Original subtitles

So thank

you very much for

joining us today.

We are very excited

to speak

a little bit

about perplexity.

Let me start

about

the beginnings,

because I

think it's

very interesting.

What inspired

you to

to create,

perplexity

as a search engine?

I've been

pretty motivated

and obsessed

about search.

In fact, like, it's

a combination

of many,

like, experiences

that came together.

My childhood

has always

been about like,

was the emphasis,

like,

be a better

knowledgeable.

And when I went

to my PhD

program, Berkeley,

I understanding of

like the import

of deciding things.

And then,

I got pretty

I was in Berkeley,

so Berkeley

is pretty close

to Silicon Valley,

so you know

and then I thought,

okay, maybe

I already finished

my undergrad.

So it's a long end.

And I used to

just be like

an intern

sleeping in

the office

all the time.

And you

read all

these books,

and I was wearing

inspired by their

PhD background,

translating

to a company.

So I kind of wanted

to do

something

like that.

And it was

also interesting

how like,

citations

influence

Larry Page

to build a Google.

Because the idea

of page rank

was based on

academic citation.

So given

all these life

experiences, my,

in the

in the mid 2022,

when I felt like

I was transitioning

from something

that was

very research

project oriented

to like

actually being,

consumer facing

applications

with like, projects

like GitHub

Copilot really

breaking out

and like,

getting a lot

of adoption.

I felt like

I could go

and realize

my dream.

And, that's

when, like,

you know,

I got to realize

my dream of

starting company

and working

on things

that truly

are exciting to me.

My investors said,

hey, like,

this is too far out

and future

specific thing

you build up

from there

and all connects.

It's very hard

to search on.

You can't go

and find

any specific post.

You can go and find

which was

having like by

which people

you can go

and find posts

of some person

like by

another person.

You can go

and find like

who follows

this person

that doesn't follow

the other person

like these kind

of searches.

So I wanted to

build something

like that

where

you could not do it

without language

models like you.

Kig any query,

you would have

to convert that

into some kind of,

structured query

language SQL.

It's called SQL

running

against the

database of

all tweets

organized

in

relational databases

and then

pull it together.

But because

language models

began to work

like coding

language models,

we built this

amazing demo.

It was working

very well that like

not only I use that

or any investor

I potential

investors

showed it to,

they all started

using it

because

something like

that never existed.

So I think that's

what gave me

the confidence

that like

we could go

and do this

for many other

domains on

the internet,

not just Twitter,

LinkedIn, GitHub,

a lot of

these websites,

but then

it would be

a very

difficult thing

to scale this up

if we did a domain

by domain

with domain,

because the

internet has so

many domains.

So we

found a more

general solution

that just

extracted links

and then let the

AI model

do the reasoning.

You, when you're

trying to create it

like it

starts from a

very simple idea

that's very

ingrained

in nature.

You scope it down

scope of now

and then, you solve

some of IT

problems,

and then when

and then you

try to scale up

and you

realize like,

oh, well,

there's a much

simpler solution.

And then

that ends up

becoming a problem.

What do you think

are the

biggest milestones

for perplexity?

This last year?

So at the beginning

of last year,

a bunch of links

given to you

for like

2 or 3 word

queries on Google.

So, at the

time of

Google's IPO

in 2004,

Google served

like 100 to

200 million

daily queries.

So, obviously

the internet

was much smaller

at the time,

which is actions.

So that's

the big focus area

for us this year.

And I think

continuing

to get better

at answering

even harder

questions,

ambiguous questions

where sources on

the web

are not sufficient

answer

when I

to go and

elicit input

from some experts

on the field

or proprietary data

sets models,

chaining all

these actions

together

to accomplish that.

Basically more

agent work

is, is kind of like

how we,

we plan to adapt

to all the fast

changes in AI.

How do you think

perplexity

nowadays?

It's, changing

the way

that the

people use AI.

Yeah.

So yesterday

we released

this thing called,

the Assistant.

It works on

Android phones.

Okay.

So the way

it works is,

like all Android

phones have,

like, a way

to activate

the system.

And, I'm

showing this

demo on a pixel

when you can

use the

power button.

So you just

click that

and you say,

can you play,

video of,

Donald

Trump singing

Baby Shark,

playing Donald.

Really useful.

Or you can,

you know,

like loud sings.

Can you play

Despacito

on Spotify?

Okay.

Playing

Despacito by Luis

Si and Daddy

on Spotify.

Like for a

searching the web,

then finding

some detail, then

going

your app, then

setting some

action on

on top of it.

Or like you're

doing research on

like when

is the next.

Oh classical match.

I like

can you make a

ticket for me

or like

can you go

and find the

tickets available.

What is the price.

Like the.

I think this is

very like.

Your regular task

that the.

People do

executive around

periodically.

Right.

Oh like

every morning

give me an alert

of like

the stock market

or whenever

Nvidia stock

goes or like let's

say Bitcoin

goes about

5% on a day

automatically,

you know,

if you find

the velocity

of increase

is pretty high,

buy some Bitcoin

on my behalf

or like anytime

Bitcoin goes below

a certain

threshold,

if you really

believe

in the future

just by like

thousand dollars

with Bitcoin.

For me,

I think the money

like like

you know existing.

So the users

the big tech has.

So

what new things

can we expect

for perplexity

soon in the next

the months.

Yeah.

So we,

we want to make

all those agents

working better,

give a read, access

to your emails

and, like,

calendar and,

like, remind

you about things,

cancel meetings

that are

not productive

for you.

Like, like, book

flights on

your behalf

and all stuff like,

get truly

personalized,

like, understand,

like what

you really want

and do stuff

on your behalf.

I think that's

that's very one.

Good. Okay. So

let me put in, in

other contexts,

let's think about

five years. Yeah.

So I think it's,

it's a

lot in AI but.

What do you think

perplexity could be

in five years?

I think it can be

one of the world's

best, assistants.

Like giving

the life

of a billionaire,

the average person.

You know,

let me

give you, like,

some examples,

right?

Like,

once upon a time,

like vacations

was a luxury.

Okay. Right.

In fact, like,

only really

rich people

could afford

a vacation

to, like,

even places

like Hawaii

in the US and,

because it

just cost

a lot of money

to go there.

Earlier

flights was

full time,

but with the

indulgence of

an Einstein.

Taking a look

at your life,

your problems,

and your

constraints

and trying to

make it better.