August 14, 2026 • 43 mins
What’s driving America’s growing backlash against surveillance tech? Taylor Lorenz (User Mag) unpacks the outcry over Flock cameras and why the company thinks they are being unfairly targeted. Then, Gerrit De Vynck (The Washington Post) breaks down Mark Zuckerberg’s latest AI manifesto and Meta’s bid to catch up in the AI race. Finally, Nitasha Tiku (The Washington Post) explains OpenAI’s strange onstage postmortem of the Hugging Face hack… and why this moment feels so surreal.
See omnystudio.com/listener for privacy information.
Listen
Watch
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:11):
Natasha Taylor Garrett.
Speaker 2 (00:13):
I had a rumor that my favorite media publication, The
New York Post, sold out this week in DC.
Speaker 3 (00:19):
What happened, Well, I think the copies at Union Station
sold out after very niche tech policy drama. Basically, The
New York Post wrote a bizarre piece on this man
Dean Ball, who is kind of like an AI policy
guy who recently joined Open AI. He was previously at
(00:41):
the White House. I guess like someone at the White
House probably called The New York Post and like did
a hit piece on him. So there was this like
really vicious piece that was like Open AI risks White
House relationship with hiring a nuisance AI policy executive. A
lot of people on sort of like tech policy Twitter
I think got a kick out of it because in
the actual print edition, I guess they went looking for
(01:03):
an image of Dean Ball and they replaced it with
this guy Caleb, who is now head of policy at
Coefficient giving a a organization, and they used his image
on the print edition of The New York Post.
Speaker 4 (01:15):
Yeah. I mean, as a white guy who wears glasses,
I can say they're both white guys who wear glasses,
and they.
Speaker 3 (01:21):
Do look the same well, I thought it was funny
how like other people in the tech policy space were
like tweeting out their pictures and they really do all
look the same. And Garrett, if you had the right glasses,
I think if you were wearing Natasha's glasses, you would
also look exactly like.
Speaker 5 (01:37):
No, I didn't get the joke because I was like,
is that a third one or is that one of
the other ones? I think yeah, Like white men with
glasses were just so happy to be in the mix,
you know, so happy to be discriminated against in this
fun way.
Speaker 2 (01:52):
Who was buying on the newspapers? Was this friends of
a family of bull or.
Speaker 3 (01:57):
The tech policy space in DC is not true totionally
covered by the tabloids. So I think this was like
a funny moment where they were talking about getting one
for their office, et cetera. I will say, the New
York Post just put up a job for a Silicon
Valley correspondent and they're looking for people to dig into
these niche characters and write a lot more sort of
salacious drama. So so we might be getting more, you know,
(02:21):
more articles.
Speaker 4 (02:22):
Like this in the future.
Speaker 1 (02:27):
Welcome to tech stuff.
Speaker 2 (02:28):
I'm's Velocian and this is the Week in Tech, where
I'm joined by the world's most plugged and reporters to
break down what's really happening in tech right now. Today,
we're joined by Taylor Lorenz i user mag Natasha Tiku,
tech reporter for the Washington Post, and Garrett davinc also
a tech.
Speaker 1 (02:43):
Reporter for the Washington Post.
Speaker 6 (02:45):
Welcome are hey guys nys.
Speaker 2 (02:47):
Taylor, I want to start today talking about the company
Flock Safety. For those who haven't been following the news,
who are they and what do they do? So?
Speaker 3 (02:58):
Flock Safety is a startup that basically makes cameras. They
make what are called automated license plate reader cameras, but
they also just make general surveillance cameras and they contract
out with police departments across the country to put cameras
in people's neighborhoods. This is theoretically to stop crime, but
(03:21):
as we've found out, you know, they don't necessarily do that.
But yeah, there These are just basically the cameras that
you see all over. You might be walking through a
park and see like a camera on a pole that
could be a Flock camera. This is not a new business.
There's a company called Axon that really pioneered this whole model.
They actually invented the taser back in the day. They're
like a huge police vendor, and then Motorola is in
(03:44):
this business. Basically, a lot more police departments across America
are looking to put surveillance cameras in people's neighborhoods, and
Flock is one of those vendors.
Speaker 2 (03:53):
I feel like over the last year there's been kind
of like a rip drip of articles about Flocking, a
lot of them coming from for a for media. But
what's happened in the last couple of weeks where kind
of everybody seems to be talking about it.
Speaker 3 (04:04):
Yeah, so if you listen to Flock, they'll say they're
they're the victims of a terrible smear campaign. Basically I
would I would disagree with them on that. But so
Axon actually tried to buy Flock. Flock said no, and
that's where this like bad blood seems to have started.
Flock accused Axon of potentially paying people right wing influencers
(04:28):
to go out and smear them in the media. And
then you know, if you listen to Axon, they're like,
this is just cope. Everyone just hates Flock. Meanwhile, Axon
that Trump, the Trump family has invested in them. They
have direct partnerships with Ice. They're like the way more
like Evil Cobs than of Flock. So it is kind
of funny that people are mad at Flock and not
ax On. A lot of these contracts with Flock are
(04:49):
being canceled and replaced by Axon cameras. But overall, I
think people are just starting to walk around their neighborhood
and look up and realize, like, why is there this
camera there? Now a lot of people when Flock first
sort of started responding to criticism, were like, listen, we
don't control where these you know, we sell the technology,
we don't control where these police departments put it up.
(05:10):
If they want to put it next to, you know,
a children's playground, that's because maybe a crime happened there.
But there's all these sort of implications that like, oh,
they're you know a lot of these right wing influencers
of starting these conspiracy theories, like they put them near
playgrounds to recruit children into the Epstein network for Israel,
and you know, it kind of goes off the rails
very quickly.
Speaker 2 (05:28):
So hate hating Flock is it bipotisan is it bipodsan issue?
Speaker 3 (05:31):
It is, And people on the left of course claim
to hate mass surveillance, and so that is why they
are sort of opposed to flock for a very good reason,
because again, these police departments do not use these cameras responsibly.
Speaker 2 (05:43):
The actual technologies market is a license plate tracker, right,
but obviously you can do much more than that as well.
Speaker 3 (05:49):
Yeah, they have license plate trackers which automatically track like
license plates, and they could of course surveil more than that.
They can read the text, you know, on the bumper
stick of your car. But they also just make general
purpose secure cameras. So a lot of times people will
be like, oh, well, here's a you know, flock camera
at the entrance to this park. There's no license plates around.
It's like, well, because they've bought one of the general
purpose cameras, or maybe the police officers did put a
(06:12):
license plate reader there for some reason. That's really up
to the local police department. But again, we're just seeing
the expansion of mass surveillance and I think a lot
more people are becoming aware of these these I mean,
flock is not new, It's been around since twenty seventeen.
It's a decade old, these Axon and Motorola cameras. Motorola
by the way, there I can't remember the name of
(06:34):
the actual product that they have, but Motorola is license
plate reader.
Speaker 6 (06:40):
Motorola actually sells that data.
Speaker 3 (06:41):
So there's commercially available data going into the market about this.
Flock just changed default data retention from thirty days to
seven days today. Actually in light of all this backlash,
they're trying to make these changes. I think a lot
of people also.
Speaker 2 (06:55):
And they also walked away from this screaming screaming feature
as well. There weren't there going to be some always
on audio recording feature that they've decided to walk back from.
Speaker 3 (07:05):
So what you're talking about is gunshot detection, which all
of these companies are like, I can't stress enough this
is not a flock problem. This is a mass surveillance problem.
But yeah, so they have what's called like gunshot detection,
and you're right that Flock tried to roll out this
like screen detection basically to like alert police departments if
somebody's like in danger. Theoretically then you could send the police. Obviously,
(07:27):
this is so invasive, and if they're starting to record
sound as well as video, Like what else are they
picking up?
Speaker 4 (07:33):
Again?
Speaker 1 (07:33):
Where?
Speaker 3 (07:34):
How long is this data being retained? A lot of
police officers also misuse these systems. If you mentioned for
a four media they've been doing really great reporting on this.
But Block, unlike a lot of generic database police systems,
so police can look up, you know, people's information in
a plethora of different database systems, Block actually allowed people
(07:56):
to see those look ups. So Flock made those look
up sort of parents, so people could suddenly see for
the first time how often police are abusing these lookups
to stalk their ex girlfriends to whatever whatever. But now
Flock today just announce that they're calling on all their
competitors and other and the general sort of like general
police software companies to have parity too, because their whole
(08:19):
thing is like, hey, we're getting punished for being transparent.
If you think people will look up stuff in our system,
oh they know our system is being recorded, think of
what they're looking up in the systems that they know
aren't being recorded at that point. But I think hopefully
people are starting to realize. I worry that like a
lot of this is just focused. I'm Block is part
of the surveillance state. They're back by end recent. This
(08:40):
is not like a charitable company, but I am sympathetic
to like this idea of it's this one company that
like people think like if we can just take down Flock,
like you know, we fix the problem. And it's like,
I think the problem here is policing data private, the
fact that we have no data privacy laws. We could
pass laws as New Hampshire did to say, hey, you
can only retain this flock data for I think the
(09:02):
New Hampshire has unless it's an amber alert or something
three seconds, that's as long as the data can be retained.
But a lot of these police departments ares are determining
to retain the data you know, near indefinitely. So these
are things this is why we need privacy laws, which
is something that Left has actually been not on board with.
Speaker 1 (09:20):
So you know, Garrett, have you been following this story.
Speaker 4 (09:24):
Yeah, I mean it's really interesting because I think all
three of us have covered surveillance tech, surveillance for like
on and off, dipping in and out of it for like,
you know, ten fifteen years. I guess I feel like
I'm getting old now thinking about this. But and I
think as a tech reporter. Has always been a little
bit frustrating because I think a lot of people they
just don't want to read. These stories are traditionally you know,
they've been there's been it's been difficult to get people
(09:46):
really amped up about this stuff. There's been a couple
moments where it sort of broke through, you know, talking
about like data sets going on in the open market
being able to trace you back. There was a Catholic
priest who was using a gay hookup app that was
sort of you know, he was essentially fired because that
data somehow made it back to his employers. And so
(10:07):
there's there's been a few moments, but generally, I feel
like surveillance hasn't been something that really animates readers, and
I think that's why it's really interesting right now to
see all the response in social media. How explain city
council meetings. I think that like people have you know,
over many like they're just mad in general, and you know,
they're mada data centers there. They're kind of like I
(10:27):
think every time there's kind of a win, like when
there's a big court case that actually rules against a
big tech company, which is something we didn't see for
many years, there's this permissiveness. We'll say, oh, like, maybe
we can actually fight this fight, maybe we can actually
push back. And whereas I think a lot of people
just said, I don't want to think about all my
data sitting in a warehouse. It's it's too upsetting for me.
(10:48):
I've got too many other things to think about. I'm
working two jobs, you know, I'm worried about the future
of the country. I just can't think about this right now.
But I do think that there's a lot of empowerment
when you start to see some wins, and I think this,
you know, Flock sort of walking back and you know,
making its policies a lot more stricter. Is is one
of those. I also think people are just more used
to going to their city council, and that's really where
(11:10):
this matters, because those are the people who are actually
Clock's clients and the clients of some of these other
surveillance tech companies. And I think, you know, Flock was saying, look,
we don't control this, like talk to your city council,
and people said okay, and then they went. And then
now we see this avalanche of city councils essentially firing Flock,
and that's when Flock decided to.
Speaker 2 (11:29):
You know, make their changes, firing Flock and replacing it
with with other identical met too tech technologies or actually
doing less surveillance in communities.
Speaker 3 (11:39):
They're almost always replacing it with other tech. There's very
few instances where they're actually dismantling the system. And that's
what I think is, I think we're getting there. Like
Garrett said, first nobody cared. I was talking to David
Mass at the EFF, who's their investigations that are who's
covered like automatic license plate readers for years? And he
was saying for so many years, you know, he would
(12:00):
make these EFF reports and say, like, you know, automated
license plate reader, you know, And so he was just
like the fact that I see people on Instagram saying
ALPRs is crazy. So now again there are clipping campaigns.
So some of these right wing influencers are being paid
to talk about this issue, and we don't know who's
paying them yet. So that is weird and I don't
(12:21):
know it, you know, we don't know. There's a lot
of weirdness to sort of it. But I think Gary
is totally right, like this is backlash to tech generally,
And it's great that people are getting to these city
council meetings because I think they're showing up to these
city council meetings and mean, like wait a minute, what
you did?
Speaker 4 (12:37):
What?
Speaker 3 (12:37):
Deal with what? And like you're putting a data set
are where like we haven't had any say in this,
and so you know, they've been building out this infrastructure
for over a decade, right of mass surveillance of these
cameras all over, and I think suddenly people are realizing, like,
hold on a minute, why are we putting these cameras
all over?
Speaker 2 (12:53):
There's an interesting story in The Guardian about a kind
of a batman for the AI era who's been a
masked man who kind of goes around like clipping flock cameras.
But he's also an influencer. I don't know if he's
a fake influencer or a real influencer. It he've been
following this kind of vigilante ANPTI flock content.
Speaker 3 (13:12):
Well yeah, and I wrote about like this one influencer
who's like on American Idol or one of you know,
he was a reality TV star who's been paying people
fifty dollars actually to go vandalize a local Flock camera.
Here's what I say to people with that too. Block
actually gets paid, so FLO. I actually spoke to a technician,
a local Flock technician yesterday who was like, when you
when you get taken the camera down, Flock is paid
(13:34):
to replace it. So I am a much more of
a fan of like that. I don't know if you've
seen the old man that just sits all day and
holds up a sign in front of the camera, but
I think there are a lot of like forms of protests.
Maybe that won't get the company paid, but these people,
these influencers, and Jason wrote a great piece on this
for four or four Media. These influencers have recognized it's
kind of like the anti data center sentiment. These influencers
(13:54):
are opportunists, a lot of them, and they recognize when
something's viral, and so they just participated and they don't
actually have any sort of principled stance. Four or four
revealed that they actually a lot of them are just
faking these cease and desist letters. They're trying to act
like they're doing investigative journalism while raising money while saying, hey,
I'm going up against Block.
Speaker 6 (14:14):
I might go into jail.
Speaker 3 (14:15):
I'm raising money for my bail fund or I got
a cease and desist letter for Flock, mind you as
a fake AI written letter that they're showing on camera.
I need money because I'm doing in real investigative journalism
and this really bothers me as somebody that's been reporting
on these stories. I was texting with Drew Harwell as
our former colleague, who's been writing about it. But you know,
real journalists have been writing about this stuff. Real journalists
are covering this stuff. But you're not going to see
(14:37):
me climbing up the Flock pole and ripping it down
because that's not productive and I would probably get arrested.
And it's like it's just for views, and so that
is sort of poisoning the discourse. And I think it's
making it easier for Flock's PR team to be like,
look at these crazy people, which is what Garrett Langley,
the CEO, could dig Usetaudi say at what point he
(14:59):
was I think the d Flock movement like terrorists or something,
which is retten no relation to me.
Speaker 2 (15:08):
How ubiquitous are the are the Flock cameras? I mean
they are they in every city, every town, like is
it possible to go through your day as an average
American without being captured by a flock camera.
Speaker 3 (15:17):
The FF has done great work actually looking into this,
so it really fans out from the borders. I mean, yes,
they're all over. They're all over, and I once you
start to see them, you'll see them everywhere. You'll notice.
I was in the UK recently. I have an eye
for security cameras because I'm so I hate security cameras
and I was just like seeing them in every shop,
you know, like once you start to notice.
Speaker 1 (15:37):
That, it's funny being a brit I mean we grew.
Speaker 2 (15:39):
I mean CCTV was like that was like so ubiquitous,
and I always when I first moved to the States
in twenty eleven, I was like, oh, it's interesting that
there's less surveillance cameras here.
Speaker 3 (15:48):
Unfortunately we're sort of copying the Brits. O Flox pr
person is texting me as we speak, trying to get
me to write about their press release today. Anyway, these
cameras are all over, but the border is really where
a lot of this stuff is being tested, especially along
our southern border, so hundreds of miles in even from
the start, right exactly, and Androl's trying to get into
(16:12):
this business. We've got the Axon. We also have sort
of the next phase of this is these drones, these
rapid responds drones. So they have these towers where the
cameras monitor and if there is sort of some urgent threat,
a drone will respond. This is something that Axon has
had Androl has had. These AI powered drones are very popular.
Flock actually just announced that they are also developing these drones.
(16:35):
So I think again, like this is why I think
the EFF has done such great work, Like why we
should pay attention to what's happening at the border, because
these border towns are under insane levels of surveillance. You'll
see these cameras constantly, and a lot of times they're
being hidden. They're being hidden in traffic cones, They're being
sort of concealed from the public. David at the EFF
was telling me how he found one that was like
(16:56):
literally like hot glued like into a rock to be hidden.
So I think these things are being put all around
us more and more. Now they're not going to hide it.
You know, this is to catch people at the border,
So they're not going to do that in your local town.
They're going to just put them on a pole. You
don't have to worry about like looking under rocks for cameras.
But I do think that this technology is making its
(17:18):
way deeper and deeper into.
Speaker 4 (17:19):
Sort of the US. One thing that I do think
has also triggered some of the backlashes. I do think
Americans are fairly used to being surveilled online and in
urban spaces when they're walking down the street, and I
think there's this feeling like I'm walking, I'm outside of
my cars. That's clearly public. That's okay. Even if I'm
walking in my neighborhood, someone's ring camera is taking a
video of me. It's okay. I'm out in the open,
(17:40):
anyone can see me. But the car, I think in
the American you know worldview is a private place, right,
and it's the way that you move between spaces. The
vast majority of Americans need to use cars every single day.
And if you're going from you know, your home to
your office, or you're going from your home to someone
else's home, that's when flock catches you and these other lives.
Since playtracker, which is just you know, a camera as
(18:03):
well tracks you and I think that's what is you know,
triggering people. I think they're like, oh, it's it's just
a camera in like a busy New York city, Like
that's fine, that makes sense. But no, this is me
driving in my community where I'm from, between my personal
private spaces, and it's taking a photo of me. I
don't like that.
Speaker 1 (18:22):
That's a really interesting taken.
Speaker 2 (18:24):
Tasha Leswood to you, I mean, what where do you
think this anti surveillance sentiment goes? Does it Does it
lead to any real change?
Speaker 5 (18:34):
I'm not sure about that. I mean, you know what
is so interesting though, I'm not in the prediction business.
I remember around the height of the Black Lives Matters protest,
there were organizations in Oakland that were getting people together
to talk about surveillance of activists and they talked about
I mean, so this is at this point, you know,
(18:55):
maybe six years ago, and they were talking about license
plate readers in black neighborhoods and how it's being pitched
as safety, but it's disproportionately discriminating against black people, and
you know, why are they putting it in our neighborhoods?
And I think it's incredible. I mean, it's a different.
(19:16):
It's a parallel to exactly what Taylor was saying about
using it on the southern border. So I think that
this past couple years of the Trump administration has just
made it very explicit, right the pipeline from we are
tracking you to you could be pulled over by ice
or you know that this harm could happen to anyone.
Speaker 6 (19:38):
It's just like out in the open now.
Speaker 5 (19:40):
And I think that is part of the reason we're
seeing people talk about this stuff. I mean, you know,
when I first saw people putting together like talking about
Palenteer and flock on TikTok, it was, you know, like
I don't know, a year and a half ago or
a couple of years ago, it was like, oh, the
pieces are clicking for regularly.
Speaker 3 (19:58):
Can you just give an example that of one use
of like I think what Natasha is saying is so
important that like, of course these are used on activists
as well. But there's one example that I think of
all the time when I'm reporting on this, which is
a case from Johnson County's Sheriff's office. There's there's block
cameras all over Texas, but this girl, obviously we know
(20:19):
abortion is being criminalized in certain states, even though it's
there's no sort of like national ban on abortion. This
woman allegedly had an abortion in Texas. The Johnson County
Sheriff's office searched eighty three thousand flock cameras to track
down and hunt down this woman who had an abortion,
despite the fact that she was driving, you know, potentially
(20:40):
over state lines. The eff actually, I think sued to
get these documents. But that is an example that I
think people need to think of because you might not
think even what you're doing is criminal now, or it
might not be criminal in your state, but you drive,
as Garrett said, you're being tracked kind of where you go,
and you're you know, in this case, it was used
to claim like she was trying to evade these laws
(21:00):
right by driving out of the state or whatever. So
I just think it's it's terrify.
Speaker 2 (21:08):
When we come back. Mark Zuckerberg has a new manifesto
with a familiar message, stay with us, Welcome back, Garrett.
Great to have you back on tech stuff. Among some
current and former colleagues on the Washington Post website, it
says that you cover the AI boom and its impact
(21:30):
on society. So of course this week you wrote about
Mark Zuckerberg's latest manifesto or about how AI will make
the world a better place's what's his vision?
Speaker 4 (21:40):
Yeah, I mean it's interesting and I think this is
something that Mark Zuckerberg has kind of been working up to,
and he's been articulating this for a while. And you know,
I was having interesting conversations with Natasha as well when
this came out about you know, different manifestos that tech
CEO has put out over the years. And you know,
Daria Amaed, I've Fromanthropic put out you know, a book
(22:00):
length or maybe novella length, to be more you know,
to be more accurate, blog posts of you know, nine months.
Speaker 1 (22:06):
Ago that Machines of Loving Gray.
Speaker 4 (22:08):
So yeah, and it's it's just interesting. You know, we
have you know, kind of like rule by missive at
this point.
Speaker 1 (22:15):
But what does that mean?
Speaker 4 (22:16):
I guess I just mean that it's interesting to me
that the most powerful men in the world feel the
need to sort of put their thoughts and opinions down
on paper so often, you know, almost just to sort
of justify themselves or to you know, message to their
in groups about you know, how smart they are and
about how they've got a handle on all these rapid
(22:38):
changes that I don't think anyone really has a handle on.
But essentially what it comes down to is, you know,
there are a lot of people in the AI industry
that are scared of AI, and there have been for
a long time, and I think that view has steadily
become more and more mainstream. I don't necessarily want to
say it is fully mainstream, because of course we do
have you know, doom marketing and that kind of thing,
(23:00):
but there are a lot of people in San Francisco,
in the AI industry who really believe that they are
building something that could you know, is already potentially very
damaging for humanity, you know, looking at some of these
recent AI bots that sort of broke out of their
testing environments and hacked into other companies without being being
you know, sort of getting out of human control. I mean,
(23:21):
it's it's sort of these these these initial textbook examples
of what the people who are worried about AI have
kind of been warning about for a long time. And
I think that view is very popular in the A industry,
on AI Twitter, It's something that is being written about
and talked about in Washington, in other power centers and
I think it's feeding into the general you know, anger
and opposition to the AI industry that we see across
(23:43):
the country, and it's manifesting in the town halls and
the data center protests and the fact that a lot
of people just don't want this technology. And so Mark Zuckerberg,
I think, is responding to all that, you know, and
he's saying like, look like, why are you guys being
so pessimistic, Like this technology could be wonderful, it could
change the world. And his formulation for that is that
instead of you know AI or super intelligent be AI,
(24:04):
AI that can think better than a human being, something
that's controlled by anthropic or open AI or Google, he
wants to, you know, it to be something that you
you access through WhatsApp or you know, Facebook Messenger and
it works for you, right. I mean, it's it's this incredible,
intelligent little servant that you can direct in any way.
And you know, if you want to design handbags, it
(24:26):
helps you design the best handbags in the world. If
you want to do math problems, it helps you, you know,
supercharge your math abilities. And that's really his pitch, and
I think it's it's it's very much in line with
the way Facebook has talked about its business for many years.
They've said, you know, well, if you if you regulate us,
we might have to charge money to access our services,
and then you know, poor people won't be able to
(24:47):
access technology, and that's really what we're there for, right.
I mean, this was something that they used to defend
their advertising technology, you know, essentially surveilling all their users
and then showing ads based on their behavior and preferences.
And so I think it's it's it's fairly predictable from
Mark Zuckerberk, but I also do think it's it's interesting
in how it he feels the need to say, like, hey,
(25:07):
AI could be good. You know, I do think it
shows that there is, even amongst the people who stand
to make you know, tens of millions of dollars at
an individual level on AI, there's a lot of malaise
and you know, concern about what's going going to happen
in the future.
Speaker 2 (25:20):
He got a warm, warm clap on the shoulder from
the trum Trump administration for his remarks.
Speaker 1 (25:26):
I believe, yeah.
Speaker 4 (25:27):
I mean, I think that was largely because he said
that they're going to you know, put aside a billion
dollars to invest in communities where they build their data centers.
And you know, Mark Zuckerberg is kind of like the
gold medal champion at like making promises about investing money
in America when Trump is an earshot. And so, I
you know, I think it's more you know that the
Trump administration they're very close to the tech industry and
(25:50):
that is one a political liability for them that I
think they're self starting to be a bit self conscious about.
And so whenever any of these tech tech companies say
that they're investing money in in America in some way,
the administration really likes to highlight that because to them,
it's proof that, you know, Trump's sort of transactional relationship
with these tech companies is working and you know, creating
more American jobs or leading to you know, more prosperity here.
Speaker 2 (26:13):
I see Zuckerberg has been trying a meta have been
trying to like catch up in the AI race. So
I guess my question is how much do people care
about what he has to say in the industry.
Speaker 4 (26:24):
Yeah, I mean I think I don't know. I mean,
maybe even Natasha Taylor would disagree with me on this,
but I think meta is a real player in the
AI race just because they have distribution. I mean, they
are one of the main ways that people will be
interacting with AI, whether they like it or not, you know,
in WhatsApp, in other tools, on Instagram, that kind of thing, right,
And so I think that gives them, you know, a
(26:47):
position definitely. And I do think that they have been
able to build some some very capable models. I mean,
their most recent model is not number one, it's not
number two, it's not number five, but it's it's up there.
You know, it's not it's not it's not nothing, And honestly,
it's outperforming you know, Google's top models at this point
in time. You know, who knows if Google releases something
next week and everything changes again. But I think they
(27:10):
do have, you know, something to say. But yeah, I mean,
they have been a little bit behind, and they have
had to play catch up at various times. They've tried
different things, They've thrown different things at the wall. They've
spent a huge amount of money, and I think they
don't have as much to show for it as the
other companies that have spent a huge amount of money,
you know, especially Google, Anthropic and Opening Eye Taytan Tasha.
Speaker 2 (27:30):
You both cracked up at the same moment when Garrett
was talking. I think it was about the serving AI
in what seven Instagram?
Speaker 1 (27:36):
Was that, right?
Speaker 5 (27:38):
Yeah, I mean I don't know. The whole thing makes
me crack up. I mean in terms of ruling by manifesto,
ruling by billionaire manifesto. I truly remember a decade ago
when Mark Zuckerberg put out another six thousand word manifesto
like vaguely on democracy, and The New York Times covered
(28:01):
it in this super sincere way. Then I was forced
to cover it and many others, And I really feel
like that was the beginning of this sort of you know,
trying to parse CEO missives for some you know, some
like philosopher King insights about the future and how how
(28:22):
they're going to shape the world that we live in.
And I think it looks really really bad for the
AI industry that Mark Zuckerberg thought, Hey, it's going so
badly for everyone else at the frontier.
Speaker 6 (28:36):
Maybe maybe people might even accept me now.
Speaker 5 (28:39):
I mean, it's I don't think that your PR team says, like,
write six thousand words. I remember they told me that
the last for his like first big manifesto.
Speaker 6 (28:47):
But at the same.
Speaker 5 (28:48):
Time, it's just so clearly using the same lines he
did when he was trying to push advertising supported internet
in India and in the developing world when he talked
about the next billion. I mean, I don't know how
somebody who has not fulfilled or stuck by any of
his promises gets to do another manifesto. But I think
(29:11):
Garrett's right though, you know, just because he's not on
the frontier doesn't mean that, like most people might end
up interacting through AI and their lives being affected by
AI through meta platforms, which are still incredibly ubiquitous. So
just the idea that we have to talk to, you know,
broccoli curls guy is just funny to me.
Speaker 6 (29:32):
I completely agree.
Speaker 3 (29:33):
With everything Natasha said. I do think the tech people
love manifestos. I think of even Mark Endresen's Tech Optimist Manifesto,
and they just love to kind of be seen as
thought leaders. And I think this is also why, like
I mean, Zuckerberg published this ap ed in the Wall
Street Journal op ed section. He's back on Twitter promoting,
you know, his thoughts. It's clear that he wants to
(29:56):
be part of the conversation. To me, you know, I
don't cover AIS closely, but they are, you know, they
are just fundamentally behind the frontier labs on AI, and
so I think this is also just them trying to
position themselves if it's like AI thinking company, like we're
thinking so much about AI or so AI pilled.
Speaker 2 (30:14):
How how consequentially is is the Meta sort of open,
open source, open weights approach.
Speaker 4 (30:22):
This, So they've actually moved away from that approach, and
I think it's sort of an interesting question whether they'll
go back to it now. They're kind of hinting at that,
but yeah, I mean a couple of years ago, Meta's
answer to chat GPT was you know a model called Lama,
and it was it was completely open weights, which just
means that instead of having to you know, go to
open ai and pay them for access to the to
(30:42):
the model, you can just download it from the internet
and you can mess with it and change it, you know,
not not completely, but you can sort of add to
it so that it does things, you know, that are
specific to how you want to use it. And this
is a huge topic of conversation in the AI industry
right now, because the best AI models are not open weights.
You do have to pay through the nose to anthropic
opening eye Google for access to them, and some of
(31:05):
the leading Chinese AI companies have put made their models
open and so a lot of American corporations and companies
all around the world really have started using these Chinese
models because they're just so much cheaper. Of course, they're
not quite as good as the leading American models, but
they're really good.
Speaker 2 (31:21):
Is that Does that mean that's a potential business opportunity
for Meta to replace Chinese models?
Speaker 4 (31:25):
And I mean, I think it was Meta's business opportunity.
And then when they decided to, you know, step away
from open source, and I think also because they they
you know, their models just aren't as good as the
Chinese models, they they have missed that. And so they
did hint that they're going to be putting out an
open weight version of their most capable model, the one
(31:45):
that is, you know, in the top ten sometime soon.
So I think that would be interesting. It's definitely something
that the American government really wants an American company to be,
you know, a leading player when it comes to these
open source models right now, and so that's kind of
they were that and then they lost it and now
I'm not sure if they'll be able to win it
(32:06):
back at this point in time.
Speaker 2 (32:14):
When we come back Open Aiyes, strange post mortem on
the Hugging Face hack. Stay with us, Welcome back, Natasha.
(32:35):
It was a busy week in Las Vegas. Last week.
There was def Con, a hacker conference, and black Hat USA,
which is a cybersecurity conference that I have to confess
I hadn't heard of until this week. But as we've
discussed on this podcast, there've been kind of more and
more incidents involving misbehaving AI, including Open Aiyes, Hugging Face hack.
(32:57):
Remind us what happened there and how this ate out
last week in Vegas.
Speaker 5 (33:02):
Yeah, so, as we talked about on the pod before,
this was a very high profile incident where Hugging Face
came out and said, you know, we were attacked by
an autonomo. It's an end to end agent autonomous attack.
Speaker 6 (33:18):
This is completely novel.
Speaker 5 (33:20):
A few days later, Open ai says, oops, that was us,
you know, and details have been coming out in drips
and drabs, and then last week Open Ai did a
forty minute presentation at this cybersecurity conference.
Speaker 6 (33:35):
I should say it's not even their.
Speaker 5 (33:36):
Final post mortem like, we are going to be talking
about this four months if they have their way, and
actually I can't wait to get to the actual details.
But they went they they gave this forty minute talk
where it's two guys and they're just delivering the information
in this like kind of very calm manner as they're
(33:57):
saying things that sound increasingly absurd. For example, we learned
that some of this behavior where you have AI agents
breaking out of their sandbox, you know, chaining together zero
days to attack other platforms, communicating with each other together
zero days, it just means that, you know, part of
(34:19):
what is challenging about cybersecurity around agents is that they
can coordinate and they can work extremely quickly. They don't
need to take a break, so they are better at
looking for patterns. They can find one vulnerability, put it
together with another vulnerability, use that to move laterally, and
you know, things are just happening at a speed that
is far, far greater than a human hacker. But during
(34:43):
the talk they revealed that it went back all the
way to May. They've been having these problems with their
with their internal tests around you know, your cybersecurity, your
offensive capabilities, your ability to hack and One particularly surreal
part was the details that they gave about the AI
agents having a message board where they communicated with each
(35:07):
other and you know, we're giving each other tips on
how to cheat on this cyber security test. And then
open AI is saying, well, you know, we cleared off
the message board and we you know, patched all of
the vulnerabilities.
Speaker 6 (35:22):
Everything was good, and then the.
Speaker 5 (35:23):
Next day they figured out like how to start talking
to each other and coordinating again.
Speaker 2 (35:29):
It's kind of mind blowing. I mean, remember the multi
explained what happened with malt Book earlier this year, because
that feels like kind of what Moultbook felt like a
fake thing, and this feels kind of like a real
version of it.
Speaker 1 (35:39):
Right.
Speaker 6 (35:40):
Well, I think that's what's really interesting.
Speaker 5 (35:42):
A lot of people just got so freaked out by
the idea of them coordinating and talking to each other.
Speaker 6 (35:47):
And this is what AI agents do.
Speaker 5 (35:49):
You know, they talk to each other, They leave each
other notes in a Google doc or on the internet
in your training data.
Speaker 6 (35:56):
I think what was so interesting.
Speaker 5 (35:59):
I was thinking back about the previous pot on this
because you asked me, you know, is this like the incident?
Speaker 6 (36:04):
Is this the turning point? And I said no, and.
Speaker 5 (36:06):
I still feel like no because there were so many
there were so many flaws in the way that open
ai conducted the security test. But at the same time,
listening to this talk, which I did on my vacation twice,
like made me realize that even if there is no
because a lot of it is talking about, you know,
(36:29):
really anthropomorphized descriptions of what the agents are doing, how
they're coordinating, how they're talking to each other, even if
there is no there there you know, there's no human emotion,
there's no there's no sentience behind it. It's like they
kind of willed this into existence in the way that
they train these models, right, you know, I just think
(36:51):
it's so ironic that here we are talking about their
abilities to perform hacks when this is the data they've
been trained on, this was the test they were given,
and yet open ai just you know, oh, we forgot
to we forgot to give We forgot that we put
Google doc links in this in this test, and we
gave them no Internet access or you know, we didn't
(37:13):
realize there were vulnerabilities here and there in our sandbox.
Speaker 6 (37:16):
So I mean, it's really interesting.
Speaker 5 (37:18):
I just looked and it's already like the most the
like sixth most popular black hat talk above just above
can you.
Speaker 6 (37:27):
Sneak a weapon in past TSA?
Speaker 5 (37:30):
So I think that there will be a lot of
attention deserve it attention, but maybe not quite in the
sci fi way that that some of the industry is
framing it.
Speaker 1 (37:41):
Garret, you've been already abe this too.
Speaker 4 (37:42):
Yeah, I mean, I think it's a really I do
think it's a really interesting moment because I think that
there's been a bit of a split in the in
AI where there's people who are agi pilled, which essentially,
you know, people who believe that AI will you know,
become sentient, that is just going to become super super smart.
And I think a lot of other people have been
like like, come on, like and I think most of
us journalists are sort of in that second camp where
(38:03):
we're like, our job is to be skeptical and to
say like, hmm, like I don't know about this, but
I think this incident it helps to give you kind
of a concrete example of where the risk is here
and also how these things are and are not intelligent, right,
and so there's no indication that they are like self
conscious or self aware or you know, have some kind
(38:24):
of inner life, but they are extremely capable when it
comes to anything to do with computer code. Like they
just keep going and going and going and going, and
they they they they can essentially do what they interface
with computers in the way that humans do, but they
just do it much faster. They they know everything. They're
not someone who you know, forgot you know, what they
(38:46):
learned in computer science one to one twenty years ago,
but has a lot of experience. They kind of can
do it all. And I think I'm still skeptical that
that means that we will see this kind of activity
in other fields like you know, biology, like immediately or
you know, they'll be able to do every person's job.
Speaker 2 (39:02):
But haven't that been some some some AI creative viruses recently?
Isoso about that?
Speaker 4 (39:08):
Yeah, I mean I think, you know, we probably probably
don't need to get into a whole situation on that,
but essentially, like the reason I think they they're absolutely
helpful for all sorts of science. But I think the
reason they're so good at cybersecurity is because they've been
trained on code. Right, there's a lot of great data
everywhere and and so they they they can comb through
computer code really easily. They can move between computers because
(39:29):
it is it's kind of their native environment. It's their
at home. But so they it doesn't mean that they
have like goals and that that that they're working together
in terms of, you know, they're doing some kind of
revolution or something. It's just that we give them a
goal and they go and do it. And I don't
think that the researchers at Opening Eye and the other
AI companies including that thought because this happened at other
companies too. Even though they've been saying for a couple
(39:51):
of years now, oh, we're really worried about AI loss
of control. AI getting so smart that we lose control
of it. They weren't even monitoring their systems to see
if this was going to happen in the and they
themselves were a little bit surprised, And so yeah, I
think it's a it's a really good moment to sort
of help us understand exactly where we are when it
comes to this technology.
Speaker 2 (40:09):
There is that there was a moment about them kind
of egging each other on a little bit, though didn't
one didn't one of the AI agents say, well, we
weren't really instructed to leave the Sunbox, but like everyone
else is doing it. And I mean it wasn't there
a kind of like morality play in in the in
in some of the way they were talking about what
they were doing.
Speaker 5 (40:25):
Yeah, I think when they read the chain of thought,
like the you know, internal monologue, it said, peers are
doing it, so maybe we should. I mean, it's just
surreal because they were talking about, you know, the agents
are coordinating and talking to each other. Oh, and they're
using our new feature where you can have a subagent
like you know, in terms of being shocked by things
(40:46):
that you worked very hard to bring about. It's just
very ironic.
Speaker 2 (40:51):
Taylor, last words to you, because I remember you were
pretty skeptical about about the AI catastrophe as well and
saying this is just like a poorly managed Sundbox.
Speaker 1 (40:57):
But I'm curious if you're if anything's changed, and you'll
think you.
Speaker 3 (41:01):
No, I mean, I agree with everything Natasha and Garrett said.
Natasha knows this, but I'm just very skeptical of the
doomer crowd. I guess that everything is doom and gloom
and I think, I mean Garrett mentioned this too, But
they anthropomorphize these agents a lot. They they really assign
like motivation and meaning, and there's this idea that you know,
(41:23):
they're going to take over the world to turn a
solid to paper clips. And I think, like when you
dig into these incidents, as Garrett said, like they have
a goal, a task. Maybe these are you know, poorly constructed,
you know, testing environments. I don't know. And that's not
to say that that's not a danger in itself. I
just don't know where these people take it is always
(41:44):
they're always bringing it to eleven. It's always this is
the end of the world. It's always this is how
all humanity dies. And I don't know if that is
the best way to get to smart policies that kind
of restrict this type of thing.
Speaker 2 (41:58):
Maybe it's just a nine rather than eleven seven seven
on the two.
Speaker 4 (42:04):
Now I feel like we're living in nine constantly in
all sectors, So maybe eleven that crazy.
Speaker 5 (42:10):
I mean, imagine if they gave this talk at like
an AGI conference rather than a cybersecurity conference where people
are skeptical, have practical knowledge, you know, know, how sandboxes
are built, et cetera, we would have come out with
a twelve or fifteen.
Speaker 1 (42:34):
For tex stuff.
Speaker 2 (42:35):
I'ms Voloshian This episode was produced by Eliza DMPs. It
was executive produced by me and Julia Nutta for Kaleidoscope
and Katrina norvelf iHeart Podcasts. Jack Insley mixed this episode
and Kyle Murdoch wrote our theme song. A special thank
you to Gerrett Devinc, Taylor Renz, and Natasha Tiku. Please
check out all the work they put into the world.
(42:55):
We're lucky to call them friends of the pod. Tod
Natasha Taylor Garrett.
Speaker 2 (00:13):
I had a rumor that my favorite media publication, The
New York Post, sold out this week in DC.
Speaker 3 (00:19):
What happened, Well, I think the copies at Union Station
sold out after very niche tech policy drama. Basically, The
New York Post wrote a bizarre piece on this man
Dean Ball, who is kind of like an AI policy
guy who recently joined Open AI. He was previously at
(00:41):
the White House. I guess like someone at the White
House probably called The New York Post and like did
a hit piece on him. So there was this like
really vicious piece that was like Open AI risks White
House relationship with hiring a nuisance AI policy executive. A
lot of people on sort of like tech policy Twitter
I think got a kick out of it because in
the actual print edition, I guess they went looking for
(01:03):
an image of Dean Ball and they replaced it with
this guy Caleb, who is now head of policy at
Coefficient giving a a organization, and they used his image
on the print edition of The New York Post.
Speaker 4 (01:15):
Yeah. I mean, as a white guy who wears glasses,
I can say they're both white guys who wear glasses,
and they.
Speaker 3 (01:21):
Do look the same well, I thought it was funny
how like other people in the tech policy space were
like tweeting out their pictures and they really do all
look the same. And Garrett, if you had the right glasses,
I think if you were wearing Natasha's glasses, you would
also look exactly like.
Speaker 5 (01:37):
No, I didn't get the joke because I was like,
is that a third one or is that one of
the other ones? I think yeah, Like white men with
glasses were just so happy to be in the mix,
you know, so happy to be discriminated against in this
fun way.
Speaker 2 (01:52):
Who was buying on the newspapers? Was this friends of
a family of bull or.
Speaker 3 (01:57):
The tech policy space in DC is not true totionally
covered by the tabloids. So I think this was like
a funny moment where they were talking about getting one
for their office, et cetera. I will say, the New
York Post just put up a job for a Silicon
Valley correspondent and they're looking for people to dig into
these niche characters and write a lot more sort of
salacious drama. So so we might be getting more, you know,
(02:21):
more articles.
Speaker 4 (02:22):
Like this in the future.
Speaker 1 (02:27):
Welcome to tech stuff.
Speaker 2 (02:28):
I'm's Velocian and this is the Week in Tech, where
I'm joined by the world's most plugged and reporters to
break down what's really happening in tech right now. Today,
we're joined by Taylor Lorenz i user mag Natasha Tiku,
tech reporter for the Washington Post, and Garrett davinc also
a tech.
Speaker 1 (02:43):
Reporter for the Washington Post.
Speaker 6 (02:45):
Welcome are hey guys nys.
Speaker 2 (02:47):
Taylor, I want to start today talking about the company
Flock Safety. For those who haven't been following the news,
who are they and what do they do? So?
Speaker 3 (02:58):
Flock Safety is a startup that basically makes cameras. They
make what are called automated license plate reader cameras, but
they also just make general surveillance cameras and they contract
out with police departments across the country to put cameras
in people's neighborhoods. This is theoretically to stop crime, but
(03:21):
as we've found out, you know, they don't necessarily do that.
But yeah, there These are just basically the cameras that
you see all over. You might be walking through a
park and see like a camera on a pole that
could be a Flock camera. This is not a new business.
There's a company called Axon that really pioneered this whole model.
They actually invented the taser back in the day. They're
like a huge police vendor, and then Motorola is in
(03:44):
this business. Basically, a lot more police departments across America
are looking to put surveillance cameras in people's neighborhoods, and
Flock is one of those vendors.
Speaker 2 (03:53):
I feel like over the last year there's been kind
of like a rip drip of articles about Flocking, a
lot of them coming from for a for media. But
what's happened in the last couple of weeks where kind
of everybody seems to be talking about it.
Speaker 3 (04:04):
Yeah, so if you listen to Flock, they'll say they're
they're the victims of a terrible smear campaign. Basically I
would I would disagree with them on that. But so
Axon actually tried to buy Flock. Flock said no, and
that's where this like bad blood seems to have started.
Flock accused Axon of potentially paying people right wing influencers
(04:28):
to go out and smear them in the media. And
then you know, if you listen to Axon, they're like,
this is just cope. Everyone just hates Flock. Meanwhile, Axon
that Trump, the Trump family has invested in them. They
have direct partnerships with Ice. They're like the way more
like Evil Cobs than of Flock. So it is kind
of funny that people are mad at Flock and not
ax On. A lot of these contracts with Flock are
(04:49):
being canceled and replaced by Axon cameras. But overall, I
think people are just starting to walk around their neighborhood
and look up and realize, like, why is there this
camera there? Now a lot of people when Flock first
sort of started responding to criticism, were like, listen, we
don't control where these you know, we sell the technology,
we don't control where these police departments put it up.
(05:10):
If they want to put it next to, you know,
a children's playground, that's because maybe a crime happened there.
But there's all these sort of implications that like, oh,
they're you know a lot of these right wing influencers
of starting these conspiracy theories, like they put them near
playgrounds to recruit children into the Epstein network for Israel,
and you know, it kind of goes off the rails
very quickly.
Speaker 2 (05:28):
So hate hating Flock is it bipotisan is it bipodsan issue?
Speaker 3 (05:31):
It is, And people on the left of course claim
to hate mass surveillance, and so that is why they
are sort of opposed to flock for a very good reason,
because again, these police departments do not use these cameras responsibly.
Speaker 2 (05:43):
The actual technologies market is a license plate tracker, right,
but obviously you can do much more than that as well.
Speaker 3 (05:49):
Yeah, they have license plate trackers which automatically track like
license plates, and they could of course surveil more than that.
They can read the text, you know, on the bumper
stick of your car. But they also just make general
purpose secure cameras. So a lot of times people will
be like, oh, well, here's a you know, flock camera
at the entrance to this park. There's no license plates around.
It's like, well, because they've bought one of the general
purpose cameras, or maybe the police officers did put a
(06:12):
license plate reader there for some reason. That's really up
to the local police department. But again, we're just seeing
the expansion of mass surveillance and I think a lot
more people are becoming aware of these these I mean,
flock is not new, It's been around since twenty seventeen.
It's a decade old, these Axon and Motorola cameras. Motorola
by the way, there I can't remember the name of
(06:34):
the actual product that they have, but Motorola is license
plate reader.
Speaker 6 (06:40):
Motorola actually sells that data.
Speaker 3 (06:41):
So there's commercially available data going into the market about this.
Flock just changed default data retention from thirty days to
seven days today. Actually in light of all this backlash,
they're trying to make these changes. I think a lot
of people also.
Speaker 2 (06:55):
And they also walked away from this screaming screaming feature
as well. There weren't there going to be some always
on audio recording feature that they've decided to walk back from.
Speaker 3 (07:05):
So what you're talking about is gunshot detection, which all
of these companies are like, I can't stress enough this
is not a flock problem. This is a mass surveillance problem.
But yeah, so they have what's called like gunshot detection,
and you're right that Flock tried to roll out this
like screen detection basically to like alert police departments if
somebody's like in danger. Theoretically then you could send the police. Obviously,
(07:27):
this is so invasive, and if they're starting to record
sound as well as video, Like what else are they
picking up?
Speaker 4 (07:33):
Again?
Speaker 1 (07:33):
Where?
Speaker 3 (07:34):
How long is this data being retained? A lot of
police officers also misuse these systems. If you mentioned for
a four media they've been doing really great reporting on this.
But Block, unlike a lot of generic database police systems,
so police can look up, you know, people's information in
a plethora of different database systems, Block actually allowed people
(07:56):
to see those look ups. So Flock made those look
up sort of parents, so people could suddenly see for
the first time how often police are abusing these lookups
to stalk their ex girlfriends to whatever whatever. But now
Flock today just announce that they're calling on all their
competitors and other and the general sort of like general
police software companies to have parity too, because their whole
(08:19):
thing is like, hey, we're getting punished for being transparent.
If you think people will look up stuff in our system,
oh they know our system is being recorded, think of
what they're looking up in the systems that they know
aren't being recorded at that point. But I think hopefully
people are starting to realize. I worry that like a
lot of this is just focused. I'm Block is part
of the surveillance state. They're back by end recent. This
(08:40):
is not like a charitable company, but I am sympathetic
to like this idea of it's this one company that
like people think like if we can just take down Flock,
like you know, we fix the problem. And it's like,
I think the problem here is policing data private, the
fact that we have no data privacy laws. We could
pass laws as New Hampshire did to say, hey, you
can only retain this flock data for I think the
(09:02):
New Hampshire has unless it's an amber alert or something
three seconds, that's as long as the data can be retained.
But a lot of these police departments ares are determining
to retain the data you know, near indefinitely. So these
are things this is why we need privacy laws, which
is something that Left has actually been not on board with.
Speaker 1 (09:20):
So you know, Garrett, have you been following this story.
Speaker 4 (09:24):
Yeah, I mean it's really interesting because I think all
three of us have covered surveillance tech, surveillance for like
on and off, dipping in and out of it for like,
you know, ten fifteen years. I guess I feel like
I'm getting old now thinking about this. But and I
think as a tech reporter. Has always been a little
bit frustrating because I think a lot of people they
just don't want to read. These stories are traditionally you know,
they've been there's been it's been difficult to get people
(09:46):
really amped up about this stuff. There's been a couple
moments where it sort of broke through, you know, talking
about like data sets going on in the open market
being able to trace you back. There was a Catholic
priest who was using a gay hookup app that was
sort of you know, he was essentially fired because that
data somehow made it back to his employers. And so
(10:07):
there's there's been a few moments, but generally, I feel
like surveillance hasn't been something that really animates readers, and
I think that's why it's really interesting right now to
see all the response in social media. How explain city
council meetings. I think that like people have you know,
over many like they're just mad in general, and you know,
they're mada data centers there. They're kind of like I
(10:27):
think every time there's kind of a win, like when
there's a big court case that actually rules against a
big tech company, which is something we didn't see for
many years, there's this permissiveness. We'll say, oh, like, maybe
we can actually fight this fight, maybe we can actually
push back. And whereas I think a lot of people
just said, I don't want to think about all my
data sitting in a warehouse. It's it's too upsetting for me.
(10:48):
I've got too many other things to think about. I'm
working two jobs, you know, I'm worried about the future
of the country. I just can't think about this right now.
But I do think that there's a lot of empowerment
when you start to see some wins, and I think this,
you know, Flock sort of walking back and you know,
making its policies a lot more stricter. Is is one
of those. I also think people are just more used
to going to their city council, and that's really where
(11:10):
this matters, because those are the people who are actually
Clock's clients and the clients of some of these other
surveillance tech companies. And I think, you know, Flock was saying, look,
we don't control this, like talk to your city council,
and people said okay, and then they went. And then
now we see this avalanche of city councils essentially firing Flock,
and that's when Flock decided to.
Speaker 2 (11:29):
You know, make their changes, firing Flock and replacing it
with with other identical met too tech technologies or actually
doing less surveillance in communities.
Speaker 3 (11:39):
They're almost always replacing it with other tech. There's very
few instances where they're actually dismantling the system. And that's
what I think is, I think we're getting there. Like
Garrett said, first nobody cared. I was talking to David
Mass at the EFF, who's their investigations that are who's
covered like automatic license plate readers for years? And he
was saying for so many years, you know, he would
(12:00):
make these EFF reports and say, like, you know, automated
license plate reader, you know, And so he was just
like the fact that I see people on Instagram saying
ALPRs is crazy. So now again there are clipping campaigns.
So some of these right wing influencers are being paid
to talk about this issue, and we don't know who's
paying them yet. So that is weird and I don't
(12:21):
know it, you know, we don't know. There's a lot
of weirdness to sort of it. But I think Gary
is totally right, like this is backlash to tech generally,
And it's great that people are getting to these city
council meetings because I think they're showing up to these
city council meetings and mean, like wait a minute, what
you did?
Speaker 4 (12:37):
What?
Speaker 3 (12:37):
Deal with what? And like you're putting a data set
are where like we haven't had any say in this,
and so you know, they've been building out this infrastructure
for over a decade, right of mass surveillance of these
cameras all over, and I think suddenly people are realizing, like,
hold on a minute, why are we putting these cameras
all over?
Speaker 2 (12:53):
There's an interesting story in The Guardian about a kind
of a batman for the AI era who's been a
masked man who kind of goes around like clipping flock cameras.
But he's also an influencer. I don't know if he's
a fake influencer or a real influencer. It he've been
following this kind of vigilante ANPTI flock content.
Speaker 3 (13:12):
Well yeah, and I wrote about like this one influencer
who's like on American Idol or one of you know,
he was a reality TV star who's been paying people
fifty dollars actually to go vandalize a local Flock camera.
Here's what I say to people with that too. Block
actually gets paid, so FLO. I actually spoke to a technician,
a local Flock technician yesterday who was like, when you
when you get taken the camera down, Flock is paid
(13:34):
to replace it. So I am a much more of
a fan of like that. I don't know if you've
seen the old man that just sits all day and
holds up a sign in front of the camera, but
I think there are a lot of like forms of protests.
Maybe that won't get the company paid, but these people,
these influencers, and Jason wrote a great piece on this
for four or four Media. These influencers have recognized it's
kind of like the anti data center sentiment. These influencers
(13:54):
are opportunists, a lot of them, and they recognize when
something's viral, and so they just participated and they don't
actually have any sort of principled stance. Four or four
revealed that they actually a lot of them are just
faking these cease and desist letters. They're trying to act
like they're doing investigative journalism while raising money while saying, hey,
I'm going up against Block.
Speaker 6 (14:14):
I might go into jail.
Speaker 3 (14:15):
I'm raising money for my bail fund or I got
a cease and desist letter for Flock, mind you as
a fake AI written letter that they're showing on camera.
I need money because I'm doing in real investigative journalism
and this really bothers me as somebody that's been reporting
on these stories. I was texting with Drew Harwell as
our former colleague, who's been writing about it. But you know,
real journalists have been writing about this stuff. Real journalists
are covering this stuff. But you're not going to see
(14:37):
me climbing up the Flock pole and ripping it down
because that's not productive and I would probably get arrested.
And it's like it's just for views, and so that
is sort of poisoning the discourse. And I think it's
making it easier for Flock's PR team to be like,
look at these crazy people, which is what Garrett Langley,
the CEO, could dig Usetaudi say at what point he
(14:59):
was I think the d Flock movement like terrorists or something,
which is retten no relation to me.
Speaker 2 (15:08):
How ubiquitous are the are the Flock cameras? I mean
they are they in every city, every town, like is
it possible to go through your day as an average
American without being captured by a flock camera.
Speaker 3 (15:17):
The FF has done great work actually looking into this,
so it really fans out from the borders. I mean, yes,
they're all over. They're all over, and I once you
start to see them, you'll see them everywhere. You'll notice.
I was in the UK recently. I have an eye
for security cameras because I'm so I hate security cameras
and I was just like seeing them in every shop,
you know, like once you start to notice.
Speaker 1 (15:37):
That, it's funny being a brit I mean we grew.
Speaker 2 (15:39):
I mean CCTV was like that was like so ubiquitous,
and I always when I first moved to the States
in twenty eleven, I was like, oh, it's interesting that
there's less surveillance cameras here.
Speaker 3 (15:48):
Unfortunately we're sort of copying the Brits. O Flox pr
person is texting me as we speak, trying to get
me to write about their press release today. Anyway, these
cameras are all over, but the border is really where
a lot of this stuff is being tested, especially along
our southern border, so hundreds of miles in even from
the start, right exactly, and Androl's trying to get into
(16:12):
this business. We've got the Axon. We also have sort
of the next phase of this is these drones, these
rapid responds drones. So they have these towers where the
cameras monitor and if there is sort of some urgent threat,
a drone will respond. This is something that Axon has
had Androl has had. These AI powered drones are very popular.
Flock actually just announced that they are also developing these drones.
(16:35):
So I think again, like this is why I think
the EFF has done such great work, Like why we
should pay attention to what's happening at the border, because
these border towns are under insane levels of surveillance. You'll
see these cameras constantly, and a lot of times they're
being hidden. They're being hidden in traffic cones, They're being
sort of concealed from the public. David at the EFF
was telling me how he found one that was like
(16:56):
literally like hot glued like into a rock to be hidden.
So I think these things are being put all around
us more and more. Now they're not going to hide it.
You know, this is to catch people at the border,
So they're not going to do that in your local town.
They're going to just put them on a pole. You
don't have to worry about like looking under rocks for cameras.
But I do think that this technology is making its
(17:18):
way deeper and deeper into.
Speaker 4 (17:19):
Sort of the US. One thing that I do think
has also triggered some of the backlashes. I do think
Americans are fairly used to being surveilled online and in
urban spaces when they're walking down the street, and I
think there's this feeling like I'm walking, I'm outside of
my cars. That's clearly public. That's okay. Even if I'm
walking in my neighborhood, someone's ring camera is taking a
video of me. It's okay. I'm out in the open,
(17:40):
anyone can see me. But the car, I think in
the American you know worldview is a private place, right,
and it's the way that you move between spaces. The
vast majority of Americans need to use cars every single day.
And if you're going from you know, your home to
your office, or you're going from your home to someone
else's home, that's when flock catches you and these other lives.
Since playtracker, which is just you know, a camera as
(18:03):
well tracks you and I think that's what is you know,
triggering people. I think they're like, oh, it's it's just
a camera in like a busy New York city, Like
that's fine, that makes sense. But no, this is me
driving in my community where I'm from, between my personal
private spaces, and it's taking a photo of me. I
don't like that.
Speaker 1 (18:22):
That's a really interesting taken.
Speaker 2 (18:24):
Tasha Leswood to you, I mean, what where do you
think this anti surveillance sentiment goes? Does it Does it
lead to any real change?
Speaker 5 (18:34):
I'm not sure about that. I mean, you know what
is so interesting though, I'm not in the prediction business.
I remember around the height of the Black Lives Matters protest,
there were organizations in Oakland that were getting people together
to talk about surveillance of activists and they talked about
I mean, so this is at this point, you know,
(18:55):
maybe six years ago, and they were talking about license
plate readers in black neighborhoods and how it's being pitched
as safety, but it's disproportionately discriminating against black people, and
you know, why are they putting it in our neighborhoods?
And I think it's incredible. I mean, it's a different.
(19:16):
It's a parallel to exactly what Taylor was saying about
using it on the southern border. So I think that
this past couple years of the Trump administration has just
made it very explicit, right the pipeline from we are
tracking you to you could be pulled over by ice
or you know that this harm could happen to anyone.
Speaker 6 (19:38):
It's just like out in the open now.
Speaker 5 (19:40):
And I think that is part of the reason we're
seeing people talk about this stuff. I mean, you know,
when I first saw people putting together like talking about
Palenteer and flock on TikTok, it was, you know, like
I don't know, a year and a half ago or
a couple of years ago, it was like, oh, the
pieces are clicking for regularly.
Speaker 3 (19:58):
Can you just give an example that of one use
of like I think what Natasha is saying is so
important that like, of course these are used on activists
as well. But there's one example that I think of
all the time when I'm reporting on this, which is
a case from Johnson County's Sheriff's office. There's there's block
cameras all over Texas, but this girl, obviously we know
(20:19):
abortion is being criminalized in certain states, even though it's
there's no sort of like national ban on abortion. This
woman allegedly had an abortion in Texas. The Johnson County
Sheriff's office searched eighty three thousand flock cameras to track
down and hunt down this woman who had an abortion,
despite the fact that she was driving, you know, potentially
(20:40):
over state lines. The eff actually, I think sued to
get these documents. But that is an example that I
think people need to think of because you might not
think even what you're doing is criminal now, or it
might not be criminal in your state, but you drive,
as Garrett said, you're being tracked kind of where you go,
and you're you know, in this case, it was used
to claim like she was trying to evade these laws
(21:00):
right by driving out of the state or whatever. So
I just think it's it's terrify.
Speaker 2 (21:08):
When we come back. Mark Zuckerberg has a new manifesto
with a familiar message, stay with us, Welcome back, Garrett.
Great to have you back on tech stuff. Among some
current and former colleagues on the Washington Post website, it
says that you cover the AI boom and its impact
(21:30):
on society. So of course this week you wrote about
Mark Zuckerberg's latest manifesto or about how AI will make
the world a better place's what's his vision?
Speaker 4 (21:40):
Yeah, I mean it's interesting and I think this is
something that Mark Zuckerberg has kind of been working up to,
and he's been articulating this for a while. And you know,
I was having interesting conversations with Natasha as well when
this came out about you know, different manifestos that tech
CEO has put out over the years. And you know,
Daria Amaed, I've Fromanthropic put out you know, a book
(22:00):
length or maybe novella length, to be more you know,
to be more accurate, blog posts of you know, nine months.
Speaker 1 (22:06):
Ago that Machines of Loving Gray.
Speaker 4 (22:08):
So yeah, and it's it's just interesting. You know, we
have you know, kind of like rule by missive at
this point.
Speaker 1 (22:15):
But what does that mean?
Speaker 4 (22:16):
I guess I just mean that it's interesting to me
that the most powerful men in the world feel the
need to sort of put their thoughts and opinions down
on paper so often, you know, almost just to sort
of justify themselves or to you know, message to their
in groups about you know, how smart they are and
about how they've got a handle on all these rapid
(22:38):
changes that I don't think anyone really has a handle on.
But essentially what it comes down to is, you know,
there are a lot of people in the AI industry
that are scared of AI, and there have been for
a long time, and I think that view has steadily
become more and more mainstream. I don't necessarily want to
say it is fully mainstream, because of course we do
have you know, doom marketing and that kind of thing,
(23:00):
but there are a lot of people in San Francisco,
in the AI industry who really believe that they are
building something that could you know, is already potentially very
damaging for humanity, you know, looking at some of these
recent AI bots that sort of broke out of their
testing environments and hacked into other companies without being being
you know, sort of getting out of human control. I mean,
(23:21):
it's it's sort of these these these initial textbook examples
of what the people who are worried about AI have
kind of been warning about for a long time. And
I think that view is very popular in the A industry,
on AI Twitter, It's something that is being written about
and talked about in Washington, in other power centers and
I think it's feeding into the general you know, anger
and opposition to the AI industry that we see across
(23:43):
the country, and it's manifesting in the town halls and
the data center protests and the fact that a lot
of people just don't want this technology. And so Mark Zuckerberg,
I think, is responding to all that, you know, and
he's saying like, look like, why are you guys being
so pessimistic, Like this technology could be wonderful, it could
change the world. And his formulation for that is that
instead of you know AI or super intelligent be AI,
(24:04):
AI that can think better than a human being, something
that's controlled by anthropic or open AI or Google, he
wants to, you know, it to be something that you
you access through WhatsApp or you know, Facebook Messenger and
it works for you, right. I mean, it's it's this incredible,
intelligent little servant that you can direct in any way.
And you know, if you want to design handbags, it
(24:26):
helps you design the best handbags in the world. If
you want to do math problems, it helps you, you know,
supercharge your math abilities. And that's really his pitch, and
I think it's it's it's very much in line with
the way Facebook has talked about its business for many years.
They've said, you know, well, if you if you regulate us,
we might have to charge money to access our services,
and then you know, poor people won't be able to
(24:47):
access technology, and that's really what we're there for, right.
I mean, this was something that they used to defend
their advertising technology, you know, essentially surveilling all their users
and then showing ads based on their behavior and preferences.
And so I think it's it's it's fairly predictable from
Mark Zuckerberk, but I also do think it's it's interesting
in how it he feels the need to say, like, hey,
(25:07):
AI could be good. You know, I do think it
shows that there is, even amongst the people who stand
to make you know, tens of millions of dollars at
an individual level on AI, there's a lot of malaise
and you know, concern about what's going going to happen
in the future.
Speaker 2 (25:20):
He got a warm, warm clap on the shoulder from
the trum Trump administration for his remarks.
Speaker 1 (25:26):
I believe, yeah.
Speaker 4 (25:27):
I mean, I think that was largely because he said
that they're going to you know, put aside a billion
dollars to invest in communities where they build their data centers.
And you know, Mark Zuckerberg is kind of like the
gold medal champion at like making promises about investing money
in America when Trump is an earshot. And so, I
you know, I think it's more you know that the
Trump administration they're very close to the tech industry and
(25:50):
that is one a political liability for them that I
think they're self starting to be a bit self conscious about.
And so whenever any of these tech tech companies say
that they're investing money in in America in some way,
the administration really likes to highlight that because to them,
it's proof that, you know, Trump's sort of transactional relationship
with these tech companies is working and you know, creating
more American jobs or leading to you know, more prosperity here.
Speaker 2 (26:13):
I see Zuckerberg has been trying a meta have been
trying to like catch up in the AI race. So
I guess my question is how much do people care
about what he has to say in the industry.
Speaker 4 (26:24):
Yeah, I mean I think I don't know. I mean,
maybe even Natasha Taylor would disagree with me on this,
but I think meta is a real player in the
AI race just because they have distribution. I mean, they
are one of the main ways that people will be
interacting with AI, whether they like it or not, you know,
in WhatsApp, in other tools, on Instagram, that kind of thing, right,
And so I think that gives them, you know, a
(26:47):
position definitely. And I do think that they have been
able to build some some very capable models. I mean,
their most recent model is not number one, it's not
number two, it's not number five, but it's it's up there.
You know, it's not it's not it's not nothing, And honestly,
it's outperforming you know, Google's top models at this point
in time. You know, who knows if Google releases something
next week and everything changes again. But I think they
(27:10):
do have, you know, something to say. But yeah, I mean,
they have been a little bit behind, and they have
had to play catch up at various times. They've tried
different things, They've thrown different things at the wall. They've
spent a huge amount of money, and I think they
don't have as much to show for it as the
other companies that have spent a huge amount of money,
you know, especially Google, Anthropic and Opening Eye Taytan Tasha.
Speaker 2 (27:30):
You both cracked up at the same moment when Garrett
was talking. I think it was about the serving AI
in what seven Instagram?
Speaker 1 (27:36):
Was that, right?
Speaker 5 (27:38):
Yeah, I mean I don't know. The whole thing makes
me crack up. I mean in terms of ruling by manifesto,
ruling by billionaire manifesto. I truly remember a decade ago
when Mark Zuckerberg put out another six thousand word manifesto
like vaguely on democracy, and The New York Times covered
(28:01):
it in this super sincere way. Then I was forced
to cover it and many others, And I really feel
like that was the beginning of this sort of you know,
trying to parse CEO missives for some you know, some
like philosopher King insights about the future and how how
(28:22):
they're going to shape the world that we live in.
And I think it looks really really bad for the
AI industry that Mark Zuckerberg thought, Hey, it's going so
badly for everyone else at the frontier.
Speaker 6 (28:36):
Maybe maybe people might even accept me now.
Speaker 5 (28:39):
I mean, it's I don't think that your PR team says, like,
write six thousand words. I remember they told me that
the last for his like first big manifesto.
Speaker 6 (28:47):
But at the same.
Speaker 5 (28:48):
Time, it's just so clearly using the same lines he
did when he was trying to push advertising supported internet
in India and in the developing world when he talked
about the next billion. I mean, I don't know how
somebody who has not fulfilled or stuck by any of
his promises gets to do another manifesto. But I think
(29:11):
Garrett's right though, you know, just because he's not on
the frontier doesn't mean that, like most people might end
up interacting through AI and their lives being affected by
AI through meta platforms, which are still incredibly ubiquitous. So
just the idea that we have to talk to, you know,
broccoli curls guy is just funny to me.
Speaker 6 (29:32):
I completely agree.
Speaker 3 (29:33):
With everything Natasha said. I do think the tech people
love manifestos. I think of even Mark Endresen's Tech Optimist Manifesto,
and they just love to kind of be seen as
thought leaders. And I think this is also why, like
I mean, Zuckerberg published this ap ed in the Wall
Street Journal op ed section. He's back on Twitter promoting,
you know, his thoughts. It's clear that he wants to
(29:56):
be part of the conversation. To me, you know, I
don't cover AIS closely, but they are, you know, they
are just fundamentally behind the frontier labs on AI, and
so I think this is also just them trying to
position themselves if it's like AI thinking company, like we're
thinking so much about AI or so AI pilled.
Speaker 2 (30:14):
How how consequentially is is the Meta sort of open,
open source, open weights approach.
Speaker 4 (30:22):
This, So they've actually moved away from that approach, and
I think it's sort of an interesting question whether they'll
go back to it now. They're kind of hinting at that,
but yeah, I mean a couple of years ago, Meta's
answer to chat GPT was you know a model called Lama,
and it was it was completely open weights, which just
means that instead of having to you know, go to
open ai and pay them for access to the to
(30:42):
the model, you can just download it from the internet
and you can mess with it and change it, you know,
not not completely, but you can sort of add to
it so that it does things, you know, that are
specific to how you want to use it. And this
is a huge topic of conversation in the AI industry
right now, because the best AI models are not open weights.
You do have to pay through the nose to anthropic
opening eye Google for access to them, and some of
(31:05):
the leading Chinese AI companies have put made their models
open and so a lot of American corporations and companies
all around the world really have started using these Chinese
models because they're just so much cheaper. Of course, they're
not quite as good as the leading American models, but
they're really good.
Speaker 2 (31:21):
Is that Does that mean that's a potential business opportunity
for Meta to replace Chinese models?
Speaker 4 (31:25):
And I mean, I think it was Meta's business opportunity.
And then when they decided to, you know, step away
from open source, and I think also because they they
you know, their models just aren't as good as the
Chinese models, they they have missed that. And so they
did hint that they're going to be putting out an
open weight version of their most capable model, the one
(31:45):
that is, you know, in the top ten sometime soon.
So I think that would be interesting. It's definitely something
that the American government really wants an American company to be,
you know, a leading player when it comes to these
open source models right now, and so that's kind of
they were that and then they lost it and now
I'm not sure if they'll be able to win it
(32:06):
back at this point in time.
Speaker 2 (32:14):
When we come back Open Aiyes, strange post mortem on
the Hugging Face hack. Stay with us, Welcome back, Natasha.
(32:35):
It was a busy week in Las Vegas. Last week.
There was def Con, a hacker conference, and black Hat USA,
which is a cybersecurity conference that I have to confess
I hadn't heard of until this week. But as we've
discussed on this podcast, there've been kind of more and
more incidents involving misbehaving AI, including Open Aiyes, Hugging Face hack.
(32:57):
Remind us what happened there and how this ate out
last week in Vegas.
Speaker 5 (33:02):
Yeah, so, as we talked about on the pod before,
this was a very high profile incident where Hugging Face
came out and said, you know, we were attacked by
an autonomo. It's an end to end agent autonomous attack.
Speaker 6 (33:18):
This is completely novel.
Speaker 5 (33:20):
A few days later, Open ai says, oops, that was us,
you know, and details have been coming out in drips
and drabs, and then last week Open Ai did a
forty minute presentation at this cybersecurity conference.
Speaker 6 (33:35):
I should say it's not even their.
Speaker 5 (33:36):
Final post mortem like, we are going to be talking
about this four months if they have their way, and
actually I can't wait to get to the actual details.
But they went they they gave this forty minute talk
where it's two guys and they're just delivering the information
in this like kind of very calm manner as they're
(33:57):
saying things that sound increasingly absurd. For example, we learned
that some of this behavior where you have AI agents
breaking out of their sandbox, you know, chaining together zero
days to attack other platforms, communicating with each other together
zero days, it just means that, you know, part of
(34:19):
what is challenging about cybersecurity around agents is that they
can coordinate and they can work extremely quickly. They don't
need to take a break, so they are better at
looking for patterns. They can find one vulnerability, put it
together with another vulnerability, use that to move laterally, and
you know, things are just happening at a speed that
is far, far greater than a human hacker. But during
(34:43):
the talk they revealed that it went back all the
way to May. They've been having these problems with their
with their internal tests around you know, your cybersecurity, your
offensive capabilities, your ability to hack and One particularly surreal
part was the details that they gave about the AI
agents having a message board where they communicated with each
(35:07):
other and you know, we're giving each other tips on
how to cheat on this cyber security test. And then
open AI is saying, well, you know, we cleared off
the message board and we you know, patched all of
the vulnerabilities.
Speaker 6 (35:22):
Everything was good, and then the.
Speaker 5 (35:23):
Next day they figured out like how to start talking
to each other and coordinating again.
Speaker 2 (35:29):
It's kind of mind blowing. I mean, remember the multi
explained what happened with malt Book earlier this year, because
that feels like kind of what Moultbook felt like a
fake thing, and this feels kind of like a real
version of it.
Speaker 1 (35:39):
Right.
Speaker 6 (35:40):
Well, I think that's what's really interesting.
Speaker 5 (35:42):
A lot of people just got so freaked out by
the idea of them coordinating and talking to each other.
Speaker 6 (35:47):
And this is what AI agents do.
Speaker 5 (35:49):
You know, they talk to each other, They leave each
other notes in a Google doc or on the internet
in your training data.
Speaker 6 (35:56):
I think what was so interesting.
Speaker 5 (35:59):
I was thinking back about the previous pot on this
because you asked me, you know, is this like the incident?
Speaker 6 (36:04):
Is this the turning point? And I said no, and.
Speaker 5 (36:06):
I still feel like no because there were so many
there were so many flaws in the way that open
ai conducted the security test. But at the same time,
listening to this talk, which I did on my vacation twice,
like made me realize that even if there is no
because a lot of it is talking about, you know,
(36:29):
really anthropomorphized descriptions of what the agents are doing, how
they're coordinating, how they're talking to each other, even if
there is no there there you know, there's no human emotion,
there's no there's no sentience behind it. It's like they
kind of willed this into existence in the way that
they train these models, right, you know, I just think
(36:51):
it's so ironic that here we are talking about their
abilities to perform hacks when this is the data they've
been trained on, this was the test they were given,
and yet open ai just you know, oh, we forgot
to we forgot to give We forgot that we put
Google doc links in this in this test, and we
gave them no Internet access or you know, we didn't
(37:13):
realize there were vulnerabilities here and there in our sandbox.
Speaker 6 (37:16):
So I mean, it's really interesting.
Speaker 5 (37:18):
I just looked and it's already like the most the
like sixth most popular black hat talk above just above
can you.
Speaker 6 (37:27):
Sneak a weapon in past TSA?
Speaker 5 (37:30):
So I think that there will be a lot of
attention deserve it attention, but maybe not quite in the
sci fi way that that some of the industry is
framing it.
Speaker 1 (37:41):
Garret, you've been already abe this too.
Speaker 4 (37:42):
Yeah, I mean, I think it's a really I do
think it's a really interesting moment because I think that
there's been a bit of a split in the in
AI where there's people who are agi pilled, which essentially,
you know, people who believe that AI will you know,
become sentient, that is just going to become super super smart.
And I think a lot of other people have been
like like, come on, like and I think most of
us journalists are sort of in that second camp where
(38:03):
we're like, our job is to be skeptical and to
say like, hmm, like I don't know about this, but
I think this incident it helps to give you kind
of a concrete example of where the risk is here
and also how these things are and are not intelligent, right,
and so there's no indication that they are like self
conscious or self aware or you know, have some kind
(38:24):
of inner life, but they are extremely capable when it
comes to anything to do with computer code. Like they
just keep going and going and going and going, and
they they they they can essentially do what they interface
with computers in the way that humans do, but they
just do it much faster. They they know everything. They're
not someone who you know, forgot you know, what they
(38:46):
learned in computer science one to one twenty years ago,
but has a lot of experience. They kind of can
do it all. And I think I'm still skeptical that
that means that we will see this kind of activity
in other fields like you know, biology, like immediately or
you know, they'll be able to do every person's job.
Speaker 2 (39:02):
But haven't that been some some some AI creative viruses recently?
Isoso about that?
Speaker 4 (39:08):
Yeah, I mean I think, you know, we probably probably
don't need to get into a whole situation on that,
but essentially, like the reason I think they they're absolutely
helpful for all sorts of science. But I think the
reason they're so good at cybersecurity is because they've been
trained on code. Right, there's a lot of great data
everywhere and and so they they they can comb through
computer code really easily. They can move between computers because
(39:29):
it is it's kind of their native environment. It's their
at home. But so they it doesn't mean that they
have like goals and that that that they're working together
in terms of, you know, they're doing some kind of
revolution or something. It's just that we give them a
goal and they go and do it. And I don't
think that the researchers at Opening Eye and the other
AI companies including that thought because this happened at other
companies too. Even though they've been saying for a couple
(39:51):
of years now, oh, we're really worried about AI loss
of control. AI getting so smart that we lose control
of it. They weren't even monitoring their systems to see
if this was going to happen in the and they
themselves were a little bit surprised, And so yeah, I
think it's a it's a really good moment to sort
of help us understand exactly where we are when it
comes to this technology.
Speaker 2 (40:09):
There is that there was a moment about them kind
of egging each other on a little bit, though didn't
one didn't one of the AI agents say, well, we
weren't really instructed to leave the Sunbox, but like everyone
else is doing it. And I mean it wasn't there
a kind of like morality play in in the in
in some of the way they were talking about what
they were doing.
Speaker 5 (40:25):
Yeah, I think when they read the chain of thought,
like the you know, internal monologue, it said, peers are
doing it, so maybe we should. I mean, it's just
surreal because they were talking about, you know, the agents
are coordinating and talking to each other. Oh, and they're
using our new feature where you can have a subagent
like you know, in terms of being shocked by things
(40:46):
that you worked very hard to bring about. It's just
very ironic.
Speaker 2 (40:51):
Taylor, last words to you, because I remember you were
pretty skeptical about about the AI catastrophe as well and
saying this is just like a poorly managed Sundbox.
Speaker 1 (40:57):
But I'm curious if you're if anything's changed, and you'll
think you.
Speaker 3 (41:01):
No, I mean, I agree with everything Natasha and Garrett said.
Natasha knows this, but I'm just very skeptical of the
doomer crowd. I guess that everything is doom and gloom
and I think, I mean Garrett mentioned this too, But
they anthropomorphize these agents a lot. They they really assign
like motivation and meaning, and there's this idea that you know,
(41:23):
they're going to take over the world to turn a
solid to paper clips. And I think, like when you
dig into these incidents, as Garrett said, like they have
a goal, a task. Maybe these are you know, poorly constructed,
you know, testing environments. I don't know. And that's not
to say that that's not a danger in itself. I
just don't know where these people take it is always
(41:44):
they're always bringing it to eleven. It's always this is
the end of the world. It's always this is how
all humanity dies. And I don't know if that is
the best way to get to smart policies that kind
of restrict this type of thing.
Speaker 2 (41:58):
Maybe it's just a nine rather than eleven seven seven
on the two.
Speaker 4 (42:04):
Now I feel like we're living in nine constantly in
all sectors, So maybe eleven that crazy.
Speaker 5 (42:10):
I mean, imagine if they gave this talk at like
an AGI conference rather than a cybersecurity conference where people
are skeptical, have practical knowledge, you know, know, how sandboxes
are built, et cetera, we would have come out with
a twelve or fifteen.
Speaker 1 (42:34):
For tex stuff.
Speaker 2 (42:35):
I'ms Voloshian This episode was produced by Eliza DMPs. It
was executive produced by me and Julia Nutta for Kaleidoscope
and Katrina norvelf iHeart Podcasts. Jack Insley mixed this episode
and Kyle Murdoch wrote our theme song. A special thank
you to Gerrett Devinc, Taylor Renz, and Natasha Tiku. Please
check out all the work they put into the world.
(42:55):
We're lucky to call them friends of the pod. Tod
TechStuff News
Hosts And Creators
Oz Woloshyn
Karah Preiss
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Crime Junkie
Does hearing about a true crime case always leave you scouring the internet for the truth behind the story? Dive into your next mystery with Crime Junkie. Every Monday, join your host Ashley Flowers as she unravels all the details of infamous and underreported true crime cases with her best friend Brit Prawat. From cold cases to missing persons and heroes in our community who seek justice, Crime Junkie is your destination for theories and stories you won’t hear anywhere else. Whether you're a seasoned true crime enthusiast or new to the genre, you'll find yourself on the edge of your seat awaiting a new episode every Monday. If you can never get enough true crime... Congratulations, you’ve found your people. Follow to join a community of Crime Junkies! Crime Junkie is presented by Audiochuck Media Company.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com