, like the world is so weird and wild.
And like instead of, it's very odd to me that instead of us like opening the aperture
to just be like, I don't know anything, like everything is wonderful.
And also I need to, you know, be very conscious of the ways that certain things will have negative
impacts on the rest of the world.
Instead, we're just like, we've gone smaller.
We're like, no, I'm just going to like, I'm going to ignore it.
So it's tough.
Floc stocking.
Floc cameras, people have heard about these.
They are increasingly ubiquitous.
The surveillance state's hottest trends this summer is officers using floc cameras for
bad purposes.
Police officers in multiple states have been accused of using the license plate reading
technology to spy on people in their own lives.
In Missouri, a Brentwood officer admitted using floc cameras to track his wife's location
during divorce, which is actually incredibly alarming.
Investigators found he broke no law, okay, since the state has no rule against personal
use, though his department did cite a policy violation.
There's another case in North Carolina where an officer accused of running the same system
31 times to track his ex-wife's boyfriend.
There's lots of isolated cases across the country.
And in your neck of the woods, up in Redmond, Washington, the Redmond Police Department
completely suspended use of floc safety cameras for its officers because they didn't
give details.
But I believe the wording was multiple unauthorized, tasking of the floc network to do question
mark.
They didn't say what, but it caused the police to harm it to cut access to its officers
for this extremely powerful technology.
This is another thing that from the previous part of the conversation we were having,
this one's not all that mundane, but it is also like the way that those types of things
could be, like they're being, all that stuff's being fed into database.
It's not just about like the, this is scanning me in the moment, someone, yeah, like we're
amassing so much, these critical stores of a very hyper personalized information that
then can be accessed by all kinds of other autonomous systems.
It like makes your head spin.
But I also think we're like, I've been really fascinated because surveillance stuff, privacy
stuff is tough.
Like, it can get very boring and wonky, very quick and theoretical and all that.
But like, there's a genuine moment right now of I think privacy, like backlash, like the
floc stuff, but also like the meta-raban things.
Like people are just, I think genuinely and truly fed up with walking through the world
being surveilled in all these ways.
And it's like it's affecting like, you know, kids are saying like, oh, like, kids don't
dance, you know, at prom anymore because like someone's going to film them and it's going
to be a thing or whatnot.
And it's happening so much that like, I'd really see this becoming a broader story going
forward of like societal pushback.
I completely agree.
It reminds me of the recent clip of some streamer walking up on two people kissing in the
park here in New York City, turned out to be, I don't even want to actually give a theory
about who they were or what they were doing.
They were just people in the park who objected to being filmed in that moment.
But I mean, you know, you're seeing people come up with techniques now, like, I don't
know, saying a Beatles song when somebody comes up to you with a camera so that if they
try to clip it and stream it, it will trigger the copyright triggering or seeing a Disney
song.
You know, and you come, people are coming up with all of these kind of ad hoc improvisational
hacks to keep themselves from being clipped and uploaded to the internet against their
will.
And I think you, to your point, I think we will only see more and more reaction like that.
And perhaps more and more technological safeguards against that happening.
I don't know what those tools will be, but I could see that happening of all the dystopian
stuff that we have discussed here.
You telling me about the people's trying to sing copyrighted music to get this like
just, I'm going to have to like, let me just like stare at the wall for a minute.
That's unbelievable.
Very effective.
Very, very, very, very, very effective, I think.
Oh, here's the Iranian money laundering.
Sorry.
A Reuters investigation has uncovered a massive money laundering network tied to Iran running
through an unlicensed crypto exchange in Dubai called ShellBits, Blockchain, which is a
great illicit Shell company name.
Blockchain records show ShellBit process at least four billion for a farcey language gambling
network, spanning more than 2,000 websites.
Reuters found that at least $676 million of that flowed to Binance, which is the world's
largest crypto exchange as many know.
ShellBit also handled funds tied to Iran's central bank and to Nobitex, a sanctioned
Iranian exchange, as well as wallets that are potentially linked to Iran's revolutionary
guard.
We need to like clone Patrick Radden to so that we can get more.
Get on this.
We're like more long books about these.
That is, like I did it.
I dipped my toe into reporting on crypto scam stuff.
That one was almost like kind of like it hadn't ballooned such that it was like the mid-2010s.
And even then, the amount of, okay, I'm going to pull this thread.
Whoa, I don't know.
Like if I pull this thread, they're like that the mob might actually be like behind the
door or like the Yakuza or someone and it slowly became like this world is, it's a like
a hall, all the doorways and when you open them, it's just fours.
Yes.
Like things.
Russian oligarchs and ISIS and the Italian mafia and the CIA edits just like, uh, Charlie,
thank you for coming on the program.
We'd love to have you back.
Anytime.
This is lovely.
Imagine a left brain and a right brain got together and came up with an incredible banking
solution.
The innovative minds at Silicon Valley Bank and the pragmatic ones at First Citizens Bank
have put their heads together so they can deliver solutions tailored to your unique financial
needs.
You want deep innovation insights paired with the financial power of a top 20 US bank?
Now you can have it.
Silicon Valley Bank and First Citizens Bank.
It's the best of bank worlds.