0:00

ABC Listen, podcasts,

0:02

radio, news,

0:04

music and more. What

0:10

do Sam Kerr, Snoop

0:13

Dogg and Kendall Jenner all have

0:15

in common? Well, this week on Download

0:17

This Show, they've all lent their personalities

0:19

to Meta's new range of AI

0:21

chatbots. So exactly how

0:23

will these chatbots with personality work?

0:26

Also on the show, ChatGPT, the

0:29

online AI service can now browse

0:31

the internet in real time. Plus,

0:34

the social media platform TikTok has launched

0:36

its own e-commerce wing, TikTok

0:38

Shop, but at least one

0:40

government in the world has moved to block it. But

0:42

why? All of that and much more coming up.

0:45

This is your guide to the week in media, technology

0:47

and culture. My name is Mark Finnell and

0:50

welcome to Download This Show.

1:01

Yes,

1:01

indeed, it is a brand new episode of Download

1:04

This Show and a very big

1:06

welcome to our guests this week. The

1:08

co-founder

1:08

of PatientNotes, Sarah

1:10

Moran. It's been far too long. Far

1:12

too long, indeed. So happy to be here. And

1:15

Jessica

1:15

Sire, technology reporter with the AFR,

1:17

the Australian Financial Review. Hi,

1:19

how are you? It's lovely to have you back. Good

1:21

to be here. All right. Who's

1:24

ready to meet Meta's new AI

1:26

chatbots? Meta, who of course, the company

1:28

behind Facebook and Instagram, they've unveiled

1:30

some chatbots, but

1:31

chatbots with some slightly famous personalities. Walk

1:34

me through this. They're rolling out this suite of

1:36

products where you can chat with

1:38

a celebrity sounding

1:41

bot. So you can chat with

1:44

Kendall Jenner. I don't

1:45

necessarily know what I'd ask Kendall Jenner, but

1:48

they've sort of programmed these bots so that they can

1:51

mimic the voice or the tone of somebody

1:53

else and you can

1:55

just sit and chat with them. Is Kendall Jenner the

1:57

one who couldn't chop the...

2:00

the vegetable, I would ask for salad recipes.

2:02

Okay, Jenna just put it out there. Yeah, she might know.

2:05

Sarah, why is this a thing? Well,

2:08

they're leaning into the AI like everyone

2:10

else, but their main chatbot is called

2:12

Meta AI. And so it allows

2:14

users to, you know, ask questions

2:16

as they're doing things like a little bit of a co-pilot.

2:19

And yeah, it'll help them apparently

2:22

with all the subjects like cooking advice

2:24

and questions, but also

2:27

for entertainment. So I imagine that the celebrity

2:29

piece is about making it entertaining,

2:31

making it more human. So you reckon the celebrity

2:34

thing is about kind of taking the creepy

2:36

out of it, Sarah?

2:37

Yes, I have strong thoughts about this.

2:39

And this is why

2:41

it's fun to have this on the show.

2:43

Mark Zuckerberg really needs

2:45

help making things not creepy and not robotic.

2:48

And so he's like, I have lots of cash. I

2:50

will call my human friends and

2:52

invite them along to the party. And so I really think

2:54

that's what this is about. If he's the

2:57

face of chatbots, I think we're all

2:58

in a bit of trouble. I love that you think he's

3:00

friends with Kendall Jenner. That's

3:04

something that if AI delivered to me, I'd be like,

3:06

no, thank you. How much should we reckon

3:08

the celebs are making out of this? Oh, heaps.

3:11

It has to be heaps. I guess to

3:13

some degree, it's been an issue coming for a long time. It's

3:16

also been an issue in and around

3:18

the recent writers and actors strikes that

3:20

have happened in Hollywood, which is about what happens

3:22

with your rights and movie studios and

3:24

game developers and whatnot. Often one, you

3:26

know, likeness rights for imp perpetuity. I'm kind

3:29

of curious about what people are negotiating

3:32

here. What's in it for, you know,

3:34

there's a footballer Tom Brady, the Snoop Dogg,

3:36

like what's in it for them to us?

3:38

I suppose there is like a brand

3:40

awareness thing for a new audience. Like

3:43

these celebrities have their pockets of

3:45

people who know who they are. So perhaps

3:48

that's part of the strategy. I mean, it feels

3:50

really like

3:51

I'd love to see one of these contracts.

3:54

I'd love to see like what data

3:56

is being fed into the model so that it sounds

3:59

like Snoop Dogg.

3:59

Does he have control

4:02

over what that information is or

4:04

is it like stuff that exists in the public

4:06

domain? And I'd love to

4:08

see how long these bots can sort

4:11

of exist because it's an identity thing, right? Like you're

4:13

giving away the rights or rather your licensing

4:15

parts of your identity. Once upon a time we

4:18

would have had dolls that mimicked people

4:20

that said

4:21

Woody from Toy Story

4:22

is like reach for the skies and if he was a real

4:24

man then like his likeness is put

4:27

into a doll whereas now we have these bots that

4:29

are like self-learning and can create I

4:33

don't know. Maybe the identity changes over

4:34

time. That's a really interesting point like this

4:36

technology is it's generative by

4:38

its nature which means that it's not just a bunch of recorded

4:41

messages from a famous person it will say something

4:43

that this person has not yet said and

4:45

what kind of strictures would you put in place?

4:48

Like the bot can say this

4:51

this this and this it can never say this. There would

4:53

absolutely be clauses like that wouldn't this era?

4:56

Oh a hundred percent. I mean we

4:58

see that when people are licensing stock images

5:01

or you know like that they can never be placed

5:03

in certain places or if you sign an advertising

5:05

agreement that ads won't be placed

5:08

in certain places because you can't

5:10

monetize it in that way. I think

5:12

what's really fascinating about this is that

5:15

from a business perspective it's really smart for

5:17

these people to be first to market with their

5:19

own bots because those bots

5:21

are going to exist anyway. Like we see

5:24

that famous people are often

5:26

imitated or there are you

5:28

know parody

5:31

accounts that spring up that mimic

5:33

these people anyway and to Jess's

5:35

point there is so much content available in

5:37

the public domain that could be used

5:39

to train AI to represent these people anyway.

5:42

Like why not do it yourself and make money out of

5:44

it? I think it's incredibly savvy

5:47

to have that ownership and then you can build those

5:49

clauses in. So being at the table allows

5:51

you to say don't use this for

5:53

pornography. Don't use this in a way that's going

5:56

to speak badly or damage my brand. Incredibly

5:59

smart to be.

5:59

be first in with this one. Do

6:02

you think

6:03

this is one of those sort of gateway drug things

6:05

where people will start using it and then eventually

6:07

become more comfortable? Because that would appear to

6:09

be the underlying philosophy, Jess.

6:11

Yeah, I was just thinking then I would

6:13

like to set up Snoop

6:15

Dogg so I can chat with him all

6:18

day and then over time just develop this

6:20

kind of like online relationship with my little

6:22

Snoop bot. And

6:24

it's strange, like I would much prefer to do that than

6:26

necessarily train the co-pilot

6:29

helper which I can see is like the strategy

6:31

of most of these major tech companies is to develop

6:33

the virtual assistant. It's

6:36

funny, it's like establishing a relationship

6:38

with a pseudo person. It feels kind of metaverse-y.

6:40

It feels very like developing

6:42

online avatars and a real

6:44

crossover between my like existing

6:46

identity and identities that exist in these

6:49

online

6:49

worlds. Well it's also a bit of a shortcut,

6:51

right? Because people already have this, you

6:53

know, like you see this when a famous person dies,

6:55

right? Like a parasocial relationship

6:58

you have where it's like you're sort of halfway there

7:00

with that relationship. And I wonder

7:02

if that aids somewhat,

7:04

Sarah, in kind of building the relationship

7:07

you may or may not want with

7:09

an AI.

7:10

Yeah, and I think for those of you who don't know

7:12

what a parasocial relationship is, it's this

7:15

idea of a one-sided relationship that you

7:17

have with celebrity or with

7:19

media. And I think we always look

7:21

at it as being a bad thing, you know. It's like,

7:24

oh, you know, this one person doesn't

7:26

even know that I exist and I am their biggest

7:28

fan. But you can also,

7:31

this seems to be leveraging that to

7:33

the positive. So with

7:35

Girl Geek Academy, which is one

7:37

of the things that I work on, we actually started doing

7:39

this. So we worked with Microsoft to start

7:42

training an AI

7:44

to scale Girl Geek Academy. It's like how

7:46

do you take our voice and

7:48

how do you take the way that we present

7:50

material and put that as a layer

7:53

over the things that we're trying

7:55

to teach? So if someone wants to learn to code the

7:57

way that someone might teach it might sound a certain way. way

8:00

that Sarah teaches it sounds

8:02

completely different. And so if you're in

8:04

class and you need to learn maths, heck

8:07

yes, sign me up to hear how Kendall

8:09

Jenner is going to run my maths class. Like I would

8:11

love to learn that way, not like

8:14

maybe, but I might

8:16

be a bit of girl maths thrown in there, but

8:18

I think that that's a really important thing

8:21

to think about. Well, what are the positive ways that we could

8:23

use this to relate with people and allow

8:25

them to access information?

8:27

How did curiosity Sarah, how

8:29

did the AI go? Like what was

8:31

the end result of that process?

8:32

I mean, all of these tools are so

8:34

new, like they're bringing out new tools

8:37

every week. And so I guess the hardest part

8:39

is that when you build an iteration that

8:41

it's outdated so quickly. And so it kind

8:44

of, it makes it hard to invest

8:46

because you think you're investing in something that's going to be

8:48

so flashy, so new. And then two weeks

8:50

later people go, oh, you're dirt, that's like so old.

8:53

So when you're dealing with young people. You're doing Kendall Jenner then,

8:55

because that felt like perilously close to

8:57

Kendall Jenner.

8:59

Yeah, look, one of the great developments

9:01

about this stuff is that it's open source

9:04

as well. Like in doing what Sarah and the guys

9:06

at Girl Geek have been doing, like experimenting

9:08

with the technology for their own ends.

9:11

Like, I mean, sure, maybe in a couple of weeks it might

9:13

feel like it's outdated, but this is the most exciting

9:16

part is that you can experiment. I mean, you

9:18

guys have done it through a partnership with Microsoft,

9:20

but I mean, this meta stuff is all

9:22

open source. So people can start to build

9:25

their own bots and use a lot of the

9:27

training data that met is put together,

9:29

which I think is just

9:30

very cool. Stick around, because a little

9:32

bit later in the show we're going to be talking about how to spot

9:35

AI images. There's been a very interesting

9:38

use of it in the video the last couple of weeks. We'll talk more

9:40

about that in a sec. Right now though, you are listening

9:42

to Sarah Moran, co-founder of Patient

9:44

Notes, and Jessica Sire, technology reporter

9:46

with the Australian Financial Review. This is

9:49

Download This Show. And we're on a bit

9:51

of an AI tear at the moment. Chat

9:53

GPT, which has become very well

9:55

known as an AI service, which

9:58

will kind of mimic different writing stages. styles

10:00

and clean up your grammar and all kinds of things,

10:02

particularly around the world of writing, can

10:05

now scan the internet? Sarah,

10:08

that doesn't feel sky-netty at all. Talk

10:10

to me what's happened.

10:11

Well, it also feels like it's overselling it

10:13

to me. So the way that

10:16

ChatGPT has worked up until now is it

10:18

was based off a large

10:20

language model trained that

10:22

had only been connected to the internet till September 2021. So

10:26

anything after that or newer than that, including

10:28

a lot of the AI tools itself, ChatGPT

10:31

doesn't know. It hasn't been trained on that in

10:33

terms of information. And so what this is

10:36

doing is it's allowing ChatGPT

10:38

to connect to the internet and

10:40

to be able to reach

10:43

for that information that it doesn't have in

10:45

its database already. To me, though,

10:47

I'm not terribly optimistic about it.

10:49

Have I said anything that's given them any consideration

10:52

for what happens

10:53

when

10:54

ChatGPT is faced with the fire

10:57

hose all of us have to deal with with the internet?

10:59

That was my initial thought. It was like, oh,

11:01

great. Now I can surf the internet. That's cool.

11:03

At least it's got information that's current or maybe

11:05

it has methods of managing

11:07

the Google or rather being indexation

11:10

pages in a way that

11:11

I can't.

11:13

Do you remember a couple of years ago, there was a story where,

11:15

and this was years ago, Microsoft

11:18

set up an AI and it

11:20

immediately became a white supremacist racist? Yeah,

11:22

yeah. Because it was exposed to the internet.

11:24

That's a very truncated version of the story. There's

11:27

a little part of me that's just like, if you give it uncontrolled

11:29

access to the accessible of the internet, what will it become?

11:32

It's funny to hear you both

11:34

reflect because I tend

11:37

to think of this as being more like you've got this

11:39

great powerful tool in open AI and

11:41

then you stuck on a slow Bing search

11:43

on the end.

11:44

So like it's

11:47

because what first comes to mind

11:49

is that, oh, great. Now the

11:51

whole internet is in ChatGPT. Yeah,

11:54

that's what I thought. Yeah, right. But

11:57

no, so the way it works is ChatGPT.

12:00

has, so it's a large language model, it

12:02

has been trained on so

12:04

many words, bajillions of words, but

12:06

it doesn't now be trained on the rest

12:08

of the words that it missed out on since September 2021. So

12:12

it then just tacks that on, right? Like it's just,

12:14

it can do a little search. And I say

12:17

little very specifically, because the way

12:19

these tools work is that they're based

12:21

on tokens. And so you

12:23

get a certain amount of tokens

12:25

that you can use within a context

12:28

window. And it maxes

12:30

out at 8,096 tokens. So to get very technically specific, 1000 tokens

12:33

is about 750 words. And so you kind of have

12:42

limits on how much it can actually be searching.

12:45

And then also after that point, it

12:47

gets lost. So if you've hammered

12:50

chat GPT, like I have, you know, if you put

12:52

too much information in there, it kind

12:54

of, it doesn't quite know what it's talking

12:56

about itself anymore. So it gets harder

12:58

and harder to use, which is just very, very

13:01

complex and

13:01

annoying. So literally like me at the end of a

13:03

long day, I'm like, I don't know what

13:05

I'm saying. Don't ask me things involving

13:08

facts.

13:09

That's exactly how it is. Are

13:11

you chat GPT in disguise? I tell you.

13:13

I could be. I

13:15

could be. You mentioned earlier,

13:17

Bing. So it is using

13:20

Microsoft's search engine Bing

13:23

to search the internet. Is that right, Jess?

13:25

As I understand it, yeah, that's the partnership.

13:27

So this will be the most used Bing has ever

13:29

gotten in the life of Bing. I'm sorry. This

13:31

is going to give Bing a reason

13:33

to exist. Well,

13:34

I think it's like the meta partnership

13:36

is with Bing as well. And so is

13:38

this chat GPT thing. But I think you can

13:40

see Microsoft's

13:41

strategy here. Yeah, we've given

13:43

up on humans and we have the search engine

13:45

just for AI. They spent years

13:48

selling Bing as like the alternative

13:50

to Google. There was placement in movies

13:52

and things like that. Unless it was sort

13:55

of baked into your computer when you bought

13:57

it. People didn't really use

13:59

it.

13:59

It's definitely like a search

14:02

for the Microsoft products. So a

14:04

lot of their development was based

14:06

off Bing as

14:09

opposed to like Google, for instance.

14:10

And Microsoft has been looking for a way to compete

14:12

with Google on search

14:14

as one thing, but also on AI as another.

14:17

So are they right? Is

14:19

this going to be the thing that finally makes Bing

14:22

sort of a viable product?

14:23

I just don't think we'll actually know

14:26

it's using Bing. Like I don't think that it'll

14:28

redirect us to Bing. I think Bing will

14:30

become like the enterprise search engine. You

14:32

know, it's like powering things in the background and

14:35

us as users won't know

14:37

that Bing is the thing behind it. Totally.

14:40

And I think like that's from a like strategic

14:43

point of view, I think that's really smart

14:44

from Microsoft's like strategy.

14:47

Going B2B is obviously is a lot easier

14:50

and a lot more lucrative generally

14:52

over time. One

14:54

of the issues with chat GPT

14:57

and it's become quite apparent as more

14:59

people have gotten online and sort of tested its limits to your

15:01

point earlier, Sarah, is that it's

15:04

great at sounding confident. It's

15:06

less great at being right. Is

15:09

there a sense that giving it access

15:11

to the wider internet, a

15:13

current internet might solve that

15:15

problem, Sarah?

15:17

I mean, it's an illusion, you

15:19

know, and it's something to aim for. I think

15:22

we have to remember that in terms of how old this

15:24

technology is, these are just little toddlers if

15:26

they're even up and walking. So that connection

15:29

is great. I think we'll learn a lot from having

15:31

it. But out of the box,

15:33

it doesn't suddenly make things accurate or

15:35

true or up to date. That

15:38

accuracy is still in question. And I think that

15:40

really highlights the role that humans

15:43

play in the use of AI. One

15:45

of the remarkable things I've noticed definitely in the last

15:47

like six months talking about this chat GPT

15:49

stuff with as many people as possible

15:51

is like the user awareness of the

15:54

inaccuracy of the tool is like

15:56

red hot. Every user knows that

15:58

chat GPT could be wrong. And I think that

16:01

it's such an interesting training

16:03

ground for humans to

16:05

interact with this technology because,

16:09

okay, the tech could be wrong. And so how do

16:11

I query it in a way for it to,

16:13

one, give me something that's accurate and how do

16:15

I develop really fast fact-checking models

16:18

for myself

16:19

to

16:20

get the most out of this thing? And

16:22

I just think that's such a good way to go about its

16:25

mass adoption.

16:26

I haven't had this level

16:28

of media literacy, critical

16:31

thinking since the early

16:33

days of things like Wikipedia, where people sort

16:35

of did the same thing. They were like, Wikipedia

16:37

is great. Don't trust it. But

16:39

I think it's possible Google got so good at

16:41

giving us information that we came

16:44

to trust it more. Whereas I think this kind

16:46

of throws back to an earlier era of the internet in a way where

16:48

you were a bit like, I'll

16:51

just double-check

16:51

that. Yeah. Demographically,

16:54

people talk about Generation

16:56

Z and the way that young people

16:58

are digitally literate and fluent. And

17:01

one of the things that I've observed is their

17:04

media literacy or their critical thinking when it

17:06

comes to stuff on the internet

17:09

is so fine-tuned. They

17:12

are cynical, skeptical about everything

17:14

they read online. It's the boomers and things

17:17

that we kind of got to worry about when it comes to media consumption.

17:20

And to your point about media literacy, I think that the

17:22

young generations who are growing up with these tools

17:25

have an inbuilt radar,

17:28

I think. And the advent

17:30

of chat GPT and the way that people interact with

17:32

it

17:32

is fine-tuning that. Sarah,

17:34

you must have seen this in spades over the years

17:36

with Girlgate.

17:37

Yeah. And I think,

17:39

to Jess's point about the younger generation,

17:42

it's both true and untrue. There is still a digital

17:44

divide that happens with young people because

17:46

it can be manipulated. And I think

17:49

that's where the dangerous part happens.

17:52

You know, it's like hacking the algorithm. But you're

17:54

hacking the algorithm of how people determine

17:56

what is true and what isn't. And I think right

17:59

now, it's... this time, we're seeing a lot

18:01

of that really take off. And

18:03

that's where the risk is, right? And so chat

18:05

GPT are the first people to come out and say, hey,

18:08

our stuff isn't going to be accurate. It's up to you

18:10

to test. Like, it's up for you to check those things. But

18:12

then there's not really that extra layer

18:15

built in, which is that media literacy. So you say, well, how

18:17

do I check if something is accurate or not? And

18:19

I think it's been the case, oh, since I was a young person,

18:22

that we don't teach that. And

18:24

it's also really hard to keep media literacy

18:27

up to date with all the trends as they move.

18:29

So the way I download this show is what you're listening to. It is

18:31

your guide to the week in media, technology and

18:33

culture. Earlier in the show, we were talking about

18:35

AI, which seems to be a weekly occurrence

18:38

at the moment. Hey, this is what's happening. But

18:40

I tell you this idea of

18:42

can you actually pick

18:45

an AI picture? Because

18:47

the AFR produced its power list.

18:50

And I'm guessing access to people like Rockland

18:52

Murdock was a bit hard, because there's

18:55

a whole bunch of images in there that are of these of

18:57

these people that made the list that are AI

18:59

generated. What was I

19:02

know that I'm not necessarily saying you were responsible

19:04

for this decision, but it seems silly not to ask you because you're

19:06

here. What was the logic of it?

19:08

I think it was just to explore

19:11

this, use this technology

19:13

and see what you could do with

19:16

it creatively. I mean, notwithstanding that

19:18

you're getting access to local Murdock, so Sam

19:20

Kerr at this time is easy. But

19:23

it was just like AI was driving

19:26

every single business decision, every investment

19:28

decision over the last 12 months has been like

19:31

people boards are talking about investors are talking

19:33

about a stock prices are going bananas over

19:35

companies that used to be blockchain companies

19:39

are also just like this was part of the zeitgeist.

19:42

And all right, well, if it's part of the zeitgeist, then we're talking

19:44

about power and AI as a powerful

19:47

force in Australia. Well, what's sort of like a

19:49

subversive way that we could use

19:51

this technology to illustrate

19:55

at least the people in our palace? How

19:57

quickly did people pick it?

20:00

Well,

20:01

pretty quickly, I think. I mean, some

20:03

of the hands are pretty mangled

20:05

if you look pretty closely. I think, and also

20:08

like the AFR was at pains to say this is an

20:10

AI-generated image because there

20:12

are not necessarily clear guidelines

20:15

around what you can and can't do with image-based

20:18

AI at the moment.

20:21

But yeah, I mean, some of them look great and

20:23

some of them are like, huh, there's something really off

20:26

about that person. So this is the game.

20:29

And

20:29

yes, of course, the AFR were very, very

20:31

clear about pointing it out. But I also think there's

20:33

been plenty of issues we've seen recently where images

20:36

are proliferated online where it's not as clear,

20:38

right? And if you're not attuned for it,

20:40

it is actually getting quite hard to tell some of these

20:42

images apart. So let's

20:45

do AI 101. If you're looking at a

20:47

picture, what's the first thing, Sarah,

20:49

that you look for to go,

20:51

oh, are you real?

20:54

How many fingers does the person

20:56

have? That is absolutely the first

20:58

test. And are they sitting

21:01

in a natural pose or are they half missing their

21:03

fingers? So for some reason, we actually

21:05

don't know why, AI is terrible at

21:08

being able to draw hands. So

21:10

if someone hands is behind their back or they're hidden

21:12

in some way, that may be a tip

21:14

off that it might be an AI-generated

21:16

image.

21:17

Why is it that AI can't do hands?

21:20

It doesn't seem like of all the things,

21:22

it doesn't seem like the most complicated

21:23

part. You know, it's peculiar talking

21:26

with the creative director at the IFR

21:28

magazine about this process. So they worked

21:30

with an AI artist

21:33

and he's got all these different images

21:35

and a set of scenes that he's trying to basically

21:38

blend together. But he is programming

21:40

his creative

21:43

bot to spit

21:45

out an image. The

21:48

first iteration looks nothing like Anthony

21:50

Albanese on the decks,

21:51

for example.

21:52

The second one, he sort of told

21:54

it to do, yeah, put his eyes

21:57

closer together, but also do this with his hands.

22:00

input that code into the machine,

22:03

it

22:04

sometimes breaks or spits out some crazy

22:06

stuff. So what's really funny is you look at the

22:09

iterations in between of these images

22:11

and like Anthony's Elvenees,

22:13

his ear might just have shot right off

22:15

and you're just not really sure like where in the code base

22:18

that bug is. So it's

22:20

actually like quite a painstaking process

22:23

to get AI to make images

22:25

like real people.

22:29

It's also like fascinating that AI

22:31

prompts has become its own industry

22:34

like in the last couple of months, it's become

22:36

a whole new skill set. And

22:39

I guess one of the things I'm fascinated by is in

22:41

that process, like are there particular tricks

22:43

that you think people should know?

22:45

Oh, well look as a prompt guru,

22:48

I'm very good at generating images and

22:50

the most convincing were actually my own.

22:53

And the ones that weren't so good really

22:56

highlighted where some of these issues come out. So

22:58

my teeth, so mangled, very, very

23:00

terrible. But to Jess's point,

23:04

to all of our point, I think what really underpins this is what

23:06

prompts are you using to

23:08

be able to generate the content?

23:11

And I think for all of the stories we've talked about

23:13

today that there is this underlying thing about the

23:16

next stage is us being prompt ready,

23:18

really getting our heads around.

23:21

How do we, because we're observing this tech,

23:23

even as builders of the technology. So at our company

23:26

where we're using this stuff to build, we're

23:28

observing what gets spat out the other end

23:31

and then having to figure out prompts to turn it around. And

23:33

also to stop the AI from hallucinating.

23:36

So this idea of it just

23:38

makes stuff up. You're like, I didn't prompt that. I

23:40

don't know where you got that from, but hey, go

23:42

for it, AI. So it's fun days,

23:45

fun times.

23:45

Well hopefully after this, you'll be able

23:47

to identify AI photos from

23:49

now until the time they get much better, which

23:52

is probably going to happen by the time we get out of the studio. The

23:55

sharing service TikTok of course has become massive

23:57

over the last couple of years, but it would also like to become

23:59

an AI. commerce platform. What does that mean? Basically,

24:02

you can pay with it. Interesting news

24:04

out of Indonesia, Jess, that Indonesia

24:06

have, which has, you know, quite

24:09

a significant TikTok user base

24:11

has stepped in and gone, actually, no, we don't want

24:14

you to use this to pay. Talk me through this.

24:16

Yeah, the Indonesian government's basically

24:19

put in a law restricting

24:22

TikTok shop

24:23

from

24:24

accessing its entire population.

24:27

They've basically said, we want to protect our

24:30

real life stores and our offline

24:32

merchants. And so we put these restrictions

24:34

around the TikTok e-commerce platform.

24:37

There's things like products that

24:39

are imported from overseas need a base

24:41

price of at least $100, which

24:43

is significant when you're buying things online.

24:47

And TikTok, I mean, has already

24:50

invested billions of dollars in establishing

24:53

a pretty like solid presence

24:55

in Indonesia. I think there's like 125

24:57

million people use TikTok in Indonesia

25:00

and they're rolling out TikTok shop. And

25:02

the government's basically said, we're going to

25:04

put parameters on the

25:06

commercial access you have to our population. Sarah,

25:09

why would Indonesia

25:11

not want this app to kind

25:13

of follow that pathway that so

25:15

many Chinese social media apps do, which is

25:17

this sort of like these all purpose, live

25:19

your life, pay for your life online services?

25:22

Well, what I've taken from it is that there

25:24

is an intent to protect offline

25:27

merchants as well, and their

25:29

marketplaces that already exist from predatory pricing.

25:32

So if you think about what happens if you've

25:34

suddenly got 125 million users

25:36

that all can become sellers and merchants

25:38

in one go, what happens to your

25:41

current existing economic

25:43

infrastructure? It can really put

25:45

it under threat in a way that potentially

25:47

other things haven't before. I think also

25:49

one of the huge risks that the government's

25:51

probably trying to

25:52

protect against, and this

25:54

is obviously an issue when we talk about

25:57

globalization and monopolies setting

25:59

up in different markets.

25:59

They're trying to protect the value chain that already

26:02

exists in Indonesia, right? So if you've

26:04

got like a super app, which has e-commerce

26:07

payments Communications

26:10

like that's three major verticals

26:13

within Indonesia that is As

26:15

all of a sudden competing with this Chinese company and

26:18

if the product if the tiktok

26:20

shop or the tiktok app or super app Is

26:23

more convenient easier to use and more reliable

26:25

than the existing government infrastructure

26:28

then

26:30

People will choose that and it's been a huge

26:32

risk for currencies for

26:34

governments for ages like when the

26:36

idea of do you remember when Facebook wanted to bring out

26:38

Libra that

26:39

Current like they've been able to do that Experiments of

26:41

so many times with like Facebook credits and then leave

26:44

these different ways of doing currency

26:45

and the US government was really Not into

26:48

that idea because what if it

26:50

became easier to pay your staff in Libra

26:52

on this, you know Platform and they could buy

26:54

enough stuff with Libra They just stopped using

26:56

the US dollar and that debases the currency

26:59

of an entire

26:59

economy and that's a serious problem

27:01

Do you know what it strikes me Sarah? It's

27:04

like the governments around the world are learning

27:06

the lessons of watching cryptocurrencies So

27:09

yeah, they're

27:11

trying to get ahead of it. And I think it

27:13

is a real risk that You

27:15

know, you've talked about a global company

27:17

who has a big opportunity

27:20

to manipulate things including

27:22

the value of any of that currency at any particular

27:24

time and they've probably been thinking

27:26

about it and planning about it for a long

27:29

time as the country that's

27:31

been subjected to US

27:34

And and other foreign apps coming in, you know

27:36

They would have seen the impact of things like Airbnb

27:39

and other sorts of players coming into

27:41

the market I think this has been part of

27:43

their strategy for a while to think about standing up to

27:45

it

27:46

And with that I shall leave you huge. Thank you

27:48

to our guest this week Just because I from the Australian financial

27:50

review purveyors of AI powerful

27:52

people Thank you. It's a pleasure

27:55

was entirely ours and Sarah

27:57

Moran from patient notes. Thank

27:59

you for joining Thanks for having me. And that's it for the

28:01

show. My name is Mark Finnell and thank you for listening

28:04

to another episode of Downloaded History.

28:07

Thanks for having me. And that's it for the show.

28:29

History Podcast.