April 2, 2025 • 47 mins
In this special episode, hosts Mark and Shashank take a break from their usual news coverage to explore the rapidly evolving world of AI agents. They define what agents are, examine their current capabilities, and make bold predictions about how these technologies will transform our lives over the next 1, 5, and 10 years.
The hosts discuss how AI agents are already revolutionizing software development, research, and reasoning tasks, while exploring the imminent impact on knowledge work, self-driving vehicles, and scientific breakthroughs. Looking further ahead, they predict the widespread adoption of humanoid robots, the "YouTubeification" of product creation, and fundamental shifts in employment patterns.
Whether you're new to AI or a seasoned professional, this episode offers fascinating insights into how exponential growth in AI capabilities will reshape our society, economy, and daily lives in ways we're only beginning to imagine.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(59:59):
today we are gonna do something a little
bit different. So normally we would be
talking about the news and stuff, but
Shashank and I were thinking that we
wanted to make a video where we kind of
brought everybody up to speed on what an
agent is, and also what our predictions
are gonna be for the future. So we aren't
(01:00:20):
gonna talk about any of the news this
week, but we're just going to speculate
on what we think(...) is gonna happen in
the future. So we are gonna predict not
just one year in the future, but we are
gonna predict up to 10 years into the
future. So we're gonna try to keep this
short, shorter than our normal podcasts,
and just try to get right to the facts.
(01:00:41):
So Shashank, you ready?(...) Let's do it.
All right. So we wanted to present what
really is an AI agent, what we have
experienced so far, what we have today,
and where we're headed. So when we talk
about agents in the broad general term,
what is an agent? An agent is someone,
something that can carry out a complex
(01:01:04):
task for you with relatively minimal
supervision. So in the traditional world,
pre-LLMs, you can think of a travel agent
who does a booking for you, scours
through multiple websites, finds the best
price, the best layovers, the minimal
flight time, and then presents a couple
options to you. So a couple examples
(01:01:26):
would be stock predictions,
order taking, trip
booking, travel agents, et cetera.
Yeah, so these agents in the past were
actually pretty good and they showed like
a ton of potential.(...) So some of these
agents that maybe you guys have heard of
are things like baby AGI or auto GPT or
(01:01:48):
GPT engineer. Actually, I created GPT
engineer, so you guys
hopefully probably use that one.
My GPT engineer was a thing that you
could use to do programming.
So you could put it a prompt, say like,
"Give me a Tetris game," and it would be
able to create something like Tetris or
like something simple, right?
(01:02:10):
And this worked okay, right? So a lot of
these agents they had,(...) they were
able to do basic coding and tool calling
abilities, right? You could do basic
stock analysis. Maybe you could do a
little bit of customer service.
You could do RAG, for those who don't
know, RAG is retrieval augmented
(01:02:30):
generation. It's just when you have the
agents or like the AI access some sort of
source documents, but there was a lot of
problems with the agents.
So one of the big problems is a lot of
them had like really tiny context
windows, meaning they
couldn't fit a lot of information
inside of their model, right? So you
(01:02:53):
could probably make a Tetris game, but it
was really challenging to make a really
large software program, right? So, and
even today, right? Like the programs that
we've been able to make have gotten
bigger, but you still can't have all of
Android being created by AI agents yet.
(01:03:15):
So there's a lot of problems that are,
like have been in the past that are
needing to work on being solved.
Which brings us to today. So I don't
know, Shashank, do you want to talk about
like where we are like in the current
state of the world with agents?
Yeah, so things have gotten a little bit
(01:03:35):
better, but not by much. So I think we've
mostly benefited from the models getting
better. There are capabilities just, you
know, exponentially growing month over
month, year over year, the token context
limit has dramatically increased with
Gemini touting a million token context
window. And in my experience, it's
(01:03:56):
actually really good. It's one of the
best ones out there if you're dealing
with large amounts of context. And to
broadly understand why that is important,
it's not just about you pouring the
entire Harry Potter series into the
context window. If you expect these
agents to carry out complex tasks for
(01:04:17):
you, you want to give them tools. You
want to give them the ability to call
functions outside of their neural
network. The ability to use a calculator,
maybe check the weather, hook up into
Yahoo Finance and check the stock prices
or check the options, trading volume, et
cetera. So each one of those tools that
(01:04:39):
you add into a model is stored in its
context window. So we're learning that it
requires a lot of information for an
agent to be like a truly general purpose
agent that can solve anything.(...) So
today we have
specialized agents that somehow,
(01:05:01):
but that somewhat mitigate the challenges
of having a generalist agent.(...) We
have amazing coding agents, Claude Code
just recently came out. Cursor has been
doing a fantastic job after the big
splash that GitHub CoPilot made.
We've had truly independent AI engineers
(01:05:24):
(...) from Devon and OpenAI also
reportedly working on a fully independent
AI coding agent that is going to cost
something close to $20,000 a month, which
is orders of magnitude more than their
$20 a month plus subscription.
(01:05:44):
And we have other kinds of agent in
limited scope like deep research. It does
a bunch of search for you.(...) So deep
research, if you don't know, is a feature
available in most models today actually.
Started from the Gemini models where they
allowed you to give it a complex task and
(01:06:05):
this model wouldn't just spit out an
answer. It would think for a while on the
order of several minutes,
go out there, do a bunch of Google
searches,(...) analyze all of its results
and figure out what you were trying to
ask, break down the reasoning in multiple
steps and then come back several minutes
later after thinking for a lot, after
(01:06:26):
prompting itself over and over again
multiple times to give you
a well thought out answer.
Now moving back to some of the coding
examples, you have, apart from the coding
IDEs, the software tooling for
developers, you have entire solutions for
(01:06:47):
non-engineers to build applications. So
we have really popular solutions like
vzero.dev, which is the parent company of
Next.js, a popular web development
framework. There's a couple other
independent companies like Lovable Dev,
Bolt.new, Replit, which is also a very
popular company that
(01:07:07):
builds tooling for developers.
Another tangent is the reasoning models.
We have amazing capabilities that have
come out from multiple companies that
allow these LLMs to not just give you one
single answer from all the knowledge that
it contains, but actually think about a
problem and give you a solution like
(01:07:29):
counting how many R's
there are in Strawberry.
LLM can't do that because it's only
generating tokens into the future,
whereas a reasoning model can look at the
word Strawberry, break it down into
individual letters, pass it through some
kind of a calculator that tallies up how
(01:07:50):
many letters are there, and
then gives you a final answer.
And moving one step beyond just
intellectual agents,(...) we have agents
that can take action for you in the real
world. One popular example that's been
around for a few years, maybe not in its
full form, is self-driving. So, Tesla has
(01:08:13):
had its self-driving out for a while,
even though it may not be what they have
marketed it as, but it is also an agent.
It takes in information from the real
world and takes action in multiple
different areas. It can accelerate,
decelerate, adjust the steering,
(01:08:35):
and make sure that ideally you don't have
to put your hands on the wheel.
Other companies like Waymo, there was
Cruise for a while, and Nvidia is also
working on a self-driving agent that can
power multiple vehicles. So, we have
intellectual agents, we have physical
agents, we talked about self-driving, and
(01:08:56):
one last thing that multiple cartoons and
sci-fi movies have been promising us for
a long time is robotics. So, we've had
industrial robots for a long time, but
now we're seeing truly generalist robots
that can take action in the real world
from a simple prompt like, hey, make me a
cup of coffee, and the robot goes into a
(01:09:18):
random kitchen and makes a cup of coffee.
We're not there yet, but we're seeing the
seeds of what is possible from companies
like Figure AI, Google Robotics, and even
Nvidia, partnering with companies like
Disney to make little robots that are
reminiscent of the ones from Star Wars
like C3PO. So, we're seeing hints of real
(01:09:42):
robots,(...) and it is the foundation for
a lot of exciting things that we'll see
in the future, which, Mark, do you wanna
show us what we can expect?
Yes, so we don't exactly know what to
expect in the future because if anybody
can confidently tell you what's gonna be
(01:10:03):
in the future, they're probably lying or
wrong, but we're gonna be both lying and
wrong in this presentation because we are
going to tell you exactly what we think,
expect to see in the future.
So, I think there's a few things that I
expect to see, right? So, one is I think
in the very near term future, like 2026,
(01:10:27):
there's gonna be a lot less job openings
for software developers. So, Shashank and
I were kinda debating this, whether
there's gonna be more software jobs or
less software jobs. On the one hand, you
could say, "Oh, there's gonna be more
software jobs "because of Jevons
Paradox," where it's like, if you have
software become a lot more easy to write,
(01:10:48):
then you just need less or like people
might just wanna create more software and
more people are creating software then
you might need more software developers.
But I think that realistically,
these AI models are able to replace the
(01:11:11):
job of like hundreds of software
developers at a time where you used to
have schemes of people who needed to make
software projects. Now you're gonna need
just a handful of senior developers to go
and create the software. And even people
who are junior developers now are able to
(01:11:34):
create more meaningful software in the
future. And even people who don't know
how to create software are gonna be able
to create software. So, maybe like there
will be more people who will be able to
do software development, but I think that
like the job of software developer in the
future, it'll still be there. I just
think that there will be a lot less
(01:11:55):
people(...) who are only purely software
developers going forward. And also,
like realistically,(...) I think that not
just software developers, but almost all
knowledge work will have their hiring
start to slow down. Now, it won't just be
(01:12:17):
because of AI, but I think that a lot of
the hiring will slow down often because
people will be aging, there'll be
attrition,(...) there could be economic
slowdowns, all sorts of things.
And AI will just be like one such factor,
which is gonna make this hiring slow down
(01:12:37):
like quite a bit more.
And then in addition to that, we expect,
so Shashank mentioned about some of the
physical AI agents. We think that
self-driving will be more widely deployed
in more locales. So for example, Waymo is
(01:12:59):
only in I think like four
or five different cities.
Like our prediction for the end of 2026,
now we don't know, but we would suspect
that by the end, like the number of
cities that you see self-driving cars
(...) will probably double if not maybe
more, but that's the prediction like, so
(01:13:20):
there's a high likelihood a city that you
live in, if it doesn't have self-driving
cars now, next year you'll probably see a
lot more self-driving cars. And then one
other bold prediction is we think that AI
agents will start to solve
(01:13:43):
unsolved problems that humans have never
been able to solve. But not in all
domains, but we think specifically,(...)
we may see within the next year, like by
the end of 2026, AI start to solve some
unsolved problems in math and logic. So
my hope is on the Riemann hypothesis. I
(01:14:05):
think that there's a really good chance
that's able to be solved
within a year, we'll see.
And yeah, I think that because of things
like open AI is $20,000 a month, coding
agent that's a shock mentioned,
because of all of these new AI wrapper
companies, a lot of people who were doing
(01:14:25):
knowledge work like
language teaching, coding,
like crunching numbers,
a lot of those jobs, the hiring is gonna
slow down a bit. I think they'll still be
around in 2026, but people try to get
into it, may have a hard, hard time.
And yeah, honestly, really starting
(01:14:48):
today, but for sure 2026, humans, we
don't think will be really special
anymore in terms of intellect. When I say
intellect, I mean, like reading books,
like working with a computer. I think in
2026, humans will probably still be
special in terms of physical ability,
(01:15:10):
right? Like plumbers probably still be
around,(...) but just being like a smart
person who can work in
front of the computer,
that is no longer guaranteed to be like a
job for the next 40 years. Computers will
probably outpace most knowledge work
(01:15:34):
going forward. And maybe the only thing
that will separate humans is our taste,
(...) meaning like, maybe humans will be
better at determining
what other humans like.
So it's like, if you go to a museum and
you see a picture or like a painting on
(01:15:54):
the wall, and you say, I like that
painting. I have good taste. You didn't
necessarily create that painting, but you
like that painting and you bought it for
your home. Like in the future,(...) i.e.
maybe next year, the humans might just be
selecting which AI model gave them the
best output and decide which one they
wanna use.(...) So yeah, that's where we
(01:16:17):
think we're gonna be the next year. We'll
see how it turns out. Shashank, but like,
what do you think about going even
further? Like, should we try
to speculate even farther out?
Yeah, I think the same
trend is gonna continue.
If you wanna bump us to the next slide.
(...) I think the agentic craze is just
(01:16:39):
gonna skyrocket.(...) I think just as
everyone today is using chat GPT or LM,
squad, Gemini, whatever you have. I think
everyone in the next five years is gonna
be using some form of an agent, whether
it's for work, whether
it's for personal tasks,
therapy, something to help you keep track
of your tasks, emails,(...) I think
(01:17:02):
everyone is gonna be so used to handing
off complex tasks to an agent just as
they do with single problems and LLMs
today. Not just that, I do think this
agentic ecosystem is gonna be a lot more
mature than it is today. Because right
now we're still trying to figure out what
(01:17:24):
these LLMs are capable of because they
don't function as deterministic software
that we've been used to for the last 30,
40 years. I think we'll have a better
grasp of what kind of abilities and
emergent abilities these agents will have
(01:17:44):
and be better at using them throughout
society, throughout our economy. So I
think this is something that both Mark
and I agree on. I think every single
person doing intellectual work will be
doing that with agents. They'll be using
agents as junior assistants, as
(01:18:06):
researchers,(...) as junior engineers
even, to do things for them to
supercharge their work.
Yeah, I just wanna cut you off right
there. I think to be incredibly clear on
what Shashanka's saying here, our
prediction is that by 2030,
maybe not 100% of people, but 99
(01:18:28):
virtually 100% of people will be not
doing knowledge work
without the help of a computer.
They will be using agents to help them
manage their work. I think that we think
that the next couple of years, if you're
not using AI, you might be okay. But by
(01:18:50):
2030, if you're not using AI, you will no
longer have a job. Like almost 100% of
people will need to be using an agent or
like working with AI in
order to do their work.
Yeah, and I think this is gonna slowly
creep into the physical world as well.
(...) With the incredible progress that
(01:19:11):
some of these companies are making, we
have seen version one and version two of
figure AI. They are reportedly working on
version three, which their founder has
mentioned in a couple of interviews. And
the amount of progress is just
staggering. The fidelity, the
fine-grained motor controls, the battery
technology,(...) the AI powering the
(01:19:37):
movements of these robotics has gotten
significantly better. And I think this
trend will continue. So we'll see immense
progress in humanoid robots from figure,
from Tesla,(...) from Chinese
manufacturers who are able to bring the
cost down significantly to a point where
I think individuals in maybe developed
(01:19:57):
countries will be able to afford a
humanoid robot to keep in their homes to
assist them with day-to-day tasks. I
think we will start seeing robotics in
industrial environments first because it
is a lot simpler. The needs of industrial
warehouses, for example, are a lot more
straightforward and repeatable compared
(01:20:18):
to an individual's, like a consumer's
home, because the amount of things that
are in a individual's homes can be very
diverse and each person's home is very
diverse.(...) So I think that may take a
little bit longer, but we'll start seeing
humanoid robots enter the scene a lot
(01:20:39):
more than we do today. I think there'll
be significant progress in self-driving.
We'll have self-driving in most major
cities because we have full self-driving
from Tesla available right now, although
it doesn't work everywhere all the time,
but it is getting better. They're on
(01:20:59):
version four of their hardware.(...) And
we did just mention how Mark Rober, a
popular YouTuber, showed that the Tesla
cannot detect like a Looney Tunes cartoon
wall that looks like a real road if you
prop it up with some cardboard and
printed paper. But the latest version,
(01:21:21):
which another YouTuber just put out a
couple of days ago, has shown that the
Tesla can detect a fake wall and doesn't
just go straight through. So their
capability is also improving.
We think that they'll, apart from
robotics, we do think there'll be a
(01:21:41):
significant breakthrough in fundamental
research in basic sciences. I think
people will start using AI agents to
assist them in physics, chemistry,
biology, to conduct experiments, to run
simulations, to help them supercharge
their work and speed up the amount of
time for scientific
breakthroughs by a significant amount.
(01:22:04):
Yeah, and I think there's a high
likelihood, we didn't put this in the
slide, but I think that with a lot of
that AI research, we may see by 2030 that
we've finally reached longevity escape
velocity. Meaning if
you're alive in 2030,
there's a good chance that you could just
(01:22:26):
continue to live indefinitely because of
the rapidly increasing
healthcare ability.
Essentially, if you live till 2030,
there's a good chance you might be able
to live long enough to live forever.
That's a tricky one. I do think longevity
(01:22:46):
and developments in longevity will be
significantly better than they are now. I
think we'll understand the human body a
lot better. I think we'll understand
causes of aging, disease, and how to
optimize our nutrition, our gut
microbiome, health, exercise, et cetera.
But we did look at the top 100 oldest
(01:23:09):
people alive, and there does seem to be
some kind of a biological threshold
around 114 to 117 years where the top 195
of the top 100 people
max out at 114 to 117. So, at the very
least, I think that more people will
(01:23:29):
start reaching that threshold as opposed
to what it is today,
which is about high 80s.
Yeah, I agree. I don't think that we're
necessarily gonna see people live to 125
or something. Because I think the oldest
person now is 114, 115. So, I think it's
(01:23:50):
unlikely that they're necessarily gonna
make it to 120. Because I think that if
they are,(...) even if we have advancing
medical technology, those people may not
want to try some sort of fancy new AI
thing, right? But I think what I'm trying
to say is that the people who are alive
(01:24:13):
in five years, maybe in their 40s or
maybe in their 50s, who are maybe more
open to trying some of
these advanced therapies,
they might be able to live for another
100 years because they're actually trying
to use these things. I think that it's
not that everybody is gonna live like a
(01:24:33):
really long time.(...) Like a lot of
people will just eat Big Macs all day and
not get enough sleep. But I think for the
people who are actually really paying
attention to their health and are
following the latest therapies and are
willing to try,(...) those are the people
who may be able to reach longevity escape
(01:24:55):
velocity and actually see some of the
pace of aging reach close to zero. So,
like actually like a very specific
prediction is that right now there is the
longevity leaderboard. I think like Brian
Johnson is running like one of these
(01:25:17):
longevity leaderboards where they have
these different pace of aging tests. And
I think that right now, according to the,
some of these aging pace tests, you are
around, like I think for every 12 months
that have passed, you according to these
aging tests have only aged like six
(01:25:38):
months. I think that there's a high
likelihood by 2030 that your pace of
aging will reach zero. For every year
that has passed, you will not have aged
at all. And that's my prediction. So
we'll see, we'll see. I could be wrong,
but that's my prediction
that we'll see that by 2030.
(01:26:00):
All right, so now we're getting into the
realm of science fiction, which is crazy
given that we're only five years out, but
let's look at the next 10 years, at least
our predictions for the next 10 years.
What do you think we
have coming up, Mark?
I think there's a lot. And I think before
(01:26:20):
we get into that, I think that we just
need to, I want to tell a little bit of a
story. So my mom, when I was a kid, she
wanted to teach me about exponential
growth. And if you guys don't know about
the beauties of compound interest, it's
really cool, right? So it's like,
(01:26:42):
if you save a little bit of money,
it compounds that money, and then you can
make a lot more money over time, right?
So it's like, if you start with $100 and
you make a 10% rate of return, after year
one, you have $110, not crazy, right? But
then after year two, you're gonna make
10% on your 110, right? So instead of
(01:27:03):
making $10, you'd actually make $11. And
that actually continues. And it actually
only takes, I think, about seven years to
double, right? So after seven years,
you'd have $200, assuming you just grew
that $100 every time. And then after 14
years, you'd have $400, et cetera, right?
(01:27:24):
So imagine you had a lake, right?
And on day 30, that lake was filled with
water and it was filled with algae.
On what day, Shashank, I mean, this is
probably an easy question for you, but on
what day is the lake
(01:27:45):
half full with algae?
What is the rate of algae, blue?
Let's say it doubles every day, right? So
let's say the algae doubles every single
day. And on day 30,
it's 100% full. And with,
on day one, you just start with a 100%
(01:28:06):
full. One little cell of algae.(...) On
what day is it 50% full?
I guess day 29, one
day before the end, yeah.
Yeah, day 29. So I mean,
I don't know. I think that the point I
have here is that a lot of these things,
(01:28:29):
they feel like they're happening slowly
when you are there, right? So it's like,
if you just look at the algae and it's
just like, oh, it's doubling, but it's
just getting doubling a little bit
faster, right? But, and then it's like,
oh, day 29, there's still like half the
lake that needs to be filled. But then it
just takes a year for the entire lake to
(01:28:49):
be like overrun with algae.
And that can happen with algae, that
could happen with COVID,(...) and that
can also happen with your bank accounts,
but that also can happen with AI in the
future. So I think that remember this
when we're predicting the future,
(01:29:09):
some of these predictions may seem crazy,
but like a lot of this exponential growth
happens way faster than you think.
So anyways, that's a little prelude
for why we're making
some of these predictions.
And just to add a little bit, you talked
about compound interest and how every
(01:29:31):
financial guru touts the benefits of
compound interest.(...) Compound interest
in the financial market would be like,
what is it? Like seven to 10% if you put
it in safe or like even
high growth index funds.
Moore's law and exponential growth(...)
is like 50% every year. It is like
(01:29:53):
significantly faster than what you would
expect with like compound interest with
your retirement funds at the end of a
lifetime of investing and working.
Exponential growth with Moore's law and
some of the other laws or heuristics that
we have, which is just rapidly
(01:30:13):
progressing the technologies, the
underlying chips, the algorithms, the
models, the tooling around it, the
infrastructure, everything, is just
accelerating at a pace that we've never
seen before. And they are self-improving
so that the pace itself is increasing
(01:30:34):
every year. So it's
just mind boggling numbers.
Yes.
So, all right, let's get into some of the
strategic predictions. So the first one,
which we kind of mentioned at five years
out, we think it's still gonna be true.
100% of knowledge work,
(01:30:56):
if there even is jobs,
we think that there'll be less jobs doing
knowledge work. And the jobs that
knowledge work has will be just simply
managing AI agents or teams of agents.
And a lot of that work will probably even
start to get phased out as the agents
just get really good
(01:31:16):
at managing themselves.
So we think there'll be a lot less people
doing knowledge work and the people that
are doing knowledge work are just gonna
be running companies of
tons and tons of agents.
So I did have an alternative theory for
this. One consequence of the Jevons
paradox, which was it
(01:31:38):
related to electricity?
Regardless, when some fundamental
resource becomes significantly cheaper,
the usage shoots up dramatically. So I
think we might see a
future where there are,
every single person will be an
(01:32:00):
entrepreneur. Every single person will be
able to use AI to build companies,
products, features for a very, very
specific niche use case. That way we will
see an explosion of software that is
customized and tailored to every single
thing that you can think of. And people
(01:32:22):
will be building software on the fly to
solve specific things that they want. So
either everyone will be phased out of a
job or everyone will be entrepreneur
where they're able to use these tools to
do things that they need in such minute
detail that we've
never experienced before.
(01:32:43):
Yeah, I think that's true. Like actually
Shashank and I were
talking about this before.
I think that that will happen.(...) I
don't know if everybody will be an
entrepreneur, but probably a lot of them.
And I wanna be specific when I talk about
knowledge work. I think that when I say
knowledge work, I don't mean companies,
(01:33:05):
but I mean like as an employee.(...) So I
think that there will be very, very few
companies that are hiring employees to do
knowledge work. If
you're the owner of a company,
that's a little bit different I think.
But like if you're the owner,
your first instinct is not gonna be to
(01:33:26):
hire somebody. It's gonna be to see if
you can agent to do something for you and
if you wanna delegate, you're just gonna
somebody like run a team of agents.
But yeah, I think to kind of add on to
Shashank saying is I think that we're
gonna see something like YouTube. So if
you think of YouTube,
it is how a lot of people, actually I
(01:33:48):
think it's the most watched video
streaming site in the world. Like more
people watch YouTube than they watch
broadcast TV. More people watch YouTube
than they watch Netflix. More people
watch YouTube than they watch ABC, NBC,
all of the media. But what's different
about YouTube is it's not a big company
(01:34:09):
that's making it. I mean, sure like
Google manages and maintains YouTube. But
Google isn't posting a bunch of content
on YouTube. They aren't controlling what
you see. I mean, they kind of are with
the algorithm. But at the end of the day,
like what you're seeing is you're seeing
videos made by individual people, like
(01:34:29):
oftentimes just with their cell phone. So
like if you have a cell phone and an
internet connection or a laptop and
internet connection, you can make a
YouTube channel and spread your message
to the entire world. Whereas in the past,
there used to be only a few major
broadcasting. It was like, in the past,
it was like three channels that you could
(01:34:49):
watch and everybody
saw the exact same thing.
And we're able to see like such
personalization with YouTube that you can
find whatever niche that you're
interested in and just watch that. So for
me, like I really like
watching content about AI
or like I like content about like watches
(01:35:11):
and stuff or Japanese learning. Like
these are some of the things that I'm
interested in. And if I'm watching
regular TV, I can't find that, right? But
on YouTube, there is a bunch of content
creators making content for exactly what
I want. And I think to add on to what
Shashank's saying is we're gonna see this
trend continue where it's not just to be
videos, but it's gonna be for every type
(01:35:34):
of software,(...) games, and we're even
gonna probably start seeing it for like
physical things, right? We will have 3D
printers that are gonna be able to make
like really, really specialized doodads
that you can use that it's like, you say,
(01:35:54):
oh man, I really wish I had a thing that
could fit this just ever so perfectly.
And there's gonna be some guy who's gonna
make a company selling just that like one
little widget that you need for your side
project. And if you can't find it, you'll
be able to easily make it and maybe even
have AI created for you. So that's
(01:36:15):
happening and that's gonna happen soon.
So there's probably gonna be a lot more
companies that are just a lot smaller.
Yeah, and to kind of summarize what Mark
was trying to say, I think the
YouTubeification is basically
democratizing access to all of these
esoteric technologies(...) because we can
(01:36:37):
do some of these things today. It just
takes a lot more effort. It might take
more effort than an individual(...) can
realistically put within their lifetime.
But within the next 10 years, I think
natural language will become the mode of
work. We'll be able to talk to machines,
(01:36:57):
to embodied physical robots, or even
virtual agents to do tasks, knowledge
work, physical work for us and get things
done like immediately, which would today
maybe take an entire company to do.(...)
So moving on to some concrete examples,
(01:37:18):
I think a lot of scientific research will
have AI agents assisting them. And we
will see maybe significant scientific
research be done by AI agents and maybe
see the first Nobel Prize being won by an
AI as opposed to a physical person.
(01:37:40):
We talked a lot about self-driving.(...)
We think this will become the norm.
Elon has been really bullish on his
Tesla, what is it, RoboTaxi.
I think that is a significant
opportunity. It is a potential massive
market that could be created because
(01:38:02):
right now, if you think about the
utilization of a car, it
is maybe 10% of the day,
whereas the other 90%, it's sitting
parked in a parking lot in a garage.
Definitely no one is using at night when
you're asleep. But even during the day,
when you're at work, when you're at home,
(01:38:22):
it's just sitting there idling, waiting
for somebody to use it, as opposed to a
self-driving fully autonomous vehicle, it
could be out on the roads a majority of
the time, maybe even having a battery
swap that just allows it to hop in and
out of some kind of a station and keep
going. Or we've seen how some of these
(01:38:42):
Chinese EV manufacturers able to have
insane amounts of charging where they're
able to charge a car(...) in the same
amount of time that it would take you to
fill a tank of gas. I do think that the
self-driving industry will mature to a
point where cars will become a service as
(01:39:04):
opposed to something that you own.(...)
Similar to, I guess, the uber-fication of
things where people can just door-dash
anything that they want. You can go live
in amazing vacation rentals that you
don't own where you're sharing someone's
home. And similarly,
share access to a vehicle.
(01:39:28):
Do you wanna take us through some of the
advancements in robotics, Mark?
Yeah, for sure. So,
and just to add on to what Shashank was
saying, in terms of the cars will become
a service,(...) we think that every
single new car will be self-driving.
Except cars that are just purely for
(01:39:55):
hobbyists. If you're doing a, I don't
know, a car that's gonna race around the
track, or you're buying a Ferrari or
something, and you really
just wanna drive that thing,
most cars that aren't
those, like your Toyotas,
(01:40:15):
your Honda Pilots of the world, those will become a service.
Those will be all self-driving.
So, it's gonna be weird to buy a car and
then drive it yourself.
There probably even, maybe even laws that
prevents people from driving cars that
aren't self-driving in the future. Now,
(01:40:36):
we're not saying all cars will be
self-driving because the cars today will
still be on the road in the future, but
all the new ones, those will be
self-driving. And also, we think that
humanoid robots are gonna be
super common around the home.
You are going to take the money that you
saved(...) from your car and your
(01:40:59):
robotaxi, and you will probably use that
to buy your humanoid robot, butler,
gardener, cleaner thing.
It will,
like, my prediction is it will probably
be, treated something like a car where
it'll be kind of expensive. It might be,
(01:41:20):
I don't know, like $100,000 for one of
these things, you know, with inflation.
Maybe like 20 to $50,000 in like today's
money,(...) similar to that of a car. And
you'll probably pay for it over time, but
it will save you so much time
that you will not have imagined life
(01:41:42):
about it, right? It'll cook for you,
it'll clean for you, it'll
make your life a whole lot easier.
And these things are gonna be super
common. Not everybody's gonna have them.
They will be expensive. And there'll
probably be some sort of rental options
to have it, but they'll
probably be everywhere.
And also, we think that you would
(01:42:05):
probably interact with more of these
robots than you will with, like,
actual human beings,
like, on average, or maybe like a similar
amount. I mean, if you're a really,
really social person, maybe you'll
interact more with, like, human beings,
but it'll probably be that you're
interacting with these robots all the
(01:42:26):
time. You're gonna go to a restaurant,
your food is gonna be delivered by robot.
You're gonna go and go to the gas
station. Like, you may not pump the gas
anymore. I mean, like, it might just be
the robot arm that comes and puts the gas
or just, like, plugs your car in, because
your new car may be electric, right?(...)
(01:42:46):
It's gonna be, you go to a restaurant,
the entire staff in the back may not be
actually, like, humans cooking anymore,
but it'll be the robots. Honestly,
probably the only places that you'll see
humans is where people prefer to interact
with humans over robots. So, for example,
if you're at a restaurant, you don't
(01:43:07):
necessarily care about how your food is
made. Like, it doesn't really matter if
it's, like, a human chef versus a robot
chef.(...) Like, you don't actually see
who's cooking your food anyways, right?
But, like, if you're at a bar, a lot of
people prefer a human bartender just
because they like talking to a person
who's there, right? Like, you're probably
not gonna see too many robot massage
therapists just because people like to
(01:43:29):
get a massage from a human, right? Like,
even though we have massage chairs,
people still go and get a massage from a
human just because they
prefer to have humans, right?
And, like, so some of that stuff will
still be there, right? Like, I think
human art will still be around, but these
humanoid and non-humanoid robots are
(01:43:50):
gonna be everywhere in the future, and
you're gonna interact with
them a lot. Definitely a lot.
I mostly agree with you, although since
we're talking 10 years out, I just think
the actual eventuality will be so wildly
out of our expectations right now,
(01:44:11):
because thinking about a massage robot, I
could envision a future where you have
the robot hooked up to your neural link,
and you know exactly which part of your
body needs to be massaged to give you the
right kind of release. So in some form, I
think you could potentially massage
(01:44:32):
yourself with the aid of a robot with
better precision that another
intelligence could do. So I do think
people will still value the human
experience, the human interaction, human
touch, and there'll be a lot of jobs
where that will still continue to exist,
(01:44:55):
regardless of how good the robots can
get. But you know, 10 years out, so it's
hard to imagine what could be possible.
Yeah, we kind of did briefly touch upon
the idea of smaller niche companies,
(01:45:16):
where the YouTube-ification of everything
will allow people to build software,
build tools, physical brick and mortar
businesses,(...) and just literally
anything with such ease that the barrier
to entry to do anything is gonna be
significantly lower.
So I think people will be supercharged
(01:45:38):
with so much power and agency that we're
gonna see a lot more individuals creating
things than we've ever seen before.
Yeah, for sure. And I think the final
prediction that we have is that, or at
least this is my prediction, I don't
think Shoshan completely agrees, but it's
(01:45:58):
my guess that less than 50% of the
population will be working. I think that
we will reach an inflection point where
there will be more people not working
than working. And it's gonna be
different. And this trend of less people
(01:46:19):
working has been going on for a very long
time. And I think by 2035, it's gonna be
something like 51, 49, where 51% of
people are just retired and just not
going to a job every day. So we'll see if
that happens, but I think that there's a
high likelihood that that
will happen in the next 10 years.
(01:46:42):
Yeah, so there you have it, folks. This
is our set of predictions from a little
bit back when chat GPT first came out,
some of the early signs of AGI.
And today where we have some really
useful tools and seeing into the future
where we're gonna have incredible, but
(01:47:04):
seemingly very realistic possibilities
where we have so much more
agency than we did in the past.
Yeah, so let us know what you think. Do
you think our predictions are accurate?
Are they completely lost base? If you
think that we're right, let us know in
the comments. If you think that we're
wrong, let us know in the comments. And
(01:47:26):
if you have any of your own predictions,
let us know. If you're not subscribed, be
sure to subscribe(...) and make sure you
hit like the bell icon too, so that like
you can get all of our updates. We're
posting stuff all the time.(...) And
yeah, we'll catch you in the next one.
today we are gonna do something a little
bit different. So normally we would be
talking about the news and stuff, but
Shashank and I were thinking that we
wanted to make a video where we kind of
brought everybody up to speed on what an
agent is, and also what our predictions
are gonna be for the future. So we aren't
(01:00:20):
gonna talk about any of the news this
week, but we're just going to speculate
on what we think(...) is gonna happen in
the future. So we are gonna predict not
just one year in the future, but we are
gonna predict up to 10 years into the
future. So we're gonna try to keep this
short, shorter than our normal podcasts,
and just try to get right to the facts.
(01:00:41):
So Shashank, you ready?(...) Let's do it.
All right. So we wanted to present what
really is an AI agent, what we have
experienced so far, what we have today,
and where we're headed. So when we talk
about agents in the broad general term,
what is an agent? An agent is someone,
something that can carry out a complex
(01:01:04):
task for you with relatively minimal
supervision. So in the traditional world,
pre-LLMs, you can think of a travel agent
who does a booking for you, scours
through multiple websites, finds the best
price, the best layovers, the minimal
flight time, and then presents a couple
options to you. So a couple examples
(01:01:26):
would be stock predictions,
order taking, trip
booking, travel agents, et cetera.
Yeah, so these agents in the past were
actually pretty good and they showed like
a ton of potential.(...) So some of these
agents that maybe you guys have heard of
are things like baby AGI or auto GPT or
(01:01:48):
GPT engineer. Actually, I created GPT
engineer, so you guys
hopefully probably use that one.
My GPT engineer was a thing that you
could use to do programming.
So you could put it a prompt, say like,
"Give me a Tetris game," and it would be
able to create something like Tetris or
like something simple, right?
(01:02:10):
And this worked okay, right? So a lot of
these agents they had,(...) they were
able to do basic coding and tool calling
abilities, right? You could do basic
stock analysis. Maybe you could do a
little bit of customer service.
You could do RAG, for those who don't
know, RAG is retrieval augmented
(01:02:30):
generation. It's just when you have the
agents or like the AI access some sort of
source documents, but there was a lot of
problems with the agents.
So one of the big problems is a lot of
them had like really tiny context
windows, meaning they
couldn't fit a lot of information
inside of their model, right? So you
(01:02:53):
could probably make a Tetris game, but it
was really challenging to make a really
large software program, right? So, and
even today, right? Like the programs that
we've been able to make have gotten
bigger, but you still can't have all of
Android being created by AI agents yet.
(01:03:15):
So there's a lot of problems that are,
like have been in the past that are
needing to work on being solved.
Which brings us to today. So I don't
know, Shashank, do you want to talk about
like where we are like in the current
state of the world with agents?
Yeah, so things have gotten a little bit
(01:03:35):
better, but not by much. So I think we've
mostly benefited from the models getting
better. There are capabilities just, you
know, exponentially growing month over
month, year over year, the token context
limit has dramatically increased with
Gemini touting a million token context
window. And in my experience, it's
(01:03:56):
actually really good. It's one of the
best ones out there if you're dealing
with large amounts of context. And to
broadly understand why that is important,
it's not just about you pouring the
entire Harry Potter series into the
context window. If you expect these
agents to carry out complex tasks for
(01:04:17):
you, you want to give them tools. You
want to give them the ability to call
functions outside of their neural
network. The ability to use a calculator,
maybe check the weather, hook up into
Yahoo Finance and check the stock prices
or check the options, trading volume, et
cetera. So each one of those tools that
(01:04:39):
you add into a model is stored in its
context window. So we're learning that it
requires a lot of information for an
agent to be like a truly general purpose
agent that can solve anything.(...) So
today we have
specialized agents that somehow,
(01:05:01):
but that somewhat mitigate the challenges
of having a generalist agent.(...) We
have amazing coding agents, Claude Code
just recently came out. Cursor has been
doing a fantastic job after the big
splash that GitHub CoPilot made.
We've had truly independent AI engineers
(01:05:24):
(...) from Devon and OpenAI also
reportedly working on a fully independent
AI coding agent that is going to cost
something close to $20,000 a month, which
is orders of magnitude more than their
$20 a month plus subscription.
(01:05:44):
And we have other kinds of agent in
limited scope like deep research. It does
a bunch of search for you.(...) So deep
research, if you don't know, is a feature
available in most models today actually.
Started from the Gemini models where they
allowed you to give it a complex task and
(01:06:05):
this model wouldn't just spit out an
answer. It would think for a while on the
order of several minutes,
go out there, do a bunch of Google
searches,(...) analyze all of its results
and figure out what you were trying to
ask, break down the reasoning in multiple
steps and then come back several minutes
later after thinking for a lot, after
(01:06:26):
prompting itself over and over again
multiple times to give you
a well thought out answer.
Now moving back to some of the coding
examples, you have, apart from the coding
IDEs, the software tooling for
developers, you have entire solutions for
(01:06:47):
non-engineers to build applications. So
we have really popular solutions like
vzero.dev, which is the parent company of
Next.js, a popular web development
framework. There's a couple other
independent companies like Lovable Dev,
Bolt.new, Replit, which is also a very
popular company that
(01:07:07):
builds tooling for developers.
Another tangent is the reasoning models.
We have amazing capabilities that have
come out from multiple companies that
allow these LLMs to not just give you one
single answer from all the knowledge that
it contains, but actually think about a
problem and give you a solution like
(01:07:29):
counting how many R's
there are in Strawberry.
LLM can't do that because it's only
generating tokens into the future,
whereas a reasoning model can look at the
word Strawberry, break it down into
individual letters, pass it through some
kind of a calculator that tallies up how
(01:07:50):
many letters are there, and
then gives you a final answer.
And moving one step beyond just
intellectual agents,(...) we have agents
that can take action for you in the real
world. One popular example that's been
around for a few years, maybe not in its
full form, is self-driving. So, Tesla has
(01:08:13):
had its self-driving out for a while,
even though it may not be what they have
marketed it as, but it is also an agent.
It takes in information from the real
world and takes action in multiple
different areas. It can accelerate,
decelerate, adjust the steering,
(01:08:35):
and make sure that ideally you don't have
to put your hands on the wheel.
Other companies like Waymo, there was
Cruise for a while, and Nvidia is also
working on a self-driving agent that can
power multiple vehicles. So, we have
intellectual agents, we have physical
agents, we talked about self-driving, and
(01:08:56):
one last thing that multiple cartoons and
sci-fi movies have been promising us for
a long time is robotics. So, we've had
industrial robots for a long time, but
now we're seeing truly generalist robots
that can take action in the real world
from a simple prompt like, hey, make me a
cup of coffee, and the robot goes into a
(01:09:18):
random kitchen and makes a cup of coffee.
We're not there yet, but we're seeing the
seeds of what is possible from companies
like Figure AI, Google Robotics, and even
Nvidia, partnering with companies like
Disney to make little robots that are
reminiscent of the ones from Star Wars
like C3PO. So, we're seeing hints of real
(01:09:42):
robots,(...) and it is the foundation for
a lot of exciting things that we'll see
in the future, which, Mark, do you wanna
show us what we can expect?
Yes, so we don't exactly know what to
expect in the future because if anybody
can confidently tell you what's gonna be
(01:10:03):
in the future, they're probably lying or
wrong, but we're gonna be both lying and
wrong in this presentation because we are
going to tell you exactly what we think,
expect to see in the future.
So, I think there's a few things that I
expect to see, right? So, one is I think
in the very near term future, like 2026,
(01:10:27):
there's gonna be a lot less job openings
for software developers. So, Shashank and
I were kinda debating this, whether
there's gonna be more software jobs or
less software jobs. On the one hand, you
could say, "Oh, there's gonna be more
software jobs "because of Jevons
Paradox," where it's like, if you have
software become a lot more easy to write,
(01:10:48):
then you just need less or like people
might just wanna create more software and
more people are creating software then
you might need more software developers.
But I think that realistically,
these AI models are able to replace the
(01:11:11):
job of like hundreds of software
developers at a time where you used to
have schemes of people who needed to make
software projects. Now you're gonna need
just a handful of senior developers to go
and create the software. And even people
who are junior developers now are able to
(01:11:34):
create more meaningful software in the
future. And even people who don't know
how to create software are gonna be able
to create software. So, maybe like there
will be more people who will be able to
do software development, but I think that
like the job of software developer in the
future, it'll still be there. I just
think that there will be a lot less
(01:11:55):
people(...) who are only purely software
developers going forward. And also,
like realistically,(...) I think that not
just software developers, but almost all
knowledge work will have their hiring
start to slow down. Now, it won't just be
(01:12:17):
because of AI, but I think that a lot of
the hiring will slow down often because
people will be aging, there'll be
attrition,(...) there could be economic
slowdowns, all sorts of things.
And AI will just be like one such factor,
which is gonna make this hiring slow down
(01:12:37):
like quite a bit more.
And then in addition to that, we expect,
so Shashank mentioned about some of the
physical AI agents. We think that
self-driving will be more widely deployed
in more locales. So for example, Waymo is
(01:12:59):
only in I think like four
or five different cities.
Like our prediction for the end of 2026,
now we don't know, but we would suspect
that by the end, like the number of
cities that you see self-driving cars
(...) will probably double if not maybe
more, but that's the prediction like, so
(01:13:20):
there's a high likelihood a city that you
live in, if it doesn't have self-driving
cars now, next year you'll probably see a
lot more self-driving cars. And then one
other bold prediction is we think that AI
agents will start to solve
(01:13:43):
unsolved problems that humans have never
been able to solve. But not in all
domains, but we think specifically,(...)
we may see within the next year, like by
the end of 2026, AI start to solve some
unsolved problems in math and logic. So
my hope is on the Riemann hypothesis. I
(01:14:05):
think that there's a really good chance
that's able to be solved
within a year, we'll see.
And yeah, I think that because of things
like open AI is $20,000 a month, coding
agent that's a shock mentioned,
because of all of these new AI wrapper
companies, a lot of people who were doing
(01:14:25):
knowledge work like
language teaching, coding,
like crunching numbers,
a lot of those jobs, the hiring is gonna
slow down a bit. I think they'll still be
around in 2026, but people try to get
into it, may have a hard, hard time.
And yeah, honestly, really starting
(01:14:48):
today, but for sure 2026, humans, we
don't think will be really special
anymore in terms of intellect. When I say
intellect, I mean, like reading books,
like working with a computer. I think in
2026, humans will probably still be
special in terms of physical ability,
(01:15:10):
right? Like plumbers probably still be
around,(...) but just being like a smart
person who can work in
front of the computer,
that is no longer guaranteed to be like a
job for the next 40 years. Computers will
probably outpace most knowledge work
(01:15:34):
going forward. And maybe the only thing
that will separate humans is our taste,
(...) meaning like, maybe humans will be
better at determining
what other humans like.
So it's like, if you go to a museum and
you see a picture or like a painting on
(01:15:54):
the wall, and you say, I like that
painting. I have good taste. You didn't
necessarily create that painting, but you
like that painting and you bought it for
your home. Like in the future,(...) i.e.
maybe next year, the humans might just be
selecting which AI model gave them the
best output and decide which one they
wanna use.(...) So yeah, that's where we
(01:16:17):
think we're gonna be the next year. We'll
see how it turns out. Shashank, but like,
what do you think about going even
further? Like, should we try
to speculate even farther out?
Yeah, I think the same
trend is gonna continue.
If you wanna bump us to the next slide.
(...) I think the agentic craze is just
(01:16:39):
gonna skyrocket.(...) I think just as
everyone today is using chat GPT or LM,
squad, Gemini, whatever you have. I think
everyone in the next five years is gonna
be using some form of an agent, whether
it's for work, whether
it's for personal tasks,
therapy, something to help you keep track
of your tasks, emails,(...) I think
(01:17:02):
everyone is gonna be so used to handing
off complex tasks to an agent just as
they do with single problems and LLMs
today. Not just that, I do think this
agentic ecosystem is gonna be a lot more
mature than it is today. Because right
now we're still trying to figure out what
(01:17:24):
these LLMs are capable of because they
don't function as deterministic software
that we've been used to for the last 30,
40 years. I think we'll have a better
grasp of what kind of abilities and
emergent abilities these agents will have
(01:17:44):
and be better at using them throughout
society, throughout our economy. So I
think this is something that both Mark
and I agree on. I think every single
person doing intellectual work will be
doing that with agents. They'll be using
agents as junior assistants, as
(01:18:06):
researchers,(...) as junior engineers
even, to do things for them to
supercharge their work.
Yeah, I just wanna cut you off right
there. I think to be incredibly clear on
what Shashanka's saying here, our
prediction is that by 2030,
maybe not 100% of people, but 99
(01:18:28):
virtually 100% of people will be not
doing knowledge work
without the help of a computer.
They will be using agents to help them
manage their work. I think that we think
that the next couple of years, if you're
not using AI, you might be okay. But by
(01:18:50):
2030, if you're not using AI, you will no
longer have a job. Like almost 100% of
people will need to be using an agent or
like working with AI in
order to do their work.
Yeah, and I think this is gonna slowly
creep into the physical world as well.
(...) With the incredible progress that
(01:19:11):
some of these companies are making, we
have seen version one and version two of
figure AI. They are reportedly working on
version three, which their founder has
mentioned in a couple of interviews. And
the amount of progress is just
staggering. The fidelity, the
fine-grained motor controls, the battery
technology,(...) the AI powering the
(01:19:37):
movements of these robotics has gotten
significantly better. And I think this
trend will continue. So we'll see immense
progress in humanoid robots from figure,
from Tesla,(...) from Chinese
manufacturers who are able to bring the
cost down significantly to a point where
I think individuals in maybe developed
(01:19:57):
countries will be able to afford a
humanoid robot to keep in their homes to
assist them with day-to-day tasks. I
think we will start seeing robotics in
industrial environments first because it
is a lot simpler. The needs of industrial
warehouses, for example, are a lot more
straightforward and repeatable compared
(01:20:18):
to an individual's, like a consumer's
home, because the amount of things that
are in a individual's homes can be very
diverse and each person's home is very
diverse.(...) So I think that may take a
little bit longer, but we'll start seeing
humanoid robots enter the scene a lot
(01:20:39):
more than we do today. I think there'll
be significant progress in self-driving.
We'll have self-driving in most major
cities because we have full self-driving
from Tesla available right now, although
it doesn't work everywhere all the time,
but it is getting better. They're on
(01:20:59):
version four of their hardware.(...) And
we did just mention how Mark Rober, a
popular YouTuber, showed that the Tesla
cannot detect like a Looney Tunes cartoon
wall that looks like a real road if you
prop it up with some cardboard and
printed paper. But the latest version,
(01:21:21):
which another YouTuber just put out a
couple of days ago, has shown that the
Tesla can detect a fake wall and doesn't
just go straight through. So their
capability is also improving.
We think that they'll, apart from
robotics, we do think there'll be a
(01:21:41):
significant breakthrough in fundamental
research in basic sciences. I think
people will start using AI agents to
assist them in physics, chemistry,
biology, to conduct experiments, to run
simulations, to help them supercharge
their work and speed up the amount of
time for scientific
breakthroughs by a significant amount.
(01:22:04):
Yeah, and I think there's a high
likelihood, we didn't put this in the
slide, but I think that with a lot of
that AI research, we may see by 2030 that
we've finally reached longevity escape
velocity. Meaning if
you're alive in 2030,
there's a good chance that you could just
(01:22:26):
continue to live indefinitely because of
the rapidly increasing
healthcare ability.
Essentially, if you live till 2030,
there's a good chance you might be able
to live long enough to live forever.
That's a tricky one. I do think longevity
(01:22:46):
and developments in longevity will be
significantly better than they are now. I
think we'll understand the human body a
lot better. I think we'll understand
causes of aging, disease, and how to
optimize our nutrition, our gut
microbiome, health, exercise, et cetera.
But we did look at the top 100 oldest
(01:23:09):
people alive, and there does seem to be
some kind of a biological threshold
around 114 to 117 years where the top 195
of the top 100 people
max out at 114 to 117. So, at the very
least, I think that more people will
(01:23:29):
start reaching that threshold as opposed
to what it is today,
which is about high 80s.
Yeah, I agree. I don't think that we're
necessarily gonna see people live to 125
or something. Because I think the oldest
person now is 114, 115. So, I think it's
(01:23:50):
unlikely that they're necessarily gonna
make it to 120. Because I think that if
they are,(...) even if we have advancing
medical technology, those people may not
want to try some sort of fancy new AI
thing, right? But I think what I'm trying
to say is that the people who are alive
(01:24:13):
in five years, maybe in their 40s or
maybe in their 50s, who are maybe more
open to trying some of
these advanced therapies,
they might be able to live for another
100 years because they're actually trying
to use these things. I think that it's
not that everybody is gonna live like a
(01:24:33):
really long time.(...) Like a lot of
people will just eat Big Macs all day and
not get enough sleep. But I think for the
people who are actually really paying
attention to their health and are
following the latest therapies and are
willing to try,(...) those are the people
who may be able to reach longevity escape
(01:24:55):
velocity and actually see some of the
pace of aging reach close to zero. So,
like actually like a very specific
prediction is that right now there is the
longevity leaderboard. I think like Brian
Johnson is running like one of these
(01:25:17):
longevity leaderboards where they have
these different pace of aging tests. And
I think that right now, according to the,
some of these aging pace tests, you are
around, like I think for every 12 months
that have passed, you according to these
aging tests have only aged like six
(01:25:38):
months. I think that there's a high
likelihood by 2030 that your pace of
aging will reach zero. For every year
that has passed, you will not have aged
at all. And that's my prediction. So
we'll see, we'll see. I could be wrong,
but that's my prediction
that we'll see that by 2030.
(01:26:00):
All right, so now we're getting into the
realm of science fiction, which is crazy
given that we're only five years out, but
let's look at the next 10 years, at least
our predictions for the next 10 years.
What do you think we
have coming up, Mark?
I think there's a lot. And I think before
(01:26:20):
we get into that, I think that we just
need to, I want to tell a little bit of a
story. So my mom, when I was a kid, she
wanted to teach me about exponential
growth. And if you guys don't know about
the beauties of compound interest, it's
really cool, right? So it's like,
(01:26:42):
if you save a little bit of money,
it compounds that money, and then you can
make a lot more money over time, right?
So it's like, if you start with $100 and
you make a 10% rate of return, after year
one, you have $110, not crazy, right? But
then after year two, you're gonna make
10% on your 110, right? So instead of
(01:27:03):
making $10, you'd actually make $11. And
that actually continues. And it actually
only takes, I think, about seven years to
double, right? So after seven years,
you'd have $200, assuming you just grew
that $100 every time. And then after 14
years, you'd have $400, et cetera, right?
(01:27:24):
So imagine you had a lake, right?
And on day 30, that lake was filled with
water and it was filled with algae.
On what day, Shashank, I mean, this is
probably an easy question for you, but on
what day is the lake
(01:27:45):
half full with algae?
What is the rate of algae, blue?
Let's say it doubles every day, right? So
let's say the algae doubles every single
day. And on day 30,
it's 100% full. And with,
on day one, you just start with a 100%
(01:28:06):
full. One little cell of algae.(...) On
what day is it 50% full?
I guess day 29, one
day before the end, yeah.
Yeah, day 29. So I mean,
I don't know. I think that the point I
have here is that a lot of these things,
(01:28:29):
they feel like they're happening slowly
when you are there, right? So it's like,
if you just look at the algae and it's
just like, oh, it's doubling, but it's
just getting doubling a little bit
faster, right? But, and then it's like,
oh, day 29, there's still like half the
lake that needs to be filled. But then it
just takes a year for the entire lake to
(01:28:49):
be like overrun with algae.
And that can happen with algae, that
could happen with COVID,(...) and that
can also happen with your bank accounts,
but that also can happen with AI in the
future. So I think that remember this
when we're predicting the future,
(01:29:09):
some of these predictions may seem crazy,
but like a lot of this exponential growth
happens way faster than you think.
So anyways, that's a little prelude
for why we're making
some of these predictions.
And just to add a little bit, you talked
about compound interest and how every
(01:29:31):
financial guru touts the benefits of
compound interest.(...) Compound interest
in the financial market would be like,
what is it? Like seven to 10% if you put
it in safe or like even
high growth index funds.
Moore's law and exponential growth(...)
is like 50% every year. It is like
(01:29:53):
significantly faster than what you would
expect with like compound interest with
your retirement funds at the end of a
lifetime of investing and working.
Exponential growth with Moore's law and
some of the other laws or heuristics that
we have, which is just rapidly
(01:30:13):
progressing the technologies, the
underlying chips, the algorithms, the
models, the tooling around it, the
infrastructure, everything, is just
accelerating at a pace that we've never
seen before. And they are self-improving
so that the pace itself is increasing
(01:30:34):
every year. So it's
just mind boggling numbers.
Yes.
So, all right, let's get into some of the
strategic predictions. So the first one,
which we kind of mentioned at five years
out, we think it's still gonna be true.
100% of knowledge work,
(01:30:56):
if there even is jobs,
we think that there'll be less jobs doing
knowledge work. And the jobs that
knowledge work has will be just simply
managing AI agents or teams of agents.
And a lot of that work will probably even
start to get phased out as the agents
just get really good
(01:31:16):
at managing themselves.
So we think there'll be a lot less people
doing knowledge work and the people that
are doing knowledge work are just gonna
be running companies of
tons and tons of agents.
So I did have an alternative theory for
this. One consequence of the Jevons
paradox, which was it
(01:31:38):
related to electricity?
Regardless, when some fundamental
resource becomes significantly cheaper,
the usage shoots up dramatically. So I
think we might see a
future where there are,
every single person will be an
(01:32:00):
entrepreneur. Every single person will be
able to use AI to build companies,
products, features for a very, very
specific niche use case. That way we will
see an explosion of software that is
customized and tailored to every single
thing that you can think of. And people
(01:32:22):
will be building software on the fly to
solve specific things that they want. So
either everyone will be phased out of a
job or everyone will be entrepreneur
where they're able to use these tools to
do things that they need in such minute
detail that we've
never experienced before.
(01:32:43):
Yeah, I think that's true. Like actually
Shashank and I were
talking about this before.
I think that that will happen.(...) I
don't know if everybody will be an
entrepreneur, but probably a lot of them.
And I wanna be specific when I talk about
knowledge work. I think that when I say
knowledge work, I don't mean companies,
(01:33:05):
but I mean like as an employee.(...) So I
think that there will be very, very few
companies that are hiring employees to do
knowledge work. If
you're the owner of a company,
that's a little bit different I think.
But like if you're the owner,
your first instinct is not gonna be to
(01:33:26):
hire somebody. It's gonna be to see if
you can agent to do something for you and
if you wanna delegate, you're just gonna
somebody like run a team of agents.
But yeah, I think to kind of add on to
Shashank saying is I think that we're
gonna see something like YouTube. So if
you think of YouTube,
it is how a lot of people, actually I
(01:33:48):
think it's the most watched video
streaming site in the world. Like more
people watch YouTube than they watch
broadcast TV. More people watch YouTube
than they watch Netflix. More people
watch YouTube than they watch ABC, NBC,
all of the media. But what's different
about YouTube is it's not a big company
(01:34:09):
that's making it. I mean, sure like
Google manages and maintains YouTube. But
Google isn't posting a bunch of content
on YouTube. They aren't controlling what
you see. I mean, they kind of are with
the algorithm. But at the end of the day,
like what you're seeing is you're seeing
videos made by individual people, like
(01:34:29):
oftentimes just with their cell phone. So
like if you have a cell phone and an
internet connection or a laptop and
internet connection, you can make a
YouTube channel and spread your message
to the entire world. Whereas in the past,
there used to be only a few major
broadcasting. It was like, in the past,
it was like three channels that you could
(01:34:49):
watch and everybody
saw the exact same thing.
And we're able to see like such
personalization with YouTube that you can
find whatever niche that you're
interested in and just watch that. So for
me, like I really like
watching content about AI
or like I like content about like watches
(01:35:11):
and stuff or Japanese learning. Like
these are some of the things that I'm
interested in. And if I'm watching
regular TV, I can't find that, right? But
on YouTube, there is a bunch of content
creators making content for exactly what
I want. And I think to add on to what
Shashank's saying is we're gonna see this
trend continue where it's not just to be
videos, but it's gonna be for every type
(01:35:34):
of software,(...) games, and we're even
gonna probably start seeing it for like
physical things, right? We will have 3D
printers that are gonna be able to make
like really, really specialized doodads
that you can use that it's like, you say,
(01:35:54):
oh man, I really wish I had a thing that
could fit this just ever so perfectly.
And there's gonna be some guy who's gonna
make a company selling just that like one
little widget that you need for your side
project. And if you can't find it, you'll
be able to easily make it and maybe even
have AI created for you. So that's
(01:36:15):
happening and that's gonna happen soon.
So there's probably gonna be a lot more
companies that are just a lot smaller.
Yeah, and to kind of summarize what Mark
was trying to say, I think the
YouTubeification is basically
democratizing access to all of these
esoteric technologies(...) because we can
(01:36:37):
do some of these things today. It just
takes a lot more effort. It might take
more effort than an individual(...) can
realistically put within their lifetime.
But within the next 10 years, I think
natural language will become the mode of
work. We'll be able to talk to machines,
(01:36:57):
to embodied physical robots, or even
virtual agents to do tasks, knowledge
work, physical work for us and get things
done like immediately, which would today
maybe take an entire company to do.(...)
So moving on to some concrete examples,
(01:37:18):
I think a lot of scientific research will
have AI agents assisting them. And we
will see maybe significant scientific
research be done by AI agents and maybe
see the first Nobel Prize being won by an
AI as opposed to a physical person.
(01:37:40):
We talked a lot about self-driving.(...)
We think this will become the norm.
Elon has been really bullish on his
Tesla, what is it, RoboTaxi.
I think that is a significant
opportunity. It is a potential massive
market that could be created because
(01:38:02):
right now, if you think about the
utilization of a car, it
is maybe 10% of the day,
whereas the other 90%, it's sitting
parked in a parking lot in a garage.
Definitely no one is using at night when
you're asleep. But even during the day,
when you're at work, when you're at home,
(01:38:22):
it's just sitting there idling, waiting
for somebody to use it, as opposed to a
self-driving fully autonomous vehicle, it
could be out on the roads a majority of
the time, maybe even having a battery
swap that just allows it to hop in and
out of some kind of a station and keep
going. Or we've seen how some of these
(01:38:42):
Chinese EV manufacturers able to have
insane amounts of charging where they're
able to charge a car(...) in the same
amount of time that it would take you to
fill a tank of gas. I do think that the
self-driving industry will mature to a
point where cars will become a service as
(01:39:04):
opposed to something that you own.(...)
Similar to, I guess, the uber-fication of
things where people can just door-dash
anything that they want. You can go live
in amazing vacation rentals that you
don't own where you're sharing someone's
home. And similarly,
share access to a vehicle.
(01:39:28):
Do you wanna take us through some of the
advancements in robotics, Mark?
Yeah, for sure. So,
and just to add on to what Shashank was
saying, in terms of the cars will become
a service,(...) we think that every
single new car will be self-driving.
Except cars that are just purely for
(01:39:55):
hobbyists. If you're doing a, I don't
know, a car that's gonna race around the
track, or you're buying a Ferrari or
something, and you really
just wanna drive that thing,
most cars that aren't
those, like your Toyotas,
(01:40:15):
your Honda Pilots of the world, those will become a service.
Those will be all self-driving.
So, it's gonna be weird to buy a car and
then drive it yourself.
There probably even, maybe even laws that
prevents people from driving cars that
aren't self-driving in the future. Now,
(01:40:36):
we're not saying all cars will be
self-driving because the cars today will
still be on the road in the future, but
all the new ones, those will be
self-driving. And also, we think that
humanoid robots are gonna be
super common around the home.
You are going to take the money that you
saved(...) from your car and your
(01:40:59):
robotaxi, and you will probably use that
to buy your humanoid robot, butler,
gardener, cleaner thing.
It will,
like, my prediction is it will probably
be, treated something like a car where
it'll be kind of expensive. It might be,
(01:41:20):
I don't know, like $100,000 for one of
these things, you know, with inflation.
Maybe like 20 to $50,000 in like today's
money,(...) similar to that of a car. And
you'll probably pay for it over time, but
it will save you so much time
that you will not have imagined life
(01:41:42):
about it, right? It'll cook for you,
it'll clean for you, it'll
make your life a whole lot easier.
And these things are gonna be super
common. Not everybody's gonna have them.
They will be expensive. And there'll
probably be some sort of rental options
to have it, but they'll
probably be everywhere.
And also, we think that you would
(01:42:05):
probably interact with more of these
robots than you will with, like,
actual human beings,
like, on average, or maybe like a similar
amount. I mean, if you're a really,
really social person, maybe you'll
interact more with, like, human beings,
but it'll probably be that you're
interacting with these robots all the
(01:42:26):
time. You're gonna go to a restaurant,
your food is gonna be delivered by robot.
You're gonna go and go to the gas
station. Like, you may not pump the gas
anymore. I mean, like, it might just be
the robot arm that comes and puts the gas
or just, like, plugs your car in, because
your new car may be electric, right?(...)
(01:42:46):
It's gonna be, you go to a restaurant,
the entire staff in the back may not be
actually, like, humans cooking anymore,
but it'll be the robots. Honestly,
probably the only places that you'll see
humans is where people prefer to interact
with humans over robots. So, for example,
if you're at a restaurant, you don't
(01:43:07):
necessarily care about how your food is
made. Like, it doesn't really matter if
it's, like, a human chef versus a robot
chef.(...) Like, you don't actually see
who's cooking your food anyways, right?
But, like, if you're at a bar, a lot of
people prefer a human bartender just
because they like talking to a person
who's there, right? Like, you're probably
not gonna see too many robot massage
therapists just because people like to
(01:43:29):
get a massage from a human, right? Like,
even though we have massage chairs,
people still go and get a massage from a
human just because they
prefer to have humans, right?
And, like, so some of that stuff will
still be there, right? Like, I think
human art will still be around, but these
humanoid and non-humanoid robots are
(01:43:50):
gonna be everywhere in the future, and
you're gonna interact with
them a lot. Definitely a lot.
I mostly agree with you, although since
we're talking 10 years out, I just think
the actual eventuality will be so wildly
out of our expectations right now,
(01:44:11):
because thinking about a massage robot, I
could envision a future where you have
the robot hooked up to your neural link,
and you know exactly which part of your
body needs to be massaged to give you the
right kind of release. So in some form, I
think you could potentially massage
(01:44:32):
yourself with the aid of a robot with
better precision that another
intelligence could do. So I do think
people will still value the human
experience, the human interaction, human
touch, and there'll be a lot of jobs
where that will still continue to exist,
(01:44:55):
regardless of how good the robots can
get. But you know, 10 years out, so it's
hard to imagine what could be possible.
Yeah, we kind of did briefly touch upon
the idea of smaller niche companies,
(01:45:16):
where the YouTube-ification of everything
will allow people to build software,
build tools, physical brick and mortar
businesses,(...) and just literally
anything with such ease that the barrier
to entry to do anything is gonna be
significantly lower.
So I think people will be supercharged
(01:45:38):
with so much power and agency that we're
gonna see a lot more individuals creating
things than we've ever seen before.
Yeah, for sure. And I think the final
prediction that we have is that, or at
least this is my prediction, I don't
think Shoshan completely agrees, but it's
(01:45:58):
my guess that less than 50% of the
population will be working. I think that
we will reach an inflection point where
there will be more people not working
than working. And it's gonna be
different. And this trend of less people
(01:46:19):
working has been going on for a very long
time. And I think by 2035, it's gonna be
something like 51, 49, where 51% of
people are just retired and just not
going to a job every day. So we'll see if
that happens, but I think that there's a
high likelihood that that
will happen in the next 10 years.
(01:46:42):
Yeah, so there you have it, folks. This
is our set of predictions from a little
bit back when chat GPT first came out,
some of the early signs of AGI.
And today where we have some really
useful tools and seeing into the future
where we're gonna have incredible, but
(01:47:04):
seemingly very realistic possibilities
where we have so much more
agency than we did in the past.
Yeah, so let us know what you think. Do
you think our predictions are accurate?
Are they completely lost base? If you
think that we're right, let us know in
the comments. If you think that we're
wrong, let us know in the comments. And
(01:47:26):
if you have any of your own predictions,
let us know. If you're not subscribed, be
sure to subscribe(...) and make sure you
hit like the bell icon too, so that like
you can get all of our updates. We're
posting stuff all the time.(...) And
yeah, we'll catch you in the next one.
Popular Podcasts
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.
Therapy Gecko
An unlicensed lizard psychologist travels the universe talking to strangers about absolutely nothing. TO CALL THE GECKO: follow me on https://www.twitch.tv/lyleforever to get a notification for when I am taking calls. I am usually live Mondays, Wednesdays, and Fridays but lately a lot of other times too. I am a gecko.