my name is Bill Gurley I've been a

venture capitalist for 25 years at

Benchmark Capital before that I was a

saleside analyst for four years um I've

someone referred to me recently as an

elder Statesman of the VC industry which

I didn't particularly like but um

Perhaps Perhaps that label will stick um

I'm going to ask these two gentlemen to

introduce each other um and while

they're doing it um I'd love for them to

talk a little bit about the unique

relationship that they have with each

other it's not super common for two

different fund managers to spend as much

time together as they do and perhaps

they can explain part of that as we go

do want to go first Kev uh sure yeah uh

thank you Bill and I'm just going to say

you're a Statesman of venture capital

thank um you leave the senior off I

appreciate that yes um

no I I I would say a few things um

Antonio and I met nearly 10 years ago or

maybe 10 years ago um and we kind of

instantly became very close friends um

and to this day Antonio is is one of my

closest personal friends uh but beyond

that um has a crossover firm

um you know it is very important to

think about your partners um who are the

VCS going to be um that are um you know

on the boards of the companies you're

invested in and something that I have um

that I've realized is I feel safer when

Antonio's firm

Valor is invested and on the board and

the reason for that um is that they are

what they have built is a very unique

product so above and beyond being great

friends but I think of them as an

insurance policy if something goes wrong

and Adventure things do go wrong a lot

lot um and the reason that they are an

insurance policy is because Antonio's

firm is very unique a lot of growth

Equity firms talk about the value uh

that they can add uh to portfolio

companies and that basically boils down

to giving

advice hiring people and then maybe you

know helping you hire people and then

maybe hey here's like a sales plan

Antonio describes Valor has we are the

garbage man of business we do the very

we Sol the very hard problems that no

one else wants to solve or can solve um

and I've seen him say to a company like

hey what are your three hardest problems

I will solve them um and it's very very

real

um you know real operational Works

supply chain expertise sending a team to

China to stand up a supply chain uh but

what convinced me it was real uh was in

the summer of

2018 um I wanted uh to see

Antonio and and um at the time he was

working every night all night at the

model 3 factory um with Elon and so

Antonio said hey I'm super excited to

hang out but we can meet it uh you know

I could really only meet at 2 or 300

a.m. that's when my shift ends um you

know so go to the model 3 factory

there's Antonio hard hat glasses um and

I later learned that at Tesla they call

Antonio's partner the wolf when when

there's a if you SE pul fiction yes I

thought you'd get it bill um the wolf

comes in and helps and then they called

Antonio gendal because if the wolf can

solve it gendal comes and if you've seen

white hair it's white hair exactly um if

you've seen yeah I don't know you might

be Gandalf the gy

man and if you've seen uh you know the

two towers Gandalf comes at the turn of

the tide and they solve really really

hard things um and I would also submit

that probably prob Antonio and Elon are

the only two people um who have their

own plane who would pull all nighters

many nights in a row running a factory

line and Antonio I feel very comfortable

saying is the only person running a

growth Equity Firm who can run a factory

line and improve it um so supposed to

introduce yourself you just talked about

I thought I was supposed to introduce

that was great I'm gav okay now you're

gav yes so wow I'm uh B speeches Thank

You Gavin um I will say that uh it's

great to be friends with Gavin because

he's a deep intellectual and I'm a

second shift plant manager so we'd

really make great Partners um very funny

no and and the reality is that uh when I

was at when I was the on the board of

Tesla the director we had lots of

investors and part of my job was to help

talk to them um and I talked to many of

them Gavin was exceptionally thoughtful

he ran the number one growth fund in the

country uh he beat 2 200 peers sort of

year after year after year and at

Fidelity he was our primary contact so

why was he at the factory in 2018 at 3

in the morning it's because you know

your friends when things are going

really badly and uh things are going

really badly when it's existential and

you're running out of money and you

can't ramp your factory and you know who

shows up one of your investors that felt

great and so when I think about Gavin

why I do deals with Gavin it's because

when things are really dark um he's a

great adviser he's a great friend and

he's want to show up and bring you some

in an Out burger at 3:00 in the morning

which is like more than most people ever

do for you so it's been a wonderful

friendship and relationship but it won

that's earned through through respect of

our commercial relationship U and now I

get to spend time with him because he's

deep on semiconductors deepal

intelligence deep on the things I really

care about which are deep Tech we have a

shared passion and that shared int

intellectual passion has led us to lot

of deals together and to work together

and I think we have a good partnership

because I am a second shift plant

manager and he's a deep intellectual so

it works well together yeah thank you

Gavin I I'm not going to take the

compliment um yes but yes thank you yes

grateful so so what I was hoping to do

um I wanted to hit on public stocks

briefly um that's Gavin area of

expertise um and then talk about

privates talk about AI although that it

might jump in before yeah we get to to

that point and then if we have time left

we'll hit on some other more specific

things so Gavin the the my first

question for you is last year I guess we

had this weird anomaly event or it felt

like an anomaly where which people now

have branded the Magnificent 7 and you

know I even I talked to LP's you know at

Benchmark who like I say how's it going

and this was like the number one thing

they brought up because you know if you

were if you're exposed fine it was what

100% 90% or something with those seven

and if you weren't you know and but but

I think there's another question I have

so one you know what do you think

happened there was it an anomaly will it

continue and personally as a as a

venture capitalist in early stage when

we'd always counted on the the big

companies to get stodgy and slow and for

their growth rates to slowly fade and

then you know we'd laugh about them and

take share if if that's not happening

that could be an alarm Bell yeah um so I

think so first of all 2023 was not that

different from 2021 as a year there was

an equally narrow maybe even narrower

Market in 2021 because it wasn't quite

the Magnificent Seven maybe it was like

the Magnificent five or six

um but the reason that this is happening

is um a few things so first you know I

think everybody in this room understands

there are increasing returns to scale

and Technology particularly online this

relates to network effects customer lock

in effectively zero marginal cost of

distribution um it's why it's almost you

know impossible to gain share against a

number one player in a digital Market

um AI turbocharges all of that because I

think the most important thing to um

know about AI is what I used to call the

iron law of AI um but then open AI

started calling it scaling laws and

obviously open AI has um more of a more

voice and credibility um than me on this

topic so I'm going to call them scaling

laws it just means very simply if you

want to double the performance of an

algorithm and we measure the performance

of an algorithm has variance from zero

errors you know so when Tesla autopilot

goes from 99.5% accurate to

99.75 um you've doubled its quality and

that's important because human beings

operate at you know 59s like I could do

this and talk and that's actually pretty

hard for a a robot to do um and

um so what scaling laws say is if you

want to double the performance of an

algorithm you need to train it on 10x

more compute and data so the reason this

turbocharges these increasing returns to

scale is the largest companies have the

most compute they have the most data um

and this is why I think to um you know

to to quote your partner uh Eric vishria

who I'm a great great admirer of along

with you bill um

you know he he said it was a game of

Kings it is a game of Emperors um Ai and

if you are you know all these Frontier

Model companies have been funded and if

you're a frontier model company and you

are not getting a lot of what is called

rhf and you could think of AIS as just

you know they're like little human

beings that need to be grown um and

human beings need to be socialized and

so do

AIS um if you're not getting a lot of

rhf you're worthless

so all these billions of dollars have

gone into these companies and they're

almost all going to be zeros if you're a

frontier model company and you don't

have a lot of distribution to get rhf

you're going to go to zero because you

cannot compete with Google with xai with

its partnership with Twitter um with

Microsoft and open AI um why don't you

with people what rhf is uh re

reinforcement learning from Human

feedback it just literally means the I

mean what technically happens is the AI

um you know gets asked a question human

beings look at its answers and then they

they help it give better answers but you

need a lot of human beings asking

questions to do that right why do you

think just quickly the the tech

companies prior you know IBM HP deck uh

whoever son like why didn't they

recognize increasing like is it just

awareness of increasing returns it's

happening well a I think it is if you go

back to IBM it was more of a hardware

company and they're not increasing

returns to scale

Hardware um but you know the funny thing

something I often think about is maybe

one reason Tech multiples have expanded

is IBM's IBM's Mainframe business has

been shockingly stable for the last 25

years and that was the only real

software part of the business that they

had yeah yeah um as you look forward to

the the year ahead what what trends do

you think are most important to the

public markets and and what what's on

your your Worry List you look look the

Worry List is very simple um the history

you know of inflation is that when you

have um and after this I really want to

listen and not speak um the history of

the history of inflation is when

whenever you have a big episode there's

almost always a second wave there are

structural reasons that this happens um

one of them that there are all these

contractual price increases that are

based on last year's inflation that are

going into effect right now um so second

wave of inflation um is a big risk the

other big risk is for the first time in

my career geopolitical risks are um

positively correlated what I mean by

that is if there is a shooting war

between the United States and Iran over

you know what what happened two days ago

our future events the odds that North

Korea does something with South Korea

and China does something with Taiwan go

up exponentially because the US tied

down in Europe have to keep Europe safe

and you know Russia is scary we're

pretty committed in the Middle East and

we now have three carrier groups around

Taiwan makes sense um so that's those

are big risks yeah um and then you know

from the variable side I I do think the

most important thing is mag 7 or lag s

does Facebook inter search they should

does Apple inter inter search what does

on device a AI due to all of this I

think it's going to really scramble the

competitive Dynamics between the mag 7

and I think those two decisions along

with how much cost Google

Cuts so much of public Equity

performance are going to flow from those

three things are you personally on the

is Google more threatened Bay AI or or

more helped no it is obviously more

threatened yeah massively more

threatened of the S I use but it's also

it's more threatened but this been a

very mismanaged company for a long time

and we saw what Facebook did when their

back was the wall now back is against

the wall we'll see what they do and they

have every right to win but I use a

search based app called perplexity now

at least as much as Google yeah to your

point though about um big companies

being slow I mean they bought deep bind

in 2010 we were investors when they

invest Deep Mind squandered their lead

they had you know seven eight year lead

over everybody else and just blew it so

I'm no AI expert but I think the uh the

the experts I read would would argue

that had Google patented the attention

window that's in the the paper the great

deine paper these other companies

wouldn't exist maybe like you can't have

open AI without the attention window fa

attention is all you need attention is

all you need yeah yeah um they they they

had a they also had a um philosophy this

padle board game yeah of trying to

protect Trade Secrets by you know they

published that paper but they weren't

publishing a lot in that time and I

don't think they knew what they had at

the time that's the issue just the human

brain attention is all you need can't

pay attention you can't you can't win

yeah yeah yeah um all right let's shift

to the private Market

um you know there's the private markets

are always in flux and I think the thing

people want to talk about most is what's

in flux and there are there are changes

that come about systematically and there

are changes that come about

cyclically well Bill why don't I invite

you to speak first on this topic since

Antonio and I just spoke and then

Antonio will speak speak and I will say

nothing cuz I'm I'm on stage with two

experts fair enough so

on I think on the systematic side since

I entered Venture it's always gotten

more competitive in fact I was earlier

today listening to Osan he talk to Doug

Leone and he he he said it went from

being like this bespoke craft business

to a Main Street retail business like

it's just way more competitive there's

more money there's more players you can

get a check from anyone

and I and I don't see that changing if

anything this AI thing just proves the

point um cyclically I think we're in a

very unique position I think it's

different than 99 and' 09 um I think the

key difference is these companies raised

so much money in the um in the 2020 to

2022 phase that many of them were

sitting on three years of Cash When the

correction happened

um many of them did layoffs pretty

quickly which is kind of interesting

because there's systematic learning from

the past um and so some of these things

are just coming to Bear right now so

it's just is this very delayed response

the other problem is in 2020 to

2022 every single data point you would

get as a founder and a venture

capitalist board member from Wall Street

said growth at all cost doesn't matter

what Ju Just grow as fast as you can the

only thing I would just say is I know

for sure that Anon and I many times in

2021 cut Bur said cut burn no I'm

talking about the markets r large are

are are are are suggesting that this is

the best behavior and if you're in a

head-to-head competition with another

company where Money Matters yeah you

could choose to be pragmatic and lose

and so you have this game I use always

call it the game on the field problem

like um when you move this dramatically

to this new world order and the New

World Order not only wants profitability

but you have massive multiple

compression it's just very difficult for

these companies like the cultural shift

you need to go from burning a lot to

burning nothing and then just to stay

motivated when you used to think all

companies trade at 20 or 30 times

Revenue

and now you look at the pieces you have

and and somebody's coming in and telling

you oh that's probably worth two three

times Revenue like it's catastrophic

like the and and and that's another

delay Factor like it just takes the

founder two or three years to get in

touch with that reality like to finally

believe it in their brain and anyway I

it makes these correction Cycles so

difficult I wish the Venture industry

wasn't cyclical I had this lunch once

with how our marxy said explain the

business to me and I spent 20 30 minutes

a bunch of graphs and stuff he go o that

business is horrible I'm like I'm like

why and he goes well I can play every

cycle like I I have a strategy for all

Cycles he goes this thing's always going

to Boom bust yeah and and I find that

people take risk on very slowly like the

The Boiled frog component and risk off

is immediate mhm yeah so I think of more

as a Sawtooth than a sineway great I I

love that analogy and just maybe I'll be

a little moderator for this part of the

panel Antonio how do you think about the

state of the Venture cycle and then more

importantly how do you think um about

building competitive Advantage for your

firm yes so I was going to answer this

question um a little bit like Bill but

structurally what's happening with our

firm is uh and I'm going to blame you

Gavin so that the the the crossover the

mutual funds and the crossover hedge

funds uh destroyed the pricing in the

Venture market and here's why so there's

a bunch of hedge fund people right I'm G

to I'm be very controversial um if

you're a hedge fund manager running

across over hedge fund or a mut fund

manager you can put a private uh company

in your portfolio and it marks maybe

once a quarter maybe once a year and no

matter what happens even if its returns

are not ACC creative to your portfolio R

return your sharpen stino ratio improve

and therefore you get paid more money

this is a joke please stop allowing them

to do this yes it's literally destroying

the pricing in the private markets and

it's creating the rationality Bill's

talking about however if you have a

hedge fund that is putting you know I'll

call it the spray and prey method of let

me be in every deal cuz I've got uh I

hired a fancy consulting firm I'll never

forget the conversation with somebody

with a client when they were you know I

have just give order of magnitude I have

about 100 people working in private

markets a third in in in infrastructure

a third in operations and a third

Investments right so I can I today

sitting here today we have four teams

deployed in the field maybe five I've

got one at neural link during quality

I've got one at K Health I've got people

over all over the globe basically

working on our companies to drive value

and help your company succeed we want to

King make with our

operations um the reality is I'm sitting

in a meeting and uh the clients asking

questions to two different managers I

won't name the manager and manager says

listen I can Outsource um you know my

due diligence and operations to these

fancy Consultants that I'd pay like $400

million a year to I said well listen man

I I get it I guess we're just not that

smart I guess actually work at Mackenzie

or Bane because I can't figure out how

to Outsource them where I would have um

I it's hard this is hard work and it's

not Outsource you have to do it yourself

cand or people won't go do it and the

reality is that it doesn't work it just

doesn't work the reason you're seeing

these giant booms and bus I have news

for you is because it actually doesn't

work so please stop doing it Gavin

doesn't do it he under writes for

principles but if you look at a hedge

fund ask the question these guys

actually doing the work underwriting

this investment understanding the cash

flow it works for somebody if you can if

you can get 2 and 20 on a00 million and

you don't go on the board right they all

have fancy they all have fancy houses

down here in Florida for sure do no work

don't show up don't do anything I guess

I'm the idiot exactly exactly I mean I

was uh so that's point one point one is

that structurally the business has

changed what no matter we like it or not

returns are coming down so we have to

find a way to add more value that's

that's the answer um and I but I will

say this it's also the most interesting

time in my career so I've been doing

this for 25ish years and maybe 30

actually 25 30 years um I actually

bought my first stock Apple computer at

the age of 13 so I've been looking at

technology for a long time um that's a

long time ago by the way and the

what I'm seeing today like today's calls

okay today I had a call with a company

doing AR Aral intelligence uh applied to

biology that was making a um in the last

week I should say it was making a

computer that had DNA that could compute

so it could actually go in measure the

gradient of a cell tell you if it the

cell was cancerous or not and then send

us the bat signal out to the immune

system to kill the cell like

unbelievably mind-blowing Quantum

Computing company we're debating what

today photonics versus chip what kind of

photonics I mean this stuff is

mindblowing it's mind-blowing like way

past the internet like the internet was

I mean really really I'd say in terms of

productivity Improvement in the economy

and what we're how we're going to live

in 10 years what we're seeing today is

revolutionary relative the internet so

be optimistic I mean the returns might

not be good somebody these managers but

in terms of Our Lives it's going to be

great so it's a great time that the the

structural changes are great cyclically

listen when you have no free money it's

free money people do crazy things and we

try to avoid that by keeping our

disciplines yeah one thing I think

everyone in the world knows but I state

it just for clarity despite the most of

the industry going into a cyclical

decline AI remains as if it were 2020 20

yeah as a category for sure with with

this caveat let me tell you where the

bubble is um our view I I'll speak for G

I think he agrees with us is we are

investing in AR artificial companies

that are we call verticalized so they

have a very specific data mode that's

protectable and therefore they can

create an Aug that's protectable and

often with a hardware integration the

llm business so you know uh open eyed

for 9 valuation that makes no sense to

us because what Gavin said earlier is

true the returns to this business are

creating from Capital allocation in

giant data centers so that's it that's

the only Advantage you have is how big

is your data center how fast is your

data center how many chips how many gpus

will Nvidia give you that's it the

models are free they're open in the

world the data is free it's open in the

world you need reinforcement learning as

Gavin said if you have enough money you

can do that that's those are all things

that can be bought by giant companies

which is why these companies doing so

well relative to the small compan

companies I think those Moes don't

survive the other thing I'd add in

addition I mean I just know because our

AI strategies are highly aligned yeah in

addition to the verticalized I think of

us as investing above the Frontier Model

layer yeah for sure in verticals or

below it an infrastructure that makes AI

more efficient an infrastructure that

drives up GPU utilization true deep Tech

because the data center is being

fundamentally rearchitecturing

three years ago the other thing go

deeper on that Gavin so obviously Nvidia

has been the most obvious big winner in

this pix and shovels which is a metaphor

people love to use but but what other

what other areas yeah yeah I'm happy I'm

happy to talked to it but the other

observation I just make is it does feel

to me that growth Equity so I think

Benchmark series a specialist series B

Specialists I don't see them as being

really disrupted because to me you guys

had real value at the early stage and

you do have like a brand that is

reinforcing it's like having Harvard you

know on your resume Harvard really sorry

hey come on brother I mean I'm sorry I

think I I think poor

I I think all these firms are being

assaulted just yes every every Alo can

fall wow every and I have a son there by

the way I don't be careful yeah but I do

think um well look if you're if you're

Harvard or if you're Benchmark if you

make mistakes eventually the brand EX

but the point I would make I think the

growth Equity series C and up industry

is going to Evol evolve in exactly the

same ways that private Equity evolved

you know so 25 years ago you know you

know Henry kravis and um others were

running around and you know it was an it

was a relatively inefficient Market um

and then now everything is done in an

auction price is set in a manner much

closer to public markets than the way

Venture and growth Equity is set um so

that I I think there's probably more pay

in coming but what private Equity firms

realized is they had to add a vast

amount of value yeah and to me that's

kind of what I what I admire about Val

strategy but for you know a GPU an

Nvidia GPU is half the size of this

iPhone okay it is made with three

tablespoons of sand a Taiwan

Semiconductor in Taiwan Invidia buys

this from TSM for

$700 and then sells it to Google or

whoever open AI for

$50,000 and you can think of a data

center as like a restaurant um and the

GPU is the single most important and

expensive part of that restaurant it's

the star Chef you want that guy being a

star all the time instead because the

systems that surround the GPU have not

kept up the gpu's gotten 50x faster in

less 4 years years the rest of the data

Center's gotten four times faster um the

GPU does nothing 70 to 80% of the time

and while it's doing nothing it is

sucking full power a vast amount of

power even when it does nothing it's

sucking power and you're depreciating it

and that's very expensive um and the

reason is is that the food that gets

delivered the Sue Chef they the you know

the the chef has gotten a lot faster and

the rest of the restaurant has not so um

a thesis I have had in Venture growth

equity and public equities for four

years now just cuz yeah I was actually

the Nvidia Analyst at Fidelity um 25

years ago I used to talk to jinen weekly

we owned 15% of the company um is that

this was a problem that needed to be

solved by fundamentally rethinking

storage networking and memory every time

we see a platform shift of this nature

we always go to infrastructure and

Gavin's right I mean this is this is the

key problem is how to make INF go faster

behind the data center for sure you see

any public winners that take advantage

of that oh yeah and I mean we like we

own Sienna coherent Optics are going to

have to come inside um the data center

in 2025 no later than 26 when that

happens

that's an unimaginable change to the

architecture of a data center you know

coherent Optics sit on the outside of a

data center today they're going to have

to come inside um all hard drives are

going to be replaced by flash storage um

you're going to actually

you know storage you can think of it

sitting the layer below something like

snowflake um storage used to be thought

of as the best way to invest in data if

you go back you know the days of EMC

when you were writing reports how do you

want to invest in data storage EMC those

days are coming again um and you know

lots of opportunity yeah oh yeah public

and private so so so going to AI which I

knew we'd already go to so that's fine

um if if the if the Standalone llms are

are you know fast depreciating

commoditized whatever who do your

vertical successful vertical players end

up buying you are they buying it you

know by the drink by the hose are they

running models internally you're asking

where they getting the data no where are

they getting their llm type Tooling in

the long run well I mean the vertical

self word investing and like example

give is um we called zip line which is

has more autonomous flowing miles than

any company on Earth because they went

to Africa built a system to deliver

blood and Drugs in five countries AF

including Rwanda they power the entire

Rand U System I mean it's extraordinary

I they lower the maternal mortality rate

at the edge of their wed system so out

in the field by 88% because they deliver

a package of drug blood and drugs to a

woman who's had a baby and she's

bleeding out they can coagulate the the

bleeding and then give her some plasma

um they Ed those miles and all that

autonomy come back to the US they the

first autonomous system approved in the

US to deliver packages and now they're

building another platform to drone that

will be more of a quadcopter but there

that is full autonomy right that was

built custom by them in Africa they've

got millions and millions of miles

that's a custom data set K house a

custom data set I mean there's there's

there's a spectrum of this that is fully

custom and almost impossible replicate

you know Google's trying with with wing

but almost impossible POS replicate all

the way to I train on the internet and

get some humans to do reinforcement

learning that's the Spectrum we're

trying to be on this side of the

spectrum and we have a bunch of

companies are doing this you agree with

all of that but I would just say I do

think there is you are going to open

source llms are going to be a winner and

you know it looked like it was going to

be llama now Mr straws come out of

nowhere um but I think it is very hard

if you're Frontier Model company that

raised at 5 to 15 billion like I think

you're lucky if you get sold for the

preference because you're going to be

squeezed between these Giants with

internet distribution who are getting a

lot of rhf Y and really high quality

open source and a you know uh Mark

Zuckerberg is a brilliant man but I

often wonder why he does not enter cloud

computing and say Hey you can use llama

for free but just you need to run it in

my data center um I often yours running

llama right now what's that Microsoft's

running llama right now as we speak but

Facebook has a lot of data centers yes

yeah yeah they're Commodities would why

wouldn't that happen I have no idea in

the same way I can't figure out why they

haven't already entered search unless

they're just not aspirational about the

web about the mean or or or the capital

what they figure out you look out 3 to 5

years the capital is a commodity it's

all a commodity So eventually

you know I was in the connector business

in uh in in 2000 I saw this happen in

the internet where all the uned backbone

stuff everyone triped everything you

overbuilt and then it crashed at some

point someone will build a competitor

GPU and this is a commodity so if you

can buy commodity Focus your resource in

other areas they higher return I could

see that yeah I'm not saying he's right

but I can see the cation question of the

board yeah can ask Peter ch's coming on

next I um Gavin you know this I have a

bit of a pet peeve about an issue that

relates to this which is I don't

disagree with you about how this is

going to play out does that mean you

agree no yes I agree with you you get

rid of the double negative so I I agree

with you about how this is going to play

out but the some of the players that are

running these Standalone models these

companies um have organized politically

in a way that the only other human

that's organized this way early this way

that I've seen was SBF actually um but

they have like three different super

packs Politico wrote an article and

they're literally trying to talk

legislators into doing things that would

either restrict open source and some of

them have said specifically out loud you

should make open source AI illegal yeah

I so profoundly think I mean it's

obviously unethical and as a human being

what we want most is a multipolar world

where there's lots of AIS where we have

our own AI um you know one reason it's

easy to see why AI is different from the

Internet is the closer you are to AI the

more worried you are about it has an

extinction level risk for Humanity

nobody thought that you know the

internet might lead to the extinction of

humanity and you know there's all sorts

of people like Yan laon who runs

Facebook's lab who says this is

ridiculous but there are a lot of people

who are deeply knowledgeable including

Jeffrey Hinton who invented deep

learning was basically the kind of

chairman of AI guy at uh Google um and

left Google cuz he is so worried about

the extinction risk um and so what we

want to mitigate that risk as human

beings is a multi-polar world where we

have many AI the most dangerous world

for human beings is one where there's

only two or three and so if you were

worried about AI safety we should have

like there are racing Dynamics we're

going fast we're not slowing down maybe

we should have but we're not um you

should be in favor of Open Source let

let me out of this which is um the

internet I think when in 2009 20 2001

nobody thought the internet would enable

political volatility and Revolution and

actually uh allow governments to

constrain and manipulate their

populations we all thought it would be

tool of Freedom it wasn't be very

careful here this technology not only I

think regulatory capture that you're

describing is not only unethical it's

unamerican it's anti-American it's anti-

Freedom why because our Che competitors

around the world are building the same

models and they're building them and

weaponizing them as fast as they can

trust me this is happening if we don't

have the best technology in the world we

will not be safe this technology is so

powerful it makes a nuclear weapon look

like a small fry small fry I mean I'm

telling you this and Gavin's right about

this the closer you are to it the more

worri you are if you want America to be

safe and our our allies around the world

we need to be competitive strategically

this requires us to have an open

platform so we can develop the best

models and the best training data and

have the best technology that's kept us

safe in the entire portal period we

should continue with that program

I yeah I can't help but think that it's

um driven by an attempt at regulatory

capture for sure it is it's pure greed

and it's anti-American and I probably

aided by Foreign adversaries yes of

course and I would I would add there

could be one more motivating factor on a

generic text llm so different from the

work that zip lines do on their own data

center um open eye and whatever yeah

just just text you know and a machine

that can talk to you you know I think

the parameter count has diminishing

marginal Returns the attention window

has diminishing marginal returns and the

data sets have there's no what what what

new data sets well you know I think we

are going to see a new data set I I

spoke about this yesterday but Yan laun

observed that you know gp5 has been

trained on the entirety of human content

ever created the entire internet every

book everything but human beings we have

an IO problem input output um and

through vision we take so much more

information I heard this so an infant by

the time it's 4 years old has taken in

100x more data than gbt 5 I actually

think a synthetic data is going to help

scaling laws continue but B I think

you're going to see video games be an

incredibly useful place for training AI

so you we we actually um our earliest

exposure of ourselves was actually uh

expert systems deploying Vision in

factories this was a form of Ral

intelligence was it was data science and

vision and um they kind of got the the

language stuff got jumped vision went

much further ahead what Gavin's saying

now you for all of you here in the

audience it looks like that language

looks a lot better than Vision but

vision is actually very very Advanced

which is why we can do autonomous

driving it' be interesting to see how

you embed that into the intelligence

part and like how you learn from video

if if you were an agent you know and

we're going to go from Models to agents

Frontier models Foundation models to

foundational agents I think that's the

trend of the next 18 months you don't

care whether you're looking at the

internet you're in a video are you're

embodied an Optimus robot right and I

think there's going to be this is going

to be an enormous source of advances and

I do think I don't know if it's 24 or 25

but Robotics are going to have an iPhone

chat GPT like moment sure you mean this

is an analogy human brain right just

think about that always remember it's an

artificial neuron Network this is a real

neuron Network and it learns this way it

starts with vision then language and

then it integrates the vision language

to cognition and then Consciousness and

we are on the same path you know where

we are in the emergence who knows but

we're on the same path at or

intelligence fair by the way and this is

why it's so interesting you know if you

talk to gp4 if you want to get a good

answer speak to it the way you would

your high school kid before a test go

slow yeah yeah yeah no no just hey go go

slow double check your answers be

careful that will make the llm perform

better

because they really are very human in

many ways but please before we run out

of time and I I I would love just um if

you have a moment Antonio to talk about

Elon you know you've had a very special

relationship with him it goes way back

um talk about like how you met why you

think that relationship is bidirectional

and why it works so well and and and

maybe you only have three minutes yeah

what makes him so unique I mean only

have three minutes three minutes um so

yeah I first met ELO invested in PayPal

in like 99 2000 so a long time ago and

you know we've invested in many

companies with him over the years and

I've gotten a chance to work with him

and it's been I'd say in many ways a

career defying for me because being near

him and probably true for you too Gavin

it I I have this saying I told which is

you can only achieve the thing you

conceive and it has moved the conception

curve the possible for for us for sure

that we're close and for everyone

probably here the things that are

possible Landing Rockets human brain

Earth fit isn't this is insane right so

what we have done and what we're good at

is like making stuff I'd say the old

economy things right making Factory run

making a process run making a sales

process run and so when a great engineer

world you know once in 100 Year engineer

like that is building companies guys

like us who are the garbage men the joke

I make we you know you got to take the

garbage out we cannot be uh genus

Engineers but the garbage got to take

out and we're good at that um I would

say what makes him special I'll tell you

I've had a couple of moments that I've

seen in 2018 in the factory in in uh at

Tesla I saw him you know in at about 3:

in the morning in the body shop sit down

with the engineers our body shop wasn't

working and and and take about 30 welds

out of the car by saying we don't need

this window this one he was running the

math in his head and we were going to

run the cars inventory um the bodies and

if we had been wrong if he had been

wrong we'd have gone bankrupt nope don't

worry run it guys R in the simulation

came back it was right it was like

watching Michael Jordan play basketball

times 10 and we just had one of these

right so the neuralink interface um

there is a there's a thing called Utah

Ray it's been around for a long time

it's a brain computer interface and it

causes scarring the brain because the

mechanical interface Electro too is too

big um the thing Elon did there is this

is uniquely him he's a the martial

Engineers with his mind they Shrunk the

mechanical interface to a size and made

it flexible that would fit between the

in the dead between the synapse little

space in the synapse so that it would

not Scar the brain that's uniquely him

that's the one 100e brain that is his

engineering mind Material Science

mechanic engineering software that

integration I've never seen anything

like it it's exceptionally special and

at the same time he's a guy who's deeply

compassionate who cares about humanity

and wants to do the right thing like all

these things are true you know and

having all those things in one mind I

think makes them very unique Gavin yeah

know I'm I'm grateful to Elon for so

many things not least of which is

introducing US 10 years ago um yeah but

um I I would say two things like first

to his his motivations you can think of

every single company that he has been

involved with has adding value to the

world in a profound way you know the

internet obviously added value to the

world and you know the payments

associated with that that was PayPal

um he was very concerned about global

warming that was Tesla and Solar City um

you know he wants Humanity to be

multiplanetary that's

SpaceX um you know the boring company is

actually another way to reduce um to

help make the planet green and make

cities green because you know we've

essentially given all our space in

cities um to roads um nurlink exist to

help bring about this multi-polar AI um

future because you know the ideal state

for human humans is you know we have a

on device AI that loves us and we

communicate with it through our neur

link We merge we merge we kind of

co-evolve with AI and then the reason he

started xai is to you know make sure

that it is not just dominated by um open

AI Google um and to lesser extent you

know two open source models you know you

want more of these true models X is a

new corporate entity yeah yeah for grock

which I actually think is great grock's

going to be grock actually is in a

commodity because and Twitter owns a

piece of it Twitter owns 25% of it yeah

awesome well we are out of time thank

you gentlemen thank you Bill thanks

[Applause]

everybody