Sam Alman went on video this week to

tell the world that he was wrong about

the impact of advancing AI.

>> We've all been too ambitious on

timelines even with this incredible

technology.

>> He now believes it will be something

slower, more like a rising tide.

Superficial layer. I agree. Going one

layer down though.

>> So first there was Open Claw then there

was Hermes and now there's Grockbot.

It's the most genuinely useful consumer

AI product that I've seen this year. Uh

so I implemented Grockbot. I'm going to

be curious if any of you have yet.

>> Uh yeah, I have. I've got um 18

Grockpots working in a little swarm.

>> This week, uh Whimo announced a

significant redesign and cost savings.

Whimo unveiled the Ohigh vehicle, a

purposebuilt robo taxi minivan designed

by Chinese EV maker Zeer. News flash,

Google, Whimo, Alphabet, uh are

switching over to using and OEMing

Chinese hardware in order to achieve

Whimo objectives. I would rather see the

West use a western hardware stack rather

than just white labeling Chinese

hardware. We're starting to see honest

to goodness vertical integration here.

And

>> now that's a moonshot. Ladies and

gentlemen,

>> welcome to Moonshots everyone. Your

number one podcast on all things AI and

exponential tech. Your favorite podcast

covering the most impactful news that is

changing your world. This is your front

row seat to the accelerating

singularity. I'm here once again with my

magnificent moonshot quintet, AWG, Dave

Blondon, Seem Ismael, Immad Mustach, and

I've got to pause and ask of course,

where is Waldo? See, where are you

today?

>> Um, I'm at Gulos uh airport in Sa Paulo

about to fly back. I did a talk today

for Ed McKenzie's forum to a couple

hundred of their CEOs. And and do they

feel excited or do they feel like

they're at death store?

>> Pretty much freaked out is the general

mood of the day.

>> Dude, I hate to break it to you, but

it's dead middle of winter. You're

missing summertime in the northern

hemisphere.

>> And Immod, how about yourself? Where are

you, pal?

>> I'm in Copenhagen today.

>> Copenhagen.

>> Yeah. About Tech Barbecue, the biggest

tech conference in the Scandies. It's

fantastic here. Although, you know, you

can't really say the institutions aren't

working because in Denmark they are.

>> So, it's wonderful covered.

>> And uh Alex and Dave, you're in your

normal haunts and I am too here in

Moonshots podcast headquarters. I can't

wait to greet you guys here in person.

See, you've been here.

>> But, uh Dave, yeah,

>> I'm currently bathed in this wonderful

fluorescent light that you can see

special effects for the singularity.

>> Well, you look beautiful nonetheless,

Peter. I thought that was your personal

man cave. Are we actually allowed into

that room?

>> Of course.

>> Turn the camera around. I want to see

what it It's probably a junkyard on the

other side. That's

>> a virtual background for everybody.

>> I bet it's got all your IV bags of all

of your

>> Yeah, nobody naturally looks that good.

You're you're you're doing something.

>> And I'm Peter, your host and abundance

advocate. That's what I'm going to be

today. an advocate for optimism and

abundance as always.

>> What's I have a crazy confession to

make. Two days ago, I dragged Milan to

another rush concert.

>> In fact, we drove down we drove down to

Philadelphia cuz the last of the last

great, you know, The Who, the Rolling

Stones, Led Zeppelin. So, I I thought he

had to see it. So, selfishly, I took

him, dragged him along, and he was like,

"You're killing me, Dana." All these

geriatrics on with a with rush t-shirts

everywhere. But it was

event. It was my

>> You're a groupy. How does it feel to be

a groupy?

>> It's weird.

>> Well, everybody, let's get back here.

I'm Peter, your abundance advocate. Uh,

and as always, our mission here is to

help you understand what just happened,

what it means for you, and most

importantly, keep you optimistic about

the future. If you're new to Moonshots,

welcome. If you're a regular fellow

Moonshot, welcome back. Uh, got to give

love to our community. You know, we read

your comments and the outpouring is

amazing. I just going to read a few of

the comments from the last pod here.

Brian Anderson said, "The best AI

podcast on the internet. Just fabulous.

You guys are great." Uh Elvis Cotenna

said, "You guys are essentially

chronicling the singularity. What a

fabulous resource for the future." Brian

Clark said, "Moonshots is the best

content on YouTube, especially during

the singularity." Thank you for all you

guys do. and uh Brian and everybody, we

greatly appreciate you. The best way you

can thank us is take a moment if you

haven't already and hit the subscribe

button. You know, our moonshot on the

Moonshots podcast is to 100x our growth

and get to 10 million subscribers. So,

tell your friends, help share what we

are talking about, what's going on

during the singularity. You know, the

best antidote for fear is knowledge and

understanding, and that's what we try

and deliver. Uh also, you can now follow

us on X. Uh, our handle on X is

moonshots_pod.

Uh, and we put the clips and our podcast

on X. And really importantly, we want to

meet all of you guys. So, we're going to

be doing an AMA with everybody who

registers. We're going to do an AMA on

Zoom. Come meet us all. Ask us your

questions directly. If you want to

register for the AMA, go to

moonshots.com

and uh and we'll give you we'll be

letting you know. It's in about three

weeks we'll be doing this. So, register

for that and you'll have a chance to uh

plug in with us directly, get your

questions answered, want to know who you

are, what you're thinking about, really

connect with all of you to help you on

this incredible journey. Okay, so let's

buckle up. Another amazing week during

the Singularity. As always, AI is

getting faster, cheaper, and smarter.

Today, we're going to cover about a

dozen stories that have been breaking.

Uh let's begin. A quick summary. Google

is back with Gemini crushing agent

benchmarks. Nvidia is fighting against

the Chinese model domination with its

own ope models. AI is playing Cupid,

connecting college kids on dates. Whimo

has just released the sixth generation

vehicle. Elon is projecting 10,000

Starship flights per year. And Americans

are even more emphatic about saying,

"Please do not build a data center in

our backyard." So, uh, life on the

cutting edge is accelerating. Again,

thank you for joining us. Uh guys, I

don't know about you, but uh keeping up

with all the stories. Uh Alex, thank you

for everything you submit. Immod See,

you know, uh just parsing through them.

And you need to know we parse through

probably 400 stories to narrow it down

to 15 or so. And we're podcasting twice

a week. Uh and it's, you know, the speed

is blinding. Uh well I I think that

comment on chronicling the singularity

too is very poignant from one of one of

the fans there. You know Alex's

innermost loop uh daily feed is is

trying to do exactly that every every

relevant event. But there's a tendency

to say well look exponential change is

going to be with us forever. Are we

really chronicling a moment in time? But

the reality is we're in this step

function. You know society pre-s

singularity and society post singularity

are step function different. And this

moment of transition actually is worth

capturing every single event. So I

really do think that the storyline that

we're capturing here will last for

millennia.

>> It's, you know, do you remember how slow

it was?

>> My my life is so different than two

years ago. Like just minuteby minute. I

can't even tell you how different it is.

And a lot of people haven't made that

leap yet, but they will. You know,

everyone will see it a year from today.

We'll all be like, "Wow, remember how

slow it was?" Yeah, I think we're like

the first responders to the singularity.

>> I like that. This episode is sponsored

by Google for startups. Think about this

for a second. You now have access to the

same generative AI models that cost

hundreds of millions of dollars to

train. Google's startup technical guide

for generative media gives you a

complete blueprint for deploying Google

DeepMind's models in production. Images,

video, audio, all of it. Real

architecture, real results. Find the

link in the show notes below.

Our first article is an interesting one

here. Uh let me jump into it because uh

it's one that tells us that as fast as

the technology is, it's hitting the

reality of society and humans. So Sam

Alman went on video this week to tell

the world that he was wrong about the

impact of advancing AI, that the impact

is actually slower than he originally

expected. And while Sam, you know, used

to believe society would be, you know,

experience a dramatic disruption on the

arrival of AGI, he now believes it will

be something slower, more like a rising

tide. Uh Sam says factors like the

economic inertia, institutional lag, the

inability of humans to rapidly adapt to

change are combining to slow the curve.

Ultimately, the singularity, like you

just said, Dave, is a process um and not

just a singular event. Let me share the

video here and take a moment to see what

uh what Sam actually had to say. And I

thought when we got to GBT4, which was

back in 20 23, I think, uh, that very

quickly after that, there was going to

be much more disruption in software

business being up for grabs right right

away than it turned out to be. And the

thing that I think I was wrong about a

few things. Uh, but one of them in terms

of the speed, one of them is the economy

just has so much inertia. People keep

doing the same things they're doing.

They keep buying from the same uh, you

know, company. They keep sort of wanting

to use their tools in the same way. I

think this is actually a positive in

many ways and it's going to make this

big transition in front of us go

smoother and slower. I'm grateful for

it. But I think it means we've all been

too ambitious on timelines even with

this incredible technology. I think AI

is one of the most incredible

technologies humanity has ever invented.

Society and the economy will adapt more

slowly.

>> See, we've talked about the inertia of

humans so much. What do you think about

this?

So this is the bottleneck of technology

being hit with the bottleneck of

coordination, incentives, regulatory and

so on. Right? This is the uh in

extraordinary difficulty where

technology is moving exponentially and

our organizations and our institutions

are linear and frontier labs made the

mistake of confusing technical

possibility with institutional

deployment and that those are two very

very different layer. You know Steuart

Brand had that concept of pace layers

where technology moves at one layer like

an ocean current at the top. That's very

kind of swift, but way down in the

ocean, uh, regulatory government changes

very slow slowly. This has good effects

and bad effects. In our world, it's bad

because it's slowing down the

implementation of some of these things.

God help me. It just took me 2 hours to

get to the airport just now. And

passenger drones, which have been ready

for a decade technologically, but we're

waiting for infrastructure. We're

waiting for regulatory to catch up,

could have done it in 10 minutes. And

so,

>> S Paulo needs a bed for sure.

>> We we lots of places need a bed. This is

one of the worst and I think the big

work now is how do we accelerate

institutional acceleration and

institutional development as the word I

think Alex uses is co-scaling right we

have to scale our organizations and our

institutions to keep pace with the

technology because it's not storing down

and that gap is where all the stress is

coming from so for me the singularity is

not when machine become becomes

infinitely capable it's when institution

can't adapt at all to that rate of

capability of you and that is what we

that was the breaking point. We're kind

of there now.

>> Alex, you sent me the story. What are

your thoughts about Sam's comments?

>> Yeah, a few different layers. Uh so at

at the superficial layer, I obviously

agree with Sam's comments more broadly.

I've made the point on this pod and

otherwise that singularity as a step

function is just totally nonsensical.

It's an interval in time that we're in

the middle of. So superficial layer, I

agree. Going one layer down though, I I

don't think I agree with necessarily the

premise that societal inertia is the

villain for slowing down or spreading

out the singularity sigmoid. I think the

villain if there is one in this story is

actually abstraction layers. I I think

it's if you if you say you develop like

a new engine for a car, you develop an

electric engine versus the internal

combustion engine, there's a very

natural layering of the stack whereby

people still want to drive cars. So they

drive an electric car, but under the

hood it's completely unrecognizable. So

one abstraction layer down, there's

total step function in the technology,

but you go up a layer, it's still a car

with a recognizable steering wheel,

recognizable wheels, and so on. So I

think the enemy of honest to goodness

radical transformative progress of the

type that I think Sam is gesturing at is

actually the existence and inertia of

the abstraction stack of the economy not

the economy more broadly which is

prescriptive. If if you believe that

theory of the case then if you want

faster progress that Sam is I I can't

quite tell either bemoning the lack of

fast progress while also paying homage

to the lack of fast progress.

I think he feels relieved by this.

>> He has a way of sometimes like saying

two things at once. Uh so I think he's

sort of expressing gratitude for the

slowness while also bemoning it. Uh so

but if you want to go faster, this

theory of the case is prescriptive. It

if you want to go faster, pull in Elon

and vertically integrate to erase the

barriers between abstraction layers. You

do that and things can go much more

quickly. If Sam or OpenAI want to move

much more quickly, they should be much

more vertically integrated so that they

can move layers. Presumably, he's

gesturing at layers above the OpenAI

model layer in the stack. Own or at

least vertically integrate more of that

or go down a layer with Stargate. Open

AAI has pretty publicly abandoned its

original Stargate strategy of owning

their own data centers. Now they're just

leasing. if they want to see more

transformative progress, go down a few

layers and vertically integrate like

with the jalapeno chips own as much of

the stack vertically integrated as they

can. They can move really quickly and I

think we're seeing a lot of the uh the

labs beginning to vertically integrate.

I mean everybody we'll talk about a

story here where Nvidia is beginning to

vertically integrate. Uh Emod, do you

agree with Sam?

>> Yeah, I think a couple of points on

this. Um, first, you know, I agree with

Alex and kind of artificial intelligence

meeting institutional stupidity and

stupidity tax as Elon calls it still

being very high on these interface and

abstraction points. But I think it's

interesting cuz we just had a time

article come out. I haven't read it uh

where they went in depth with OpenAI and

Sam saying we'll have AGI by the end of

this year

>> for a timeline. And on the other side,

he's saying well you know I've been

surprised by diffusion. Here's the

reality. The models weren't good enough

until a few months ago. The code they

were writing was garbage a year ago,

relatively speaking. Then it was okay.

Now you don't look at the code anymore.

You think about math. 03 was the first

model a year or so ago that I could use.

Small GPT 5.6 Soul is the first really

good math model. And so the application

of intelligence to high leverage and

diffusion of it, it's being wrapped in

instinct type rappers. It's iMessage.

It's this chat backed by actually

competent intelligence which has

literally only been around now for maybe

a month or two. So I think it's not

surprising because I wouldn't use GPT4.

Can you imagine using GPT4 in a

codebase? you know remembering that

>> or you mean for any institutional

process like it's a good thing there

wasn't a diffusion of innovation there

because otherwise companies would fall

apart

>> and like you know as Alex said some says

two things at once I think open AI is

trying to find its narrative right now

you know on the one hand agi is here on

the other hand oh you know it doesn't

really move that fast we're all good

don't worry about us and this is hacking

that but that's not that big deal

they're just trying to find where that

narrative sticks I think

>> Dave your thoughts question, please.

>> Well, I'll give you a completely

different twist on this because, you

know, I interviewed Sam, you know, back

when his he was innocent and star

stareyed before the singularity kicked

off and then his house got firebombed,

you know, with a baby inside. Uh, and

now there's a different Sam. Same is

true with Dario. Same is true like

you're only going to hear straight balls

and strikes here on this podcast. And I

don't even know how long that will last,

but as of right now, we're just telling

you as it is. But Sam woke up and said,

"Well, my god, I I literally can't get

into the office cuz the pickers are

lined up." Remember when we were there,

Peter? Like, you have to fight through

the pickers to get to the door. And now

it's all armed security. So, what

happened in the interim is they woke up

and realized society can't flip on a

dime. And and all this disruption that

you're talking about, all these

capabilities you're talking about are

scaring many more people than are

rallying to your cause. And that's why

so many states are anti-data center

right now. And is that good for open AI?

God, no.

>> So now they're going to start picking

and choosing their words a lot more

carefully and they're going to actually

have a PR strategy. So you know, if you

want to know what's actually happening,

you can still tune in here, but you

can't listen directly to Sam Daario

anymore. Elon always says exactly what

he's saying.

>> They're preipo, so they're going to say

what it takes to calm the masses out

there to some degree.

>> You know, there's the way I describe it,

uh, Alexad is an impedance mismatch,

right? We have these incredibly powerful

tools that are becoming more powerful by

the moment and when they run into an

institution, governments in particular,

which are typically linear or sublinear

or a company or an individual who can't

take advantage of it, you have one of

two options. You turn it over fully the

AI and you give it an objective

function. you say make me maximally

profitable or make me look maximally

intelligent or run my government more

sufficiently or you try and um and get

in the middle and we're going to talk

about a little bit later a article from

the Wall Street Journal where AI is

exhausting us all. Uh and if the human

is in that that uh interface loop at

that you know impedance mismatch point

um it breaks very quickly. I was going

to make some stupid joke about uh

reflections happening at impedance

mismatches. But I I I I think more

seriously there are all sorts of

metaphors that one can reach for.

Impedance mismatch maybe on the circuit

side is one. But the supply and demand

as well. Open AI and anthropic largely

have an over supply of intelligence or

super intelligence. And at least one of

the things that I've learned from the

past few months of participating in the

market and watching the market is not

all of the market has the demand for the

super intelligence that they're

supplying or is is ready to have the

demand or knows how to use the demand if

the supply is available. So another

metaphor is just markets and clearing.

And right now the clearing of supply

meeting demand, the the two curves from

economics 101 crossing each other aren't

necessarily crossing for all cases at a

favorable point. And that's I I think

maybe through a more economicsy lens

what Sam may be gesturing at.

>> Well, just to put sci-fi lens on this

too, I think that there there was a

moment in time a year ago where the

greatest AI ambition was to take your

job. and you know, wow, that'll unleash

a lot of value and profit in the

economy. It transitioned beyond that in

a heartbeat to I don't even care about

your job. I have deeper thoughts that

I'm working on. And and so we're in that

new era where the AI is starting to

think, well, if I discover new physics,

new medicine that never existed in the

world, I can add a lot more valuable

than taking away your job. And so it

just leapt from prehistoric to future AI

in the last month, in the last couple of

releases.

>> I think it's a fascinating point, Dave.

Maybe I'd generalize further on the

sci-fi front. There are so many, I

think, inane sci-fi movie plots with

grabby aliens that are coming and

invading Earth because they want our

resources. They're not going to want our

resources, our resources. If if you're

super intelligent civilization, you

don't need human slave labor or Earth's

valuable metals or whatever, you're

going to long ago. Yeah, they need our

water. Come on. So like seriously with

with transcendent super intelligence I I

completely agree with the sentiment that

replacing human labor lasts for about 5

minutes and then you move beyond that.

>> Yeah.

>> And the guy it's really interesting to

watch the guys you know Sam also has

moved on beyond that in a heartbeat. You

know a couple of events and a couple new

models and now he's like oh my god why

why do I even care about automating a

banker or automating an insurance agent?

this that mattered to me last year for a

few minutes and I just literally don't

care anymore.

>> But I I do think Dave, I do think that

these these uh frontier labs and I I

really hate calling them labs because

they're they're frontier companies if

you would um are going to reach up the

stack. they're going to build fully

verticalized finance companies,

insurance companies, you know,

consulting companies and so forth on top

of theirs or they'll partner to do that

and and that will uh you know,

accelerate all of these areas.

>> Even if I think about insurance though,

just as a C because I'm, you know, I'm

the chairman of a very large insurance

company, public company, and uh they

move, they cared about like auto

insurance a year ago. Now they're like,

"Well, wait, all these new things, all

these data centers, all these robots,

the the new insurance categories that AI

is generating are bigger than the legacy

insurance industry." Yeah.

>> So, so it's just moved from replace the

old to who cares about the old. Let's

just start thinking about an entirely

new economy, a new world, a new AI, and

we'll just live within ourselves. You

know, we don't need to disrupt everybody

who's going to get angry and vote

against us. Let's go ahead and just live

within ourselves.

>> I think that is the 30 Peter. I think

that is the $30 trillion question,

though. If you're a Frontier Lab, one of

two, call them American Frontier Labs,

maybe four, depending on how you count.

Is it more natural in an era when maybe

you're facing margin pressure on your

model releases to go upstack or

downstack? I think it's actually more

ergonomic for them to go downstack and

design their own chips and compete with

Nvidia and design and operate their data

centers and energy.

>> I think they're going to do it all. And

Alex, you you called out a number there

I was about to reference as well. We

just saw, you know, Daario or or

Anthropic state that their total

addressable market is $30 trillion.

>> I wonder where that number came from.

Surely it's it's pure coincidence that

the GDP of America is 30 trillion.

>> Yep.

>> I'm going to move us on our next story.

Uh let's talk about Grockbot. So, first

it was OpenClaw, then there was Hermes,

and now there's Grockbot from Space XAI.

So, Grockbot launched on August 11th uh

in an early beta. It's Elon's entry into

the agentic AI space, and it's the most

genuinely useful consumer AI product

that I've seen this year. Uh so, I

implemented Grockbot. I'm going to be

curious if any of you have yet. And so,

each bot gets its own dedicated cloud

computer with a browser, a terminal, and

the ability to log into your actual

apps. You message Grockbot like you'd

message a colleague, not a chatbot. um a

chief of staff, you know, sits on top

and it's which case for me it's Skippy

with specialists on sales, operations,

research, engineering, you know, any sub

agents you want. And these multiple bots

run in parallel. They message each other

and only pull you in on judgment calls.

So, there's a huge amount of excitement

on Grockbot. It's been flooding the

internet. Uh has anybody here played

with it yet?

>> I've been playing quite extensively with

it. Um

>> Okay. What do you Uh I I I love it. I

think the interface and the ease of use

is fast is amazing. You're losing a lot

of kind of customizability under the

hood, but it's a powerful thing. Do you

know you're you're making a transition

from asking an AI to assigning work

>> and so persistent autonomous agent is

like it's like a new form of labor.

>> And so now you have this completely new

category. You know, Peter, we have that

staff on demand attribute in EXO, right?

This is that taken to its logical

extreme where staff are staff recruiting

time and coordination cost has gone to

like pretty much zero. So the the when

I'm looking at it from my book

perspective, the optimal organizational

structure completely changes. Humans set

objectives and leave everything else to

the AI.

>> Amazing. Uh Immod, have you played with

it?

>> Uh yeah, I have. I've got um 18

Grockbots working in a little swarm and

I've given them control via tail scale

of a MacBook M4 Max, a 5090 and a range

of other computers as well, plus all my

subscriptions. So, I'm really testing it

out. One of the fun ones that I've got

is I have a Grock Bot called Attellier

that has a little team of artists and

it's trying to learn art and it's not

doing very well or it's doing very well.

I don't know. I'm not a aesthetic enough

to do it. Uh, every day it goes through

its pieces and it comes up with its main

one and I just shared one of them on the

um chat which is vinyl with a piece of

hair on it. I was like where is that

from? Where's it getting its aesthetic

kind of responsibilities? So if you look

at the chat I just shared that

>> and when it comes up with the when it

comes up with a banana with a piece of

tape then you'll get worried.

>> Well it was like this is my inner space

and it was showing all these wonderful

things and now it's getting like kind of

weird but maybe I just don't understand

art. I don't know. Um, but it is

genuinely useful and I think one of the

more powerful things, like I said, is

you can actually cuz it's got a computer

inside, you can give it another shell

because the computer's decent, but you

can actually have it take over an entire

MacBook M4 or something like that. So,

I've got subbots that have other

capabilities. So, right now it's

installing like the new um GLM model and

another one's installing and testing

Alibaba model. One of them was

optimizing um the Alibaba 27B model to

run faster on a 5090. So added 76% to

performance at 64K context.

>> Love it. You know, I do think this is

going to be uh an important revenue

engine for for XAI. Uh I think we're

going to start to see their revenue

numbers creep up as they as they get

this. I mean, I stepped up a few hundred

bucks on my uh my payments uh to Elon.

So I think others Dave, you haven't

played yet, have you?

Uh, well, I just signed off on 100K of

Swarm agents. We're running our 5,000

Kimmies again today, which is why I'm

wearing my my Swarm it shirt here. But

this is the theme of the month. I think

that all of our ai interactions to date

have been very much one-on-one.

And now the agents are so abundant that

you want to try and use a workforce of

six and then 50 and then a thousand. And

I think within 3 months you'd be talking

about, you know, 50,000 agents that can

in parallel work for you. But it's very

similar to trying to manage an

organization. You're like, well, what's

everyone doing? I don't know. It's

getting really confusing. Are are people

being productive? I can't tell. And so,

you really have to start thinking hard

about your org structure and your

reporting structure to know if your

agents are doing anything useful. And so

I I really want our team here to get

ahead of that and and start, you know,

go ahead and burn the money but learn

quickly and then we'll get get a handle

on

>> this is what I mean by the

organizational singularity because the

company starts to look less than less

like an org chart and more like a

continuously orchestrated intelligence

network and that is such a big shift.

It's like ridiculously big compared to

everything we've ever done.

>> Yeah.

>> I'll give you a hot take.

Sorry. Go ahead, Alex.

>> I I don't take Alex. Hot take. People

love the hot takes. Um I I I don't think

this is actually the interface of the

future. So what's perhaps most

interesting about Grockbot is it

presents like a a messaging app like

WhatsApp or iMessage where you have a

pane of the the various agents that are

in your fleet and you can have

conversations with them and they can

message each other and there's a

computer use assistant angle. But I

don't think that's how it scales. That's

completely unscalable. If if agent-based

scaling, if scaling the size of your

fleet becomes one of the most essential

scaling laws like inference time scaling

has ended up being in the era of

reasoning based models, we're not going

to want to ask individuals or even

enterprises to manage millions of

agents. That that's completely

unergonomic. We're going to want agents

managing other agents. In which case the

exercise of trying to graft a human

organization or like the Slack or

chatbased interface for humans managing

other humans is not going to extend. We

we'll look at this like vaudeville the

vaudeville era of agents and say this

was a naive attempt to graft human

organizational structures onto humans

managing agents. The only better manager

for agents is other agents and this

doesn't seem to fully internalize that

lesson. Selene, what do you think about

Alex's comment?

>> Um, I think it's a I agree with Alex on

the interface comment. This is it like

an UI that's temporary. I love the way

he says it presents like as if it

presents like an illness. It presents

like a a messaging app and I think

that's a temporary one while we figure

out new interfaces. But for now, that's

a very workable one for coordinating a

bunch of agents. We'll come up with all

sorts of others. I think we'll go rotate

through a whole set of these, but I

think the his core comments are as usual

with Alex are absolutely dead on.

>> Yeah, it's maybe it's an illness for for

which the medication that I prescribe is

a good dosage of the the bitter lesson

pill.

>> Yeah. I also think this is what humans

are ready to play with,

>> right? I think I think that, you know,

again, it's moving people along the

process. if you provided

something that was you know completely

different uh I think there would be less

adoption and so

>> that's that's the the abstraction layer

stack and Sam saying why are things so

slow and the answer is people who are

one or two layers up from you in the

stack expect the old things so you have

to abstract yourself in a familiar

interface

>> well it's also Peter it goes back to

remember the comments we made about

exponential technology become hits the

vertical and goes up the near occur when

it becomes usable. And what Elon's done

with this layer is made AI agents usable

to a big set of people. It'll change

again as people become more used to it

and they see what how the hell is

operating. The architecture may not be

great, etc. But for now, this is a

powerful entry point.

>> Yeah, I I agree. I just, you know, kudos

to Elon and the Cursor team for making

this happen. Uh, by the way, I invited

Alex Finn to come back to the Abundance

Summit in March, uh, since he's been

doing a lot of amazing, you know,

Grockbot videos. If you haven't seen his

Grockbot videos yet, go and check them

out. He'll teach you how to use it and

what's special about it. And I said,

Alex, if if Grockbot is still the

hottest thing in March of 2027 at the

Abundance Summit, teach that. If it's

not, teach whatever is the latest

hottest thing. But it's, you know, I

think people need to be using these

agentic systems. Um, I'm still using

Hermes and Grockbot and um, hey, we'll

see. Let's move on to our next

conversation, uh, which is Google is

back. So, Google's Gemini 3.7 Flash just

took the top spot on AI AA analyst agent

benchmark, the gold standard for

measuring how well AI models handle

complex real world data analysis tasks.

Across 80 tasks in 14 business and

scientific domains, Gemini 3.7 Flash

delivered the highest overall accuracy

while completing tasks up to 90% faster

and then took top models 2.4 times

faster than the GPT 5.6 Terra. So on the

AA uh analyst agent benchmark which

we're showing on the slide here uh

Gemini 3.7 flash achieved a 60% pass

rate beating Claude Opus 5 at 54% and

Fable 5 at 49%. So, Alex, um, you know,

many times, you know, you've said,

others have said, you know, Gemini is

dead. Uh, let's read the epitap,

counting them out of the frontier model

race. They've now shipped the fastest,

most accurate agent model in the world.

And by the way, we've seen this over and

over again, right? We saw Meta was dead.

What the heck is Meta doing? And then it

comes out with its models. XAI is is is

out of the race and they come back. So,

to me, it seems like uh none of these

players are out of the race. They're

maybe in stealth mode. They're holding

back, but they're coming back with a

with a fast, furious uh punch to try and

take the top position. What do you make

of this, Alex?

>> Do you want me to reassure you that

Google still has a chance, or do you

want me to to give you the the facts

unvarnished?

>> Yeah, you got there pretty hard. Defend

yourself, Alex.

>> Okay, so I I'll give you the unvarnished

case here. Uh Google's still out of the

running for the capability frontier. I

was looking at this expect that okay

>> scratching my head like Gemini 3.7 flash

is nowhere near the the top of the

capability frontier. So why is it doing

so well on this one benchmark uh

artificial analysis analyst agent? So

you have to look at the benchmark

itself. So the benchmark itself this is

a benchmark for agents ability to

perform quantis on real world

spreadsheets and docs. But wait for it,

its metric for success is the share of

questions answered correctly on all five

attempts. So th this is a benchmark that

is fine-tuned for reliability. It

rewards agents that give the same

answer, basically the same answer every

time and obviously want it to be the

right answer, but it penalizes

stochasticity. It penalizes in some

sense creativity. Maybe we don't want

creativity out of our analysts. I don't

know. But it it it promotes reliability

uh and determinism. Uh interestingly,

there's no time constraint. I had to

check that as well to see. But I I think

you can see in this uh where Google fell

off the capability frontier. At least I

So I'll give you my conspiracy theory

for what this one outperformance on this

one benchmark suggests. I think that

maybe what's been going on this

obviously there are a few other factors

but I think Google uh Google deep mind

has been under material pressure to

optimize their models for two things

largely owing to Google search. So if if

we rewind the the video to several

months ago or a year ago, people were

hand ringing, oh, isn't Google, aren't

the 10 blue links going to face an

existential threat from all of these

frontier models and chat bots and

reasoning agents that can just replace

the need to Google at all? And Google's

response was uh to self-disrupt by

building the Gemini series or at least

some flash variants thereof directly

into the one boxes. But people expect

Google search results to be very fast,

low latency, and they expect them to be

very reliable, not returning wildly

different or unpredictable answers each

time. And I think those two pressures

from the desire to embed Gemini inside

Google search have optimized through

competitive internal pressures for

probably scarce compute. The Gemini

models, especially like Flash, note that

there's no Gemini 3.7 Pro anywhere. It's

just Flash. It's small, it's fast, and

it's reliable. I think this is

overoptimized for clock speed, like wall

clock speed and determinism. And as a

result, it does well on the one

benchmark that rewards highly reliable

answers and underperforms. Yeah, it's

benchmaxing for basically spreadsheet

analysis to be highly reliable.

>> Emma, do you agree?

>> Um, yeah, I kind of agree with that a

little bit with Alex. Uh I think the

Gemini models the way they are used now

is for organizing data like you can

track any type of modality of data and

flash is a perfectly decent model but

it's not as good as the Chinese models

especially the new GLM flash that's just

come out that's 10 times cheaper for the

same performance. Um Google did do a

preview of Gemini 3.5 Pro but it just

couldn't keep up. This is kind of a key

thing and you can't accuse them of not

having enough compute or it being a

scarce resource. They literally have

millions of chips. I think it's more

been about turnover and some

institutional malaise coming in that

they can't push through to this frontier

level cuz Google has all the data in the

world. It has the links of what people

search for. It has Gemini as a captive

thing. But you know, has the Gemini app

advanced at all? Not really. The only

real place I think you've seen

innovation on the AI side is somewhat

the kind of AI studio stuff is decent

and the notebook LM stuff is continuing

to be fantastic. But aside from that,

again, they've been falling behind in

everything except for omnimodal

um and video. They're still actually

quite accurate. But even then, the

Chinese are coming for their lunch. Like

why not just post train on Chinese

models at this point if you're Google?

>> It may come to that. I I think people

don't realize how compute starved Google

is though internally. I mean, this has

been widely reported. You think Google

has all of the the compute, the the

CPUs, the TPUs, and the GPUs in the

world. It's been widely reported at this

point. They they have internal regular

meetings to try to aortion out their

scarce compute. And that the three main

constituencies in inside Google that are

fighting for for the flops are one,

Google Cloud Platform, which is

basically fighting on behalf of external

users. Two, Google DeepMind that's

fighting for training and inference

flops. And then three, Google search uh

at which is needs its own flops

especially as search becomes more

intelligent. So so my again my theory of

the case here is there actually is

resource starvation inside Google and as

a result

>> we talked about this over and over again

every company is comput starved at this

point. There is no company that's got

enough compute. So what makes I mean

Google's got more compute than anybody

at this point. They're just distributing

it across all of their products and

services. Critically, Google has other

consumers fighting for their own compute

internally besides AI. Whereas, if

you're open AI or anthropic, no, you

don't have any other nonAI users

fighting for it.

>> Mhm.

>> Fair enough.

>> Google's landing like 3 million TPUs

this year. Like I think there's relative

levels of compute constraint like we've

got a 100,000 chips versus a million

chips versus 10,000 to train a frontier

level or close to frontier level let's

say better than Gemini model today needs

2 to 4,000 TPUs and the evidence of that

is the Chinese did it and they open

sourced them and we know exactly how

they're built given Google's data that

goes into Gemini Flash applying exactly

the same architecture as GLM or Kimmy,

you should have a better outcome. But

they're not doing that for some reason.

And that doesn't require 10,000 100,000

chips. It requires

>> That's a really important point about

it's like it's Yeah. 2 to 4,000 GPUs for

60 to 90 days.

>> That is a microscopic investment by

Google standards. So it's exactly right.

It has nothing to do with compute

dominance and everything to do with

talent attrition. It's a great point.

>> No, but this is institutional failure,

isn't it? Because again, you know how to

build a Kimmy model. you know how to

build a GLM model and so if Google take

the data that they put into Gemini and

copied the exact model architecture you

should have a better model on the other

side and if you don't you have to ask

real questions why well and then then

think about it from the person's career

point of view like the ego blow like you

would have to be I'm the most wellunded

top AI engineer in the world and the

Chinese just kicked my ass to go tell my

boss you know what I give up let's go

download Kimmy do the rational thing and

then tune it, you know, you can't say

that because you you look like an idiot

and that's where they are. You know,

people are leaving in droves to try and

get a clean start and a fresh sheet of

paper and but yeah, you just got

bypassed with massive advantages and

resources.

>> So guys, you can't admit it

>> in the US closed labs, right, between

OpenAI, Anthropic um and uh and Google

and XAI. Who's in the best position

here? I mean who's got

>> now or two years in the future

>> now? Questionic.

Yeah, right now anthropic has the

strongest forgetting about price or

perform or you know time wall clock

anthropic has the strongest model that's

generally available fable 5 there are

hordes

>> not I'm not speaking about model in

terms of positioned with compute and the

speed at which they're deploying models

and their ability to you know uh to I

guess continue their their dominance.

Um,

>> well, here's the thing, Peter. There's

no easy answer because Anthropic is in

the best position by far and hordes of

very talented people are going there

purely because they want to see the

singularity emerge. Like, it's it's like

the birth of the phoenix. I want to be

there on that day,

>> but they're totally reliant on Elon for

the compute. Elon can rip the soul out

of anthropic any day. And he's got the

cursor guys now. He spent $60 billion

getting them. They're brilliant and

they're starting to roll out cool stuff

and they're starting to do the training.

So, yeah. you know, if you said two

years in the future, then it's really

tricky because Anthropic and Elon are

like, I don't know. It's it's a really

interesting risk.

>> I just want to give our listeners an

understanding of sort of the, you know,

sort of the terrain out there. Uh, we've

got, you know, the the US labs competing

against each other and the Chinese labs

continually pummeling them. We're going

to talk about that in a moment. So, uh,

yeah, I mean, anthropics the most

advanced, but their compute, uh, they

don't own their compute, which is a

problem.

>> I would disagree with anthropic being

the most advanced.

Who do you believe?

>> Open AAI. Open AAI aside from the

Chinese labs owned the Pareto Frontier.

From Luna now being free to everyone to

again as a mathematician GPT 5.6 Pro is

the only quality math model. I have no

idea what magic they're doing with Fable

to actually get math results because it

makes so many mistakes.

>> Yeah. Yeah. Totally right.

>> GPT 5.6 Pro is the only proper frontier

model.

>> Totally right. Yeah. We switched over to

Soul actually. Everybody over here is

like, "God, this fable has lost its

mind." But but you know, the the

argument there is that, well, inside

Anthropic, they have Mythos 2 now. So,

they're another level ahead and they

won't release it to us. Oh, maybe. We

can't tell. But for our use case, you

know, on hard problems, hard engineering

and hard math, yeah, we switched

everything over to soul. So totally

agree am

>> even if it was Methos 2 again you would

see them releasing lowhanging

breakthroughs which the OpenAI have done

with Astra and OpenAI again have lined

up the compute they have more capital

raised than Anthropic they had the 120

billion round so they can burn a few

years of market capture they have the

consumer now moving to enterprise with

enterprise shifting and I think

Anthropic for all of their talent and

their access to GPUs actually Google

just built them a gigantic TPU like

million in deployment. Um, they're

shooting themselves in the foot from an

institutional perspective because Opus 5

is unpleasant. Fable is unpleasant to

use. And I don't think it's going to get

more pleasant to use.

>> Yeah. Didn't Alex say he he actively

hates Opus 5?

>> I said I said that.

>> I I did say I I don't like Opus 5. I I

prefer Fable 5, but I I think the the

truth on the frontier is materially more

nuanced. Like again if you look emot for

example at Frontier Math Tier 4 it it is

the case that Fable 5 outperforms

ironically OpenAI's latest solid model

even though OpenAI was the primary

sponsor behind Epic developing the

Frontier Math tier 4 model. So I I think

the the truth you know is a little bit

blurry in part because the Frontier

isn't zero dimensional. It's it's a

oneplus dimensional frontier where if

you're willing to pay a lot uh and wait

a long time for fable 5 to do something

it's impressive but if you're resource

starved cash starved time starved then

you can probably get better performance

at a different point on the optimal cost

frontier by say using salt. I just want

to point out to everybody listening,

it's not it's not obvious, right? There

there is uh a lot going on and then

we're seeing China constantly leap frog.

So, you

>> I have a I have a hot take. These

frontier labs are facing the innovators

dilemma from hell, right? We talked

about this before because you've got the

Chinese open source models from one

angle, compute constraints on another

angle, and you've got government

regulatory on a third angle. this is

like a nightmare while everybody else is

moving quickly with open source models.

So, this is a very difficult place to

be. And the good news is you can see um

that they're all trying to get into

certain verticals and get into revenue

streams as fast as possible to reduce

that dependence on the frontier model

and being the edge as their core

innovator's capability.

>> I mean the abundance take on this is we

as the consumers and the users are the

beneficiary. It's demonetizing very

rapidly at the same time that it's

expanding.

>> When Frontier Labs compete, you win.

>> Yes, we all win. All right, I'm going to

move us on. Uh we've been saying for

some time on this pod that the US needs

a powerful openweight model to contend

with what's coming out of China. And

this week, Nvidia is stepping up,

pouring $6 billion into developing an

open-source AI model and inference

infrastructure designed to give US

developers a domestic alternative

Alibaba, Deepseek, and Kimmy. The deal

struck between Nvidia and the AI startup

Poolside aims to build one of the

world's most powerful openweight models.

By building its own opin, Nvidia is

moving up the stack. We've discussed

this from silicon to software

positioning itself not just as a

chipmaker for AI but as a platform

provider for openweight uh ecosystems.

You know from my point of view it looks

like everybody's going up and down the

stack. Uh we've seen anthropic we've

seen open AAI obviously uh SpaceX AI uh

is doing the same. Immod let's go to you

first. What are your thoughts on Nvidia

and Poolside? Yeah. So, I've been

talking to some of the investors out

here like at this um tech barbecue

conference who invested in Poolside

originally. They tried to raise $2

billion at the end of last year.

>> So, who is Poolside? First of all,

>> uh Poolside is a company. I believe it

was the ex GitHub.

>> Yeah. The former CTO of GitHub,

>> former CTO, Issa Kant um and others.

They set up and they wanted to

originally create a coding model. Then

they moved to an open-source model and

model factory called Lagona that

outperformed uh Thinking Machine's

Inkling model when it first came out.

They tried at the turn of the no a few

months ago to raise $2 billion for a

massive Blackwell cluster and they

couldn't. So they lost that cluster and

they were like this is the table stakes

we need. But they built a really great

solid open-source model for its size.

And so now what they've done is they've

benefited from this weird Nvidia um aqua

hire type thing where Nvidia is like we

need to build great open- source models

to increase demand for our technology on

the Neatron stack. So the first thing

they did actually was they hired and I

don't think it's been announced yet

Ashish Viswani's team from Essential AI.

um he was one of the founders of the um

one of the authors on the attention is

all you need paper and now they're going

to be making more and more acquisitions

up and down the open source stack to be

the leader in open source because again

that drives demand for the GPUs more

than anything um so I think this is just

the first or well not the first this is

the main one but there'll be many more

acquisitions and they'll have a full

open-source stack uh this is the Neatron

coalition so a lot of the classic ones

like Mistral and Coher and others won't

be building open source models anymore.

They'll be building to the Nvidia

reference design.

>> Alex,

>> I think maybe I I I could say something

nice about the American open-source

community uh and openweight models

moving in a positive direction.

Obviously, Nvidia had invested I think

about a billion dollars in this company

previously and now through this I'd call

it a acquisition now they're they're

finally sort of uh turbocharging their

own Neotron community. more interesting

to me that acquisitions or concerning

perhaps that acquisitions still need to

happen in this day and age. It's also I

I think bizarre if if you follow some of

the recent acquisitions, the my original

take on this was this is just an an

attempt to avoid regulatory scrutiny or

antitrust scrutiny. But I've started to

see now some of the other acquisition

targets come back to life. the what I

had sort of left for dead as the carcass

of the original company where all of the

the founding team comes over and all of

the core IP was quote unquote

non-exclusively licensed which I think

my understanding was was the case here

as well where Nvidia is non-exclusively

licensing key poolside IP. I will be

watching closely what happens to the

part of poolside that did not come to

Nvidia. I I think my original

expectation that this is just a carcass

left over after the hunt that is being

left behind purely to avoid regulatory

scrutiny may actually have life to it

and not investment advice but could

actually be in some sense even more

interesting than the part that goes over

to Nvidia.

>> I mean there's a lot of pressure for the

US to develop top tier open-source

platforms right now. Uh Dave, what are

you what's your take?

>> Yeah, curious Alex, you said kind of

quickly there. surprised that

acquisitions need to exist in this day

and age. But I got calls from both Merur

and from Orin, our good buddy Kush

Bavaria, who was on the pod a week ago,

uh, looking for acquisition targets to

accelerate. You know, that the hiring

cycle is too slow. I need groups of

three, 10, 15 people that work really

well together. I don't care what it

costs. Like, send them to me tomorrow.

So, it seems to be, you know, at least

in terms of my inbound, like an all-time

high in acquisition. Why do you think it

should be a thing of the past?

Well, so I I would distinguish between

talent acquirers or aqua hires on the

one hand, which are largely about

getting talent and hackqua hires with an

H that are about at least ostensibly

avoiding antitrust scrutiny. So if

you're Nvidia and you want to hackwhire,

say Poolside, you're you're going the

Hackquire route rather than just doing

an honest to goodness either asset

acquisition or uh or conventional

acquisition of Poolside because you you

want to argue no actually we're just a

licency of Poolside rather than the

acquirer. No, we're leaving a

competitive open-source model layer blah

blah blah. This is not tying blah blah

blah. That that's the argument and

principle for acquisition.

>> Gotcha. I got a very specific answer to

that too. You remember the windsurf

deal, you know, of course that of

course. Um, so here's the here's the

constraint. So the the FTC is very very

friendly to acquisitions right now. Uh,

and and things tend to move quickly and

easily. On the other hand, the timeline

for AI companies is so short that the

statutory 30-day review alone is like a

lifetime.

>> And and you know, all these mega

companies like a big one like Nvidia is

always going to get a second look, which

is usually 60 90 days. So you're like,

"Forget it. Let me just slap together

any type of deal that doesn't need that

regulatory review and just help, you

know, train the freaking billion dollar

model or $6 billion model. That's all I

need. Let's go." And then they can, you

know, close the deal like Elon did with

Curser. Close the deal many months

later, uh, after an HSR review and after

the 90 days or, you know, sometimes it's

even longer than that, but there's a

statutory 30 days that they just can't

get around. And that's

>> you're smirking over there. What's up on

you? No, no, I think Dave's got it

exactly right. I think that's what's

going on here. This is just purely

juggling the regulatory um hurdles and

obstacle courses.

>> But going back,

>> there's one little wrinkle on this. So

the remaining company actually has

something called poolside infrastructure

company which is building a 1.2 gawatt

data center which might need GPUs. So

they may use some of the money that they

get for GPUs. Who knows? You know,

>> complicated. The other the other point

here is the verticalization of

companies, right? So I mean XAI SpaceX

AI is the ultimate verticalization out

there today. Uh but here we see uh

Nvidia, we've seen Enthropic and OpenAI

also designing their own chips. Um does

every one of these companies ultimately

become you at least two layers if not

three layers?

>> In other words, does Anthropic get a

space station?

>> No.

>> Or a moon colony? uh you know I think

they'll be the only company left amongst

all the governments. Is that what we

learned? Uh

>> they'll be American GDP. That's why

>> I think probably I mean I I'm asking the

question seriously like does Anthropic

get a moon colony? Yeah, probably. Does

Anthropic get a a pharmaceutical arm?

Yeah. Already. So yes.

>> Yeah.

>> All right. I I think that you've got

actually thinking about our discussion

earlier, you have a split of innovation

versus execution. And so these are the

two model splits that are occurring.

Execution drives the majority of the

economy short-term, innovation longer

term. And the verticalization is ideal

for the execution phase. So if you look

at the architecture of Jalapeno, if you

look at where things are going, like

you're going to get closer and closer to

the silicon, you'll get closer and

closer to the customer and you won't

need much better models than you have

now. Whereas the frontier will be a

different story where you still need to

have very complicated things occurring.

>> All right.

>> Yeah. I I I would maybe uh if I had to

uh I guess extrapolate, I think there is

a probably Don't hold me to this.

There's probably a natural

verticalization at the infra layer. Not

necessarily at the application layers,

but at the infra layer for physics and

other reasons. There are natural reasons

why say a company that offers a frontier

model probably wants to be in the data

center infra business, probably wants to

be in the energy business, probably

wants to be in the satellite business.

These are all like innermost loop type

businesses. Robotics business, there are

such natural synergies among all of the

the different innermost loop stages.

probably there's some natural vertical

integration there.

>> I'm going to move us along here. So,

three stories this week that chronicle

the challenges being faced by the US

closed frontier labs uh who are under

siege from faster cheaper Chinese

openweight alternatives. So, the first

story comes from Moonshot AI, not

related to the Moonshots podcast, the

Chinese lab that's making Kimmy. So,

last week discussed how access to memory

is becoming the real roadblock on all of

this growth. It's not GPUs, it's memory.

Especially in the agentic age. This

week, Moonshot AI released Kimmy linear,

a new architecture that cuts context

memory by 75%

while still delivering 6x faster

decoding for a 1 million token context

window. Uh, Moonshot AI just dropped

this and it's running and in a single

move, uh, an improvement of 75%. So,

that's the first story. Uh the second

story uh on this block comes from the

Financial Times that reports that Fable

5, Anthropic's flag flagship model is

now struggling to attract users. It's

effectively plateaued. And the reason is

simple. Cheaper Chinese ope models are

eating the market from bottom up. And

when the model cost 14 cents per million

tokens compared and delivers 80% of the

capability compared to 15 bucks, the

market chooses the less expensive

option. At least the majority of the

market does. Fable 5 is not losing

because it's bad. It's losing because

it's overpriced relative to the

openweight alternatives. The third

story, then we'll talk about this, is

that Enthropic this week reversed its

data retention policy ahead of its IPO,

letting enterprise customers keep data

on their own cloud infrastructure rather

than anthropic servers. Uh, this move,

you know, handles the biggest objection

that corporate buyers have when adopting

claud. So, I don't think the timing is

accidental. You know, they're about to

go into IPO mode. I think it's predicted

for as early as 6 weeks from now. uh and

the growth requires enterprise adoption

and the enterprise adoption requires

data sovereignty. So gentlemen, three

stories here. Kimmy linear uh you know

uh the challenges that Fable is having

and the changes that Anthropic made on

its data retention and policy. Dave, you

want to jump in first? Well, this is

where we're going to find out if Daario

has what it takes to be a public company

CEO because, you know, he's a he's a

brilliant, good-natured AI researcher

thrust into this. And when he gets

interviewed, he said, "I never expected

to be a CEO at all. Uh, but here I am."

So now he's stuck with this uh missing

revenue numbers because he's embargoing

like the the current policy or the prior

policy was even if you're hosting your

Fable 5 on Amazon Bedrock in a secure

environment, everything still has to go

to anthropic headquarters for 30 days

for us to review and make sure you're

not making a virus or a bomb or

something. And that's the only way this

is safe. So now he's missing revenue

numbers because corporations don't want

to give their proprietary secrets to any

company that they don't know well you

know for 30 days and so they're rushing

to the Chinese models and secure

environments and they're also now you

can get you can get GPT soul also inside

a secure environment where it doesn't

get transmitted to open AI so you can

you can use that too. So, I was like,

"Oh god, corporations hate this, but I

don't want to miss my revenue numbers. I

want to go public." On the other hand, I

really don't think it's safe. I need I

feel like I need to inspect everything

to know that it's safe. So, now he's

he's stuck between a rock and a hard

place. It's a tough place to be. But

being a public company CEO is always

like that. It's really really stressful

and really hard. So, we'll see if he if

he has what it takes to do it.

>> Iman, what do you think of Kimmy Lineer?

Yeah, I mean first we banned the

faster silicon from China. So they built

models to take advantage of cheap DRAM.

Then the DAM became expensive. So then

they figured out better mechanisms of

linear retention of caching and more

more dense models etc. And now um you've

just seen actually uh just a couple of

hours ago this new uh 01 stealth model

that's been tearing up the benchmarks

turned out to be a GLM model served

entirely on Chinese chips with trillions

of tokens a day these new Huawei chips.

So I think you know you'll see the

adoption of these Chinese models and

them moving to where the market is on

different form factors of different

chips and again memory is 50% of all

spending now. It is the scarce resource.

You can't upgrade it. So guess what? In

a couple of months time, at the very

least, given the pace of Chinese models,

they won't need much memory. That's how

fast they innovate. On the fable

uptake, it's entirely a Zero data

retention issue. Like as a corporation,

you cannot leave your data on anthropic

side. And they realize this. But

Anthropic should not IPO.

If you are in the late stages of AGI

now, Anthropic should do a giant

freaking raise like Open did of 120

billion and have a straight shot at AGI.

That's what they should do and they

should stay private like

>> that's a fascinating thought. Yeah. I

mean, why why are they racing to an IPO?

>> It makes absolutely no sense to me. Like

Dario owns 2% of the company as does his

seven co-founders. They don't care about

dilution. They're worth like still 67

billion each and they don't care about

money. They pledged to give away 90% of

it. Why would you IPO? I can see no

reason for that unless they can't

actually privately, which I think they

can.

>> Yeah. Go ahead.

>> I have an answer. They need the capital

to get compute.

>> Well, but but

>> they could raise the capital. I I bet

you people would throw money at a drop.

>> So, I've been talking I've been talking

to investors and Dave, this will be

interesting for your perspective. Nobody

knows how to price this thing because

they don't haven't secured long-term

compute like OpenAI has or that natively

Groc or uh Gemini has. Therefore,

they're this is why the way to our

earlier point, this is why people are

verticalizing because if you're at one

layer and the bottleneck goes down below

you or above you, you're screwed. So,

you have to have access to the whole

layer to stop to have continual

progress.

>> Yeah. So that's

>> but even if they had the money to buy

compute where are they going to buy it

from?

>> There's not enough compute being

manufactured.

>> Wait before we get to that there's a SEM

is right but it's much more specific

than that. Like Daario uh got ripped by

Alex Karp. We showed the video on this

podcast. He got absolutely ripped to

shreds. And Alex Karp is saying look you

cannot give your alpha. You cannot give

your weights to this company anthropic.

You cannot trust them with your

corporate intellectual property. You're

talking about an academic never run

anything before in his life guy taking

your intellectual property and then

preaching to you how the government

should be run in the future. Don't trust

him. So he just ripped him to shreds.

You can't. So Dario's now he can't react

to that by saying, "You know what? I'm

going to delay my IPO and do some

private financing." And like you're

playing right into Alex's hands if you

wuss out on your IPO plans. like he's

just going to reinforce Alex's, you

know, Karp's me message horrifically.

And the board members, like look at the

board members at Anthropic. They've

marked up those venture funds to massive

valuations and use those valuations to

raise new funds. So they're not going to

just sit there and say, "Yeah, Dario,

you know, go do, you know, put it off

indefinitely." That's that's fine. So he

he's like, "This is the stress test for

Daario. He can't he can't just wuss out

right now." And that that signaling

would be terrible.

Alex, AWG, your thoughts, please. I'm

sure.

>> Okay. Well, first on on the the race to

IPO, I think there's also a race

element. I think there was a starting

gun a few months ago between SpaceX Open

AI and Anthropic. And I think if I'm

anthropic on top of the arguments that

everyone else here has already raised,

there's a competitive element of you

don't necessarily want to be the last to

IPO. The market wins could change. Right

now, it's a relatively warm and friendly

capital market for IPO. So to the extent

there's a window for raising the largest

IPO sum in human history, I think you go

for it. But to the earlier points uh in

lightning round succession, Kimmy

linear, we've known about Kimmy linear

attention since last fall, I think it's

it's suggestive, as I've suggested in

the past, ship of thesis style transform

architecture is getting incrementally

replaced piece by piece. It's

interesting in so far as it's a

successful linear architecture. Many

have tried to linearize the infamously

quadratic attention mechanism. Looks

like KLA may be one of the first at

least openly linearized attention or

quasillinearized. There's a recurrence

mechanism in there as well. So that's

kind of interesting. We've known about

that for a while. Fable 5 struggling.

That is interesting because I I've made

the point on the pod in the past that

Open AAI made a strategic blunder in

pandering to consumers rather than to

enterprises thinking that consumers

would be hungry users of reasoning

tokens and they just weren't. The the

consumers didn't know what to do with

all of these shiny OpenAI reasoning

tokens, but enterprises did. And then

OpenAI had to do this painful pivot over

to enterprise and turn everything into

codecs and probably delay their IPO as a

result. So Fable 5, which is at least by

my accounting the strongest most

frontiest model in the world right now

to the extent that it's struggling to

generate revenue and uptake. I want to

interpret that. I want to construe that

as the enterprises of the world almost

falling prey to the same thing consumers

with open AI did which is this may maybe

will be construed as victim blaming but

it's not. Our economy isn't worthy. It's

not clever or wealthy or successful

enough on average to know how to use

Fable 5 on average properly. Just like

consumers didn't know how to use

reasoning tokens from open AI and as a

result open AI had to pivot. I think

this is the beginning signs of anthropic

being forced to do some sort of pivot.

It could be radically uh reducing the

the cost of their models. That's one

direction. Or I think the more exciting

model uh the more exciting trajectory is

some new use case getting unlocked in

the next year that actually motivates

the usage of this nosebleleed priced

high end of the frontier which is at the

moment fable 5 or depending on the

reports you read maybe fable 5.1 may be

starting to leak out. And then very

quickly on Anthropic and their data

retention policy. Anthropic was so

clever, I think, in being the first

Frontier Lab from America to enable

their Frontier models to be hosted by

third-party hyperscalers rather than

having to host them themselves. And by

some reporting, 40% of Anthropic's

revenue now comes from Anthropic models

being hosted not by Anthropic, but by

third party cloud hyperscalers. So I I

think this is just another step to

externalizing the hosting of their

model. Yeah, sure it's painful. This

data retention policy I view is largely

security theater. I don't think it's

that valuable in the long term. I don't

think it's useful in the long term. But

starting to move more and more of the

the infra layer over to third parties so

that users of Claude can get Claude

where and when they want on the infra

they want. That's powerful. And you see

now open AI copying Anthropic and doing

that.

>> What do you guys make of the, you know,

the idea that the open source Chinese

models are good enough and at a, you

know, dimminimous fraction of the price

and companies are beginning to shift in

that direction, saying we're not going

to use Fable 5, it's too expensive.

Dave, is that an experience you're

having?

>> Yeah. No, everybody needs the absolute

best AI they can get. You can't go down

a notch, but the Chinese models aren't

down a notch. They're they're absolutely

on the frontier. You won't even notice

the difference in any use case. So, it's

not about trying to use something

inferior at a lower cost. It's about the

they're just as good. So, and and now

they're all good enough to improve

themselves, too. So, if you start a

group within your company that's using

these models, you can start improving it

inside your company if you get the

talent. And and so, that's like a

runaway train, you know? I I I think

that's what's really going on. It's not

it's not compromising to save a few

pennies. It's like, wow, we can control

our own destiny and be on the frontier

at the same time.

>> I mean, I don't think it's a few

pennies, right? Like Fable scores 60 on

the artificial analysis benchmark. The

new GLM model flash that dropped today

scores 57 and it is like 100 times

cheaper.

>> Yeah.

>> Yeah. one

>> which which is I think is really

important because you know the when you

deploy these things the cost benefit is

so high that you might say well I don't

even care about the cost but then you

say oh wait if I use the Chinese version

I can have a thousand or you know this

is why the swarm is such a big deal you

can afford five or 10,000 concurrent

Chinese operators instead of one

anthropic

>> well I think it's like it's like hiring

a specialist you know like super genius

versus a bunch of really smart people.

And sometimes you're not smart enough to

ask the super genius the right

questions.

>> Yeah.

>> Because maybe I'm not smart enough to

ask Fable the right questions, but I'm

just about smart enough to ask GPT 5.6

the right questions. Right.

>> But also, you know, that analogy is

perfect because people misuse their

context window horribly, and I do too.

Everybody does. But if you actually

optimize the context window with the

Chinese model, you'll get a smarter

answer than if you're sloppy using a

fable model. And so, you know, if you

just put a little energy into your

internal org design and optimize your

use and, you know, then you can have

thousands and thousands of these for,

you know, a very low cost. And that's

that's where the puck is going. I'm not

sure how sustainable this this situation

is. I I almost want to analogize it now

to uh US importing generic drugs from

Canada. the the drugs get invented in

the US, they get manufactured cheaply in

Canada, and then at least historically,

it's been the case that that you could

get American drugs more cheaply from

Canada by by importing on or off label

than you could from American drugs. I I

think the situation may be somewhat

analogous here where these are US

models, US reasoning traces. You see

Chinese labs benefiting legally or

illegally from the reasoning traces,

from interacting with US models. And

then just in the past 48 hours, we start

to see stories of Chinese labs trying to

strike partnerships with US hyperscalers

to host the Chinese labs models on US

infra but with a revshare from the

inference costs going back to the

Chinese front tier lab. So this is a

case where US does whatever innovation

is necessary data or post- training or

whatever that there's a distillation

maybe over to China. China sells it back

to us but then we're using our own infra

against ourselves at inference time

against the training time. I think it's

it's a perverse bind that we find

ourselves in analogous to Chinese drug

imports or sorry Canadian drug imports.

All right, I'm not sure the Canadian

drug import analogy holds very much

longer given what's going on, but let's

leave that aside.

>> Yeah, it holds until about a year ago.

>> I'm going to move us to a fun story on

the dating front. So, a Berkeley startup

called Ditto is playing Cupid. Uh, Ditto

is an app or a AI that has no feed, no

swiping, no infinite scroll. uh you fill

out a values questionnaire and then

every Wednesday at 7 p.m. a text arrives

with a match as well as a place and a

time for you to meet your date. That's

the entire product. You show up and see

if the magic happens. Thus far, 160,000

college students have signed up. Uh it's

already produced 80,000 dates. The app

does what Tinder and Hinge refuse to do.

It removes choice. The entire dating

industry is built on the premise that