Episode Transcript
Transcripts are displayed as originally observed. Some content, including advertisements may have changed.
Use Ctrl + F to search
0:00
Over the last twenty five years the world
0:02
has witnessed incredible progress from dialup modem survive
0:04
G connectivity from massive Pc towers to a
0:06
I enabled Microchips innovators, a rethinking possibilities every
0:08
day through it all in West Coast Qq
0:10
each yeah has provided investors access to the
0:12
world of innovation. Be a part of the
0:14
next twenty five years of new ideas by
0:16
supporting the find that gives you access to
0:18
innovative companies Invesco, Kinky Kill Let's rethink possibility
0:20
there was one of getting any just including
0:22
possible. Lots of money it has risks are
0:25
similar to those of stocks investors in the
0:27
tech sector or some Twitter, as can relativity
0:29
the more diversified investments. Before investing, heavily beaten
0:31
consider fun investment Doctors risk started expenses and more
0:33
in perspective that invesco.com invesco distributors and. What's.
0:36
Going on with you? Well you know I'm
0:38
having sort of a weird week. So I
0:40
came in on Monday him and to the
0:42
office of the New York Times, the San
0:44
Francisco and someone said there's and there's a
0:46
naked man outside the office who's ranting and
0:49
I said well, how is care. About
0:54
aware that you should know that weekly podcast
0:56
on Wednesdays and on Mondays and then. And
1:00
then I learned that I was actually the part of
1:02
the seat stupid day parade you know about. Way I've
1:04
not heard of this. This is. Apparently an annual
1:06
tradition in San Francisco where some big goofballs
1:09
get together on April Fools Day April first
1:11
and they parade through the streets of San
1:13
Francisco holding like nonsensical signs and some of
1:15
them apparently are naked eye and this is
1:18
what was going on. I love this town
1:20
so much there's a tell it as a
1:22
cell back and fix it happened in San
1:24
Francisco you wouldn't even believe and they're wonderful.
1:32
I'm Kevin or it's A on us
1:34
to New York Times. I'm Casey Noom
1:36
from Platform Or and this is our
1:38
for this week How Ai is affecting
1:40
the economy. Ten artist Paul Trillo joints
1:42
to discuss how we used Open the
1:44
Eyes Sore Actual to explore the future
1:46
of filmmaking, and finally a cyber sneak
1:48
attack that could have brought down the
1:50
web was caught. In the nick of
1:52
time. not by govern, not by me, A
2:07
before we get a nursery this week we just
2:10
wanted to remind people were on you tube some
2:12
always talk about in the so this week is
2:14
going to be better to see them to Here
2:16
we're gonna talk with someone who uses a I
2:19
to make films If you'd like to watch those
2:21
films are frankly if you'd like to watch the
2:23
podcast instead of just listening to it's you can
2:25
go to Youtube channel on youtube.com/hard Fork or deserves
2:28
Hard Fork on you tube. Aren't know?
2:30
Onto the shell. Casing this
2:32
week I want to talk about A I
2:34
and Jobs because this is a topic that
2:36
we get asked about all the time by
2:38
listeners of the show that I hear about
2:41
all the time from readers. Is this technology
2:43
that you guys are always talking about? Actually
2:45
going to take people's jobs is a good
2:47
to help people at their jobs. And how
2:49
long do we have to wait before our
2:51
jobs? Your jobs get affected by all this
2:54
stuff. That's where and when Blisters as we
2:56
always say. don't worry if we find out
2:58
that A is gonna take your job we
3:00
will email you individually and until then he
3:02
said well as websites Cbt, Emily's as befits
3:04
so I thought we should. said. Break.
3:07
This out into a couple pieces. One of them is
3:09
sort of. About. The Present: What
3:11
do we know about how Ai is
3:13
already affecting jobs and companies and their
3:15
plans to employ or unemployed people in
3:17
the near future? And and then I
3:19
think we should talk about some of
3:21
the various theories that are coming out
3:23
about how and whether generative A I
3:25
Will actually leads to major seems is
3:27
in the job market the right? Let's
3:29
do it. So this story I'm is
3:31
by my colleagues Jordan Home and and
3:33
genius my like it's called will Ai
3:35
boost productivity Companies for Hope So And
3:37
it's sort of a survey of what's
3:39
going on. Out there in the world,
3:41
at companies like Walmart and Wendy's and
3:44
Abercrombie and Fitch which is apparently using
3:46
a I to write and product descriptions
3:48
an answer trying to take stock of
3:50
like, what effect does all this cumulatively
3:52
having on the economy and the job
3:55
market? Well, And I'm curious about this
3:57
because there was that story the other
3:59
week about how Wendy was thinking about
4:01
during surge pricing and that was sad.
4:03
repeat. his idea is the bad rap
4:05
and I have yet so digging into
4:08
the data a little bits of so
4:10
far, it just seems pretty early for
4:12
any of this to start showing up
4:14
in official economic statistics, but we do
4:16
see in some of the most recent
4:18
data a bump in productivity and this
4:21
data has been a little volatile since
4:23
covert, But some economists are starting to
4:25
wonder if this is real and if
4:27
this productivity increase might actually stick around.
4:29
Okay, In addition to that sort
4:32
of aggregate data that economists can see
4:34
at the national level of, there's also
4:36
just been a bunch of examples of
4:38
companies that are starting to use entered
4:40
of Ai in some cases to pair
4:42
jobs among their own workforce. So I
4:44
recently duolingo the company that makes the
4:47
language learning app not to be confused
4:49
with do a Lipo dresses and Er
4:51
has Min Suk let her out at
4:53
another, had us and yes you the
4:55
best of duel lingo or recently said
4:57
that it was cutting about ten percent
4:59
of it's contractors. Are Not. Laying
5:02
off any full time employees, but basically
5:04
just at in a paring down the
5:06
number of people it needs to create
5:08
content. A spokesperson for the company said
5:11
we no longer need as many people
5:13
to do the type of work some
5:15
of these contractors were doing. Part of
5:17
that could be attributed to a I
5:20
U P S recently cut about twelve
5:22
thousand managerial jobs. The Ceo I mentioned
5:24
how ai machine learning and could reduce
5:26
the need for pricing experts among other
5:29
jobs, but they've also said they're not
5:31
replacing workers. With a i've. Been
5:33
there was the company Florida which is one of
5:35
the things didn't do. Not far enough actually do
5:37
know corner foreigner is one of these by now
5:39
pay later companies have they have said that their
5:41
Ai assistant did the work of seven hundred customer
5:44
services. I'm always so curious though when I hear
5:46
that it's like yeah I'm for from foreigners perspective
5:48
it's great but I would love to hear from
5:50
the people who actually had to use the Ai
5:52
chat bots. Do you think it was as good
5:54
as a person whom I've used to this A
5:56
I chat bots and I'll say I think people
5:59
are better. Coroner said in their
6:01
experiment that they actually found that the Ai
6:03
chat bots were rated just as good as
6:05
the human agent all right, and that they
6:07
solve their problems faster. So these
6:09
are the kinds of experiments and test the
6:12
we're starting to see play out at various
6:14
companies. We haven't seen sort of mass layoffs
6:16
yet as a result of generative A Ice,
6:18
but these are the kinds of experiments you
6:21
would expect companies to be running on this
6:23
technology trying to figure out where can we
6:25
save down you know, maybe you know thirty
6:27
percent of the the accounts payable department or
6:30
maybe a few engineers who we may be
6:32
don't need anymore and replacing those people with
6:34
a i Will It as A is A
6:36
is a read through. All of us have
6:39
been. I find myself wondering if maybe there
6:41
never will be a mass layoff moment at
6:43
these companies. Maybe it will just be a
6:45
steady erosion as they figure out bit by
6:47
bit how to make do with fewer of
6:50
your people. You know This is sort of
6:52
one of the fascinating things that I'm observing
6:54
as I go out and talk to people
6:56
who run businesses you have. No one wants
6:58
to be seen as sir have a heartless
7:01
capital as to is just like you know,
7:03
wantonly are laying off workers and replacing them
7:05
with robots. but they are doing a lot
7:07
of things around the edges to try to
7:09
may be so lay off people for assistance,
7:12
see and then replace some of those people
7:14
not with other people but with software that
7:16
have it both ways defined. a signal to
7:18
Wall Street. Hey look look at how clever
7:20
were being and how much more efficient were
7:23
getting and how much worth cutting costs with.
7:25
are trying to avoid the Pr backlash that
7:27
Woods couple I when they say that we
7:29
no longer things humans have value in the
7:31
enterprise exactly. One. Of things have been
7:34
really surprising to me is the reluctance that
7:36
some corporate leaders have been having to pay
7:38
embrace this new Ai technology and use it
7:40
to replace workers even when the technology is
7:42
fully capable of releasing the workers. So I
7:45
had a conversation a few months ago with
7:47
a guy of were met at a at
7:49
an Ai event he he runs a big
7:51
commercial real estate from they be developed real
7:53
estate all over the country and he was
7:56
telling me that you know for years he's
7:58
had these junior analysts who will. Go
8:00
out the visit various cities you know and
8:02
they'll come back and or produce reports about
8:04
the local commercial real estate market in those
8:06
cities you go to Jacksonville and it'll come
8:09
back with a ten page report about all
8:11
of the various commercial real estate friends in
8:13
Jacksonville News and he said basically once had
8:15
C B T came out he started giving
8:17
those assignments to chat bots and seen whether
8:20
they could do them any. Found that the
8:22
reports it's had to be T would give
8:24
him about the local real estate markets wesley
8:26
better than the ones his junior analyst were
8:29
giving him. And. So I
8:31
said, well. Okay so then what
8:33
happens to the junior analyst like dude just lay
8:35
them off and replace them all with a I
8:37
and he said something that for a surprise me
8:40
which is no because that's how they learned the
8:42
job so he was not. Just. Doing
8:44
these junior analyst has sort of you
8:46
to help or monkeys who go out
8:48
and produce these reports see viewed this
8:50
process of going to a city, getting
8:52
on the ground, talking to local businesses,
8:54
examining the real say market up close.
8:57
As. A part of the training process and
8:59
how he builds sir future leaders for his
9:01
business. He basically was telling me like yes,
9:03
I could replace those people with a I,
9:06
but then I'm actually cheating myself in the
9:08
long run and I think that's one of
9:10
the kind of intangible things that is hard
9:12
to get when you just look at kind
9:15
of. Overall economic data is like. There
9:17
are many reasons that people have. Jobs.
9:19
At their companies have and there are many
9:22
types of. Serve incentives that are operating
9:24
at these companies. and so even if it
9:26
is sort term profitable to replace a bunch
9:28
of people with a eyes, there might be
9:30
other reasons that you don't want to do
9:32
that. So I think that's part of what's
9:34
going on his that makes us to me.
9:36
But I would also just note that we
9:38
are still sort of in this very early
9:40
phase with Generated A I were If you
9:42
believe the people working at the big companies
9:44
making the largely was models, they're telling us
9:46
within a generation or two it's these models
9:48
are going to be exponentially better. and then
9:50
I wonder if some of that feeling of
9:52
need to keep. His people around and train them
9:54
so they can take on the next job up
9:56
the ladder. I wonder if that feeling starts to
9:58
diminish? It's possible it's all. If. He
10:01
thinks to make mistakes are still not totally
10:03
predictable. They're still pretty weird, frankly, and so
10:05
you might not wanna throw them into the
10:08
core of your business right away. At least
10:10
not without a lot of supervision. So that's
10:12
kind of. Where. We are now
10:14
today in the job market with a
10:16
I. We have lots of companies running
10:18
lots of experiments, spending lots of money,
10:20
hiring lots of consultants, trying to figure
10:22
out how can the stuff make us
10:24
more productive. We don't see it kind
10:26
of it in the economic data just
10:28
yet, but there are signs that people
10:30
are starting to figure out ways to
10:32
use the stuff to automate jobs. absolutely,
10:34
but you know, Cabinet the same time
10:37
we're starting to seats studies that suggest
10:39
that perhaps the middle class will actually
10:41
thrives in a world where generative a
10:43
Ice is ascendant. Am I think that
10:45
finding surprise us A we should talk about
10:47
this year. So there was an interesting a
10:49
paper that got written up this week by
10:51
my colleague Steve Laura The Times that was
10:54
based on the up some work by an
10:56
economist at Mit named David. Artur. Davis is
10:58
someone whose work I've been falling for a
11:00
long time. He's one of my favorite economists
11:03
who looks at a I and the labor
11:05
markets and. Last. Month he
11:07
came out with a paper that had
11:09
what I would consider like a pretty
11:11
contrarians ceases which is that he actually
11:13
thinks that a I could if used
11:15
well assist with restoring and of the
11:17
middle class of the labor market that
11:19
has been hollowed out by things like
11:22
automation and globalization move more one that
11:24
be nice to fix. Yeah so basically
11:26
his arguments is it's not that he's
11:28
like observing this is already happening, this
11:30
is a sort of something he thinks
11:32
will happen Which is that basically you
11:34
have this economy now where you have.
11:36
Kind of like a eat a
11:39
missing Middle East or have like
11:41
wage workers and people who are
11:43
lower earning workers and then you
11:45
have kind of this expert class
11:47
of people who in make decisions
11:50
about you know medical decisions, legal
11:52
decisions of corporate management decisions and.
11:55
That one of the affects the
11:57
Gen I could have is basically
11:59
empowering a lot of people at
12:01
the bottom the end of that
12:03
market of that labour markets to
12:05
develop the kinds of expertise and
12:07
make the kinds of decisions that
12:09
previously required at highly paid professionals.
12:11
So one example he cites in
12:13
this paper is. Nurse
12:15
practitioners. That's. A relatively
12:18
new occupational category, there used to
12:20
be nurses and doctors, and then.
12:22
Several. Decades ago, they started to
12:24
practice at exactly exactly speed. Basically
12:27
developed this kind of middle tier
12:29
of medical professionals who were not
12:31
full doctors. They didn't go to
12:33
medical school, but they were qualified
12:36
to do things like write prescriptions
12:38
and make certain recommendations about your
12:40
health care. And so what David
12:42
Autor argues is it. Basically. A
12:45
I could allow non experts in
12:47
lots of different fields to kind
12:49
of develop the expertise and the
12:52
decision making capacity to basically take
12:54
on the nurse practitioner equivalent in
12:56
whatever their industry is. So maybe
12:58
you have paralegals who. Are.
13:00
Armed now with all the center to be I
13:03
who can actually start to make the kinds of
13:05
decisions they might have required a full lawyer before.
13:07
That's exciting. I'm imagining using sad Cbd, become like
13:09
a pair of firefighter where I can just sort
13:12
of read about how to do it and then
13:14
call down to the scene and as be like
13:16
that may and point of my resume or over
13:18
there that's exactly what you call you call the
13:20
fire department because your house is on fire and
13:23
are like well we could get a firefighter there
13:25
but it's gonna take you know an hour at
13:27
we do have Casey. ah he's got he's got
13:29
a tattoo pretty subscription. And a hose and he's ready
13:31
to go. He could be there and five minutes. Systemically
13:34
that if you have it says he beauties of
13:36
prefer the housing a very fighters like sisters I
13:38
would be like a Paris the Iau you know
13:40
where it's like you sort of have the fat
13:42
salary and the prestige but were able to do
13:44
it with only half the trainings and you know
13:46
you are mostly does asking questions have a chat
13:48
bot which as far as i can tell us
13:50
mostly what see as are doing any way. Through
13:53
the i mean this is this is one
13:56
exciting possibility I think and I I I
13:58
love David are tours op. Them about
14:00
your restoring the middle class through Jenner
14:02
they ice. I think there are lots
14:05
of reasons it might not work In
14:07
practice, there's all sir licensing regimes in
14:09
various occupation, so it's like that. There
14:12
are some barriers to add the sort
14:14
of optimistic feature that David Artur envisions,
14:16
but it's just one sort of. Interesting.
14:20
Theory about where all this could be
14:22
had. Answer and I But I mean
14:24
I do buy something fundamental about that
14:26
which is that if you believe that
14:28
the as generative A I tools will
14:31
be com strain of counselors, coaches, guides
14:33
and there is a field that you're
14:35
interested in, this and that technology can
14:37
just kind of live alongside you understand
14:40
what you're working on, continually make suggestions.
14:42
It should actually accelerate people's rate of
14:44
learning and the development of their expertise
14:46
and I can see that having an
14:49
effect. On the middle class yeah, I think
14:51
that's that's server Optimistic vision. I do think that
14:53
there are a lot of people who are not
14:55
actually using a I to become better workers. They're
14:57
using a I to cut corners I'm and do
15:00
less work and in a we see this in
15:02
schools obviously with students to using the stuff that
15:04
seat. But there's also a lot of examples of
15:06
this happening out in corporate America to people you
15:09
know, maybe not using the stuff in the way
15:11
that would be most effective for them over the
15:13
long term, to saying like i've got a pretty
15:15
this report from my boss or there's and I've
15:18
gotta put together this powerpoint presentation. I don't feel
15:20
like doing it was a sled they I metal
15:22
a threat. But if there were more economic opportunity
15:24
that could come from using these things to develop
15:26
as what is perhaps more people would develop as
15:29
priests right now people are doing What you're saying
15:31
is because there's no economic penalty for them rights.
15:33
But ah, if there was an economic advantage maybe
15:35
they would use it. Yes it does improve with
15:38
what do you think companies should be doing with
15:40
centered of a Iran? Let's let's take like let's
15:42
take it out of our industry. And to say
15:44
like in a you run let's say I have
15:46
a big guy you know restaurant same yeah the
15:49
cheesecake. Factory reset Exactly like they make
15:51
the menu ten times longer. Outlet
15:53
for the villa in there a
15:55
a a a be like what's
15:57
companies be using this stuff for.
16:00
I think it depends what kind of
16:02
what what kind of company you are.
16:04
Honestly, if you're if you're running a
16:06
restaurant chains, I don't necessarily see that
16:08
there's a ton in there for use.
16:11
Maybe you want to experiment with some
16:13
copywriting? Maybe want to experiment with using
16:15
their the Image Generation test? Consider some
16:17
new advertising campaigns, but all that stuff
16:20
feels like you know, a minor and
16:22
experimental. As a another
16:24
he added maybe if you work for like a
16:26
copywriting from then maybe you want to be using
16:28
and a lot made, Then maybe you want to
16:30
be testing out all sorts of different models and
16:32
seeing which ones are. He is working better for
16:34
you. So I think that's kind of highly dependent
16:36
on the the kind of business that you're running.
16:39
But for the most part I would say you
16:41
want to manage your expectations Here is like it's
16:43
not going to be doing at a time for
16:45
you. I don't think you, I do, I do.
16:47
I. I think that this stuff is already. Pretty.
16:50
Good any and in certain sort of
16:52
out what? I. Mean so
16:54
much of our economy does runs on paperwork
16:56
and forms and reports and presentations. that's of
16:59
is some catnip that's low hanging fruit for
17:01
dinner of a see like like a story
17:03
the other day that was like there's a
17:05
company that's creating an Ai coworker like and
17:07
if you're in, if you're working on a
17:09
software company there's gonna be says character called
17:12
doesn't see them in the up. Very confusing
17:14
name. Yes, I don't like it. I believe
17:16
it's cognition. Yes the name of this company
17:18
that has working on it's he. Are you
17:20
know the idea is you know you're not
17:23
gonna have dire. As many engineers because now
17:25
you have Devon and of and can
17:27
like sort of help you write code
17:29
us that's a very early stages when
17:31
that gets good then. Okay yes now
17:33
a lot of people I think are
17:35
going to be using something like that
17:37
and as going to have a meaningful
17:39
affects our productivity stuck his I want
17:41
to and this discussion by talking about
17:43
our own experiences miss and and I'm
17:46
it doesn't work We do support one
17:48
another physicists So you've talked before with
17:50
me about how you have started experimenting
17:52
with using Gen Vi in your. Newsletter
17:54
I'm used to use at Center of
17:56
Ai to create the images that run
17:58
on top of some your. The letters
18:00
I've noticed you doing that less recently are
18:02
going to hear about why I'm You've also
18:04
talked about using it's user organize and collect
18:06
various links that you putting your newsletter. So
18:08
how is your own use of generative ai
18:10
at work or changed over maybe the past
18:12
year? So what I would say is I
18:14
truly have tried a bunch of things and
18:16
for the most part it has banned Marshall?
18:18
Start with the images but you're right. For
18:20
a year or so I was regularly using
18:22
a I generated images of the top of
18:24
my newsletter he added. The truth is I
18:26
just got a lot of feedback from readers
18:28
that they hated it. They. Felt like I was
18:31
stealing money from artist. They felt like you know
18:33
I was using models that had been improperly trailer
18:35
copyrighted material and they hated. seeing. as I have
18:37
some people say that they refused to subscribe like
18:39
because they swaths that I was using these images.
18:41
From my perspective, it had been a way to
18:44
enhance my own creativity because I can't draw, I
18:46
can't make anything look cool, but I can type
18:48
in a box and that's really cool to meet.
18:50
But I decided to take the note from readers
18:52
of For the most part, I've sort of taken
18:54
a pause on using that kind of generative a
18:57
That's really interesting because youth, it's not like you
18:59
were. The alternative with the you were
19:01
going to go pay a human artist to make
19:03
this stuff you were probably just gonna pull like
19:05
an image from Getty images are surprisingly that's what
19:07
I that's what I was going to do ano
19:09
I when I work on a very short deadlines
19:11
the idea that I could find the illustrator answer
19:13
a make it Happen inside interviewed it just wasn't
19:15
very likely. But at the same time you know
19:17
people were not excited to see the generative they
19:19
are images that's interesting L ot no more. Interestingly,
19:21
I think I have been able to with when
19:23
I finish my columns with some reasonable amount of
19:25
time before my deadlines actually take them to some
19:27
of the large language model at us. A critique
19:29
This it. The idea is not necessarily that
19:31
it's going to make Mike Hulme hundred times
19:33
better, but. You. Know, I think all
19:35
of us writers if you could get feedback
19:38
from five or ten people before you publish
19:40
anything, you you might do it. And because
19:42
these things can analyze your work instantly. ah,
19:44
they're sort of no penalty for doing it.
19:46
I wouldn't say I've changed my writing a
19:48
lot in response to what I've heard, but
19:51
yes, it does catch grammatical errors, it does
19:53
taps typos and it's increasingly it's been able
19:55
to earth identified the tone of of however
19:57
writing about something and it's know sort of
19:59
asking me to did you mean it to
20:01
sound the Swiss What? that is The eerie
20:04
as part for me is I feel like
20:06
over the past year am I would say
20:08
Gemini Googles Gemini in particular has been the
20:10
one that has really been doing this. It.
20:12
Feels like it can get at the subtext.
20:14
Of what I my name's better than other things
20:16
in the past so those been interesting and I
20:18
do think I will keep doing it because you
20:21
know it went when you're right or that kind
20:23
of be back is a gift and I will
20:25
send you you centered of Ai at the New
20:27
York Times. So yeah we the New York Times
20:29
has some rules about using generative A I and
20:31
our allowed are not allowed to use it. You
20:33
know I'm not using it for journalism. I do
20:35
not write my columns which entered of a Ice
20:37
and I I. I frankly wouldn't do that even
20:39
if I were allowed to because I just think
20:41
that like that would be boring. I enjoy writing.
20:43
It's not like I don't I'm. Not eager to
20:45
turn that part of my job over
20:47
to dinner to be I, I have
20:49
basically found that it is the best
20:52
research assistant I've ever had. Some you
20:54
know now if I'm looking up something
20:56
for a column or preparing for a
20:58
podcast interview and I do consult with
21:00
centered of a I almost every day
21:02
for ideas and brainstorming interests. The things
21:04
like research like like new make me
21:07
at a time line of all the
21:09
major cyber attack newlands him years or
21:11
something like that and of course I
21:13
will fact check that research. Before I use
21:15
it in a piece. just like I would with
21:17
any research assistants and but that's the tennis thing
21:19
That centered of Ai for me has been really
21:21
good at. And I found it. Generally
21:24
I has actually changed my work in a
21:26
different way that I wasn't. Sure,
21:28
if perhaps expecting which is that is
21:30
is made me much more attentive to
21:32
the detail of my own writing and
21:34
trying to make sure that what I
21:36
rights does not sound like sad tbt
21:38
rodents because I think the moment as
21:40
a writer that you allow yourself to
21:42
drift in that direction you are giving
21:44
up your advantage. You are basically saying
21:46
I am replaceable. I am totally indistinguishable
21:48
from this sir generic text exterior and
21:50
I think this is that sounds actually
21:52
for the entire economy as more and
21:54
more of us have reason to use
21:56
generative A Ice. In our jobs is
21:59
how to use it to augment what
22:01
you do without making your boss think
22:03
that that augmenting technology could actually do
22:05
the whole job. I think that's a
22:07
really important topic. A I and Jobs.
22:09
I think we should continue to keep
22:11
tabs on it and in addition to
22:14
looking at said the economic data and
22:16
what economists the researchers are noticing about
22:18
kind of overall productivity. I would love
22:20
to hear what people's actual experiences of
22:22
using centered of a I successfully or
22:24
unsuccessfully at their own jobs. If you
22:26
are listening to this and you have
22:28
a really interesting story. About your own use
22:31
of Gen Vi at work or your company's experiments
22:33
with the stuff that if you they're gone super
22:35
well or super badly of I would love to
22:37
hear from you said send us an email or
22:39
Portland with a semi for that. I I particularly
22:42
want to hear from you if you tried. you
22:44
started of a i work in that when particularly
22:46
badly sister that sort of my own son of
22:48
a lot of sensibility as you have bloopers. I
22:50
mean. For
22:56
the break will take a look at a very different
22:58
part of it. I has been used Smith movies. Take.
23:14
A minute to check your social battery has
23:16
a doing. It's easy to ignore
23:18
our needs and spread ourselves too thin.
23:21
Therapy. Can give you the self awareness to
23:23
build a social life that doesn't drain your batteries.
23:26
Better. Help offers affordable online therapy on
23:28
a schedule that works for you. Start.
23:30
The process and minutes and slits therapists
23:33
any time. When you're so sweet
23:35
spot with better health. Visit. Better
23:37
hop.com/hard for today to get ten
23:39
percent off your first month. That's.
23:41
Better help. Hcl. p.com/
23:44
Hard for. I mean
23:47
A Martin the host of the Modern
23:49
Love podcast. In every
23:51
episode, repeat into an intimate corner
23:53
of someone's life and learn about
23:56
what love means to them. Thirty
23:58
five years with another. I've
24:00
never spent that much time with anyone else so
24:02
both kind of that. I love you pretty fast.
24:04
As long as they keep for dance up I'll
24:07
keep the dense Snc felt the same way. an
24:09
instant connection. It's a window
24:11
into how real people navigate all
24:13
kinds of. Love. I mean
24:16
romantic family friendship, dog
24:18
based. Their stories
24:20
of life changing moments small
24:22
do is. Big Revolutions. My
24:25
advice is that it's okay if
24:27
it's hard. A lot of the way that
24:29
you manifest your love. For your children is
24:31
some fucking and I'm under just looking
24:33
at them in or almost like wow
24:36
you know so much that I couldn't
24:38
even dream of knowing about my brother.
24:41
You can listen to Modern Love where
24:43
you get your podcasts. Okay,
24:47
that was very sober and rational discussion
24:49
about Ai in the labor market. and
24:51
now I want to get a little
24:53
weird. Let's please get a little weird.
24:55
So we've been talking for our weeks
24:57
about this new open a I video
24:59
generation tool called Sourav. This is something
25:01
that was demoed at Sam Altman was
25:03
Sir fielding requests on X. You tell
25:05
me whatever prompt you want to type
25:07
in disorder and all of see what
25:09
comes out of it's this is basically
25:11
doing for video what tools like Dolly
25:13
and me Juri did for still images.
25:16
It works much the same way as
25:18
one of these diffusion based models you
25:20
type in some tax. It gives you
25:22
back a snippet of video representing whatever
25:24
you typed. Yeah, and you hear that
25:26
and you think what you know we.
25:28
We know that making films is extraordinarily
25:30
expensive. It's very collaborative, it involves all
25:32
kinds of specialists, and the idea that
25:34
we might soon be in a world
25:36
where people can just type what kind
25:38
of movie they want to see in
25:40
a box and get something resembling that
25:42
feels like a big leap forward. So
25:44
whenever any new Ai tool or comes
25:46
out, my first question. Is always what can
25:48
I use it and for this tool sore
25:50
the answer was absolutely not Thought that the
25:52
people who weren't allowed to use this product
25:54
they called the sore losers. Up
25:57
Up Up Up Up Up. So
25:59
we couldn't. Actually use it to ourselves.
26:01
Open A I have not making this
26:03
public yet for various reasons but they
26:05
did put out a a blog posts
26:07
of showcasing the work of a bunch
26:09
of filmmakers who were given access to
26:11
the earliest versions of Sore Us and
26:13
so we're going to do the next
26:15
best thing today with his We're going
26:17
to talk to someone who has actually
26:19
been able to use or and play
26:21
around with it said. Today we are
26:24
talking with Paul Trillo. He is a
26:26
multi disciplinary artist and filmmaker who is
26:28
based in L A. It's I've seen.
26:30
Some has worked before with other Ai tools.
26:32
He's been playing around with the stuff for
26:34
a while now and organ talked him say
26:36
about what he learned, what has experience was
26:38
like and what he thinks. The implications for
26:41
Hollywood and some of the filmmakers out there
26:43
who are nervous about the stuff are that's
26:45
right. One of things were going to ask
26:47
him about his this short film that he
26:49
made with sore called the Gold and Record
26:52
which he made after being inspired by a
26:54
project that Carl Sagan undertook in the Nineteen
26:56
seventies to create a kind of audio time
26:58
capsule of humanity and broadcasted. Out into space
27:00
in the hopes that aliens would find it's
27:03
and listen to it and decide not to
27:05
destroy our entire civilization. We'll still far. Let's
27:07
say it has been successfully worked for. Delta
27:09
Record Worth hats off to call the and
27:11
the Go On record we should states is
27:13
a little hard to describe. It's a minute
27:15
long. It's kind of avant garde. You'll be
27:17
able to see it if you're watching this
27:19
on you tube. If you're just listening on
27:21
the podcasts, didn't go on the show notes
27:23
willing to at there as well. Alternatively, just
27:25
take mushrooms and think the Golden Record and
27:27
it'll be similar inside. surmise a specific. A
27:31
thing of all. Pulse.
27:38
Hello Welcome to Hard Forks! Oh Wow!
27:40
Thank you so much for having me
27:42
Mrs. Literally. The only
27:44
podcasts I can tolerate. A
27:48
Shock Jocks of Tax and It's sister
27:50
We We actually were recently voted that
27:53
most tolerable podcast which was regarded as
27:55
he have Few people have told us
27:57
they have. They've been a punch there
27:59
speakers After hearing the bad. So
28:03
paulo. I'm wondering if you could. Tell
28:05
us about the emotional experience of using
28:07
saw the first time you've typed in
28:10
a prompt into this tool and got
28:12
back a video. Do you feel
28:14
any things. I mean. I.
28:17
I was shocked I was floored. I
28:19
was confused I what I was like
28:21
a little bit unsettled because I was
28:24
like damn this is like doing things
28:26
that I didn't know it was capable
28:28
of. Steam Member: what the the first
28:30
thing you tried was that you the
28:32
has a reaction with. The first
28:35
one that really took me out
28:37
of it was the video that
28:39
ups birth is the first fifteen
28:41
seconds and appears in. there have
28:43
been a I blog posts of
28:45
this kind of real I did
28:47
where I'm zooming through time and
28:49
I'm I'm I'm saying I give
28:51
me this site. dynamic fast moving
28:53
time lapse from lights volcanic ass
28:55
going underwater and then we emerge
28:57
and to like ancient civilizations and
28:59
we're zoom in through like the
29:01
seventeen hundreds, eighteen hundreds and then
29:03
until like. Modern day time. Borrowing all
29:06
the stuff at it and it gave
29:08
me some outlook. I give a shout
29:10
and separate selloum. It was moving the
29:13
camera in a way that was never
29:15
possible with like old film technology and
29:17
it was like making edits with in
29:19
the clip so isn't it? Almost had
29:22
it's own sense of peace in and
29:24
editing and it. Sat. Really
29:26
made me think out and once you
29:29
kind of throw a lot like the
29:31
kitchen sink at the saying and you
29:33
get this really experimental affect, you can
29:35
start to experiment and ways ah, that
29:38
we've never experimented with for And so
29:40
that that really got me excited was
29:42
specifically that and a hallucinatory aspects. May.
29:45
Costs to just walk us through the basic
29:47
steps of the process of making a film
29:50
using a tool like sore us like what
29:52
prompts did you use for this film like
29:54
how long did take you to put it
29:56
all together will and just walk us through
29:58
the prime. the thought of it. So.
30:01
There's a website that you go to
30:04
and there's a tax field and translate
30:06
you know what you're used to, prompting
30:08
with other generative A I tools. And.
30:11
Then it gets sort of translated interpreted
30:13
through chassis be t So it's like
30:15
okay you want best and then it
30:18
it gives you something like that and
30:20
then you can edit the chassis be
30:22
T response bites. The. Process
30:24
of using Saura. I
30:27
feel like is akin to. Trying.
30:30
To tell a story to a
30:32
toddler with superpowers? What he means
30:34
for hits It feels a little
30:36
bit like. This. Naive
30:39
and city with black
30:42
magic superpowers, I
30:44
want to just a route this conversation in the actual
30:46
video that you pretty swiss are are one of the
30:48
videos the producers are as I think we should just
30:50
like watch it together and I will kind of describe
30:52
what we're seeing and then we're gonna ask you some
30:55
questions about it again. It's very chaotic
30:57
and. Can. Addict and dynamic and
30:59
may cause motion sickness baths as I was
31:01
kind of better than nothing to have to
31:04
sign a waiver before watching the video. The
31:06
as as. The
31:15
video that we're looking at right
31:17
now called the Golden Records and
31:20
for people who who are watching
31:22
this on video is basically showing
31:24
a a record like of bio
31:26
record made of gold that is
31:28
sort of hurtling through space. Yeah
31:30
I'm getting a little dizzy. Looks
31:33
like there's a sort of like zooming like
31:36
were zooming through space encountering all these like
31:38
golden orb. So yeah but the kind of
31:40
has tear was to see how how dynamic
31:42
can I make these cameras, How cinematic can
31:44
I. Create. Ah, and a
31:46
static that sales may be different
31:49
than when I had been seen.
31:55
Cool. So that's that's a a clip about
31:57
a minute long. This is not a full.
32:00
Your film but you'd You did make
32:02
this almost entirely with sore us. What
32:05
was The idea? Their. Yeah. So
32:07
I had been fascinated by this.
32:10
Project. Has by Carl
32:12
Sagan and Nasa off.
32:14
Like in Nineteen Seventy Six and
32:17
Seventy Seven where they essentially made
32:19
like a time capsule of as
32:21
humanity. Up until that point they
32:24
collected sounds from my bubbling my
32:26
dad to like human speech and
32:28
then they collected a bunch of
32:31
songs are from around the world
32:33
including like Johnny Be Good is
32:35
on the record and then they
32:38
included an image is into a
32:40
golden record and splash out to
32:43
to and space and hope. That
32:45
maybe Aliens Sunday was literally it
32:47
was. It was a message sent.
32:49
Aliens. In. A We talked about
32:51
sending episodes of Hard Fork out into
32:54
space as of his other alien civilizations.
32:57
Assists the I Do Not Greenlight
32:59
podcasts on your system and. To.
33:05
Ask some questions about the creative process
33:07
your sucks. So many prompts did you
33:09
use to make this. One. Minute
33:11
movie. Yeah. I
33:14
probably. Size.
33:17
But. There's like variations of that, right? So
33:20
when I first got my hands on
33:22
Saura, I was like, how do I.
33:24
Break. This thing harrowing and stick it
33:27
from this like very ai looking
33:29
video ascetic. This kind of slow
33:31
moving camera moves the things that
33:33
feel lights just three, the animation
33:35
or stock footage and so like
33:37
any delays move the camera and
33:39
even if it causes motion sickness
33:41
that was part of the task
33:43
was to see like how crazy
33:45
can can this gets, how chaotic
33:47
can it be and. Just
33:49
for the sake of comparison, how long would
33:52
it have taken you to make something like
33:54
the gold a record using conventional film tools
33:56
and then Salander take you using Saura. Higher.
33:58
I would say with. The how dynamic
34:01
the camera is, how maybe complex
34:03
the renders are, with, you know,
34:05
the materials been used, how many
34:08
shots they are. I'd
34:10
say this would take a a
34:12
a few months to make. I.
34:15
Did. The Golden Record. Maybe.
34:17
In. Two. Or three
34:19
days. Wow. Yeah, suits
34:21
time, France. Did get
34:23
open A I put any restrictions on Saura
34:25
when you were using it Like did they
34:27
tell you you you can't make this can
34:29
a video or you gotta stay away from
34:31
this kind of prompts? Did they give you
34:33
any guidance? or did they just sort of
34:35
give you access this tool and and say
34:37
go nuts. They. Specifically wanted
34:40
to be like as hands off as
34:42
possible but it was obviously an hour.
34:44
Like know, like nudity, know, extreme, yeah,
34:46
gore or that kind of staff. so
34:48
they're go. All of Kevin's ideas for
34:50
making to see was sore at the
34:52
i know it's on your eyes to
34:54
different. Era your
34:56
deflating is how hot pot as you as
34:58
you reflect on the experience as a making
35:00
their the sword Psalms that you have made
35:03
with swore I would you say that on
35:05
the whole the process felt easier than you
35:07
expected, more difficult than you expect a. Light
35:09
like well what we're we're your expectations for what
35:11
the same was gonna be like and and where
35:13
did the the result fall? I.
35:15
Actually had somewhat like tempered expectations.
35:18
I was just like this is
35:20
you know a cool tech demo
35:22
that I saw from like our
35:25
a massive company with tons of
35:27
compute power. But is this. A
35:30
clickable: The filmmaking and.
35:33
After ten of breaking and loosening
35:35
up the camera I was like
35:37
okay this can give us some.
35:40
Like experiments, you know,
35:42
wilde bold weird things
35:44
that that may be
35:46
difficult to achieve with
35:48
other tools. And sasser
35:50
when I times. Crafts.
35:53
Or a series of of words
35:55
is basically kind of like Alchemy
35:57
Woods with words. Then I was
35:59
like okay, this can allow for
36:02
the shot types that an idea
36:04
is that maybe just killed in
36:06
the process of so making sweet
36:08
what are some of the secret
36:10
words you found Success Let's see.
36:13
Ah, Thirty five millimeter Fuji film
36:15
stock, Twenty four millimeter, an Amorphous
36:17
Lands analog, warm vintage tone, chromatic
36:19
aberration, whole A since ah, things
36:22
that are like I guess words
36:24
to describe literal sounds and to
36:26
see what's in that the training
36:28
data Basically. That's. Interesting. So it's
36:31
basically like your ears are giving it
36:33
the instructions that you might otherwise give to
36:35
like a cinematographer Someone who's yeah, say
36:37
like what tailored suit on film Ram
36:39
Regulators has. Ah, D P I
36:41
give me a lesson. Chromatic Aberration Like you know
36:43
they're just gonna be like but I had a
36:45
bad case of police and chromatic aberration was but
36:47
I went to the doctor and a clear night
36:49
out. The sad. Part I
36:51
I just have a very basic nuts and bolts question
36:54
which is like you type in a prompt in the
36:56
Saura. You fill it with
36:58
all these magic words he hit, enter how
37:00
long does it actually takes to get the
37:02
video back as it insert now, but? Faster.
37:05
Then you would think it is how long are
37:08
we talking here I heard from someone else it
37:10
it's like like ten or fifteen minutes usually between
37:12
when you put in the prompter when it goes
37:14
back to videos that consistent with your experience. Roughly.
37:17
I just depends on the third, the
37:19
settings. or are you at seven, twenty,
37:21
pete and eighty? It's ah slight the
37:24
saturation. but to do a really. Simple.
37:27
You know, like shot of just
37:29
a ball on the ground that's
37:31
fifteen seconds long? Will
37:34
take just as much rendering time as
37:36
doing and like a crazy it's Golden
37:38
Records, Turtle, Interspace and Exploding and all
37:40
this stuff so that's actually really fascinating
37:42
is what it does. surrender time so
37:44
as having use this for awhile now
37:46
are you thinking about this like oh
37:48
yeah This is definitely a tool that
37:50
I want in my arsenal going forward
37:52
as I continue to make films like
37:54
and a sort of see a lot
37:56
of applications for this sort of more
37:58
in the i could take. It or leave it
38:00
out. I. Would
38:02
definitely. Keep. Using
38:04
this for this is supplements all
38:07
this is not replacing by any
38:09
means and this is like a
38:11
much better alternatives to stock footage
38:13
be role and yeah again allows
38:16
you to discover pass you may
38:18
be wouldn't have gone down by
38:20
yes I I still think do
38:22
so what control and you want
38:25
nuance and you want pay seen
38:27
you're gonna have to use the
38:29
regular tools and I. I
38:32
still find it to be more gratifying
38:34
to do things the traditional way, but.
38:36
Damn. It gives you some really crazy stuff
38:39
that. Is. Outside of the Box I think
38:41
the outside of the box staff is the most
38:43
exciting and. When the.
38:46
Demos. Of sore I went online
38:48
and people actually started to see some
38:50
of the footage that was emerging from
38:52
the system. There were a lot of
38:55
people. Especially in Hollywood who had
38:57
sort of a panic about it and Tyler
38:59
Perry that the famous director said in an
39:01
interview with the Hollywood Reporter that he was
39:03
basically bowled over by some of this footage
39:06
and that he was actually planning to put
39:08
on hold and at a plant expansion of
39:10
his studio because he was just like i
39:12
don't know what what I need right now
39:14
what he'd go. If I can just sit
39:17
in my office and creates amazing foot is
39:19
using this ai tool, you know why do
39:21
I need to go through that? It's the
39:23
no hassle of building at an expensive studio,
39:26
so. Do you think those people who
39:28
saw this and freaked out are overreacting? Is it
39:30
the case that Decks are? The closer you get
39:32
to this technology, the less impressive it is. Or
39:34
to what do you make of some of the
39:37
responses that have come out about this tool? I.
39:40
Feel like the more. You. Use
39:42
these tools the less afraid you
39:44
are of them because you do
39:46
understand their limitations and you understand
39:48
their place for them. And you
39:50
understand what separates this from. using.
39:52
Other traditional tools be affects
39:55
her in camera or actors.
39:57
I think what Tyler Perry
39:59
is same. The Is. Somewhat.
40:02
Harmful and sending the wrong message to
40:04
people that are at the top at
40:06
the studio level that are the gatekeepers,
40:08
the ones that have the money to
40:10
say hey guys, let's not spend our
40:12
money And as it's incredibly capitalistic way
40:14
of thinking, South it's
40:17
not the in think he's from
40:19
necessarily about the potential of the
40:21
technology to displace labor in, so
40:23
making it's that's enough. Basically this
40:25
is this is served someone. Same.
40:27
Acquire part out loud. Like saying we we might not
40:30
want to spend all this money on humans. I think
40:32
it's both to. I mean, he hadn't. Even
40:34
tested sore at the time. I don't
40:36
know if he has it now, but
40:38
am I think he had the wrong
40:40
interpretation of of sort of been this
40:42
kind of replacing everything? It'll create certain
40:44
efficiencies for sure, but they're all the
40:46
people. Are the people on Twitter that.
40:48
Love. To tweet the line or
40:51
Ip Hollywood I really encourage them
40:53
to go and actually watch a
40:56
movie. By seriously
40:58
don't watch a movie trailer,
41:00
Go watch a real movie
41:02
and see how much nuance
41:04
and detail and micro decisions
41:06
that are made at every
41:08
split second of Asylum. From
41:10
from an actress choice sad,
41:12
the A, Static, and everything
41:14
movies are incredibly complicated. Let.
41:16
Me ask about it at a different
41:18
way in earlier in the upset at
41:20
we are talking about the fact that
41:23
I have for a while I put
41:25
a i generated images into my newsletter
41:27
and my readers ultimately just trainers revolted
41:29
against at like I got a lot
41:31
of feedback to spend like we hate
41:33
this year you're using images that were
41:35
trained are not copyrighted material. Your taken
41:37
away money friends from human illustrators I
41:39
imagined you know that that you you
41:41
might have gotten some some similar feedback
41:43
where you can at least imagine getting
41:45
that's he backstory you think. About
41:47
Those questions. Source:
41:49
At Wealth, can I ask what? what was
41:51
your illustration budget per year for your newsletter?
41:53
I will. So that's the thing I was
41:55
A. and you know, I mean I had
41:58
access to some image libraries like Getty. Images
42:00
you know he in are all those cases
42:02
a a human being was was paid for
42:04
their labor rights to that's what I but
42:06
using but I was not commissioning standalone illustrations
42:08
for for my pieces. Exactly. I
42:10
think that's what people are missing
42:12
is that we are creating content
42:15
now that simply wouldn't have existed
42:17
before. Sure, there can be pretty
42:19
studio heads at the top of
42:21
that, will try to find ways
42:24
to lights at the bottom line
42:26
and increase their margins, but. For
42:29
the most part of people that are
42:31
using these things are making things that
42:33
that just wouldn't have existed. As you
42:35
see musicians and notoriously have zero budgets.
42:37
Sometimes they posts like I and a
42:39
i just Generated image on Instagram and
42:41
then people like our no not you
42:43
use a I to like oh My
42:45
God cancel on subscribes and it's a
42:48
site. But. They wouldn't have
42:50
posted anything that day if they didn't have
42:52
a I am and so it's It's kind
42:54
of like opening up a new channel. and
42:56
I mean yes, there's like sensitivity around. like.
42:58
Black. Box of what's been trained
43:00
on. But reality is we can
43:02
buy clothes. Pandora's Box waiting like.
43:05
Technology. Is relentless and and we
43:07
have to just kind of adapt using these
43:09
things. The way I feel like the narrative
43:11
needs to be kind of steered is that.
43:14
Ninety nine percent of scripts and
43:16
hollywood get rejected and an even
43:19
of the one percent that get
43:21
and bought. Only. About half
43:23
of those go into production and
43:25
so I think this will open
43:27
up the opportunity for people with
43:29
these ambitious and bold ideas to
43:31
resurrect. Project said that when they've
43:33
existed. I'm even going back to
43:35
some music video concepts and I
43:37
pitched in the past that just
43:39
like weren't gonna work for the
43:41
budget. Yes, I
43:44
have seen some backlash pointed
43:46
at you and other filmmakers
43:48
who has been experimenting with
43:50
these tools. physically accusing open
43:52
a i was artist washing
43:54
of your face of a
43:57
using artists to sort of
43:59
test out. Tool to show all
44:01
the cool and creative uses for them
44:03
while actually certain negotiating behind the scenes
44:05
to add to replace a bunch of
44:07
labor or to to use these tools
44:09
that many artists feel have been trained
44:11
on their work or or work of
44:13
their colleagues without permission. So what else
44:16
you could just speak to that they
44:18
this idea of open a I sort
44:20
of using artists and filmmakers to try
44:22
to convince a skeptical public that all
44:24
the stuff is just going to be
44:26
good and can enhance creativity and that
44:28
it's not gonna replace anyone's. Jobs while
44:30
actually having a very different strategy
44:32
behind the scenes. Shore and. I.
44:35
May I think that is a very
44:37
fascinating points and I it's something I
44:39
kind of grapple with all the time
44:41
because I have I again I still
44:43
love to do the traditional way I
44:45
so love to employ people are then
44:48
the other side of me as a
44:50
plane with all this new tech and
44:52
I might am I just some sort
44:54
of pawn in the sights. grades master
44:56
plan as of Agee Eyes but it's
44:58
I've What is that? The opposite of
45:00
this as you don't want artists involved
45:02
in the research process. I feel like
45:04
including artists if you're developing. Things that
45:06
are that as visual technology including artist
45:08
in the process is is critical because
45:10
otherwise you're just kind of in. This
45:12
does Bubbles, Am and and young really
45:15
understand like what the purposes of your
45:17
research. A. What one question that
45:19
comes to mind for me Paul is what are you
45:21
working on next with this thing? Can you give us
45:23
a preview of what else you think you can do
45:25
with us Aura? Yeah I
45:27
mean. So. I will say
45:29
everything has to be kind of run
45:31
through open a I in order to
45:33
make it to the public. They are
45:35
being very kind of selective with that
45:38
they so they don't I can inundate
45:40
people the been careful with how much
45:42
as is released but I have my
45:44
brain has been spiraling. I've been. Working.
45:47
On. A short song
45:49
I'm I'm also I'm working on a music
45:51
video and on of that's a breaking numbers
45:53
in bringing it up as broken as has
45:55
hats so that that will add that can
45:58
say who are when or what bad. Well
46:00
let's just say Beyond Say does have a new
46:03
record out in. a lot of people are listening
46:05
to us as a specific answer is yes. That
46:08
and then the schools and record projects
46:10
and Sleaze is bigger. But I'm also
46:13
I'm I'm still exploring other routes and
46:15
I don't I don't see sore as
46:17
as oh, I'm only going to focus
46:19
on this tool to get everything out
46:21
of my head. It's just a at
46:23
the supplemental saying, but it's it's been
46:25
very liberating. I'll say that. I.
46:28
Bought we Gotta Run and
46:30
so much larger that I
46:33
really appreciated cynicism. Lox. Debate
46:40
on something sedated. This
46:54
podcast is supported by Americans United
46:57
for Separation Of Church and State.
46:59
Americans United defense your freedom to
47:01
live as yourself and believe as
47:03
you choose, so long as you
47:05
don't harm others. Core freedoms like
47:07
abortion rights, marriage equality, public education,
47:10
and even democracy itself rest upon
47:12
the wall of separation between church
47:14
and state. Pristine nationalists are attacking
47:16
these fundamental freedoms. Americans United is
47:18
fighting back freedom without favor and
47:20
a quality without exception. Learn.
47:23
More about a use work at
47:25
a You.or access and whitey. Kiss.
47:29
You know we love a caper on
47:31
this show. Off we Love a Keeper
47:33
On Saturday I have when I believe
47:36
is the biggest tech caper of at
47:38
least the past year. Okay well then
47:40
let's hear about it. And I want
47:42
to preface this caper by saying that
47:44
this is going to involve a lot
47:46
of terms from the Linux and open
47:48
source software development and various databases. And
47:50
and I need you not to fall
47:52
asleep because the payoff is gonna be
47:54
worth and will there be a quiz
47:56
at the at the typical V? Oh
47:58
yes. So this is. A A
48:00
very interesting and strange story that came
48:02
out of the world of cyber Security
48:05
over the past week. and it involves
48:07
a thirty eight year old software engineer
48:09
named Andreas Frames good for of the
48:12
pod the effort Griffin Health of I
48:14
did. He was in San Francisco, he
48:16
works at Microsoft and he stumbled into
48:19
what may be the biggest attempted cyber
48:21
attack in history, Which it's crazy that
48:23
sort of just one person could stumble
48:26
into something as big as the says
48:28
yes, I've been totally obsessed with. The
48:30
stories are basically I. I think we
48:33
should just explain off the bat that
48:35
the internet is a very sort of
48:37
rickety contracts. I think most people don't
48:40
understand this unless you talk to people
48:42
who are engineers or work and cyber
48:44
security. or sort of develop the set
48:47
of building blocks and which the entire
48:49
Internet rests. It's weirdly precarious that it
48:51
works. I had an hour. Honestly, someone's
48:54
miraculous right? because imparts so much of
48:56
the internet, the technology reliance depends on
48:58
these tiny little open. Source projects that
49:01
might be maintained by for example,
49:03
one person gets the other. There's
49:05
a famous Xkcd comic where you
49:07
have kind of like a a
49:09
a a vast must or that
49:11
is sort of resting on one
49:13
little sin Like Rob and the
49:15
elaborate machine is labeled all modern
49:17
digital infrastructure and the little peg
49:19
that it's resting on is labeled
49:21
a project some random person in
49:23
Nebraska has been sank with we
49:25
maintaining since Two thousand and three
49:27
at this story is the literal
49:29
instantiation of. That cartoon because of it
49:31
as a sort of our society bad
49:33
as yes, this is the tiny peg
49:35
that is holding up the giant machine
49:37
called the Internet. And how one guy
49:40
Andreas Freud's I discovered basically by accident
49:42
a plot to kind of mess with
49:44
the entire internet as we know it
49:46
all right. So how are we stumble
49:48
across as Plucked Oven so he works
49:50
in Microsoft. He develops this piece of
49:52
open source database software called Post Dress
49:54
Up. The details aren't important but this
49:56
a big database. Lots of companies use
49:59
it. It's open. The Worth Hundreds is
50:01
one of the people who maintains it's
50:03
database. As part of his work, he
50:05
does a bunch of tests to make
50:07
sure that various pieces of this software
50:09
are running correctly. A few weeks ago
50:11
he's doing some tests and he says
50:13
noticing some weird error messages and at
50:15
the time you know he's flying back
50:17
from Germany, was visiting his parents his
50:20
German and he's sorted, set, lagged. He
50:22
thinks okay, maybe this is not important,
50:24
I'm just gonna collect ignore these error
50:26
messages then he gets home to San
50:28
Francisco. He starts running some
50:30
more pasts and he says noticing
50:32
some other weird error. Some anomalies.
50:34
Some anomalies. Yes, he notices For
50:37
example of that, this process called
50:39
Ss H. It's running sort of
50:41
slower than possible by a little
50:43
bit is using more processing power
50:45
than I usually would. It's causing
50:47
some memory errors that usually aren't
50:49
there. Now can we say slowdown
50:52
isn't slowing down by like thirty
50:54
seconds? None of these delays, they
50:56
are. They are tiny. It's like
50:58
measured in milliseconds. But your Andreas
51:00
is a very detail oriented guy. He's
51:02
been working on this particular piece of
51:05
software for a long time. Any kind
51:07
of knows what it's all supposed to
51:09
look like and so he started noticing.
51:11
Like a little lag there a little
51:13
more Cp use uses. Here's something is
51:15
going wrong. Hey, Spidey sense starts to
51:18
tangle exactly. So he basically starts digging
51:20
in and investigating. And he traces the
51:22
issue to the set of data compression
51:24
tools called X the Utes Hills. I
51:26
wondered if it might be an axial
51:29
tilt. Member of a specific it gets So
51:31
you're basically the details of what the Sig
51:33
is are not important but it of I'd
51:35
try it would just that are to me
51:37
it's a set of data compression tools that
51:39
is all I now okay sort of like
51:41
the of the premise of the old Silicon
51:43
Valley Tv so exact I but x you
51:46
tilt is used by Linux, the open source
51:48
operating system and another piece of information that
51:50
you need to know at the story that
51:52
linux is probably the most important piece of
51:54
software and the were linux is everywhere. Yes
51:56
so Linux is used by the vast majority
51:58
of. The world. Data Centers servers
52:01
like every major really important computer
52:03
in the world runs on Linux.
52:05
Or there's a computer talking to
52:07
another computer somewhere Linux is involved.
52:09
Yes, Yeah. So this little tiny
52:11
software package exit you tills is
52:13
a very small piece of a
52:15
very important piece of software. So.
52:18
Audrey starts looking into these weird delays
52:20
and and these weird anomalies and he
52:22
eventually starts looking at the source code
52:24
for accede you tells and he discover
52:26
something that blows his mind with as
52:28
he finds a backdoor. Now a backdoor
52:30
I know you're gonna make a joke
52:32
about. I'm not going to make his
52:34
else about best or you're passing up
52:36
jokes about backdoors today. Something is wrong.
52:38
Are you okay Timothy Mcveigh This is
52:40
plenty of the keep say but I'm
52:42
not going to make a joke of
52:45
it as a fastball over the plate
52:47
of my friends. Is. That
52:50
of so it is basically a back
52:52
is a piece of malicious code that
52:55
is inserted into a piece of software
52:57
that allows and it's hacker to to
52:59
basically remotely access or control that are
53:02
sort of slip in some code to
53:04
they road space we do something malicious
53:06
it's it's kind of a seats to
53:09
unlock a piece of software of with
53:11
the intent to mess with it in
53:13
some way goddess So. Andreas.
53:16
Is not a cyber security engineer, he's
53:18
just a guy who maintains a database
53:20
and but he finds this evidence that.
53:23
X you till this tiny piece of
53:25
linux has been compromised that someone has
53:28
intentionally gone and and placed a backdoor
53:30
their that that if you are that
53:32
person you can then go in and
53:34
you can basically tamper with any computer
53:36
that is running Ssh on linux which
53:38
is to say the vast majority of
53:40
the important computers on earth which like
53:42
I'm just imagining being here Myths and
53:45
like you've noticed these this like series
53:47
of small anomalies as you have that
53:49
feeling that something is amiss yes but
53:51
I bet that even in his wildest
53:53
imagination. He did not imagine that.
53:55
He vowed a very sophisticated backdoor
53:57
know. So I talked to Andreas.
54:00
The his discovery, he's or walked me
54:02
through the whole thing any says it.
54:04
At first he was sort of like
54:06
skeptical of his own findings. He said
54:08
it's seltzer real. He said there were
54:10
moments where I was like I must
54:12
have just had a bad night of
54:14
sleep and heads and fever dreams. Basically,
54:16
this is not the kind of thing
54:18
that you find in a widely scrutinized
54:20
piece of software like Linux. And so
54:22
Andreas. He looks at this, he says
54:24
man and or know this to sounds
54:26
like too big to be true. And
54:28
how could something like this. Agnew to
54:30
get approved and make it's way into
54:32
the the release version of linux but
54:34
he keeps digging, he keeps finding new
54:36
evidence and then last Friday he basically
54:38
rights up what he found and sends
54:41
it to this group of open source
54:43
software developers and he makes a basically
54:45
says all these errors that I've been
54:47
seeing all these anomalies in this me
54:49
know in In is very obscure software
54:51
packages. It's all because this
54:53
thing has been back toward. Someone is
54:55
here messing with this release and they
54:57
are intending to use this to to
55:00
basically break into a bunch of computers
55:02
and do whatever they want. So earrings
55:04
the Alarm of Hearings. The alarm. And
55:07
immediately the entire cyber security world meltdowns.
55:09
I talked to one researcher Alex Stamos
55:11
i know you know he's a former
55:13
as Si A at Facebook is now
55:15
are involved in some the called Sentinel
55:17
One which is a cyber security research
55:20
firm and he told me this could
55:22
have been the most widespread and effective
55:24
backdoor ever planted in any software products.
55:26
while and basically you know what people
55:28
I talked to said his. Look, if
55:30
you have this back door if you
55:32
have this master teeth that lets you
55:34
get into any linux computer that is.
55:37
Running as as a it's is
55:39
very ubiquitous software package, you essentially
55:41
have a way to get into
55:43
hundreds of millions of computers around
55:45
the world. Once in their, you
55:47
can steal private information, you can
55:49
intercept encrypted traffic, you can plant
55:51
malware, you can cause major disruptions
55:53
to like big pieces of infrastructure,
55:56
and critically, you can do all
55:58
of this without being caught. Because
56:00
part of what. Andreas. Discovers as
56:02
is investigating this back door is that
56:05
whoever planted it there has taken
56:07
steps to ensure that is very hard
56:09
to detect. And you know
56:11
basically this would have worked if not
56:13
for Andreas and his very eagle eyed
56:15
detail oriented like obsessive approach to add
56:18
trying to figure out what the heck
56:20
was going on with these error message
56:22
This is why nerds are so important
56:24
to the economy and to celebrate a
56:27
bit. three celebrated are such as if
56:29
people that are like this process is
56:31
running one second to slow cancel my
56:33
afternoon, we elevate you totally. So this
56:36
discovery i think it's safe to say
56:38
it is is a huge shock waves.
56:40
Through the world of cyber security because
56:42
you know this thing was caught before
56:45
it could do any real damage. It
56:47
had not made it into the sort
56:49
of widely used versions of Linux that
56:52
all these servers run on, but it
56:54
would have any, Would have been a
56:56
potentially disastrous I'm now that are. Andreas
56:58
has kind of become dislike nerd hero.
57:01
All kinds of people are are are
57:03
praising him. Satya Nadella his served as
57:05
the Chief executive Microsoft his bosses boss's
57:07
boss as a praise him for his
57:10
curiosity and craftsmanship. Are there was
57:12
a popular are post the went around
57:14
calling Andreas be Silverback gorilla of nerds
57:16
and people are basically comparing him to
57:19
the little Peg in that a comic
57:21
than all of modern capitalism rests on.
57:23
Okay let's get the who did this
57:26
cousin to. That's what everybody wants to
57:28
know right? Who's responsible for this backdoor?
57:30
Those was as bad or bad it.
57:32
So here's what we now so far
57:35
according to some researchers I talked to
57:37
this is so elaborate. this plot was
57:39
was so sophisticated. That it couldn't
57:41
have to spend like a random group of
57:44
hackers. This had to have been like a
57:46
nice and state like a Russia or China
57:48
or and North Korea someone with access to
57:50
vast resources and very skilled teams of hackers.
57:52
Web island. I'm interested in that cabin because
57:54
as far as I can tell, the main
57:57
thing that separates this attack from many of
57:59
the other attack. That U C all a
58:01
time is just how much time they invested.
58:03
Employee lot right suits us to talk to
58:05
us about all the time involve. He has
58:07
a one of the cool things about open
58:09
source software that you can actually go back
58:11
in like see all of the changes and
58:13
who was requesting them and and what they
58:15
actually met in terms of what ended up
58:17
in the codes and so researchers have been
58:19
going back and from trying to forensically look
58:21
at all the evidence trying to set of
58:23
see how this happened and they found a
58:25
really interesting story buried in some of the
58:27
details of this software. So
58:30
back and Twenty Twenty One there's a
58:32
user who creates a get have account
58:34
and starts contributing to various open source
58:36
projects. This user uses the name Xia
58:39
Tan for various reasons. Researchers actually don't
58:41
think that's a real name, it's probably
58:43
a pseudonym. sort of. Smarter soon. Amid
58:45
the what I would have picked which
58:47
would have been backdoor wizard this s
58:50
a G, A Tan or whoever it
58:52
is they start suggesting sort of the
58:54
teams is to access you tills starting
58:56
back in Twenty Twenty Two. And
58:59
this is the ways that Open Source development
59:01
and a Works is like. People proposed, been
59:03
to changes and and and indies special our
59:05
inner tubes developers called maintain or were sort
59:07
of in charge of a projects. They look
59:09
at the proposed changes they serve, test them,
59:11
mixer, they work, see what affects they have
59:13
on performance and then if they're good they
59:15
approve them in that kind of gets like
59:17
merge into the main coed but the basic
59:19
ideas that everyone who participates is essentially a
59:21
good samaritan. right there somebody comes along and
59:23
says i use the software I noticed this
59:25
thing could be better. What am I right?
59:27
A little co depicts it's also. Minutes you
59:29
and if you like it's you can share
59:31
with everyone. Yes, and many of these projects
59:33
only have one or a handful of maintain
59:35
as a test them because these are not
59:37
like these are not like Sas moving software
59:39
objects. Rothys are not things that are being
59:41
constantly refined and redevelop. He's a specific. This
59:44
is what infrastructure that yeah like is what
59:46
you you built the A You have a
59:48
plumbing ducts for the internet and it's just
59:50
gonna sit there mostly and people are going
59:52
to build on it and use it for
59:54
stuff that they're building. but you actually don't
59:56
need much more than one person to kind
59:58
of keep tabs. On this project right? Mostly
1:00:00
done but software is never totally
1:00:02
done exactly So this person's yeah
1:00:04
Hannity. This group of people using
1:00:06
this names yet hands start kind
1:00:09
of proposing changes and then they
1:00:11
start of gradually serve social Engineering.
1:00:13
The entire team that's involved in
1:00:15
maintain this project which again is
1:00:17
mostly comes down to one person,
1:00:19
one maintain or has been doing
1:00:21
this for many years. Sergio.
1:00:23
Ten starts contributing be sort of
1:00:25
minor proposed changes to X You
1:00:27
Tells Back and Twenty Twenty Two
1:00:30
and then something interesting happens. which
1:00:32
is that, See a tan? Whoever
1:00:34
it is, whatever national the hacking
1:00:36
team it might have bands they
1:00:38
start trying to is basically take
1:00:40
over control of X The You
1:00:42
tells. And they do this by
1:00:44
essentially seizing on the fact that
1:00:46
the person who maintains this software
1:00:48
project is getting kind of tired
1:00:50
of doing it the the third
1:00:52
Don't wanna do this anymore. It's sort
1:00:54
of sounds like and so it's just Han,
1:00:56
whoever it is test sticks their hand up
1:00:58
and says well what if I was the
1:01:00
main tanner and I would if I could
1:01:03
like solve this problem for you by taking
1:01:05
over this very thankless task of maintain this
1:01:07
tiny little software like think that problem off
1:01:09
your hands. So over the course of a
1:01:11
couple of years Jets hand builds frost with
1:01:13
the other people who are involved in contributing
1:01:15
to this software tool and as and eventually
1:01:18
gets named a maintain her of this project
1:01:20
and so he is able to kind of
1:01:22
do the final approval for. This proposed
1:01:24
code seems that would insert this hidden
1:01:26
back door into the software project, effectively
1:01:29
becoming a double agent like something
1:01:31
you would read out of a novel
1:01:33
by a lake. Or it's while
1:01:35
it's a it's like honestly v be
1:01:37
trade craft. The kind of spycraft involved
1:01:40
in this is a very sophisticated
1:01:42
operation. It involves not it's a technical
1:01:44
piece of of of hacking but also
1:01:46
have a social peace where you're
1:01:48
kind of winning over the small team
1:01:51
of very harried, very under appreciated
1:01:53
developers. You're volunteering. To help them,
1:01:55
you're kind of establishing your credibility
1:01:57
in this very tiny community. And
1:01:59
eventually years. Using their credibility to install
1:02:01
a backdoor that will let whoever it is
1:02:03
have access to hundreds of millions of computers
1:02:05
and do whatever they want with them. A
1:02:08
wild story It as and now I know
1:02:10
that we don't know truly anything about the
1:02:12
real identity of G A Damn. But I'll
1:02:14
tell you, I like thinking of G A
1:02:17
Tan as as a deadly woman assassin Carmen
1:02:19
Santiago back of the Ideal Computer Games yeah,
1:02:21
sort of yo international criminal and In In
1:02:23
your is very elusive. G at hand is
1:02:26
the new Carmen Santiago. That's true, you know,
1:02:28
and I think we should devote. Many episodes
1:02:30
of podcast trying to track down to his hand at
1:02:32
the end of every episode where they were the world
1:02:35
as the attack and we're going to keep saying it's
1:02:37
a we find out at night part of that show
1:02:39
was the the do It Rock capella remember that yes
1:02:41
it was one of the only are capella theme songs
1:02:43
river Heads for a show and it was so successful
1:02:46
that they should up accessory so. I.
1:02:48
Think there are lot of things to say
1:02:50
about this story, but one of the search
1:02:52
interesting side discussions that I've seen and come
1:02:54
out of this. You know there's this whole
1:02:56
group of people in Silicon Valley who believe
1:02:59
that a I should all be open source
1:03:01
and that the reason that you would want
1:03:03
something like you know as an Ai language
1:03:05
mile to be open source is because then
1:03:07
you'll actually be safer because then you'll have
1:03:09
not just one company kind of trying to
1:03:11
keep the bad guys that you'll have this
1:03:14
kind of distributed army of volunteers who are
1:03:16
constantly sir looking through things, poking around in.
1:03:18
The source code. It's happened to the Global
1:03:20
Nerd Hives, exactly. And that's so how are
1:03:22
you get things like Linux, which are the
1:03:24
result of thousands of contributors working on their
1:03:27
little pieces of this thing? Eventually it all
1:03:29
comes out in it's pretty secure for the
1:03:31
most parts of and so that is One
1:03:33
thing that those people are now saying is
1:03:35
this episode with X the the X the
1:03:38
Back door and proved that all software needs
1:03:40
to be open source. When a mega that
1:03:42
I mean look at all. So all software
1:03:44
does not need to be open source. It's
1:03:46
perfectly fine to haven't from a. Private companies
1:03:48
making their own software. But I think to
1:03:51
the degree that a piece of software is
1:03:53
traditional to how the internet works out, there
1:03:55
is a really great case for making it
1:03:57
open source. Yes, I am saying for to
1:03:59
under is not just for saving us all
1:04:01
from doom, but also for forcing me to
1:04:03
learn about Linux development and open source repositories
1:04:06
and maintain. Or that I guess I'm just
1:04:08
struggling with one more question Kevin which as
1:04:10
we enter this been oh brave new world
1:04:12
where there are a lot of G attempts
1:04:14
out there. What are you doing to protect
1:04:16
your backdoor? And
1:04:19
that that's all we have this week's on
1:04:22
our fourth. Place
1:04:24
is. Looking
1:04:40
for new? ride? something next car from
1:04:42
the comfort of your home? One hundred
1:04:44
percent online with Carbon. Carbon is a
1:04:46
massive inventory, including thousands of cars under
1:04:48
twenty thousand dollars. So sending a car
1:04:50
this is your budget and lifestyle is
1:04:52
hassle free. Carbon A makes car financing
1:04:55
hassle free. Guess he qualified in minutes
1:04:57
by answering this you click questions Carbon
1:04:59
also of his customizable terms of payments
1:05:01
as low as zero dollars down. Because
1:05:03
payment plans to fit your plans, this
1:05:05
is provided.com or done of yesterday to
1:05:07
sign your next car without ever leaving
1:05:09
the comfort of home or anywhere else.
1:05:11
Terms and conditions may apply. Listing.
1:05:15
Before we go, we continue to be really
1:05:17
interested in how young people are using technology
1:05:19
and we've been hearing stories about snapshot causing
1:05:21
drama in middle schools and high schools and
1:05:23
we want to see or about it. Has
1:05:26
snapshots roiled your school or your kids' school?
1:05:28
In some ways, what was the snaps at
1:05:30
incident where you lived? Let us know. Email
1:05:32
us at Hard Fork and I would have
1:05:35
zoc up the mess years a better. Or
1:05:39
works for it by Resort Town
1:05:41
and Whitney. Chance of Winning Edited
1:05:44
by and points of Accept My
1:05:46
Caitlin Love Studies show was engineered
1:05:48
by Alyssa Moxley. original music by
1:05:50
Zion Wong at Mccusker and Damp
1:05:53
Towel or Audience editor is no
1:05:55
globally. Video. Production by Ryan
1:05:57
Manning and Don't Ferguson has you either.
1:06:00
I got our youtube channel the You
1:06:02
tube.com/for for special thanks to Polish Human
1:06:05
We wings him a prestige and
1:06:07
different as always. You can email us
1:06:09
at third, fourth and. You
1:06:12
know, Some
1:06:35
for quick break to Taco Bell Mcdonalds
1:06:37
we got in big lies. Did your
1:06:39
taste buds ready for Mcdonalds breakfast bagel
1:06:42
sandwiches Now just three dollars only on
1:06:44
the app to were delicious. Stake Agencies:
1:06:46
Beagle Bacon Agencies: bagel or sausage second
1:06:48
seed bagel Just three dollars when you
1:06:50
order a hit on the up, hurry
1:06:52
and sees his breath the steel before
1:06:55
it's gone. Off about of one time
1:06:57
daily March eleventh are able Seven Twenty Twenty four
1:06:59
participating Mcdonalds. Must have been to awards.
Podchaser is the ultimate destination for podcast data, search, and discovery. Learn More