0:00

Time is precious and so are

0:02

our pets. So time with our

0:04

pets is extra precious. That's why

0:06

we started Dutch. Dutch provides 24-7

0:08

access to licensed vets with unlimited

0:10

virtual visits and follow-ups for up to

0:13

five pets. You can message a vet at

0:15

any time and schedule a video visit

0:17

the same day. Our vets can even

0:19

prescribe medication for many ailments and shipping

0:21

is always free. With Dutch you'll get

0:23

more time with your pets and year-round

0:26

piece of mine when it comes to

0:28

their vet care. In

0:30

a world of economic uncertainty and

0:33

workplace transformation, learn to lead

0:35

by example from visionary C-sweet

0:38

executives like Shannon Skyler of

0:40

PWC and Will Pearson of

0:43

Ihart Media. The good teacher

0:45

explains, the great teacher inspires.

0:47

Don't always leave your team to

0:49

do the work. That's been the most

0:51

important part of how to lead by

0:54

example. Listen

0:56

to Leading

0:59

by Example,

1:02

executives making

1:05

an impact

1:07

on the I heart

1:10

radio app,

1:12

Apple Podcasts,

1:14

or wherever

1:17

you get

1:19

your podcasts.

1:23

As a reminder, you can now buy glorious

1:25

better offline merchandise. There's a link to it

1:28

in the episode, notes, the t-shirts, the tumblers,

1:30

the tote bags, all that shit, it's lovely,

1:32

you're gonna love it. Buy it today! But

1:34

if I'm honest today, I'm actually a little

1:36

bit pissed off, and that's why we've got a

1:38

two-part episode this week about how fucking stupid the

1:40

AI boom has become. I wrote Farsicle in the

1:43

script, and I'm gonna be honest need to be

1:45

a little more pointed to be a little more

1:47

pointed. Because I've written tens of thousands of words

1:49

about this now. I've recorded hours upon hours of

1:52

podcasts and still to this day. People are babbling

1:54

about the AI revolution as the sky rains blood

1:56

and crevices open in the fucking earth, dragging houses

1:58

and cars and the best... get animals into

2:00

their maws. Things are astronomically fucked outside,

2:03

yet the tech media continues to tell

2:05

me to get my swimming trunks on

2:07

and take a long nice dip in

2:09

the fucking pool. As you can tell

2:12

this is going to be a little

2:14

less reserved than usual. I'm a little

2:16

bit frustrated. I don't know why I'm

2:18

the one saying this, and I frequently

2:21

feel with I, a part-time blogger and

2:23

podcaster, and writing the things that I'm

2:25

writing. Since I put out the newsletter

2:27

open AI as a systemic risk to

2:29

the tech industry and actually a couple

2:32

weeks back I did the two-parter about

2:34

it too, I've heard nothing in response

2:36

as was the case with how does

2:38

open AI survive and how open AI

2:41

is bad business. There just seems to

2:43

be little concern or belief that there's

2:45

any kind of risk at the heart

2:47

of AI and open AI in particular

2:50

and there are companies that spent $9

2:52

billion in 2024 to lose $5 billion.

2:54

Well I'd love to add a because

2:56

here. If not, because it's important to

2:58

be intellectually honest and represent views that

3:01

directly contrast my own, even if I

3:03

do in somewhat sarcastic and sardonic fashion,

3:05

nobody seems to actually have a cogent

3:07

response to how they write this ship.

3:10

Other than hard fork, a case in

3:12

you and throwing a full-scale psychotante on

3:14

a podcast and saying I'm wrong because

3:16

inference costs are coming down. Infrants, by

3:18

the way, is when an AI takes

3:21

an input and produces an output. It's

3:23

the calculations that take place right before

3:25

Google's generative AI assistant. says that Black

3:27

Tar Heroin when enjoyed immodoration can help

3:30

you lose weight. Newton is a nakedly

3:32

captured booster that ran an infographic from

3:34

Anthropic a few weeks ago, the likes

3:36

of which I haven't seen since 2013,

3:39

it was telling you all the ways

3:41

that people use generative AI. It looks

3:43

like some shit from, I don't know,

3:45

early day mashable, no offense Christina. And

3:47

they essentially treat this company propaganda as

3:50

gospel, but he's really far from the

3:52

only one with a flimsy attachment to

3:54

reality. The information, a publication that genuinely

3:56

does some great stuff, which makes it

3:59

even more heartbreaking to say this, ran

4:01

a piece in early April that made

4:03

me even more furious than usual, claiming

4:05

that Open AI was forecasting... revenue topping

4:07

$125 billion in 2029 based on selling

4:10

agents and monetizing free users as a

4:12

driver to hire revenue. Agents I should

4:14

add are AI systems that can interact

4:16

with other systems and do stuff. So

4:19

an AI that can order pizza from

4:21

DoorDash for you is that's an example

4:23

of an agent. And when I say

4:25

it can order pizza view, I am

4:28

talking entirely theoretically as they cannot do

4:30

this right now and may never be

4:32

able to do so. Indeed, the whole

4:34

agent thing is just... What we wish

4:36

AI was, and it actually doesn't work.

4:39

And the piece reported based on things,

4:41

and I quote, told to some potential

4:43

and current investors, takes great pains to

4:45

accept literally everything that Open AI says

4:48

is perfectly reasonable, if not Gospel, even

4:50

if said things make absolutely no god

4:52

damn sense. So, according to the information

4:54

reporting, Open AI expects agents and new

4:56

products, and both of those are quotes,

4:59

to contribute tens of billions of dollars

5:01

of revenue, both in the near term,

5:03

somehow contributing $3 billion in revenue this

5:05

year, which I will get to in

5:08

a little bit, and in the long

5:10

term, with an egregious $25 billion in

5:12

revenue in 2020-29 even, projected to come

5:14

from just new products. If you're wondering

5:17

what those new products might be, I

5:19

am too. Because the information doesn't seem

5:21

to know. And instead of saying, Open

5:23

AI has no idea what the fuck

5:25

they're talking about and is just saying

5:28

stuff, the outlet continues to publish things

5:30

with the kind of empty optimism that's

5:32

indistinguishable from GPT generated LinkedIn posts. Must

5:34

be clear, the information isn't generating their

5:37

articles, the writing and fresh. I want

5:39

to be really, really clear about something.

5:41

We aren't in nearly in May 2025,

5:43

and indeed one of these will come

5:45

out. Actually, in May, the second part.

5:48

I see no evidence that Open AI

5:50

even has a marketable agent product they

5:52

can sell. Let alone one, it will

5:54

make three billion God-dam dollars off of.

5:57

And they definitely are not going to

5:59

do so in the next six or

6:01

seven months. Oh my... For context, let's

6:03

triple the revenue of Open AI. that

6:06

they made reportedly, at least from selling

6:08

access to their models via its APIs,

6:10

essentially allowing third-party companies to use GPT

6:12

in their apps, in the entirety of

6:14

2024. And those APIs and models actually

6:17

exist in a meaningful sense, as opposed

6:19

to whatever the Farkopin AI's half-baked ass

6:21

agent's stuff is. In fact, no, no,

6:23

no, no, no, no, no, no, no,

6:26

no, I'm not, I'd be calm. be

6:28

normal. I'm going to explain exactly what

6:30

the information is reporting in an objective

6:32

way, because writing it out really shows

6:34

how silly it all sounds. I'm going

6:37

to rate they believe a lot, because

6:39

I must be clear how stupid this

6:41

is. Now, according to the information's reporting,

6:43

they believe that Open AI will make

6:46

$3 billion in 2025 from selling access

6:48

to its agents. This appears to come

6:50

from Soft Bank, which has said it

6:52

will buy $3 billion worth of Open

6:55

AI products annually. Earlier this year, we've

6:57

got a... bit of extra information about

6:59

how Soft Bank will use these products.

7:01

It plans to create a system called

7:03

Crystal Intelligence, that's C-R-I-S-T-A-L, and it's one

7:06

of the most generic names I've ever

7:08

seen, and it will be a kind

7:10

of general-purpose AI agent platform for big

7:12

enterprises. The exact specifics of that will

7:15

shock you and that there are none,

7:17

but Soft Bank intends to use the

7:19

technology internally across its various portfolio companies

7:21

as well as market. it to other

7:23

large enterprise companies in Japan. I still

7:26

do not know what the fuck this

7:28

is. Crystal Intelligence. Billions of dollars. And

7:30

they just don't, they can't even describe

7:32

what it is just saying, yeah, it'll

7:35

be an agent platform that does stuff

7:37

with your business. Like, is that sound

7:39

good? Can I have three billion? I

7:41

need 40 billion dollars. Give me... Okay.

7:44

I also want to add that the

7:46

information can't seem to keep its story

7:48

straight on this issue. Back in February

7:50

they reported that Open AI would make

7:52

$3 billion in revenue only from agents.

7:55

With a big beautiful chart that said

7:57

$3 billion would come from it, only

7:59

to add that it would be Soft

8:01

Bank using Open AI's products across its

8:04

companies. Based on these numbers it seems

8:06

like Soft Bank will be the only

8:08

customer for Open AI's agent. While this

8:10

most likely won't be the case and

8:12

it isn't because it excludes anyone willing

8:15

to pay a few bucks to test

8:17

it out, it nonetheless doesn't signal good

8:19

things for agents as a mass market

8:21

product. Not that there were any good

8:24

signals beforehand, though. Agents do not exist

8:26

as a product that can be sold

8:28

at scale. Yes, OpenAI teased operator, its

8:30

first agent at the start of the

8:32

year, but it doesn't seem to be

8:35

able to do anything. The information's own

8:37

reporting from mid-April reporting from mid-April highlighted

8:39

how Open AI's... operator agent struggled with

8:41

comparison shopping on financial products and that's

8:44

a quote and how operator and other

8:46

agents are and I quote again tripped

8:48

by pop-ups or log-ins as well as

8:50

prompts asking for email addresses and phone

8:53

numbers for marketing purposes which I Think

8:55

accurately describes most websites and just the

8:57

summarize from everything I've said the information

8:59

is saying that the above product will

9:01

make open AI three billion dollars by

9:04

the end of 2025 Sounds very real

9:06

to me Sounds extremely real. I love

9:08

that the business media just prints this.

9:10

I love this. I love this so

9:13

much. I'm having so much fun. Jesus

9:15

Christ, according to the information reporting, they

9:17

believe that Open AI will basically double

9:19

revenue every single year for the next

9:21

four years and make $13 billion in

9:24

revenue 2025, more than doubling that to

9:26

$29 billion in 2026, nearly doubling that

9:28

to $54 billion in 2027, and nearly

9:30

doubling that again to $86 billion in

9:33

2028, and eventually leveling out a ridiculous

9:35

$125 billion of revenue in 2009. said

9:37

revenue estimates as of 2026 includes billions

9:39

of dollars of new products that include

9:42

free monetization. Free user monetization either. And

9:44

if you're wondering what that means, I

9:46

also am. The information does not explain.

9:48

Jessica Lesson must have been busy being

9:50

horrible to people that work for her

9:53

they do however say that open AI

9:55

will start and I'm quoting this won't

9:57

start generating much revenue from three users

9:59

and other products until next year that's

10:02

2026 in 2029 and I'm still quoting

10:04

however it projects revenue from free users

10:06

and other products will reach 25 billion

10:08

dollars or one-fifth of all revenue and

10:10

then adds that shopping is another potential

10:13

avenue. You still probably don't know what

10:15

they're doing and neither do I and

10:17

I have driven myself insane reading about

10:19

this. I really cannot express my disgust

10:22

about how willing publications are to blindly

10:24

published projections like these especially when they're

10:26

all so stupid. Let me just read

10:28

this to you all right? And I

10:31

quote. Open AI has already begun experimenting

10:33

with launching software features for shopping. Starting

10:35

in January, some users can access web

10:37

browsing agent operator as part of their

10:39

pro chatGPT subscription tier to order groceries

10:42

from InstaCar and make restaurant reservations on

10:44

Open Table. Just want to be clear,

10:46

this is a few episodes ago. I

10:48

mentioned Casey Newton not even being able

10:51

to like say this worked. I just

10:53

want to be really clear as well

10:55

what the information is saying. So they're

10:57

saying that this experimental software launched to

10:59

an indeterminate amount of people that barely

11:02

works is going to make open AI

11:04

three billion dollars in 2025 and then

11:06

somehow this is going to lead to

11:08

open AI making twenty nine billion dollars

11:11

in 2025 and then somehow this is

11:13

going to lead to open AI making

11:15

twenty nine billion dollars in 2026 and

11:17

then they're going to eventually be up

11:20

to venture capital. Time

11:28

is precious and so are our pets.

11:30

So time with our pets is extra

11:33

precious. That's why we started Dutch. Dutch

11:35

provides 24-7 access to licensed vets with

11:37

unlimited virtual visits and follow-ups for up

11:39

to five pets. You can message a

11:42

vet at any time and schedule a

11:44

video visit the same day. Our vets

11:46

can even prescribe medication for many ailments

11:48

and shipping is always free. With Dutch

11:50

you'll get more time with your pets

11:52

and year-round piece of mind when it

11:54

comes to their vet care. In

12:00

fact, I think we have real reason

12:02

to worry about whether Open AI even

12:04

makes its current projections. In my last

12:06

multi-part episode, and in the newsletter Open

12:08

AI is a systemic risk, for those

12:10

of you who will like to read

12:12

while listening to my fucking podcast, I

12:14

wrote the Bloomberg had estimated that Open

12:17

AI would triple revenue to $12.7 billion

12:19

in 2025, and based on its current

12:21

subscriber base, Open AI would effectively have

12:23

to double its current subscription revenue and

12:25

massively increase its API revenue to hit

12:27

these targets. These projections rely on one

12:29

entity, Soft Bank, spending $3 billion specifically

12:31

on Open AI services. Really shouldn't have

12:33

said specifically because they keep changing what

12:36

it means. Meaning that they'd have to

12:38

make enough on API calls, so people

12:40

plugging the models into their products, to

12:42

generate more revenue that Open AI made

12:44

in subscriptions in the entirety of 2024,

12:46

and something else that I can only

12:49

describe as an act of God. And

12:51

that I admit assumes that Soft Bank's

12:53

spending commitment is based on usage and

12:55

not like a flat fee, where Softback

12:57

just hands them $3 billion and gets

12:59

infinite levels of access. Assuming it's the

13:02

former, I'd be stunned if Soft Bank's

13:04

consumption hits $3 billion in 2025, even

13:06

with the massive cost of the reasoning

13:08

models that Crystal Intelligence will maybe be

13:10

based off of? Again, we don't know.

13:12

And Softbank announced this deal with Open

13:15

AI in February. Crystal Intelligence, if it

13:17

works, and that is possibly the most

13:19

load-bearing if, of all time, will be

13:21

a massive, complicated, and ambitious product. Details

13:23

are vague, but from what I understand,

13:25

Softbank wants to create an AI that

13:28

handles a bunch of varied tasks that

13:30

knowledge workers do. I mean, it's just

13:32

the same marketing bullshit. It's the same

13:34

thing they've been lying about before. And

13:36

to be clear, Open AI's agents cannot

13:38

consistently do well anything right now. What

13:40

I believe is happening is that reporters

13:42

are taking Open AI's rapid growth in

13:44

revenue from 2023 to 2024, when they

13:46

went from like tens of millions of

13:48

dollars a month in the beginning of

13:50

the 2023 to 300 million in August

13:53

2024, genuinely a big leap. They're taking

13:55

this to mean that the company will

13:57

always effectively double or triple revenue every

13:59

single year forever. with their evidence being

14:01

open AI has said that this will

14:03

happen in projections. It's bullshit. I'm sorry,

14:05

it's bullshit. It's bullshit. As I wrote

14:07

before in a newsletter, it's called There's

14:09

No AI Revolution and the accompanying episodes

14:12

at the time. Open AI effectively is

14:14

the generative AI industry. Nothing about the

14:16

rest of the generative AI industry suggests

14:18

that the revenue exists to sustain these

14:20

ridiculous obscenes and frankly fucking stupid valuations

14:23

and projections. What do I mean by

14:25

that, by the way? Okay, let me

14:27

get into it. ChatGPT is the only

14:29

real generative AI product with any significant

14:31

usage, or rather, the nearest rivals are

14:34

a fraction of said user base. Or

14:36

maybe I need to be a little

14:38

bit blunter. If anyone held a Google

14:40

Gemini user conference, all the attendees could

14:42

probably share a cab. Believing the open

14:45

AI growth myth, and yes, reporting it

14:47

objectively is both endorsing and believing these

14:49

numbers, is engaging in child-like logic, where

14:51

you take one of them, which is

14:53

open AI's revenue, grew 1,700% from 2023

14:56

to 24, wow! To mean another will

14:58

take place, which is that open AI

15:00

will continue to double revenue literally every

15:02

other year, another insane thing to believe,

15:04

and you're consciously ignoring difficult questions such

15:07

as... How will they do this? And

15:09

what's the total addressable market of large

15:11

language models? And their associated subscriptions, exactly.

15:13

And how does this company even survive

15:15

when it expects the costs of inference

15:18

to triple this year to $6 billion

15:20

alone? Wait, wait, wait. Sorry. I really

15:22

need to be clear with that last

15:24

one, because it's a direct quote from

15:26

the information. The company also expects growth

15:29

and inference costs. The costs of running

15:31

AI products such as ChatGPT and their

15:33

underlying models to moderate over the next

15:35

half decade. These costs will triple this

15:37

year, referring to 2025 to $6 billion,

15:40

and rise to nearly $47 billion in

15:42

2030. Still, the annual growth rate will

15:44

fall to about 30% then. Okay, thanks.

15:46

Also, are you fucking kidding me? Six

15:48

billion dollars for fucking inference. Hey, Casey

15:50

Newton! Casey! He's not here. He's not

15:53

here. He's not here. Anyway, that's not

15:55

really great at all. That's actually really

15:57

bad. The information reports that Open AI

15:59

will make about $8 billion in subscriptions

16:01

to chat GPT in 2025, meaning that

16:04

75% of Open AI's largest revenue sources

16:06

eaten up by the price of providing

16:08

it. This is meant to be the

16:10

cheap part. This is the one fucking

16:12

thing people say to me is meant

16:15

to come down in price. I've had

16:17

assholes saying to me for the last

16:19

year, custom inference is coming down. Is

16:21

it? Are we living in different dimensions?

16:23

Are there large parts of the tech

16:26

media that have fucking gas leaks? What

16:28

am I missing? Tell me what I

16:30

am missing. Now Ed, you haven't taken

16:32

people the abilities, you don't know where

16:34

they... shut the fuck up. If you

16:37

were one of these people who says,

16:39

I need to... Casey, you're included man.

16:41

Fuck. Like, I'm so sick of this.

16:43

Oh, you don't talk to people running

16:45

these things. I am sick of people,

16:48

like Casey, you and others, too, saying,

16:50

oh, you don't talk to enough AI

16:52

people. You haven't listened to them. You

16:54

mean, I haven't listened to the pabblem

16:56

of the people that make money off

16:59

of lying about this dog shit. Are

17:01

you really thinking that's what's missing from

17:03

my analysis? Interviewing people who work at

17:05

these companies and understanding how the technologies

17:07

work? I know other technologies work. I

17:10

don't need to talk to these fucking

17:12

people. There are people out there, like

17:14

Simon Wilson and Max Wolf, who know

17:16

how these things work, that I talk

17:18

to fairly regularly, and both of them

17:21

push back on me, because they know

17:23

how large language models work. Those people

17:25

matter. What doesn't matter to me. What

17:27

will never matter to me is what

17:29

Darrio Amadeday, Jack Clark and all the

17:32

other fucking people an anthropic, an anthropic-anthropic

17:34

thing. And I think it's detestable and

17:36

actively, honestly malpractice in journalism, to pretend

17:38

that there's something ethical about speaking to

17:40

these people and listening and taking in

17:43

their marketing spiel. It's actually a little

17:45

bit disgusting that this is even a

17:47

critique leveled at anyone. But you can

17:49

have to forgive me, I'm going to

17:51

be a little rude and I know

17:53

that seemed like it, but I'm not

17:56

even getting excited. Now what, I think

17:58

it's time, okay everyone, I think it's

18:00

time that I go through the most

18:02

common critiques in AI. It's time for

18:04

me to really sit down and I'm

18:07

gonna do my Kevin Roos voice and

18:09

I know a lot of you, like

18:11

my Kevin Roos voice and some of

18:13

you, not a lot of you, I'm

18:15

gonna say I'm being rude to these

18:18

people and it weakens my analysis to

18:20

which I kiss my ass. I will

18:22

turn you, I will turn you, I

18:24

will cube you like a car in

18:26

a garbage dump. But let's

18:29

start, shall we? The cost of inference

18:31

is coming down. That's one argument, okay?

18:33

Source? Where is your source? If you

18:35

are someone saying to me that the

18:37

cost of inference are coming down. I

18:39

want your source. I want you to

18:42

show me the costs. I want you

18:44

to show me the costs at scale.

18:46

Because it sure seems like they're increasing

18:48

for open AI and they're effectively the

18:50

entire user base of the entire generative

18:52

AI industry. But Ed, what about Deep's?

18:55

You sweet idiot child. Deep's you sweet

18:57

idiot child. Deep's not open AI and

18:59

Open AI's latest models only seem to

19:01

be getting more expensive as time drags

19:03

on. GPT 4.5 cost $75 per million

19:05

input tokens and $150 the entire generative

19:08

AI. industry at least for the world

19:10

outside of China. On top of that,

19:12

we actually don't know whether Deep Seek

19:14

is even profitable to run at scale.

19:16

It is definitely cheaper to run, but

19:18

we don't know if it's actually profitable.

19:21

Indeed, I don't know even know how

19:23

you'd calculate this because running a Deep

19:25

Seek modelled as just one person is

19:27

one thing. The question is whether you

19:29

could scale it up like Open AI.

19:31

We don't know. But let's get back

19:34

to the other critiques. This is a

19:36

company, it's growth stage, they can just

19:38

hit the button, all be profitable. You

19:40

have the mind of a child. If

19:42

this was the case, why would both

19:44

anthropic and open AI be losing so

19:47

much money? Why are none of the

19:49

hyperscalers making profit on AI? Why does

19:51

nobody want to talk about the underlying

19:53

economics if they're at the growth stage?

19:55

And also, little side point as well,

19:57

why have we been at the growth

20:00

stage for years and why are hyperscalers

20:02

at the growth stage? They're not startups.

20:04

Anyway, on to another one, though. These

20:06

are the early days of AI. It's

20:08

just like the early days of AI.

20:10

Wrong. Wrong. We have all the King's

20:13

horses and all the King's men, the

20:15

entire tech industry, and more money that

20:17

has ever been invested into anything piled

20:19

into generative AI, and the result has

20:21

been utterly mediocre. Nobody's making money on

20:23

AI other than invidia, and maybe chewing

20:26

a consultancy. But Ed, they're already showing

20:28

signs that the AI is going to

20:30

be powerful. No it's not. If anyone

20:32

brings these critics, you just say no,

20:34

no they're not. Show me. Show me.

20:36

Show me. Why is it the only

20:39

people? I'm given Simon Wilson credit here.

20:41

He's one of the only people who

20:43

will show you anything cool. And it's

20:45

cloud compute stuff. It's like relatively boring

20:47

enterprise stuff. It's exciting for the niche

20:49

cases, like software generally is. But it's

20:52

really not showing any power. We talk

20:54

about this powerful AI thing. Is it

20:56

in the room? Where is this powerful

20:58

AI? But then I have actually had

21:00

a few emails saying, Ed, Ed, look

21:02

at opening eyes 03 model. And I

21:05

just want to be clear that this

21:07

new and extremely expensive reasoning model also

21:09

hallucinates more. Is that AGI, by the

21:11

way? Is this AGI, is the AGI

21:13

in the room with us? Did the

21:15

AGI tell you it loved you? Did

21:18

it tell you to leave your wife?

21:20

Did it, did it, did it offer

21:22

you sex? I hope you're okay! Time

21:35

is precious and so are our pets.

21:37

So time with our pets is extra

21:39

precious. That's why we started Dutch. Dutch

21:42

provides 24-7 access to licensed vets with

21:44

unlimited virtual visits and follow-ups for up

21:46

to five pets. You can message a

21:48

vet at any time and schedule a

21:51

video visit the same day. Our vets

21:53

can even prescribe medication for many ailments

21:55

and shipping is always free. With Dutch

21:57

you'll get more time with your pets

21:59

and year-round piece of mind when it

22:01

comes to their vet care. One

24:00

guy was wrong once. One guy, he

24:02

said that the internet wouldn't be big.

24:04

And this proves that I, Ed Zetron,

24:06

was that 99, so like 20 years

24:09

later, because one guy said that the

24:11

internet wouldn't be big, that I am

24:13

wrong summer, motherfucker, have you read the

24:15

piece? That's actually the thing. All of

24:17

these are things that you can box

24:20

up and use and people who use

24:22

this half fast bullshit. Clifford Stoll basically

24:24

says that the internet at the time

24:26

was pretty limited and yes he conflated

24:28

that with the idea that it wouldn't

24:31

be big in the future. However Stoll's

24:33

piece also as Michael Hiltzig wrote for

24:35

the LA Times was alarmingly accurate about

24:37

misinformation if you think it does. One

24:39

guy being wrong in some way is

24:42

not a... response to a criticism. I

24:44

will crush you like a bug if

24:46

this is your logic. I will eat

24:48

you. I will put you in my

24:50

mouth like Kirby and I'll shit you

24:53

out and I will have the powers

24:55

of a dunce. Stoll's analysis also isn't

24:57

based on hundreds of hours of research

24:59

and endless reporting. Mine is I will

25:01

grab you from the ceiling like the

25:03

wallmaster from Zelda and you will never

25:06

be heard from again. Anyway. Another argument.

25:08

Now their argument that people are to

25:10

give me is that open AI and

25:12

anthropic research entity is not business and

25:14

that they are not focused on profit.

25:17

Okay, so just so we're clear that

25:19

if that's the case, they're just going

25:21

to burn money forever? Is that the

25:23

case? Or are they going to hit

25:25

like the B profitable button sometime? Also,

25:28

if Open AI was a research entity,

25:30

why does it need $40 billion from

25:32

Soft Bank or to change its weird

25:34

corporate structure to become a fool for

25:36

profit? Actually. Actually. Open AI as many

25:39

as 800 million weekly active users. That's

25:41

proof of adoption, right? That's going to

25:43

be an argument that people have. There's

25:45

some bloke on blue sky. It's just

25:47

been responding to me every few days

25:50

with this kind of argument saying, look,

25:52

look at all the users. And look,

25:54

I get the, you might be a

25:56

bit hoar. about this number but something

25:58

don't make no sense about this number.

26:01

On March 31st 2025 Open AI said

26:03

that it had 500 million people who

26:05

use chat GPT every week. Two weeks

26:07

later Sam Altman claimed that something like

26:09

10% of the world uses their systems

26:12

a lot. They're referring to chat GPTs

26:14

and the media took this to mean

26:16

that chat GPT is 800 million weekly

26:18

active users. I just want to be

26:20

clear about something as well. Sam Altman

26:23

didn't say that. He said the weird

26:25

vague thing about something like 10% percent

26:27

of the world. Like that's what he

26:29

said. And everyone just went, oh shit,

26:31

we've got to help Sam Oman out,

26:33

got to push this bad boy over

26:36

the edge. And there are three ways

26:38

to interpret what he said, and you

26:40

tell me which one sounds real. Number

26:42

one, open AI's user base increased by

26:44

300 million weekly active users in two

26:47

weeks. Number two, open AI understated its

26:49

user base in the announcement of their

26:51

funding announced when an open ai.com by

26:53

300 million users. Or three, number three.

26:55

How about this? Sam Altman Fucking Lied.

26:58

I get that some members of the

27:00

media have a weird attachment to this

27:02

damp little man but have any of

27:04

you ever considered that he's just fucking

27:06

saying things knowing that you'll print them

27:09

with the kindest possible interpretation. Sam Altman

27:11

is a liar. He's lied before and

27:13

he'll lie again. I wrote an entire

27:15

newsletter called Sam Altman is full of

27:17

shit. You should read it. I'm gonna

27:20

link to it. I'm gonna link to

27:22

it. I'm gonna link to it. I'm

27:24

gonna link to it. Yes, Google Gemini

27:26

has 350 million monthly active users and

27:28

that's because they started replacing Google Assistant

27:31

with Google Gemini in early March. You

27:33

are being had, you are being swindled.

27:35

If Google replaced Google Search with Google

27:37

Gemini it would have billions of monthly

27:39

active users. Jesus Christ. Jesus Christ, on

27:42

a god damn cracker. Even reading this

27:44

script out I get at, like, some

27:46

of you have suggested that this is...

27:48

At all manufactured no reading this stuff

27:50

makes me very angry because I didn't

27:53

grow up popular or intelligent anyway I've

27:55

had to pick this shit up as

27:57

like And I don't think what I'm

27:59

saying is crazy, but I am sometimes

28:01

treated that way and this episode I

28:03

realize I'm doing myself no favors, but

28:06

Anyway, back to the critics really quickly.

28:08

Open AI having hundreds of millions of

28:10

free users, each losing it money, is

28:12

proof that the free version of ChatGPT

28:14

is popular, largely because the entirety of

28:17

the media has written about AI nonstop

28:19

for two straight years and mentioned ChatGPT

28:21

every single fucking time. Yes, yes, there

28:23

is a degree here of marketing and

28:25

partnerships of word of mouth of some

28:28

degree of utility, but when you remove

28:30

the nonstop free media campaign, ChatGPT would

28:32

have peated out. by now along with

28:34

this stupid fucking bubble but edits post

28:36

somebody's doing so yeah it's proved that

28:39

something is broken in society generative aye

28:41

has never ever had the kind of

28:43

meaningful business returns or utility that actually

28:45

underpins something meaningful but it has had

28:47

enough to make people give it a

28:50

try do you not Actually, no, I

28:52

know you listening, you're going to get

28:54

this. In some ways this episode has

28:56

been, I mean, in all ways it's

28:58

been pretty ranting, the second one is

29:01

going to be even more so. What

29:03

I'm trying to do here is show

29:05

you how farcical all this crap is,

29:07

how ridiculous it is, how silly these

29:09

posits are, these projections are, the suggestion

29:12

that what we have today will become

29:14

something else when all we've had is

29:16

proof that it won't. Do you see

29:18

the obvious cracksks in the wall here?

29:20

No matter how strenuously people like professional

29:23

credulous dipshits or other big publications try

29:25

to pay over them, does any of

29:27

this make sense to you? Because I,

29:29

even when I try and steal my

29:31

own arguments, can't wrap my head around

29:34

how any of this survives, let alone

29:36

becomes an industry where the biggest player

29:38

has annual revenues greater than some major

29:40

industrialized countries. And I know some of

29:42

you, the emotions a lot, and I

29:44

know the aggressions a lot. I'm frustrated

29:47

because I truly believe this stuffs falling

29:49

apart. I truly believe that this was

29:51

never really anything. While I'm saying this,

29:53

Kevin Roos is in the New York

29:55

Times going, I believe that AGI is

29:58

my friend. I believe AGI will rise

30:00

up. the ground and hug me in

30:02

the way that no one ever has.

30:04

I think that's disgusting on multiple

30:07

levels, but I also think it's

30:09

genuinely irresponsible. I

30:12

think all of this is. I think

30:14

when this collapses, we're going to have

30:16

to look back and take inventory

30:19

of how we got here. And I need

30:21

you to, in the next episode,

30:23

listen to it, through the kind

30:25

of lens, that's how lenses work.

30:27

I need you to just stick with it and

30:30

realize that all of what this is is

30:32

trying to show you and hopefully other people

30:34

that you talk to, how silly this is,

30:36

how ridiculous this is, and that we have

30:38

a major problem in tech and business media.

30:40

We have a problem where people can

30:42

come out and just say whatever. The Charlie

30:45

Brown had hose of the tech media. And

30:47

it's disgusting to me because there are

30:49

startups that could use this money. There

30:51

are better things to be done with

30:54

this money. Perhaps they're not hyper-growth markets,

30:56

but there are things that actually exist

30:58

that could be piled into. Instead, we've

31:00

done this to make companies look like

31:02

they can grow, to make Sam Altman able

31:05

to buy another $5 million Konazig car. Is

31:07

that the one he has? Either way, I'm

31:09

not going to lower the temperature on the

31:11

next episode. I'm going to be honest, it's

31:13

going to be just as spicy. But I

31:15

want you to know all of this frustration

31:17

comes from a place of knowing that we

31:19

can do better, of knowing that the tech

31:21

industry could do better. Perhaps it won't be

31:23

as big as it is today in the

31:25

future. I don't know. But for it to

31:27

get better, this shit needs to end. Stick

31:29

around for the next part, where I'll talk

31:31

about how we actually got here, how this

31:33

bubble got inflated, and how nasty the result

31:35

could be at the end. Thank

31:44

you for listening to Better

31:47

Offline. The editor and composer

31:49

of the Better Offline theme

31:51

song is Matosowski. You can

31:54

check out more of his

31:56

music and audio projects at

31:58

matosowski.com. m-a-t-t-o-s-o-s-w-s-s-k-i.com. I can email

32:00

me at easy at betteroffline.com or

32:02

visit betteroffline.com to find more podcast

32:04

links and of course my newsletter.

32:06

I also really recommend you go

32:08

to chat.where is your ed.at to

32:10

visit the discord and go to

32:12

r slash betteroffline to check out

32:14

our read it. Thank you so

32:16

much for listening. Better offline is

32:18

a production of cool zone media.

32:20

For more from cool zone media,

32:22

visit our website coolzone.com or check

32:24

us out on the iHeart. Apple

32:27

podcast or wherever you get your

32:29

podcast. Time

32:47

is precious and so are our pets.

32:49

So time with our pets is extra

32:52

precious. That's why we started Dutch. Dutch

32:54

provides 24-7 access to licensed vets with

32:56

unlimited virtual visits and follow-ups for up

32:58

to five pets. You can message a

33:01

vet at any time and schedule a

33:03

video visit the same day. Our vets

33:05

can even prescribe medication for many ailments

33:07

and shipping is always free. With Dutch

33:09

you'll get more time with your pets

33:11

and year-round piece of mind when it

33:13

comes to their vet care. In

33:17

a world of economic uncertainty and

33:20

workplace transformation, learn to lead by

33:22

example from visionary C-sweet executives like

33:24

Shannon Skyler of PWC and Will

33:27

Pearson of Ihart Media. The good

33:29

teacher explains, the great teacher inspires.

33:32

Don't always leave your team to

33:34

do the work. That's been the

33:36

most important part of how to

33:39

lead by example. Listen to Leading

33:41

by Example, executives making an impact

33:43

on the I Heart Radio app,

33:46

Apple Podcasts, or wherever you get

33:48

your podcasts.