Please welcome Kevin
Roose and Casey Newton,
hosts of Hard
Fork podcast.
[MUSIC PLAYING]
About to make the--
What's going
on, Dreamforce?
Do we have any Hard Fork
listeners in the crowd?
We hope you're having
a good conference.
And we have so much
to get into, Kevin.
Should we get
started right away?
Everybody, please
welcome to the stage
the CEO of Salesforce
AI, Clara Shih.
It's a good looking
bunch of people.
I think this is the
only time we've ever
taped a segment with
motorcycles competing
We're hoping to win out
over the motorcycles.
My earpiece is,
like, falling,
Well, shall we
jump right into it?
So this week, Salesforce
announced the rollout
Clara, what
makes Agentforce
different from some
of the other AI
solutions on the
market that people
Well, first of
all, it's so great
I'm a huge fan of your
podcast and your writing.
We're at a time
now when agents
are going to be as
commonplace as apps
And every company
knows they
But it's hard,
and it's scary.
And what we're so excited
about for Agentforce
is that the very things
that you need to roll out
an enterprise autonomous
agent are the things
that customers already
have in Salesforce--
their roles, managing
the roles, the channels,
the data, the actions,
the trust, and security.
And so we want to make it
really easy for people.
That's why we have our
Agentforce launch zone.
And we've had thousands
of customers already go
through the last
day and a half
building their
own Agentforce.
So if I'm a
company, and I have,
let's say, 500
customer service reps
who are dealing with
customer queries,
and answering
questions, and things
like that, how
many do you think
I will need in a world
where AI agents can be
answering-- can be
interfacing with customers
Is it going to be half
the number as before,
Or maybe I'll still
need the same number,
but they'll be able to
process way more queries.
How do envision
that going?
Well, I think it really--
it's amazing to see the
role expansion happening.
So we've seen this
with so many customers
that have rolled
out Agentforces.
People who used
to be confined
to handling customer
support tickets,
Now they can help people
with their e-commerce
Now they can become
brand storytellers,
because the AI
is giving them
assistance in the moment.
So I really think
just like we
saw with the internet,
the definition of what
a job is is going to
change dramatically.
So I was surprised when
you guys rolled this out,
because Kevin and
I have been talking
about AI agents all year.
We've been waiting for
all the big players
And typically,
what we've heard
is the technology just
isn't good enough yet.
We can't actually trust
a piece of software
to go out and
perform these roles.
So how did you guys get
this ready to roll it out?
I mean, I would really
encourage both of you--
go to our agent
launch zone,
and build your
own Agentforce.
The reason it works
is really twofold.
First is you
need the data.
And companies have
their data already
They have their
business processes.
There are ways that
retailers, like Wiley,
a publisher of
books, and there
are ways that Open
Table, that they
work, in terms of handling
defective product,
in terms of an
order that's late.
That flow, that
business logic
they've built
into Salesforce.
And so they don't have
to worry about an LLM
hallucinating how to
handle a critical business
They can just invoke
what they've already
And so that's
why it works.
One of my favorite stories
of mishaps in AI this year
came when a Chevrolet
dealership in California
implemented an AI
chat bot to talk
And people discovered
that this was happening.
And they started sort of
messing with the chat bot,
and saying, hey, can I get
a new Chevy Tahoe for $1?
And the agent
would say, yeah,
How do you prevent
that kind of thing?
I mean, you said
it's grounded
We know they still
do hallucinate.
They're still sort of
gullible in some ways.
So in a sort of critical
sales or customer service
role, how do you-- what's
the technical process
under the hood
for making sure
that they don't
start offering people
I remember that
story, too.
I mean, that's exactly
why companies don't
Because you can
build a demo,
and set that up really
quickly, like that Chevy
dealership did, but
to actually deploy it
into production in a way
where you're protecting
your business, and driving
the outcomes you want,
you need to make sure
you have those safeguards
So the very first thing
that our team built
was our Einstein
Trust layer.
Every Agentforce
interaction
runs through
this trust layer.
And it has four
components.
The first is
data security.
So we have something
called prompt injection
So that would protect
against what happened
to that Chevy dealership.
If someone's trying to
inject an instruction
into the prompt it,
try to sneak it in,
we'll catch that, because
we have models actually
specifically trained
for that purpose.
The trust layer also
has data privacy,
Anything proprietary,
personally identifiable,
It has toxicity filtering,
so ethical guardrails.
Because these models were
trained on the internet,
there's a lot of
bad stuff out there.
So we have to
adjust for it.
And then the
last component
of the trust layer
is observability
and monitoring, so
that we can always
be supervising that
the AI is performing
When it comes to
the monitoring,
how much of that are you
able to do automatically?
And how much
are businesses
still going to need to be
keeping an eye on things?
We're all still
learning together.
That's why we have
this huge community
We've built tools to do it
both in an automated way,
and also so that
in the moment,
businesses can constantly
supervise the agents.
There's a part of the
Salesforce product called
We've actually had
it for many years.
And it's what VPs
of customer service
use to manage all of
the real-time calls,
and emails, and chats that
are coming into their call
So they can see what's the
volume, how is it spiking,
who's on hold, how long
have they been waiting,
which calls are
active, what's
They can see every
single person
who's working in
their contact center
We've extended that
now to also include
So they can monitor
it the same way.
They can get
the same alerts.
They can get
the same flags.
Oh, this isn't
working well.
You better escalate this
to a human team member
I mean, one model for
how AI is being deployed
in businesses is this
sort of copilot model
that we've heard so much
about, where you have--
everyone has sort
of an AI assistant
that's helping
them be more
productive and
efficient, maybe
doing some of the more
menial tasks in their job.
This really
sounds different.
This is like
you're basically
having a kind of
AI worker that
is working alongside
your human agents.
Maybe compare and contrast
the kind of copilot model
that we're hearing about
for the past couple
of years with what
Agentforce is doing.
I think we've learned
a lot collectively
about the shortcomings
of a copilot model.
I think people-- we talked
about this yesterday, some
of the security issues,
the fact that the copilot,
because it tries
to do everything,
it doesn't do
anything well.
And that's really the
difference with agents
is that it's very specific
to certain actions.
You tell the sales
development representative
agent, your job is to
answer product questions,
and to get basic
information
Anything beyond
that, the customer
asks about pricing,
you transfer that
Same thing in customer
service, right?
There's guardrails
around the actions.
And I think that's
what we're seeing
is really important
for driving
the focus, the
relevance, and accuracy.
And then the second
key difference
behind Agentforce--
we talked
about this yesterday--
is the Atlas reasoning
Atlas is the brain
of Agentforce.
And it was developed by
our Salesforce research
So a lot of big
brains in that group.
And it really is specific
to CRM, to Customer
It's specific to sales,
service, marketing,
And it's been tuned
specifically for those use
And so that's why
it's so accurate.
So I think I read today
that it might cost
customers about $2 to
complete a customer
service transaction
with the new model
I'm wondering, as
you've been doing
your testing, what
real world examples are
you seeing customers
use this for?
And what kind of ROI
are they getting?
So we asked our
customers how
they would think about the
business model for this.
And they all
told us, look,
we think $2 per
conversation is fair.
We pay much more
than that today,
if we have a customer
service call.
It could be much
more than that,
depending on where their
call center is located.
In terms of the
ROI we're seeing,
the most common
use cases right now
This is their
back-to-school fall
They have a surge in
customer support questions
from teachers and students
about their materials.
And so they've been able
to surge up their work
force with Agentforce
and handle 40% more
cases, completely
resolved using Agentforce
than their
previous chat bots,
The other use
case we're seeing
So a lot of times when a
prospect goes and visits
a company's
website, they're
not a qualified lead yet.
They have a lot of
basic questions.
And rather than deploying
your most highly paid
sales people to answering
those questions,
now you can deploy a
sales Agentforce to,
as patiently as
possible, answer
any number of questions
that prospect might have,
whether it's on a website
chat, or on a voice call,
And then when that
prospect is ready,
when they're asking about
pricing, when they're
asking about
rollout timelines,
you know they're
ready to hand off
on a silver platter
to that salesperson.
One of the big concerns
that people have in AI,
generally, right now
is about misinformation
We had Yuval Noah Harari,
the author of Sapiens,
on our podcast last week.
And one of his
ideas for how
to prevent some
of these AI harms
is that it should
be illegal to have
an AI that doesn't
identify itself as an AI.
Basically, you should
have a law that
says if you're a bot, and
you're talking to someone,
you have to proactively
stick your hand up,
your virtual hand up,
and say, I'm an AI.
Are these agents
in Agentforce,
are they going to identify
themselves to customers?
Will customers
know that they're
When there's a handoff
to a human agent,
will they be
told about that?
I mean, not just in
the Agentforce case,
When you're engaging
with an entity,
whether it's a
prospective customer,
or a prospective
girlfriend or a boyfriend,
you need to know who
you're dealing with.
And I do think that
should be regulated.
We've certainly brought
those principles
I talked earlier about
the Einstein trust layer
as our product
and technology
way of enforcing
ethics, but we also
have an acceptable
use policy for AI
that was developed by
our Office for Humane
And one of the policies
is that any Salesforce AI
product-- and this
dates back to pre-LLMs,
it has to self-identify
so that our customers know
and their customers know
that they're using it
in a self-service capacity
to that exact point.
So yesterday I was
talking with somebody
who works for one of the
big consulting firms.
I told him I was going
to have this conversation
with you, and I was trying
to get some ideas for what
And he said that when
he talks to companies,
like the folks
in the audience,
he finds that
people kind of just
have their data
all over the place.
They bought these
three companies,
and they use these seven
different architectures.
And he was sort
of confused
how your system
would work in a world
where people's data
is spread everywhere.
So I'm curious how you
think about that problem.
And maybe, what
do you think
companies need to do to
get ready for a world
where they're using
these kind of agents?
I'm so glad
you asked that,
because the
conversation about AI
really is a data
conversation, first
And I'm really grateful
to my colleague Rahul.
He's a GM of
our Data Cloud,
because that's exactly
what Data Cloud does,
is it takes all of these
islands of siloed data
that every company
has, like different
applications, different
data lakes, different data
Some of them are
in the cloud.
Some of them
are on premise.
Some of them are on
someone's computer.
And it unifies and
harmonizes that data
through a zero
copy manner.
So you don't have to pay
for moving and copying
It actually creates a
virtualization layer
that points to the data,
and connects the formats
so that it can be
usable by the AI.
So people are able to
do that sort of once.
What I'm hearing
you say is
you don't need to do
that much to get ready
If you're using
Salesforce today,
you will start on
your agent journey
with a tremendous
head start.
You're starting
like 30% of the way
depending on how many
Salesforce products
If you have Data
Cloud, you're
Imagine starting a
race at the 70% marker.
You're going
to win, right?
And so that's what
we're so excited about.
And it really does,
in this odd way,
feel like so much of what
the company has been built
on over the last 25
years-- like, obviously,
it predates me
joining the company--
from our metadata
to our low code
builders to our
Data Cloud to Slack,
which is the
perfect interface
for teams of humans to
interact and collaborate
It feels like all
of these pieces
are coming together
to make it easy.
I'm curious to know
how the human agents
and the AI agents will
kind of work together.
Obviously, there'll be
this sort of handoff
if a customer's
asking questions
that the AI feels could
be better answered
But I remember
doing a story years
I think it was a
health insurer that
had started
implementing some AI
stuff in their
call centers
to actually train
their agents
So if they detected
that they were talking
too fast, it would flash
a little sign that said,
like, slow down
on their screen,
or it would give them
feedback after a call
and say, you let this
person talk a little too
long, or you talked
a little too much,
I was going to give
you that feedback.
This could apply
to podcasts, too.
But how do you see
agents in Agentforce sort
of helping humans do
better at their jobs?
It's such a
great question.
I think it happens
before, during,
and after the
agent interaction.
Or put another
way, it happens
in the form of guardrails,
nudges, and handoffs.
So humans set
the guardrails.
Agent, you can
talk about this.
You can talk about
customer service.
You can't talk about
giving a refund over $50.
I want to make sure that--
No selling Tahoes for $1.
And no selling
Tahoes for $1.
And no talking
about random things.
If you're a car company,
don't talk about fashion.
What you can do,
what you can't do.
That happens before
you deploy the agent.
During the
interaction, then I
think agents can give
nudges to humans.
Just like you were
saying, if you're
talking too fast,
or you're not
making eye contact,
or in my case,
sometimes I have to pep
up a little bit more--
drink another
cup of coffee.
It's helping me in
the moment be better.
And then it's also
great for handoffs,
because when that prospect
talks about pricing,
the agent, because of
its guardrails knows,
it's time for me to hand
this off to someone.
Or if the agent is
confused and doesn't
have a high
confidence score,
Atlas tells it to hand off
rather than hallucinate,
That happens to
us when we're
As I'll start talking
about something,
and realize I don't know
what I'm talking about,
and I have to hand it
off to Kevin to explain.
I'm like the intelligent
agent of the two of us.
So OK, so you guys are
starting with stuff that
Let's not roil the
waters too much.
But I'm so
curious what you
think you might be able
to do a year from now,
As some of these
models improve,
you see a step change
in functionality.
What is kind of
the next pieces
of fruit on the
tree that you're
I feel like we're
living in a time warp
right now where a
year doesn't mean what
And so it's so hard to
say what time horizon.
But what I will say is I'm
seeing incredible results
And many of them
are here this week.
I mean, because
they had invested
in Service Cloud and
Commerce Cloud and Data
Cloud, they were able to
get their Agentforce up
You should go to saks.com.
And you can experience
it for yourself.
And so in terms
of what's next,
our research team
now is working
on the next
generation of Atlas.
And it involves
multi agents.
So different
agents working
with each other,
refining, evaluating,
we think that
it's inevitable
that every consumer will
have their own agent
And so what does
that world look
like of enterprise
agents dealing
And at what point
does each hand off
to their
respective owners?
If you were a sort of
fast-growing, beloved
podcast, is there
anything you would want
Sourcing what's
interesting to people,
preparing for guests
coming on the show,
generating the first
set of questions,
refining with the agent
to come up with that,
editing maybe,
and promoting.
All right, well, it
sounds like we're
going to need to get a
Data Cloud, and a data
lake, and a
data warehouse.
We would be glad to
set you up in the zone.
Well, I think that that's
all the time that we have.
Thank you so much, Clara.
Thank you for
everything you do.
And I'm excited to say
that we have one more
And we don't even
have to do an ad
The Dreamforce Hard Fork
is completely ad free.
I do want to tell you
about a new mattress I'm
And please welcome
to the stage
the CEO of
fyi.ai, will.i.am.
I knew you were
going to be dressed
Will, for folks
here who haven't
had a chance to try it
yet, what is fyi.ai?
Fyi.ai is a Web3
messenger with elliptical
cryptography
methodologies that
keeps all your
digital assets
and conversations super
secure and private.
We partnered with
IBM to ensure that.
We issue everybody
encryption keys.
So we don't have the key.
It's end-to-end encrypted,
but you own the key.
FYI is a project-based
work team flow.
Here are some of the
projects that we make.
And we have generative
AI at the core.
So if you're not
blessed to have
an amazing Rolodex,
you're going
to have a Rolodex
of expert personas.
Salesforce calls
them agents.
And for us, the difference
between a persona
and an agent is
something that's
ultra expressive,
lifelike, human-like,
clever, and
metaphorical for
And what led you from
your career in music
to wanting to build
an AI company?
I was lucky enough to
invest in companies
like Tesla in
2000-- end of 2006,
I invested in
Tesla before Elon
And I invested in Twitter
early on with Jack.
I invested in OpenAI super
early, and Anthropic.
I just learned this
about you this week.
You are, like, on the
cap tables of every AI
company in San Francisco.
You know what I'm saying?
Have you made more
money from music or AI
Well, AI companies
are just growing.
I made more
money from Beats
when we sold to Apple
than making music.
So I was a part of
Beats from the jump.
I still have my Tesla
and Apple stock,
and Beats stock, and
Pinterest, and Dropbox.
So yeah, so I like to
invest and build teams.
And so when I have
disposable income or play
around money, I'll go
and create skunkworks.
so you go to Bangalore,
or you go to Israel,
where folks are
more likely to want
to collaborate
on some new new.
So recently, you
introduced radio.fyi,
which provides audiences
with the chance
to have these two-way
communications with the AI
personas that you were
just telling us about.
Tell us a little bit
about what this is,
and what got you
excited about it.
So right now as we speak,
in Detroit, Germany,
South Korea, China,
Japan, wherever they're
making automobiles
on the assembly line,
somebody is putting an
AM-FM antenna in a car.
If you went--
right now somebody
is buying a car, either
online or at a dealership.
And you're not going to
buy a car without radio
If you didn't have
radio in the car,
But at the same
time, radio
hasn't had any innovation
since satellite radio.
I was like, yo, let me
try to transform radio,
something that's
dear to me,
and let's take it
to the next level.
What would radio
mean right now?
That means a broadcast
needs to be interruptible.
It needs to be banterable.
And you should
be able to change
the subject
to-- you should
be able to go from
sports radio to politics
because you're
interested in it.
You should be able
to go to any theme.
Radio is great at telling
you local information.
Radio should also tell
you personal information.
Like, what do you
mean, like what?
What kind of
personal information
should the radio tell you?
Say, for example,
you're on the freeway.
Radio ain't going to
tell you where to go,
It's just going to
tell you it's traffic.
And that shit's
personal if you're late.
So just like basic
stuff like that.
Radio in this
configuration
should also give you
aspirational GPS.
It should be
able to take--
we haven't done
that yet, but it
should be able to take
the things that are coming
from my feeds, my email,
and my text messages,
and throw it into
its broadcast,
and present to me
my life in a very
colorful,
entertaining way.
I love that,
because now when
I'm late for something,
I can just say,
I was using the
aspirational GPS.
Which gave me a
very inflated idea
of when I might get there.
You built this
AI platform.
Can you just show
us something?
What would you use this
for that you wouldn't
do on ChatGPT or Claude or
one of the other AI models
Just to clear it up,
ChatGPT and Claude,
And the premise of
these types of systems
are for creators
to imagine
what you do with the LLM.
What type of soup you
want to make is up to you,
And GPT does a good
version of telling you
like, hey,
here's the types
Look how we're
using our LLM.
And now they have
multi models of--
it has fusion model
on it for TTS.
Well, give us an
example of a thing
that you would ask of FYI
in your day-to-day life.
so one day, I woke up,
and I had-- from a dream.
And I had this
very vivid dream.
And so instead of calling
my nearest friend,
I started chatting
with Finn.
Hey, so I'm on stage
here at Dreamforce.
Kevin asked me like,
yo, what kind of stuff
What kind of
conversations do you have?
I was like, I have
all these types
I think it was
two months ago--
the dream that I had, and
then the banter that we
had, and then what I want
to do with that dream,
and how can I
take that dream
and materialize it
into something else?
I'm here to share
some insights
about the incredible
journey we've been cooking
So let's break it
down, the dream.
[INAUDIBLE] a while
back, our friend
here had an
intense, vivid dream
where he stayed
awake for seven days,
symbolically traveling
around the sun.
This journey wasn't
just physical.
It was metaphysical,
pushing the boundaries
Two, unlocking
the pineal gland.
During this period
of wakefulness,
his pineal gland
was unlocked,
releasing a
unique chemical.
This enhanced
his perception,
allowing him to see beyond
ordinary human visual
Three, expanded
perception--
with this new heightened
state of awareness,
he began to see hidden
dimensions of reality,
perceiving things
like X-rays,
infrared, ultraviolet,
and other electromagnetic
Four, sentient
wave beings--
in this heightened
state, he
encountered
intelligent beings
that appeared
as waves similar
These beings were
fluid, colorful,
Anyways, that was a
freaking crazy dream
I had a similar dream
at Burning Man one year.
That was a
little different.
I don't-- I
don't do drugs.
Well, you just have more--
you get maybe melatonin
before sleep or something.
Or I'm just born
with melanin.
All right, so it
remembers these things.
It can have
conversations with you.
Or how else is
it helping you?
So let me pull
up this one here.
Let's go over
here right quick.
Pressed on grocery
prices, Trump
struggles to answer again.
I just saw this
freaking article.
It says, pressed
on grocery prices
and Trump struggles again.
There's been a lot of
buzz about grocery prices
and Trump struggling
again with this topic.
Let's break it
down right quick.
Trump got pressed
about grocery prices
And he fumbled
the response,
which seems to be a
recurring theme for him.
Inflation and the
cost of living,
especially groceries,
are big topics right now.
And every-damn-body
is feeling it
when they hit the
checkout line.
According to an
article from today,
Trump struggled with
that simple question
It's not just a
one-time slip up either.
Why do you think
he's slipping up
Or do you think
he's just, like, not
in tune with
everyday people
struggles with simple
things like groceries?
And why do you think
that might be the case?
That's a real
important question.
When it comes to why
Trump keeps flipping up
on grocery prices
and inflation,
there could be a
few reasons at play.
First off, it might
be a simple matter
Look, Trump's lifestyle
and background
are miles away from what
everyday folks experience.
When you're
living in luxury
with a team handling
your daily needs,
the nitty gritty details
of grocery shopping
and managing a
household budget
might not be fresh
in your mind.
Second, it's also
about the broader
Inflation and
grocery prices
are complex topics
tied to policies.
I think we get the point.
So this is an AI
assistant that
talks like an
extremely cool person.
No, it's an AI assistant.
You have to be
careful with that.
Because there's folks
that have assistants.
And the last
thing I want to do
is be responsible
for replacing jobs.
So it's not an assistant.
We're just conditioned
to call it an assistant.
And echoing that rhetoric.
This is just a
companion that
is able to go out, fetch
from credible sources,
so I can have a deeper
dialogue and understanding
And I can push it, and
continue to push it.
I could go out
there and have it--
how does it impact me now?
What other things
should I be looking for?
I could continue to go on.
If I'm reading an
article, I just read it.
And then I have questions.
And after I read
it, I go on Reddit,
and hopefully,
somebody on Reddit
could give me an
understanding of what
it is I read, or a
deeper perspective
And then there's
all these other bots
that I don't know if
they're tuned, or narrowed
in, or honed in on
anything that is credible,
So it is a companion for
a deeper dive in media.
it's the dawn of
intelligent media, where
the media itself
is banterable and
So I'll show you
a version of that,
And radio is
that on sports.
Play the sports
station app.
Welcome back to
FYI Radio, where
It's your host, bringing
you all the excitement
Got any
heavyweights you're
rooting for this weekend?
Because the
anticipation is off
the charts for the big
heavyweight championship
Anthony Joshua
versus Daniel Dubois.
These two Brit titans are
set to light up the ring,
carrying on the legacy
that greats like Muhammad
Ali, Mike Tyson,
and Joe Louis built.
Imagine the electrifying
atmosphere, the ring
buzzing with energy,
and the fans holding
their breath as
the rounds unfold.
Will it be a
strategic masterpiece
We've just scratched
the surface here.
Let's dive deeper into
the world of boxing.
But first, let's
hit a track
to pump some energy
into our veins.
Here's DJ Khaled,
featuring T-Pain,
Ludacris, Snoop
Dogg, and Rick Ross,
bringing you "All
I Do is Win."
There's more knockout
content coming your way.
And so at any
point in that,
you could interrupt,
and chat more with it
Or ask questions
on who produced
this song, who's
the writers,
It's not Spotify, because
I can't ask questions
about the song on Spotify.
I can't ask questions
about current events,
breaking news,
historical events.
It's just streaming all
the information that's
there in the ether, and
able to then converse
So is this-- is everyone
who's here right now,
can they go download
this and start using it?
Yeah, you can
download fyi.ai.
Engage with our
personas out here.
There's Felicia,
Finn, Fiona, Fiera.
These are some of the
personas in the app?
So we have this
one called Felicia.
Well, I have sort of
a broader question
about music, the world
that you've spent
And a lot of
creators and people
in creative industries are
really worried about AI
right now, whether
it's people using
artists' likenesses
without their consent,
or without paying
them, or whether it's
some of these
new programs,
like Udio and
Suno, and how
they allow people to
create music just out
A lot of creators
are worried
that this is going to
eat into their income
or their ability to
make a living doing
You are not
worried about AI.
In fact, you've
sort of embraced it,
and made it part
of your music.
What do you tell musicians
or other creators who
come to you and say,
"This stuff is scary.
And I don't want my label
or my team using it"?
Before I answer
that question,
I want to play
you something
- Dude, my freaking
robot outfit.
- That's Will
back in the days.
I don't mean to cut
you off or anything.
[INAUDIBLE] talking about.
You know, I always
come with that next.
- This right here
is the future.
I input my voice, high
notes, my low notes.
Then the whole
English vocabulary.
What you're able to
do with that, because
of this artificial
intelligence,
like, when it's time
to make a new song,
I just type in the lyrics,
and then this thing
It says it, raps
it, talks it.
That means I don't got to
rap anymore or something?
- I mean, you're still
rapping, but-- but--
- Like, I mean,
me, physically,
So you're saying a
machine can do anything
that an artist or
a group can do.
This is what's going to
take the Peas into 3008.
This is the
future right here.
- We're not just going
to not go to the studio,
I mean, it takes
the soul out of it.
- No, you can still
go to the studio.
It could be-- go to
studio and check out
- Yeah, but it's
not real, Will.
- It's not--
we're not robots.
- Nobody's saying robots.
- You guys are taking
this totally wrong.
- You can't say
futuristic, and then
So that was from
14 years ago.
And that came--
that vision of today
came then before
there was an LLM
to be able to-- no one
knew what an LLM was
in 2009 when we made that
video, to be able to say,
hey, you take the entire
English vocabulary.
No one knew what a
fusion model was--
to then say, and my high
notes, and my low notes,
to then type it in for
the machine to sing that.
To do that, you need a
large language model.
But no one knew
that in 2010--
And the way I
was able to see
that the projection
of where we were
is because of going
to Professor Patrick
He taught the AI class
there, and the Media Lab.
And I was super inspired
by the projection
of compute, what
compute is going to be
That was 2005,
'06, '07, '08.
So 14 years
later from when
we launched this video,
that company is now Udio.
Before it was Udio,
David and team
were a part of
Demis's team.
In 2012, me and
Demis had a--
we had the same
investor, Solina Chau.
And she introduced me
to Demis back then.
And she also introduced
me to Adam and Dag,
who created Siri and
created Viv, and then sold
Viv to Samsung
to create Bixby.
So to your question,
should creatives
Especially the
hyper creatives.
Because we'll be a lot
safer because we're
going to be able to
take this technology
We're going to be able
to dream new experiences.
The reason why we
call classical music
classical music is
because the music industry
conditioned our current
understanding of music.
When classical music
was classical music,
that shit was
just called music.
And the industry
that really propelled
that style of music was
the publishing industry,
the printing press, to be
able to print new sheet
music, to sell
more violins,
and to make a violin, and
a saxophone, and a piano.
To make the
amphitheater, that
was architecture
and technology.
Now there's new
technology that
created a record industry,
and television and radio
And now here's
new technology.
You're supposed to dream
up the new industries.
And who's going
to do that?
The folks that you
should be worried about--
we should be concerned
about and protecting are
the assistants,
because there's tools--
literally companies that
are like-- assistants,
You didn't say creatives.
You said assistants, bro.
These people are the ones,
if we truly are ethical,
we need to protect
those folks.
We need to
protect the admin,
because the
copilots are going
The creatives
will always be OK.
That's what the
fuck creatives do.
Now, how do we make sure
people are all right?
How do we make sure
our morals are right,
and not everybody's
bottom line is equal?
Because if everyone's
bottom line is equal,
that means the chief
of staff is like,
Then he fucking
just sliced down
your whole freaking team.
So everyone's bottom
line can't be equal.
Have you ever
heard a piece
of AI generated music
and thought, that's
way better than
anything I could make?
Udio is an amazing,
amazing piece
We have a new member
of the Black Eyed Peas.
You're adding
an AI bandmate.
So in 2009, we released
this song, "Boom, Boom,
And we had this album
cover from the end.
And so the reason why
we said, "I'm so 3008.
You're so 2000 and
late," is because we were
experimenting with the
concept of AI back then.
And so this album cover
is like an accumulation
of all of our
faces to make
And now Vida is a new
member of the Black Eyed
Peas that you can converse
with, engage with.
And we have a
residency in Vegas
that starts in February
till May of 2025.
And Vida is now a
part of the Peas.
And "The End," the reason
why we called the album
"The End" is because it
was the end of traditional
There was no more
record stores.
Tower Records
was shut down.
Virgin Megastore
was shutting down.
And so we called it
an end of an industry.
So we were like, yo,
it's the digital age.
And new experiences
are coming.
And at the time,
"I Got a Feeling"
was the number one
downloaded song digitally
on iTunes, because we
saw this transformation.
And just like a
transformation that we
have now, this
new intersection,
but how we should not
use AI only to mimic
we're supposed to
freaking build tomorrow.
Well, we'll have
to leave it there.
Have an amazing
Dreamforce.
