Okay.

See, it's very uplifting. Um,

[laughter]

who here thinks that the job won't be

automated by AI in the next five years?

Wow. What do you do? How do I get a job?

So, who here thinks that your job will

be automated in the last in the next

five years? Okay. Okay. So, how about

the rest of you? Like you you like don't

have a job or something like [laughter]

Yeah. So, so I just did this um question

yesterday. It was just curious like what

people online would think and it seems

like a lot of people think that the jobs

are going to be automated.

So, don't despair. I think I think I was

trying to end the talk on a very

uplifting note. Uh but I think like

recently I I launched something as a

side project. It's small. I'm happy

about it. Uh it got like some some eyes

on it, right? It's like a week. It's

like 300,000 views. And um within a day,

I got an email from someone saying like,

"Hey, I love what you you built." So I

use Clico to recreate exactly that and

here's a link. And I'm just like I'm

flattered, but also like what? like

[laughter]

so so so and so like made me realize

that

whatever exists can be replicated.

So I think like it's it makes me feel

weird because at the same time like I

just fluctuate between excitement and

despair because on on one hand right I

feel like now I can build anything I

want but at the same time anyone can

build anything I want. So what is the

incentive structure for me to do

anything right? Like I think I have

stopped using a lot of what I call like

SAS light like some some products I feel

like s a very small problem and charge

me like a lot of money like per se for

example like I don't know $50 per seat

per month for single like and I feel

like why should I keep paying for you

when I could just recreate exactly that

um like with with AI um and of course

people can tell me that like okay it's

not quite the same there are things

that's like harder to build right you

can't just get AI to do like a Google

like in a day. Of course, it takes

longer, but I also noticed that like um

AI just get better and better over time.

Maybe it cannot create recreate Google

in a day, but maybe like over time it

can. And I think there about research by

um people showing that AI can accomplish

tasks like expo exponentially more

complex. So what AI can do today is

already way way more powerful than what

I imagined it could do like just a year

ago or like way way more than three

years ago. Uh so so maybe it's just a

matter of time and people used to tell

me that like okay there are like

different modes like data is a mode but

it turns out that like data is not a

mode it's just like expensive like you

have seen how easy it is for people to

replicate dips or like GBD5 it's just

it's not it's not it's not really a mode

if somebody can just throw money at it

and acquire data and build it then what

exactly a mode here like why should I

continue building uh so sometime I call

it like I call it like the deeply moment

of software

Uh so the idea is that like a few years

ago like everyone was excited about hey

you can use AI Jared pictures in any

star and somehow the Gibly studio star

became the star that people really like

and I think was by that point I realized

that if you can describe a star AI can

generate it and just sending software if

you can describe a software then AI can

build it for you and the more you build

like the more put things out there you

remove the need for imagin ation you can

say okay now I like that website do that

for me uh and and this is quite weird so

the question I want you to understand is

like so why should I continue building

so um I'm curious here like why why do

you think that we should continue

building if whatever we build can be

copied in like very very short amount of

time

>> learning process

>> learning process that's great and then

what

>> sense of purpose

So is this make you feel like he has

purpose but if I don't do it somebody

else will do it you know like it's

somehow not needed anymore.

>> It feels great.

>> It feels great. Okay you stole my punch

line. It was supposed to be like the end

of the talk. [laughter]

Um but I do think that's one thing

that's like make me want to build is

that I built because I want to s problem

right build you create a product and it

doesn't exist in a vacuum. The product

you build is your son a problem. So that

like the more you do that the better you

become at problem solving. And one thing

I do believe that it will never change

that there wouldn't always be problems

to solve. I don't think AI will just

instantly makes me a happy person or h

like I don't think AI will like

magically make me stop being annoyed at

like customer support agents like I

don't think it's going to go away ever.

Um so I think there a lot of problems to

solve and when I look at the world of

like problem I do believe that problems

like follows a longtail distributions

and just by how AI is trained it's it

will be able to do a lot of things that

like see a lot right so I think of them

as the top of the long long tail problem

like something very common issues that a

lot of people experience AI wouldn't get

really good at it and over time AI would

cover more and more edge cases but the

edge cases this would never go away. So

there are a lot of things I consider

longtail problem. Um by the way anyone

here into prediction market

>> prediction market like anyone into

betting gambling

it will never go away.

>> Huh?

>> Other than without.

[laughter]

>> Yeah. So so I'm thinking so it was so so

it was thinking about so so I I did

build a bot to do trading. Everyone has

some of my friend was like I'm shocked

that your trading phase comes so late in

life because I feel like if you're into

engineering and math at some point in

your 20ies you just have to get into

trading. Um so so I got into trading and

I realized it's like the the bigger the

market like if some higher trading

volume the more efficient it is. Like if

you get into sport um prediction, it's

almost like impossible to compete with

like all the sport um uh trading firm or

if you get into like I don't know like

um um you cannot compete with like hedge

funds because anything when when it

becomes big enough so people with a lot

of money and amazing investors get in.

So I found out like the sweet spot it

was something that's like that is a

market that is big enough to make some

profit but not too big that like the

sharks are you know ready. So, so I

think of it as the same thing as a

problem that I want to solve, right?

Like if it's a big problem like everyone

can see then only this big companies

would get into it but where is this like

there are a lot of problems it's like

smaller then maybe open AI won't be

motivated to solve it but maybe I can

like a lot of people can and I think

like what what what do these problems

look like and I think it's like human

preference is one thing I don't think

that human preference is just an

equations that people can just like

package nicely and like hey ask people

hey which of these two answers

people would prefer. It's very very

personal, very culturally dependent,

geographically dependent, age dependent.

So, uh I'm from Vietnam and recently I

came to I went back to Vietnam and talk

with a bunch of people doing AI there

and I noticed something very

interesting. So, here when a lot of

companies deploy customer support

chatbot agent stuff, right? They usually

go the text route first. Like you do

text and be and and text is easier and

then they do voice bots because like

voice is like so much more complicated.

Like instead of having like text uh in

and then text out, you first have like

transcribe from like speech to voice and

then oh sorry voice to speak oh sorry

speech to text and then you input the

text the question from users into the LM

get back the answers and then synthesize

into the voice and then get back to the

people. So it's a lot more complicated.

So it's natural that people here do tax

first. But in Vietnam and also in a lot

of other Asian countries, people are on

the move all the time. Like people are

on the motorbike all the time. So

actually really don't like typing. So

the voice like a lot of the companies in

Vietnams actually deploy voice bots

before they do uh do chatbot. And then I

talk to them and there a lot of like

cultural nuances. Uh so for example like

the response time. Uh, so I was like

just showing um so so I have a niece and

nephew who's like very young and then I

have like my godp parents here who like

a lot older and then I put them on the

call. I feel like it's a disaster when

you get like a 60 years old American

grandparents who like 10 years old

Vietnamese boys and girls and and and my

god my godmother right because she want

to avoid awkwardness she just kept on

talking and when she asked a question

didn't answer she just kept on asking a

lot of questions try to get conversion

going and then after that I told her

it's like do you know that there's a

research that show that um in the US

people expect you to respond instantly

so like when you finish sentence people

only give like 80 second 80 milliseconds

for the other person to respond

otherwise you need to continue whereas

in Asian culture like our respect like

you we actually wait a lot longer more

like 200 millisecond or like 300

millisecond to make sure the person

finish so if you just keep on throwing

to erase awkwardness it will become

awkward because the app was like wait I

can never get my voice in so the same

with being chatbot uh voice bot right

because like voice bot you want to like

balance our latency and also like

humanness of it, right? You need to like

wait for the human to rest to like

finish. But then you if you wait too

long, it would be too slow because now

you had to generate all this like

process of like parsing and then

generating. So like all of that is very

hard to solve. You have to understand

all these nuances and all the examples

that I show are just like um they like

very obvious like cultural differences

but they are a lot more things that only

when we go into like specific use cases

and specific demographics we're

targeting that we can understand.

Another thing I think is very important

is the way humans interact with AI. So I

I think there's a lot of thing we're

still trying to imagine what would be an

AIdriven world look like, right? I think

people are trying to like retrofit

whatever exist to fit what they think is

a new workflow. So for example, um like

the IDE and the terminal. So who here is

using a lot of coding in the terminal?

So who here only started using the

terminal because of coding?

Nobody. I guess you are more

engineering. But I think saw someone

here. Thank you. Appreciate. [laughter]

Uh I I have friends who a lot of them

are like PMs or doing more of like a

product. They never use terminal before

but now because of AI they they actually

like beame like wow what is this like I

have to do this now and it's terrible

because like you cannot copy and paste

you cannot like upload a a file into it.

It's just like very annoying to use but

people use it right because that is what

what we have and people was like okay

can we have a different terminal like

why is there separation between like a

terminal and a VS code for example right

like why is there's a debate what they

do is that use they take an instruction

and they produce code like or like or

like product like why should there be

difference what's a fundamentally

difference between a terminal and and

and and an IDE so I think like a lot of

that is still like ongoing questions um

that we need to figure Wow. Another

things that I think is so very important

is the human to human collaborations

because I do think that AI is getting

really good at solving like problems for

each persons but I do think that like to

build things that are meaningful right

we need to work together and now the

human to human collaboration with a

makes actually very complicated so

here's an example um so who uses GitHub

a lot right so who here like collaborate

with your team on GitHub like via PR,

right? So who here is still review every

single PR lie by lie. [laughter]

So so so like so PR or like the way it's

like it should go rail the human to

human collaboration. So the idea that

somebody could review some like your

coworker work to make sure that it meets

the standard before merging. So I talked

to a team recently and and one of the

most senior people person on the team

told me that he still does that live by

line. Now he does not review his AI

record code live by line but he reviews

his team members AIG record live by

line. And the [snorts] reason he said oh

it's not just for code quality control

but also for education like he's

mentoring his team. So he want to give

feedback so your team can get better. Uh

and then I asked his team member like

okay do you read his feedback and they

were like no. Um because the reason is

that like it's not actionable right like

because those junior members are not the

people who write the code like if you

say okay instead of writing code like

this write like this they were like yeah

but I'm not writing the code like how do

I give that feedback to my AI so you can

write code like that. So I think that

the whole workflow of like reviewing

code is very outdated like I don't think

the I think the senior member instead of

like giving feedback on the code they

should be giving feedback on like how

you give instruction to AI to produce

better. So I'm just like another example

of like how the humanto human

collaboration is going to be very

different and another things that is I

think it's very exciting is to AI

interaction with the environment. So

right now we interact with AI mostly uh

on the computer. So think like we're

just giving AI more and more access to

different things right first is a leak

code on the IDE and then terminal which

terminal is already getting a bit more

dangerous because recently for example

like just two days ago cl just wipe out

my my postgress locally and the reason

is that it was trying to create a new

app and I already have another post

running locally and was like wait a

second this port is taken let me just

remove it so to create this new I was

like dude like um yeah so so so it's

it's very scary but luckily I have a

local backup because I'm not stupid. Uh

but but yeah, so it was fine. Uh but I

think like that made me think about like

a lot of the environments that AI

currently operate in are reversible. You

can be they can be snapshot, right? Like

code if AI mess up, you can revert back

to the last comet. Like if the even if

you have a database, okay, you can just

go back to like the previous backup. But

like there are a lot of environments

where the like they where the action

cannot be reversible by us. So let's say

that we have like an agent that we

usually fill a form for us, right? Maybe

it go to a website, enter a form and

click submit. Now as a user we cannot

reverse that action because now it

belongs to somebody else computer and we

cannot do that. Or like if we give AI

like more access outside the digital

wall like in the real world, right?

Let's take an example of like a car, it

runs over a pedestrian, you cannot

reverse reverse that, right? It's just

not working. So I do think that we need

to build out the whole guard rails for

the reversible actions because that

actually where things get really really

scary. Um and another thing that could

be very interesting is that um I do

things like AI to environment

interaction is two-way street on the one

hand we have also foundation models

function labs that are making models

better at interacting with the world but

who is making the world more agent ready

right like for example like a lot of the

website I I do things for the book like

I I write books and as much I wish that

books as a format will be like will

survive I do think that people don't

read books the same way anymore more

right like I mean I don't think people

like read books and I wish they did to

my book if someone told me that I was

like you're lying you probably jump

around a little so so I do think that

like um the world is changing and we

need to come up with format like the

websites is easier for agents or like I

was reading something else like so a lot

of the world nowadays is or like digital

world is control rate limit right and a

lot of time it makes sense because as

humans we don't do things that fast

Right. But with AI now AI can just

interact with like AI to AI can be

really really fast. So also whole

concept of rate limit is quite

irrelevant to AI like it's can become

bottlenecket also the whole concept of

search. So recently I spent a lot of

time looking into how AI do web search

and it bother me. So, so when I sent a

search as uh like uh so I did a bunch of

benchmark between grock and um Gemini

and um claude and open and GPT model to

do web search and I asked a query and I

saw that for query um some of them do

like 900 and a thousand URLs visit and

like that is so much burning my credits

like crazy um and and then I look at

like how many of this URLs are unique

and

Oh shoot. Yes, [laughter]

I think that's five. How's already done?

So, so, so and it's like it was like how

many of these a thousand URLs are like

unique, right? And it turned out it's

like only 20 of them. So the AI kept

visiting all of these LL's like again

and again and to me that's stupid and

then I realized what happened because

like it first it visited LL it took the

citations it took the part that relevant

and then it do another query it found

the parts that relevant and I feel like

it's a very human way of doing web

search right because we enter things in

the Google and we see all the like

citations the quotations part but like

why would we limit AI to that if AI only

visit a page why just p the entire page

out why do you keep on doing that again

and again. It's just stupid to me. So I

feel like or maybe not stupid, I'm sure

the people who build that are smarter

than me. I'm just saying that like my

mental model is just seem off and I I

feel like there I think there must be a

more efficient way of doing things that

are less humanentric and more AIcentric

and I think like I hope that somebody

will do that. So I do think that um

there are a lot of things you build and

the question nowadays is usually less

about like how to build because if you

can describe the problem and the

solution you want usually AI can do it

maybe not today but maybe like two or

three years from now on they can do a

lot of those. The question is like what

you build because we talk about like yes

if something exists right AI can

replicate it but like who's going to

build the things that don't exist yet?

Who should be like imagining it? like do

we want AI to be able to like just

create this solution or like imagine a

future of how humans should live or like

we can also propose this idea like think

about we want to build um and going back

to like what you said about previously

like why should we why should I

contributing is I do think this is like

because fundamentally I I enjoy building

like it just bring me joy and I do

things that um I do think that I hope

that we can normalize like building

things for fun because before I found

out that I spend a lot of energy in

doing things just like just to get to

the part of building. But now it's just

so much more fun. It's so much easier. I

can do a lot more things. Um and I think

it's like if you look at our uh a lot of

economic or like industrial progress uh

so in early day for example right

clothes like we did everything by hands

and then we have like all the mass

manufacturer clothes which is great

because like people can access to

clothes cheaply like anyone can have

like fast fast fashion but then when

people have like higher like disposable

income they start looking oh actually I

don't want mass-produced stuff I want

something that's like custom made for

me. Um, I don't think we would have a

future where when where like artisan has

software. I don't think I'm not sure

people would actually be interested

like, oh, I like this app more because

it's built by hand versus like this app,

right? Um, so I'm not sure if we'll get

there, but for now, I do actually enjoy

building apps as a gift for my friends.

So, for their birthday, instead of like

buying them something, I'm like I'm

going to spend like a weekend build them

like an app because first of all, they

like tea. I'm going to build them a t

tracking app, you know, like it's just

like very very simple and it's a lot of

fun. So yeah, so I do enjoy building and

AI does make my life for now happier

except when I'm feeling very very

depressed because I'm don't know like

why should I continue building but I

think like we can figure it out. Um but

thank you so much everyone that is that

is my talk. Um, [applause]