0:00

So Mike, I'm guessing you've probably used

0:02

at least one of the big Chinese platforms.

0:04

The marketplaces that everyone is, using

0:06

nowadays. Sheen or Timu. I

0:08

don't know. Maybe is that Mike stand

0:11

from one of those platforms?

0:15

No, no, no. The mic stand is not from one of those

0:17

five. I have used, the

0:19

original one of those, which is AliExpress.

0:21

Uh,

0:23

Oh yeah.

0:23

purchased many things from AliExpress, but

0:25

I have not gotten into the, Timu

0:28

Shian, uh, elements of this world.

0:30

Okay. Well, you don't need to anymore actually, because

0:32

Amazon, you may have seen late last year

0:34

introduced Amazon hall, which is

0:36

essentially the same thing. Super cheap

0:38

goods from China. You

0:40

know, at your convenience, basically. And I

0:42

thought I'd use their user

0:44

prompt, their homepage prompt the start of today's

0:47

control or speech podcast. Amazon

0:49

hall prompts you to find new faves

0:51

for way less.

0:56

Well, given how costly,

0:58

the folks ripping apart our government

1:00

is, I'm, hoping that we can, find

1:03

a, new governmental

1:05

system technology for,

1:07

for everything that is possible. Being, uh,

1:09

ripped apart in our government so that

1:11

we might have a functioning U S federal government.

1:16

that seems like a good trade. I'm not sure

1:18

if Amazon hall has that, but

1:20

Yeah. What, what about you? What, uh, what new

1:22

fav do you want?

1:23

well, I'm actually in need of a new.

1:26

Hat, a new hat. It's pretty cold

1:28

over here in the UK. So I'm going to, buy myself

1:30

a new fave from Amazon hall. maybe

1:32

the logo that says control alt speech, we

1:35

make sense of the chaos.

1:38

There we go.

1:48

Hello and welcome to control alt speech,

1:51

your weekly roundup of the major stories about

1:53

online speech, content moderation,

1:55

and internet regulation. It's February

1:57

the 6th, 2025. And this week's

1:59

episode is brought to you with financial support

2:01

from the future of online trust and safety fund. My

2:04

name is Ben Whitelaw. I'm the founder and editor of

2:06

Everything in Moderation. And I'm

2:08

with Mike Masnick, editor and founder

2:10

of Tector, who has

2:12

had a hell of a week.

2:17

Yeah. I mean, you know, there's

2:19

a lot of stuff going on.

2:20

Yeah, yeah, I think that's fair to say, Mike, I

2:22

mean, not a politics podcast,

2:25

but we should kind of address the elephant in the room,

2:28

which is the fact that, Elon Musk

2:30

and others in his immediate circle

2:32

are getting access to the federal government. In

2:35

the States and doing crazy shit with it.

2:38

Yeah. Yeah. Which is, has been

2:40

interesting. And, and, you know, one of the interesting

2:42

things is that. Because so much of

2:44

what is happening is sort of Elon Musk driven,

2:47

the folks who spent a lot of time

2:49

following what he did with Twitter

2:52

in 2022, 2023

2:54

and so forth, seem to recognize

2:56

what's actually happening much faster

2:58

and, and much more clarity and

3:01

depth than political reporters

3:03

who are treating this as sort of an ordinary transition

3:05

in power. And. You

3:08

know, I think the best reporting on it really

3:10

come from Wired, you know, which is

3:12

famous technology magazine, old school

3:15

technology magazine. and some of the other

3:17

tech reporters as well, actually have

3:19

a deeper, better understanding of

3:21

Elon Musk, the way he works and what it

3:23

is that he's doing and how incredibly

3:25

crazy it is. Because as folks

3:28

will remember who followed this stuff, I mean, he bought

3:30

Twitter and he had every right to do this, but he

3:32

went in and just, quickly made it

3:34

clear that he had no care

3:37

and no intellectual curiosity to learn how

3:39

things worked, just assume that he knew

3:41

how it must work, was wrong

3:43

on almost all of those things, but just made a whole bunch

3:45

of very, very quick decisions, fired

3:48

a ton of people, broke systems, ripped

3:50

out servers, and just assume this is

3:52

all stupid. Nobody was here before

3:55

was smart. They were all, you know,

3:57

woke idiots. And therefore, you

3:59

know, we can just rip stuff up and nothing will go wrong.

4:02

And of course, lots of stuff went wrong and they lost all

4:04

their revenue and a whole bunch of other problems

4:06

cropped up. And he's using the exact

4:08

same playbook on the U. S. government.

4:10

and that is absolutely terrifying for

4:13

a long list of reasons.

4:15

Yeah. I mean, you wrote about it this week on tech, the

4:17

kind of Twitter deconstruction playbook.

4:19

I mean, we could spend hours and hours

4:21

talking about just that single

4:23

story. We're going to try and, branch

4:26

out a little bit because it has been covered so extensively,

4:28

but just talk a little bit about what it's been like. Kind

4:31

of writing on tech that this week,

4:33

because you have seen this huge increase

4:35

in readers and

4:37

we've seen a bunch of people come to old

4:39

episodes of the podcast and listen to that

4:41

last week's episode with Rene D'Arresta was one

4:44

of our largest yet. So

4:46

just talk a little bit about what that's been like, from

4:48

your own perspective.

4:49

Yeah. I mean, it's been, it's been overwhelming

4:51

and crazy. I'm, I'm working on less sleep than usual

4:54

and just trying to, to keep up with everything. I

4:56

mean, there's way more stuff than, anyone can possibly

4:58

keep up with. Even like, larger newsroom,

5:01

will have trouble keeping up with this stuff. but it's,

5:03

been fairly overwhelming and sort of trying to,

5:05

figure out which of the things that, you

5:08

know, The traditional media is having trouble sort

5:10

of putting into context and figuring out,

5:12

you know, can I take that and

5:14

put it into context and speak to experts

5:16

and find out what's really going on and then figure

5:18

out how to explain it in a way that makes sense. And,

5:21

that has been a lot of my week

5:24

and it's somewhat terrifying just because,

5:26

we've seen how this plays out. In

5:28

a way that is doesn't

5:31

really matter, right? You know, with just a

5:34

second tier social media network, an important

5:36

one, an important one for voice. And we've sort of

5:38

seen how that up allowed

5:40

for things like, blue sky again

5:43

on the board disclaimer, blah, blah, blah, but

5:45

also mastodon and everything else we've

5:48

seen how that enabled those things

5:50

to spring up, which was interesting, but what

5:52

is the, blue sky of the United

5:55

States, right? Like there's, you

5:57

don't have the same exit path, that

5:59

you do with, a social network. And

6:01

there's a whole bunch of really important

6:03

things. And, the fact that we're seeing

6:05

over and over again, the same thing where Musk

6:08

and his team insist

6:10

they know what's going on and don't understand

6:12

it and have no interest in learning why things are

6:14

actually done the way they are. And, you know,

6:16

it sort of culminated with one of his,

6:19

he has this 25 year old kid who worked for

6:21

him, who was given,

6:23

first of all, terrifyingly enough, any access

6:26

to the treasury nonpolitical

6:29

thing. It was run by a guy who had worked there

6:31

for 40 years and who was, somebody

6:34

in the former Trump administration said, you

6:36

know, he was. You know, at no point did I have any

6:38

sense of what his politics were because his job

6:40

there is to be the sober person in the

6:42

room. He, that person got pushed out

6:45

and instead this 25 year old, kid who

6:47

worked for Elon Musk was given

6:49

full access to the system

6:51

that handles U. S. government payments. Six

6:53

trillion dollars go through it and

6:57

was apparently given right access to

6:59

it. It's this old system that is, you

7:01

know, using COBOL code

7:04

and the kid was, was making changes

7:06

to it. without testing it. There

7:08

are all these terrifying things of things that could

7:10

go wrong. The system is set for migration

7:12

this week. Nobody knows how that's going to work. This

7:15

morning, a court stopped them and said that

7:17

they have to have read only access,

7:19

which is already terrifying enough. So he can't

7:21

write to it anymore. and I've seen

7:23

reporting saying that the right access has been turned

7:26

off, but this is, is terrifying.

7:28

It's not, It's not, Twitter,

7:30

right? Like you could break Twitter and the world will move

7:33

on. You break the entire United States

7:35

the world becomes a very different place.

7:38

And you're right. There's eerie. parallels

7:40

between what he did at Twitter

7:43

X, you know, going in, firing the whole

7:45

trust and safety team, getting rid of the kind of

7:47

adults in the room and what he's doing here

7:49

within the kind of department of government efficiency.

7:51

And, And this will probably be out of date by the time

7:53

this, podcast goes out, but like, what,

7:56

what are you thinking will happen in the next kind of seven days,

7:59

like by the time we next record, like what, what

8:01

are you kind of expecting will happen?

8:03

it's just more craziness, right? I mean, we,

8:05

we don't know, right? I mean, the other reporting

8:07

that came out last night was another,

8:09

you know, Musk employee, a kid, was,

8:13

basically put in charge of controlling the systems

8:15

from the, NOAA, which

8:17

is, you know, oceanic and handle

8:20

weather stuff and all of these things. And

8:22

then they, everybody who works for NOAA

8:24

were, was given an order not to communicate

8:26

with any foreign national, which

8:29

like. that's what they do, right?

8:31

They have to like communicate

8:34

with other people about, weather patterns

8:36

and, things that are happening and like, are

8:38

there hurricanes in the Caribbean?

8:40

Is there a typhoon or in the Pacific?

8:43

Like all of these things that's, that's

8:45

their job. And you have these kids

8:47

who are like, think they know. Everything.

8:50

I mean, at one point, you know, some idiot

8:52

on Twitter, Twitter, X, whatever,

8:55

was saying, Oh, you know, Elon Musk

8:57

is so brilliant. the first thing he did was come

9:00

in and say, like, well, we have to see

9:02

where the payments are going. No one has ever

9:04

thought that. What do you mean? No, one's ever thought that.

9:06

Of course, I've thought that like this, all

9:09

of these systems in place, they just act

9:11

as if, Everybody else must

9:13

be an idiot. And they are the only ones who

9:15

think of these things. And what the reality is, is they

9:17

don't understand all of the different

9:19

reasons why things were done. You

9:22

know, is there waste in government? Of

9:24

course. Right. Like, is there fraud and abuse in government?

9:26

Absolutely. But There are reasons

9:29

why systems are in place and if you're going to change them,

9:31

you have to understand them first.

9:33

And no interest in that. there's

9:35

not coming in to try and find out why

9:38

are these things in place. The only interest that they're

9:40

having, and again, there's like this 19 year

9:42

old kid who recently graduated high school

9:44

who was going in and demanding people explain

9:46

their jobs, but not to learn from them, but

9:48

to see whether he could fire them, basically.

9:51

I mean, whole thing is just,

9:53

it's so scary.

9:54

Yeah. There's something very apt about the United States

9:56

not knowing figuratively and,

9:59

in real terms when a typhoon

10:01

or a hurricane is going to hit the

10:05

country because of, an edict that says you can't

10:07

speak to Foreign workers abroad about this stuff. I

10:09

think that is, it is insane. I

10:11

would also just, you know, shout out to the

10:13

media organizations, as you say, that have been doing

10:16

this work this week. I mean, we

10:18

cover a lot of their work. we

10:20

analyze a lot of the reporting that comes out

10:22

of, of outlets that you've mentioned. And we obviously

10:24

you do a lot of writing yourself on this, Mike. So I

10:27

think it's worth just, you know, shouting out those guys for.

10:29

covering this in a really crazy time.

10:32

and also we spent the best part of two hours

10:34

prior to recording trying to figure out

10:36

what stories to cover and

10:39

how to make sense of it for listeners. So it's

10:41

been a crazy week or two and it's

10:43

likely to be, like that for a while. So that

10:46

I think probably is a point

10:48

where we can dive into a part of the U S government,

10:50

Mike, that we've touched on before.

10:53

and somebody we've, come back to time and time again

10:55

on controllable speech. before we do that, did

10:57

you know that Jim Jordan is a two time wrestling

10:59

champion?

11:01

Unfortunately, I do know that

11:03

you, I was reading his Wikipedia and I was like, this

11:06

guy's got range.

11:07

well, the history though is right, like, you know,

11:10

he became famous before he

11:12

was elected to Congress, he was a wrestling

11:14

coach Ohio, but

11:16

there's a huge scandal of

11:19

abuse involving his wrestlers.

11:22

That it is reported repeatedly

11:24

that he looked the other way on and

11:26

was aware of. And these are like, one

11:29

of the scandals about Jim Jordan, is

11:32

how, under his leadership

11:34

as a coach, there were some sexual

11:36

abuse scandals that. He never

11:38

took responsibility for. So yeah, he's,

11:41

you know,

11:42

okay. Okay. There's a pattern. Let's

11:44

say there's a pattern. Well,

11:45

yeah,

11:46

he's in the news this week again, for

11:48

a letter that he sent to the European

11:50

commission's Henna Verkanen, who

11:52

is, the kind of tech, uh, Boss

11:55

the EU and took over from Thierry Breton

11:57

last year. This is one of a number

11:59

of letters that he sent EU over the last six months

12:01

or so, Mike, but it represents a bit

12:04

of a change intact talk

12:06

us through kind of what you read about

12:09

the letter.

12:10

I mean, you know, I don't know. Is it

12:12

a change in tech? I don't know. I mean, it's

12:15

like, basically is, you know, screaming

12:18

at the EU about the DSA and how

12:20

it is a censorship bill. And

12:22

I mean, if you've listened to me on this podcast

12:25

and on TechTurt for the last few years,

12:27

like there's some stuff in here that, you I

12:29

am concerned about the DSA and how it can be

12:31

used for censorship. And I've called out Terry Breton

12:33

and, and his attempts to use

12:35

it. And so some of that is, repeated,

12:38

but this is a fairly aggressive letter,

12:40

which is also in sort of Jim

12:43

Jordan style, but, you know, he

12:45

talks about how. The DSA requires

12:47

social media platforms to systematic

12:49

processes to remove misleading or

12:51

deceptive content, including so called disinformation,

12:54

even when such content is not illegal,

12:56

and says though nominally applicable

12:58

to only EU speech, the DSA

13:01

has written may limit or restrict Americans

13:03

constitutionally protected speech in the

13:05

United States. And so he

13:08

sort of, you know, Calling out these

13:10

things and basically, in a

13:12

very, accusatory way

13:15

saying like, if anything

13:17

DSA related leads to the suppression

13:20

of American speech, then

13:22

the U S Congress might take action

13:24

in some form.

13:26

Yeah. he calls for a briefing. Um,

13:28

as part of the letter, doesn't he? So he says to work

13:30

and, and, you know, we need a sit down

13:32

for you to talk about what the DSA means

13:35

for us. And we need it by the 14th of February.

13:38

which I think is, you know, romantic. Yeah.

13:41

Valentine's day, uh, meeting

13:43

Yeah. maybe

13:45

we can kiss and make up. Um, but

13:47

yeah, there's a kind of like, we need to be told what

13:49

the plan I

13:51

think that's for me where there's a slight change.

13:53

It's, you know, it's on the front foot. It's trying to

13:55

demand, seize control. And I

13:57

think in the way that prior to Trump

14:00

being elected, there was maybe slightly

14:02

less of, uh, of backing from

14:04

the

14:04

Yeah. and I actually do think that element

14:06

is really important, right? So, before,

14:08

even though in the house the Republicans

14:11

were the majority and therefore had power within the

14:13

house, that power was limited

14:15

to what the House had, right? They didn't have power

14:17

in the Senate and they certainly didn't have power at the executive

14:20

branch. Now, Republicans

14:22

have all three, and so the nature of this letter is sort

14:24

of reflecting that, which is that the whole

14:26

of the US government may actually listen

14:28

to Jim Jordan rather than him

14:30

just sort of making noise and being obnoxious

14:32

in the way that he is normally obnoxious. Now,

14:35

there was another thing which is not

14:37

in, the letter itself is interesting.

14:40

And we talked about that. There's this sort of political article

14:42

that alerted us to the letter. There's another

14:44

political article, which I actually think

14:46

might've been interesting if they'd combine the two,

14:49

which is saying, Musk

14:52

and Jim Jordan have had a mind meld

14:54

that the two of them are working very closely

14:57

together And so suddenly you start

14:59

to put this letter into a slightly different

15:01

context, which is Is Jim

15:03

Jordan doing this on behalf of?

15:06

on behalf of one U. S. citizenry? Or is he doing

15:08

this on behalf of one Elon Musk,

15:11

who is, as we have just discussed

15:13

now sort of running large elements

15:15

of the government while also running

15:18

a social media platform that is being

15:20

targeted by the E U and the D S A

15:23

and to Whose advantage is this?

15:25

and, you know, there are all sorts of issues about

15:27

conflicts of interest, but this article

15:29

talks about how Elon and Jim Jordan are,

15:32

buddy buddy and are constantly working

15:34

with each other now.

15:35

Interesting. I hadn't seen that. That's a really interesting,

15:38

actually saw it like literally as we started recording

15:40

and I was skimming it during the,

15:42

the intro here.

15:45

how fresh this podcast is. and

15:47

obviously, you know, last year we talked

15:49

at various points during the year about

15:52

Jim Jordan and Elon

15:54

Musk and GARM and, you know, the reports

15:56

that, his subcommittee wrote about

15:58

the, kind of awful effects of GARM and the kind

16:00

of, again, the industrial complex

16:02

that was surrounding the advertisers. coming

16:05

away from Twitter, taking away their spending from Twitter.

16:07

And so again, there are continuations

16:10

of a theme, but there was also a kind of emboldening,

16:12

I think of a message here, which is,

16:15

DSA, and the we're

16:17

not happy, we need to be told what's what. The

16:19

irony about all of that, Mike,

16:21

is that in the very same week,

16:24

Joel Kaplan, who is

16:26

the new global policy chief at

16:28

Meta and obviously a Republican,

16:31

in terms of his political leanings and the new

16:33

Nick Clegg was in Brussels,

16:35

essentially, defining how EU

16:38

speech should, work. You know,

16:41

he, he was, he was on a live stream.

16:43

In Brussels and was declaring

16:45

to a bunch of, Brussels wonks

16:47

about how the announcement that Meta made earlier

16:49

this year around, changing

16:52

the policies on the platform, getting rid of fact checking,

16:54

adding community notes was also going to

16:56

be rolled out in the EU. He predicts in 2026

17:00

and that he was ready to work with regulators

17:02

in the EU about how that would work. So. The

17:05

timing is very funny. You know,

17:07

you have Kaplan on behalf of Meta in

17:09

Brussels saying, this is how we're going to do things.

17:12

We expect this to be happening in the next 18 months.

17:14

And then we have also Jim Jordan, requesting

17:17

more information from the DSA because he

17:19

feels, the EU is, infringing

17:21

on, you know, the way that US companies work.

17:24

What did you make of the Kaplan stuff?

17:25

Yeah. I mean, it is a massive

17:27

step up in terms of the aggressiveness of

17:30

meta on the EU. Right. I mean, we've

17:32

had years now where meta

17:34

has been, I honestly think too

17:37

compliant with the EU on stuff. Right. I

17:39

mean, they've, there've been different things and pushback

17:41

and fines, but for the most part,

17:43

like meta has, been it. really

17:45

willing to kind of bend the knee

17:48

to the EU when the EU says we're going to regulate

17:50

social media these ways, the notable

17:52

thing is, media reporting

17:54

on it has been, it's always set up as like, Oh,

17:57

you know, the EU is cracking down on meta, but

17:59

meta has been sort of a happy participant,

18:02

and, really sort of embrace the

18:04

DSA approach to regulation because

18:07

They had indicated that they could handle

18:09

it, and they sort of recognized that smaller

18:11

competitors would have a more difficult time.

18:14

And in fact, you know, within the U. S. context,

18:16

MEDA had been really kind of going around

18:18

to both the federal government various

18:20

state houses and kind of hinting at the fact

18:22

like, look, you pass a law that's kind

18:25

of the same thing as the DSA, we're

18:27

not going to complain about it. and,

18:29

the unspoken part of that was

18:31

like, look, we're already complying with the

18:34

EU. We'll comply with the U S ones.

18:36

And we know that it'll cause trouble

18:38

for upstart competitors. And that's

18:40

the thing that Meta has been scared of, right? Because their

18:42

growth had sort of stalled out and

18:44

you saw the rise of things like Tik

18:46

TOK, that came in that. Really took medicine

18:49

by surprise. So their move

18:51

on the political front had been let's

18:53

embrace regulations because

18:55

that makes it harder for upstarts to compete.

18:58

And so this is a big change. And I think,

19:00

This also came up in the, Joe Rogan,

19:02

Mark Zuckerberg interview talking

19:04

about the EU and him and Zuckerberg

19:06

saying like, any kind of foreign regulation

19:09

is the equivalent of a tariff. And

19:11

we expect the U S government to

19:13

basically have our backs and, protect

19:15

and defend, domestic industry,

19:18

which is a very different way of looking at

19:20

it. you know, the tech industry has looked

19:22

at international trade, around

19:24

technology and internet issues forever,

19:27

right? Since the beginning of the internet industry.

19:29

And so looking at it as, as industrial

19:32

policy and being able to fight,

19:34

and so Joel Kaplan, then going to the

19:36

EU and basically saying this, he doesn't

19:38

do that if he doesn't think that

19:40

the whole of the US government under

19:42

Donald Trump has his back and.

19:45

we've had a couple of, like, weird

19:47

skirmishes around tariffs, uh,

19:49

in the U. S. in the last week as

19:51

well, that, that

19:52

I've been, I've been following.

19:53

Yeah, it's just, you know, again, sort of outside

19:55

the purview of this podcast, but like, you

19:58

know, among those things, Trump did

20:00

suggest that, he's also looking

20:02

at tariffs on the EU. And so, it's all

20:04

sort of part of this negotiation. And

20:06

Kaplan and Meta seem to think like, well, we

20:09

can go to the EU, make our demands, and

20:11

we'll have Donald Trump and Elon

20:13

Musk and Jim Jordan to effectively

20:15

back us up. And so, for all their talk

20:17

over the last four years about how perfectly fine.

20:20

And perfectly in compliance they were with the

20:22

DSA and the DMA and related laws

20:25

in the EU You I think they just see

20:27

this as look we have

20:29

this very stupid very brash

20:32

willing to smash and break things

20:34

person in the white house right now Let's

20:36

take advantage of that and see if we can

20:39

smash and break All these other

20:41

things in a way that just favors us and

20:43

this is how we're seeing it play out

20:45

Yeah. I mean, I can hear almost

20:47

the European listeners of control or speech

20:49

say, Hey, actually the EU

20:52

is 27 States and almost 500

20:54

million people. Naturally matter

20:56

is going to have to kind of negotiate in the way that

20:58

it has done in the past, you know? so I

21:00

think there's, that, but you're right. The, you know,

21:02

there has been a. Compliance there,

21:05

to use a term that kind of Daphne Keller has been using

21:07

increasingly, because there are competitive

21:10

benefits to doing so. and that seemingly

21:12

changed. the other thing that was interesting to me

21:14

was how it was a note in the, reporting

21:17

on Kaplan's comments about the code

21:19

of conduct, which we've talked about in

21:21

the last couple of weeks as well, you know, he

21:23

was referring to the AI act

21:26

of code conduct that's accompanying

21:28

official legislation that was.

21:31

past last year, and it came into play this

21:33

year. there's apparently a code of conduct that

21:35

sits alongside that, which is additional voluntary

21:38

obligations that the companies can sign

21:40

up to. He's basically said that we're not going to sign up

21:42

to those. And it made me think

21:44

about the fact that we have, the hate speech

21:46

code of conduct, the disinformation

21:48

code of conduct, one of which has already been

21:51

subsumed into the DSA, the other, which is

21:53

likely to be subsumed into the DSA, and

21:55

how, again, The relationships

21:57

are becoming much more tense, much

21:59

less about voluntary agreements and

22:02

signing up to stuff willingly and

22:04

the hammer of, of regulation being used instead.

22:07

Yeah. this is all uncharted territory

22:09

and in lots of ways. And, I

22:12

obviously, part of my role on this

22:14

podcast is to be the critic of

22:16

EU regulations, because

22:18

I, I have my concerns about EU regulations.

22:20

And so I'm in this weird position where I'm reading some of this

22:22

stuff and it's There's this kernel

22:25

of truth within the complaints that

22:27

the American companies have with regulations,

22:30

their complaints that I have made

22:32

about these same regulations. And

22:34

so there is this truth, but they

22:37

are not doing this. for

22:39

honest and good purposes, right?

22:41

This is just like a smash job,

22:44

right? Like we're just going to break these things because

22:46

we have the opportunity to, and

22:48

honestly, I'm torn because like,

22:50

would I like these laws to be better? Absolutely.

22:53

Do I have concerns about how these laws work?

22:55

100%. I've been calling them out since

22:57

the DSA was first being discussed. and

23:00

those, Concerns still stand there, but

23:02

I am perhaps equally,

23:04

maybe more concerned about what

23:06

happens if the EU caves to these

23:09

demands, because.

23:11

that will embolden them to go further

23:13

and do more and make things

23:15

even worse. And so again,

23:18

this is part of my, distress

23:20

at,

23:24

at this moment, which is like, there

23:26

are reasonable arguments behind

23:28

all this, but they're going about dealing

23:30

with it in the most destructive,

23:33

damaging, horrific ways possible

23:35

that will have long term. Potentially

23:38

extraordinarily negative impact.

23:41

and so, and you can take any of these

23:43

quotes or comments from Kaplan out

23:45

of context. And, and I might

23:47

agree with it and say, yeah,

23:49

some of these EU regulations are really problematic,

23:52

but the intention here is

23:54

basically to destroy any,

23:57

any sort of regulatory state, any

23:59

sort of administrative state. And

24:01

where that leads is extraordinarily

24:04

dangerous.

24:05

yeah. And we'll hopefully get a

24:07

readout next week of this briefing that

24:09

Jim Jordan has asked for, and we'll know a bit more then.

24:11

But I also am concerned about if

24:14

Virkin and then the EU, do cave.

24:17

it seems crazy that we're going to talk about meta

24:19

again. A friend of mine, Mike

24:21

was. At the event that we ran last

24:23

week in London. And she was saying that

24:26

right now feels like a kind of meta dead zone.

24:28

We've just been talking about it forever.

24:31

And in some ways, today's kind

24:33

of second big story is a continuation

24:35

of that theme. You and Renee talked

24:37

about it last week as well, but it's the,

24:39

reporting we've seen come out this week around,

24:42

advertisers and their feelings

24:44

towards the platform. In

24:46

the wake of Mark Zuckerberg's announcement, a

24:49

couple of new stories from

24:51

Digi day and from marketing week,

24:54

essentially saying that. Within the

24:56

kind of advertising ecosystem, there's

24:58

been really no discernible impact

25:00

on spending, by major

25:02

players. And yeah,

25:04

you touched on it last week. They're unlikely

25:07

to say so. because of the risk

25:09

of being targeted, but we also saw in the

25:11

earnings call, Meta's earning call last week,

25:13

that they have also seen no change

25:16

in advertiser spending in the short time since that

25:18

announcement was made. At the same

25:20

time, there's. interesting kind of sub thread

25:22

that is in a couple of those stories and

25:24

in a story from the Wall Street Journal around some

25:27

advertisers going back to X

25:29

slash Twitter. so the Wall Street Journal

25:32

report that Amazon has started to raise its

25:34

ad spending on X.

25:36

in recent months and a few

25:38

larger, brands who, decided that

25:40

X was kind of persona non grata for a

25:42

while considering once again, spending on

25:44

a platform. And I wondered what you thought

25:47

were the, maybe the kind of drivers of this, Mike, you know,

25:49

there's a cynical take, which I imagine you might give

25:51

us. Um, but where

25:55

cynical. No,

25:58

where do you see, these shifts happening

26:00

and when do you think we'll know fully whether, you know, the

26:02

question of Where the brand safety is still,

26:05

you know, alive and well.

26:06

look, I so much of this is just sort of

26:08

marketing, right? I mean, it's for show

26:11

brand safety is still a thing. Brands. Matter,

26:14

right? And if, something bad is happening with the

26:16

brand that harms the bottom line and every company

26:18

will react that way, the major difference.

26:20

And this is what we discussed last week with Renee is

26:22

that companies aren't going to talk about it.

26:24

And so it's not a surprise to me

26:26

that companies are taking a wait and see approach. I mean,

26:28

you and I talked about this, last month at

26:30

some point around, ROI

26:33

on meta ads is way better than anything

26:35

that they ever were on X. And so they're

26:37

important. as a part of strategy. And

26:39

so I think a lot of companies and

26:41

a lot of advertisers are concerned

26:44

about this, but it was one of these things where

26:46

it's like, well, let's, not make a big

26:48

show of this because that leads

26:51

to attacks from Jim Jordan,

26:53

and, Elon Musk and potential

26:55

lawsuits and all this stuff. So let's

26:57

take a wait and see approach. Let's see how this

26:59

actually plays out. We don't need to make a

27:01

big stand. The sort of public

27:03

sentiment at the moment feels like

27:06

they don't want us taking a stand on this. In the

27:08

past, when we did take a stand, it was because public

27:10

sentiment was in one direction and now it's

27:12

in another. So I sort of understand the kind of wait

27:14

and see. But if there is like actual brand

27:16

safety stuff, I think that

27:19

companies will focus on their bottom

27:21

line. And if it is damaging to their brand

27:23

to be advertising on a certain platform, they

27:25

will start to move away. Some of that will just be

27:27

general public sentiment. That is

27:29

a different story than the X. Story.

27:32

And this story of advertisers

27:34

potentially returning X. Now,

27:36

there have been similar stories

27:39

for over a year, every few months,

27:41

there's a major publication that writes a story

27:43

saying so and so advertisers are

27:45

returning to X. And it is touted often

27:47

by Elon himself and his fans is

27:49

like, Aha, see, everything's coming back

27:51

to normal. The details

27:53

are always less

27:55

than Enthusiastic about it. They're often

27:58

examples of them returning at

28:00

much lower volumes, 5

28:02

percent of what they used to spend 10 percent

28:04

of what they used to spend. And,

28:07

you know, obviously the overlying,

28:09

issue right now is the fact

28:11

that it is all just like to get

28:13

into the good graces of the

28:15

person who is currently running the U S government,

28:17

which is Elon Musk. And

28:19

so like. Jeff Bezos has done a

28:21

bunch of things with the Washington

28:24

post lately to signal that

28:26

they're going to be much friendlier to the Trump administration.

28:29

So another way to indicate that

28:31

and to give a public signal of

28:33

that is to say, Oh yeah, we're going

28:35

to bring Amazon back to advertising on X.

28:37

We don't know how much, we don't know how involved,

28:40

but sure. let's do it. And I think that

28:42

is what other companies are doing too. It's like, well,

28:45

right now, Elon Musk has so much power.

28:47

If we don't want to be in the crosshairs, if

28:49

we don't want to get sued by him, maybe

28:52

the easiest thing is to like take

28:54

some advertisements out. Now this is again, horrifying

28:56

in general, if you think about free speech

28:58

and the fact that, people are effectively being

29:01

coerced into giving money to the world's

29:03

richest man, who is also running the government

29:05

at the same time. There are all sorts of reasons

29:07

to be terrified by that. You know, when you put

29:09

it that way, it seems pretty bad, right? it's

29:12

problematic on a variety of

29:14

levels, but I think the reasons why advertisers

29:17

are doing what they're doing on meta

29:19

versus X are two

29:21

different things. You know, one of it is sort of

29:23

currying favor and the other is kind of,

29:25

uh, you know, this platform has been okay for

29:27

us. Let's wait and see if it actually turns

29:30

as bad as it does. But if there

29:32

are moments of actual attacks on

29:34

brands or brand safety problems, then

29:36

I think Companies will act the way that they

29:38

always act.

29:39

Yeah, talking of, lawsuits

29:41

and bringing claims against advertisers

29:44

only this week, Musk and

29:46

X brought a whole bunch of new

29:48

advertisers brands into, a

29:50

filing that they'd, uh, Submitted last year.

29:53

And so we've got new conglomerates,

29:55

corporates, Shell, Nestle, Palm

29:58

Olive, Lego, who've been added to

30:00

this as well. And again,

30:02

like there's, there's a kind of, I guess, willingness

30:05

to, roll out this playbook for

30:07

whatever brands that he sees fit, you to

30:09

kind of force them the platform,

30:12

it

30:12

Yeah, it was interesting. I mean, it's,

30:14

the same lawsuit as from last

30:16

year. The one that, that effectively brought down Garm.

30:19

it's just been expanded. It's this, their proposed

30:22

second amended complaint. and

30:24

so the court still has to accept it and add in these

30:26

new defendants. but that'll probably

30:28

happen. now remember this case

30:31

was originally filed with a very friendly

30:33

judge, Reed O'Connor, who was sort of willing

30:35

to bend over backwards, but he had recused

30:37

himself from the case because he held

30:39

stock in one of the advertising

30:42

defendants. So it is

30:44

now in front of a different

30:46

judge. Uh, who is not known

30:48

as being quite so partisan,

30:52

let's say. and so, the case may go,

30:54

you know, a little bit more seriously than other

30:56

cases, but we'll have to see, it would still roll

30:58

up to the fifth circuit, which is crazy

31:00

as it gets there, it was interesting to me that among

31:03

the companies that were added here were.

31:05

Twitch, which is owned

31:07

by Amazon. We were just talking about Amazon

31:09

and Jeff Bezos and caving. So it'll

31:12

be interesting to see how that plays out

31:14

if Twitch stays in the lawsuit,

31:16

but also Pinterest. And so to

31:18

me, both Twitch and Pinterest. are

31:21

in some ways competitive to x

31:23

and the idea that they should be forced to then advertise

31:26

on x Seems really

31:28

problematic from a whole

31:31

wide variety of things And

31:33

in fact I was sort of confused

31:35

by their inclusion in the lawsuit. And even

31:37

the way it was written, like, they're talking about how

31:39

much they normally advertise and how much

31:41

they spend on promoting themselves, but

31:44

is that always on a competing platform?

31:46

I don't, I don't know. The whole thing just felt a

31:48

little weird. Like maybe this was like a backdoor

31:50

way to attack. competitors for social

31:53

media attention. and so it'll

31:55

be interesting. It is still just, it

31:57

is a ridiculous lawsuit

31:59

in so many ways. I, I need to stress that because

32:01

I feel like we sort of skipped over that part. I know we

32:03

talked about it last year, the

32:05

idea that, Choosing not to advertise

32:07

on a platform represents an illegal

32:10

boycott is absolute nonsense.

32:12

There are Supreme Court rulings on

32:14

record about how boycotts around

32:16

things like this are speech. They

32:18

are a form of speech. They are protected by the First

32:20

Amendment. The kinds of boycotts that are

32:22

illegal are ones that are done for anti

32:25

competitive purposes, which maybe that's

32:27

why he's adding Twitch and Pinterest to it

32:29

and going to try and claim that there's like an anti

32:31

competitive element to it. But that

32:33

is. just obviously ridiculous

32:35

and silly. and so, we'll

32:38

have to see. I'm hopeful that before

32:40

a more reasonable judge that, quick

32:42

work is made of this case and that it gets dismissed

32:44

quickly. But then of course it'll be appealed to the fifth

32:46

circuit where, anything goes.

32:49

yeah. Okay. So yeah,

32:51

we, we, we'll see how this pans out. I mean,

32:53

I, for one might wonder to

32:56

what extent we're going to see a big brand

32:58

safety. Event like

33:00

we did in the mid 2010s,

33:03

that's what I'm, I'm wondering is, is in the

33:05

kind of near future, if you remember, like

33:07

you would be very privy to this, but the, big

33:09

YouTube story, 2015, I

33:12

think it was 2016, 2017

33:14

was, terrorist videos, ISIS

33:16

videos being uploaded onto,

33:19

YouTube and other platforms and

33:21

having large Brands

33:23

advertised next to them in a way that was completely

33:26

unsuitable and led

33:28

to massive changes to the way that the platforms

33:31

vetted advertising and tried to provide

33:33

a kind of brand safe environment, it feels

33:35

like we're going down a similar route. It

33:37

feels like there's going to be a similar kind of event

33:40

like that, but it does rely, as you mentioned

33:42

on. Media reporting it in a way

33:44

that is kind of nuanced and recognizes

33:47

that, you know, this is happening. I

33:50

wonder it and worry actually that it's

33:52

going to get lost in the noise of everything else.

33:55

that this is no longer something that, we

33:57

can expect from platforms that if you want

33:59

to advertise in a way that is. allows

34:02

your brand to sit next to content that is,

34:04

you know, not egregious and not harmful

34:06

and not, you know, offensive to users, we've

34:09

kind of lost sight that that's even possible. so

34:12

yeah, it goes back to your point earlier about how

34:14

the reporting of the likes of 404

34:16

and, and other tech sites who do

34:18

really do good work here is going to be so key over

34:20

the next few months, I think.

34:22

Yeah. I mean, we'll have to see, right. And, and,

34:24

the whole thing with like advertising and brand

34:26

safety is that it, you know, at

34:29

times it did get overheated and, and

34:31

right. Like, I think most users

34:34

will often recognize that, Advertising

34:36

that appears next to any particular piece of content. There's

34:39

an algorithmic component to it. It's not

34:41

like, AT& T is choosing

34:43

to advertise next to

34:45

pro Nazi content or, or whatever.

34:48

And so like the actual impact brands.

34:50

maybe overstated, but

34:53

there is this general sense of like, if an

34:55

entire platform is

34:57

supporting the fascist takeover of the United

34:59

States, do you want to be helping to fund

35:02

that? and that is the type of thing

35:04

that could boomerang. And, you know, I remember

35:06

going way back into like early two

35:09

thousands where there was this big freak out

35:11

of a. programmatically

35:13

delivered advertisement for

35:16

a set of steak knives or something

35:18

showing up, I think it was on the New York post,

35:20

next to an article about a stabbing

35:23

right.

35:23

was like the beginning of the list. Like, Oh, when

35:26

we have these algorithmically driven advertisements,

35:29

They might show up in a way that it feels

35:31

inappropriate. And then the media can mock

35:33

it. And then the companies get all worried. Cause they're like,

35:35

we don't want to be responding to media

35:37

requests about how come our knife

35:39

ad is showing up next to a stabbing. So

35:42

those kinds of things will happen. It will depend

35:44

on the media, but the media has a lot of other

35:46

things that they may be focusing on that are probably

35:48

more pressing than, whose ads

35:50

are showing up where. And so we'll see

35:53

how that plays out. and

35:55

sort of where things go. I will note, this is

35:57

a very, very small sample size and

35:59

probably not indicative of anything, but,

36:02

my office, right down the hall from

36:04

my office, sort of across the way from my office,

36:07

there had been a brand safety company. and

36:09

I noticed recently that it's gone. So

36:11

maybe, maybe

36:15

they may have moved. I have no idea. I

36:17

they've tripped. They've trebled in size.

36:19

perhaps, perhaps I just noticed that they were no

36:21

longer, on my floor in my office

36:24

building.

36:24

Right. All right. Yeah. I mean, this

36:27

will be an ongoing conversation we have, I think. And

36:29

there are big brand safety summits coming up in

36:31

the next couple of months as well. So, you know, I would

36:33

love to hear from listeners if they

36:35

are attending those or they're, they're in the industry

36:38

and want to share their thoughts. Cause, it's only a bit of

36:40

a kind of knife edge as to where it goes.

36:42

yeah, and I'm sort of curious, like, you know, the way

36:44

that, Musk and Jordan have sort of attacked the whole

36:46

concept of brand safety, you know,

36:48

how that conference is going to be like, how big

36:50

is that conference or are people going to be afraid

36:52

to go to that conference and attend? Because

36:55

it feels like, you know, similar to the way

36:57

that the whole concept of trust and safety has

36:59

been sort of negatively attacked. so

37:01

too, will the space of brand safety. And that,

37:04

that seems like a concern.

37:05

Yeah. Okay. let's move on to a story that has

37:07

been kind of bubbling along in the background the last few

37:09

weeks, but we haven't necessarily talked about

37:11

actively on the pod. this is the story

37:14

of deep seek, which many of our listeners will

37:16

have heard about this, new. Model

37:18

that's been trained at a fraction of the cost of some of the

37:20

other big general LLMs. and

37:23

what we're seeing this week is a couple of parts to

37:26

the deep seek story, Mike, but you picked out

37:28

a story from 404, about

37:30

a possible lawsuit, being

37:32

filed for people who download deep seek.

37:35

Which would include you, I think.

37:37

Well, that depends. Depends on how you define

37:39

all of this. And so it's not it's not a

37:41

lawsuit. Just to clarify here.

37:43

Josh Hawley, Senator Josh Hawley, who's problematic

37:46

in all sorts of ways as well, introduced

37:48

a law, right? That that would effectively

37:51

extend, a ban on

37:53

import export. To cover

37:55

Chinese related AI, which would include

37:58

deep seek code. Now, then the question

38:00

is, what does that actually cover? And

38:02

that becomes a lot more complex because what is actually

38:05

deep seek, right? You know, what everybody is talking about

38:08

with deep seek is this model, but the model

38:10

is open source. but also

38:12

DeepSeek itself as a company

38:14

offers a hosted model, just like chat GPT

38:17

or, Google Gemini or whatever that

38:19

you can go log into. But because the code

38:21

is downloadable and you can run it, there

38:23

are lots of people who are running it. on

38:25

their own laptops, if you have a powerful

38:28

enough laptop, which is actually not that powerful,

38:30

or there are lots of people who are hosting versions

38:33

of the deep sea code. They downloaded it and

38:35

then created their own hosted version of

38:37

the same model, or perhaps a slightly adjusted

38:39

model. You can take it because it's open sourced. You can adjust

38:42

some of the weights and, mess around with it. I

38:44

know Microsoft recently announced that they're

38:46

launching a version of it. I

38:49

have played around with two different us

38:51

hosted versions of it. rather

38:53

than the Chinese hosted version. So

38:55

it's unclear to me, Under

38:57

this law, if it passes. And again, it's

39:00

just been introduced the likelihood of it actually going

39:02

anywhere. I don't know, but it would criminalize

39:05

the downloading of designated

39:08

Chinese AI. so you could

39:10

go to jail in theory for

39:12

using open source code or even for downloading

39:15

open source code, which is very

39:17

problematic.

39:18

yeah. 20 years you could be

39:20

imprisoned for, or fined

39:22

not more than a million dollars, which is,

39:25

which is nice to cap it at a million.

39:27

And, speaking of 20

39:29

years, it has been, well, it's actually been more than

39:31

20 years since we fought

39:33

some of these battles around,

39:36

import export controls and software

39:38

code over encryption.

39:40

So in the 1990s, there was

39:42

encryption software and the U S under

39:44

the Clinton administration tried to ban

39:46

the export of, high grade encryption

39:49

saying it was a form of a weapon.

39:51

and. effectively tried to block

39:53

it. And the court said no.

39:56

and basically gave us one of the first rulings

39:58

that indicates that software code is

40:00

speech. And under the first amendment,

40:02

you can't limit it in this way. And

40:05

so I think that kind

40:07

of fight and those kinds of rulings come

40:10

back into question if this passes.

40:12

But now at the same time,

40:14

we're living in the wake of the ruling on TikTok,

40:17

which we have talked about, where suddenly

40:19

the Supreme court sort of, uh, The

40:21

first amendment doesn't matter as much when we're talking

40:24

about China, so how does this

40:26

play out? If this goes into law,

40:28

does the Supreme Court say it's okay?

40:30

And then what does that mean? Because it's been

40:32

downloaded so many times and

40:34

Microsoft is offering it and other services,

40:37

multiple other services are offering a version

40:39

of DeepSeek.

40:40

mm

40:40

and then on top of all that, right. you

40:42

know, we didn't talk about like deep seek

40:45

itself in terms of how it works.

40:47

I mean, you talked about how, was. trained

40:49

in a way that was much cheaper, which,

40:52

you know, it's being referred to by lots

40:54

of people sort of like a Sputnik moment,

40:56

this sort of realization, that might

40:58

be a bit of an exaggeration, but

41:00

it is a wake up call for the

41:02

belief that, so called frontier

41:05

models know, the, the

41:07

big models, which know, uh,

41:09

open AI, anthropic, Google Facebook's

41:12

llama that only those companies

41:14

can afford to actually do the training

41:16

which cost hundreds of millions of dollars.

41:19

Whereas, deep seeks basically

41:21

saying we could do this with cheaper GPUs

41:24

because the US already bans export

41:26

of the powerful GPUs. And,

41:28

you know, we can make this model a lot simpler and

41:30

a lot lighter. Now I'll note just as an aside,

41:33

because it'll come up no matter what

41:35

among commenters, OpenAI

41:37

effectively claims that part of DeepSeek's

41:39

secret sauce is that

41:42

they distilled an OpenAI model.

41:44

Not going to get into what distillation is, but it

41:46

is an important part of DeepSeek. element of sort of how

41:49

this model was chained, trained.

41:51

And we know that they use distillation. They admitted

41:54

that the question is, did they do the distillation

41:56

on, open AI's model? And then

41:58

what does that mean? Does that violate a contract? All

42:01

those things, like we're going to leave that aside, but

42:03

the model itself is very lightweight. Very

42:05

cheap to train and then incredibly

42:08

powerful in terms of, what it does.

42:10

And as, as I mentioned, I've been playing around with it

42:12

and it's a really good AI

42:14

model. it's very impressive. It

42:17

has limitations. the default

42:19

model has somewhat famously

42:21

been directly coded in a very,

42:24

very obvious, clunky and

42:26

awkward manner to not

42:28

talk about things like Tiananmen

42:31

hmm.

42:32

Taiwan sovereignty. Um,

42:35

and you know, some of the people who have downloaded

42:38

versions and run them locally

42:40

have figured out how to extract

42:42

those restrictions from it

42:44

and, get the code to actually comment

42:46

on reality properly. But,

42:48

there are some concerns there. And then obviously

42:51

also there are some other concerns.

42:53

And I think you had found this, report

42:55

on a study on DeepSeek. Did you want

42:57

to talk about that

42:58

yeah, I think there's been a, wide conversation about

43:00

how safe is this model, when

43:03

actually, you know, it's been trained,

43:05

like you say, pretty quickly and, and so cheaply

43:07

compared to other, frontier models. And there's a,

43:10

a piece of analysis that was put out by

43:12

a company called Encrypt ai,

43:15

which calls itself, you know, like they all do the

43:17

a leading security and compliance platform

43:19

and.

43:20

leading. We're the leading podcast,

43:22

Exactly. Exactly. Um, they should come and

43:24

sponsor an episode of control or speech. Um,

43:27

but did a kind of red teaming analysis

43:29

and found that the model was

43:31

much more likely to generate harmful content

43:34

than, open AI is Oh one

43:36

model and. threw

43:38

out basically lots of bias and

43:41

discrimination. there was cybersecurity

43:43

risk. It was spitting out kind of malicious code

43:45

and viruses much more readily

43:48

than some of the other models that it has tested

43:50

in the past and was much

43:52

more likely to, produce responses

43:54

that were toxic, that contain profanity,

43:57

hate speech, and kind of extremist

43:59

narratives. so again, you know, the trade

44:01

offs between producing lightweight models

44:04

that are very kind of agile

44:06

and can do things very quickly. quickly and

44:09

safety side, which obviously some of

44:11

the big frontier models have had to focus on because

44:13

of the companies that they're coming out of is

44:16

an interesting dynamic that we're, we're seeing,

44:18

and I think we'll continue to see play out. and

44:20

it's worth going to have a read.

44:22

Yeah. And I think like The fact that

44:24

the model, it doesn't have as many

44:26

safety things built in. Didn't strike

44:28

me as surprising. I mean, I sort of expected that,

44:30

right? I mean, if it's a cheaper model, the

44:32

company is, they're not commercializing

44:35

deep seek right now in any way. And in fact,

44:37

that's raised a bunch of questions about what is the real purpose

44:40

of this? the company CEO has

44:42

sort of come out and sort of expressed like,

44:45

traditional, open source.

44:47

helping the world. We're just want to make this

44:49

available to everybody. People question

44:52

how, you know, truthful that

44:54

is there's the argument because it's attached

44:56

to a hedge fund that, maybe

44:58

they were doing this to short

45:00

sell a bunch of

45:03

American companies that may have

45:05

lost a lot of money when deep seek suddenly

45:07

became popular, there's all sorts of conspiracy

45:09

theories, who knows what the reality

45:11

is. But they have less incentive

45:13

to actually spend the time on

45:15

building safety aspects of it. And

45:18

I think the important thing to think about there is that

45:20

that is the future, right? We're going

45:22

to see more lightweight. But

45:25

very effective models and

45:27

there is going to be less incentive for them to have

45:29

the safety training and so You know the thing

45:31

that I would say folks who are listening to this who are interested

45:33

in safety stuff specifically

45:36

is that relying on the

45:38

model makers to build in safety

45:40

is probably not going

45:43

to work and

45:45

that we have to start thinking about safety at different levels

45:47

of the stack in terms of who's actually

45:50

implementing and using these models. As I mentioned, like

45:52

the deep sea code Microsoft is, hosting

45:54

a version of it and others are as well, we

45:56

might need to start looking at who is hosting these

45:59

models and who's providing access to the models

46:01

rather than the model makers themselves for

46:04

safety stuff. But on top of that, we

46:06

also need to get more.

46:08

literacy among users, right? I mean,

46:11

this is where it always comes back to, which is, you

46:13

know, at the end of the day, the users

46:15

of these products need to understand the safety

46:17

implications of what they're doing. And,

46:20

we may be over correct when we expect

46:22

other intermediaries to, stand in

46:24

the way and, be the safety cops.

46:27

Yeah. I mean, how does open source. play

46:29

into that, because so many of the, the models are open

46:32

source metas, Llamas, open source, deep

46:34

seek, you know, you kind of suggesting that

46:36

actually part of the problem is that when

46:39

somebody kind of takes the code and, and,

46:41

you know, host it themselves, that they don't have

46:43

the necessary awareness of the,

46:46

of the safety challenges, or like how does

46:48

the open source element play out against the safety

46:50

element?

46:51

that's a trickier question to answer than

46:54

you would like, right? So like open source in the AI

46:56

world is also slightly different than open source

46:59

elsewhere, right? So open

47:01

source AI models,

47:03

we still don't know the training data.

47:05

You don't, it's not like, You know, open

47:08

source code in most contexts, as you get access

47:10

to all of the underlying code, with

47:12

the AI models, it's really, it's the weights

47:14

that you're getting access to. And so you can implement

47:17

it yourself, but you still don't know the underlying, how

47:19

it was trained. And so like what biases

47:21

were built in, you still have to discover

47:23

that separately. And then you can adjust and

47:25

change some of the weights yourself. maybe

47:28

deal with the safety aspect where you can add an overlay

47:30

or filters or other things along those lines

47:33

to deal with the safety questions related

47:35

to it. but you're dealing with

47:37

multiple different things, right? So,

47:40

the underlying training is still.

47:42

a secret, right? This is why came

47:45

out and accused DeepSeq of basically

47:47

building off of their own work. and

47:49

so, again, it's just a question of like, where

47:52

and how do you insert safety

47:54

related features? And it could be in multiple places

47:56

and, how that's done, we'll see. Whereas

47:59

like, really at this point,

48:01

you have llama and, and,

48:04

meta who are trying to build in safety, or

48:06

at least were, you know, maybe now

48:08

with the new super enlightened,

48:11

Mark Zuckerberg, they will stop those, those

48:14

practices. I mean, you had other

48:16

open source models. There was mixed role, which

48:18

came out of France.

48:20

Mm hmm.

48:20

though they've moved from open source to a more

48:22

closed source approach as well, you

48:25

know, so You know, this is all sort of

48:27

in flux right now, but we'll

48:29

see. But just the fact that DeepSeq was able to do what

48:31

they did means it's almost

48:34

certain that we're going to see more open source models

48:36

that are released, that are powerful,

48:39

that are way cheaper than the

48:41

existing costs of the frontier models, and

48:43

are going to lead to. some changes

48:46

to how the market is viewed. And

48:48

there will be tremendous temptation to

48:50

make use of these models if they're much

48:52

cheaper to use on a, just use basis.

48:55

so, where the safety aspects

48:57

come in is, going to be a big deal and something people

48:59

need to think about.

49:00

Yeah. I'm waiting for the story. The, you

49:02

know, the first story where deep seekers used for some

49:04

sort of, giant, don't know,

49:06

attack on the U S federal government. Um,

49:09

it's

49:09

something like that will happen. Absolutely.

49:11

Yeah. Okay, Mike, what we've got

49:13

probably time to mention one more story very briefly,

49:16

which is one that I, stumbled

49:18

across, which takes us back into

49:20

Europe and to the regulatory regime

49:22

there. This is a story, in which

49:24

Apple is very kindly warning

49:27

us about the, perils

49:29

of having other app stores on

49:32

its app store. so there's

49:34

an app store called Alt Store

49:36

Pal. Which is actually an open

49:38

source, app, and it's designed

49:40

for kind of sideloading other apps onto,

49:43

the iPhone. It gave a story

49:46

to Bloomberg this week in which it said it was worried

49:48

about the fact that out store

49:50

has, got a porn app. so

49:52

users can now use out store to download

49:55

a porn app, which aggregates lots of pornography

49:58

from various different sites, and,

50:00

uh, you It's kind of suggesting actually, this is

50:02

a problem for context

50:05

last year. And over the last few years, Apple's

50:07

come under fire from the digital under the digital

50:09

markets act, which is suggesting

50:11

that actually Apple has a dominance in terms of

50:14

the marketplace for apps. And

50:16

in only last June did an investigation

50:18

into Apple's dominance of

50:21

the place. And so this

50:23

is essentially kind of Apple trying to push

50:25

back against that a little bit. It's

50:27

a slightly cynical attempt to make everyone worried

50:29

about the DMA. And it suggests

50:31

that actually, you know, users are going

50:33

to become less safe. If it's

50:36

allowed to do, these kinds of sideloading

50:39

apps via the app store. So again,

50:41

little bit naughty from Apple, suggesting

50:43

that this is going to be, it's going to open up a whole

50:45

range of harms to users in the EU.

50:48

What do you think about it, Mike? Very

50:49

Uh, it's, it's nonsense. I mean, I,

50:52

I, I, I have, I have all sorts of criticisms

50:55

about the EU regulatory regime, as I've talked

50:57

about. this is not it. This is, you know,

50:59

I think Apple's is, this is

51:01

the fainting couch, Oh my goodness. Like

51:03

somebody could install a porn app. You know what? You

51:05

have a browser on your phone. You can

51:07

access all sorts of pornography

51:09

if you want. Right. this is not

51:12

the concern. The fact that you have alternative app

51:14

stores, like, I don't know why Apple's making a big deal

51:16

of this. Like Android on Google Android,

51:18

you can have alternative app stores. They have existed

51:20

for a long time. All of these fears

51:23

have never been proven to be true. Most people

51:25

don't use them. Most people are not going

51:27

to alternative app stores to download dangerous

51:29

apps. There are all sorts of reasons to complain

51:31

about the EU. Maybe this ties

51:33

back to the story where like Apple sees

51:35

this as an opportunity to just

51:37

try and smash EU regulatory

51:39

regime right now, but this is not it,

51:41

it seems silly, it seems like, Oh

51:43

my, Oh, you know, very, very prudish.

51:46

Like, Oh my gosh, someone might be able to get

51:48

porn on an iPhone. Oh, come

51:50

on. Like, uh, this,

51:53

I don't know. I don't understand why Apple's doing

51:55

this. I mean, I think everybody should be able to see

51:57

through this as it's just a nonsense

51:59

complaint.

52:00

Yeah. Yeah. It's big companies pulling up the moat

52:02

again, isn't it? which is something

52:04

we've talked about, So are kind of

52:06

four stories for our listeners

52:08

this week. we've tried to kind of distill everything

52:11

that we've heard. and you've written about

52:13

this week, Mike, it's been a hell of a, of

52:15

a pod this week. thanks to trying to unpack

52:17

everything for us and sorry, you've had

52:19

such a kind of shitty

52:21

week, basically. listeners, thank

52:24

you for taking the time to, listen this week

52:26

and for staying with us all the time.

52:28

we're going to be back next week, but we're going to be recording

52:31

on Friday rather than our usual, our new

52:33

Thursday slot. So tune in

52:35

then. And, if you enjoyed this week's

52:37

episode and all of the episodes that we put out, do

52:40

rate and review us on your favorite podcast

52:42

platform really helps us get discovered. thanks

52:44

for joining us this week. Take care. Bye.

52:49

Thanks for listening to Ctrl-Alt-Speech.

52:52

Subscribe now to get our weekly episodes

52:54

as soon as they're released. If your

52:56

company or organization is interested in sponsoring

52:58

the podcast, contact us by visiting

53:01

ctrlaltspeech.Com. That's

53:03

C T R L Alt Speech. com.

53:06

This podcast is produced with financial support

53:09

from the Future of Online Trust and Safety Fund,

53:11

a fiscally sponsored multi donor fund

53:13

at Global Impact that supports charitable

53:15

activities to build a more robust, capable,

53:17

and inclusive trust and safety ecosystem.