June 3, 2026 • 34 mins
What if the fastest path to superintelligence is AI that builds itself? That's the bet Richard Socher is making — and he has the track record to back it up. A double unicorn founder and early investor in eight unicorn companies (including Perplexity and Hugging Face), Richard has spent 15 years building the foundational research that powers modern AI. Now he’s co-founded Recursive with an elite team from Google DeepMind, OpenAI, and Meta to pursue something more ambitious: a self-improving AI that generates its own scientific breakthroughs — what he calls a "eureka machine."
Richard joins Oz to unpack how recursive superintelligence actually works and why open-ended AI systems could outpace today's giants.
EXCLUSIVE NordVPN Deal ➼ https://nordvpn.com/techstuff Try it risk-free now with a 30-day money-back guarantee
See omnystudio.com/listener for privacy information.
Listen
Watch
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:15):
Welcome to tech stuff. I'm os Voloscan. For those outside
of the industry, it can feel like AI came out
of nowhere and is now taking us all who knows where.
That's why I wanted to talk to Richard Socher. He
spent the last fifteen years developing the foundational research and
building the infrastructure that powers AI. He's also, as of
recently a double unicorn, founder and CEO of two companies,
(00:37):
U dot Com and Recursive Superintelligence. Recursive launched officially in
mid May with the six hundred and fifteen million dollar
fundraise at evaluation north of four billion dollars, and Richard
is swinging for the holy grail of technology AI that
improves itself what he has called a Eureka machine. Richard,
(00:57):
Welcome to tech stuff. Thanks for having me. Eureka machine
sounds even better than a thinking.
Speaker 2 (01:02):
Machine, doesn't it.
Speaker 1 (01:03):
Yeah, what is it?
Speaker 2 (01:05):
Eureka machine? You know, it's not the official term we
definitely we'll use moving forward, but I do think it's
a great analogy. It's essentially an AI that got very
good at inventing scientific advances, and we'll start with scientific
advances on the science of intelligence and artificial intelligence. Itself,
(01:29):
but eventually believe that that will lead to a machine
that will then get even better at doing all kinds
of and making all other kinds of inventions because it
got very good at ideating, implementing and validating ideas.
Speaker 1 (01:44):
So what is recursive superintelligence? First with a small R
and a small S, and then with a big R
and as a big S in terms of your company.
Speaker 2 (01:51):
Yeah, so recursive superintelligence is essentially an AI that improves itself.
That means over time it also gets better proving itself.
So it's a self reinforcing loop that compounds with essentially
no ceiling. Like most of AI, we have like one
specific objective function and you get these very sort of
(02:12):
spiky intelligence capabilities. Here. We believe this could get much
much smoother because what we're asking the I to do
is to automate knowledge work itself, so automate the discovery
of different kinds of knowledge. And so that's why we
think it will actually lead to this discovery engine, the
(02:33):
Sercom machine. And why is this possible now? It's mostly
because AI is code. An AI can code, and so
you can actually loop that capability onto itself if you
figure out a lot of other triggy things.
Speaker 1 (02:46):
Now, I want to come back to the idea of
AI being spiky, because when you first started working on AI,
there was like ImageNet which you worked on, which recognized
things in images, and then there were language models and
basically every different human sense modality like site, you know, words, visual,
et cetera, were captured in different kind of AI approaches
(03:08):
and products. Right, and then recently in the last couple
of years, it feels like AI has already become multimodal. Right.
You know, you can generate video, you can do math,
you could do coding, like what is the thing which
is not happening now which can happen next?
Speaker 2 (03:25):
You know, people often say ideas are cheap, and it's
sort of true, right, A good idea without good execution
is not very valuable. But also like really good ideas
are hard to come by, and so what we currently
have is really really smart researchers who worked for many
years in AI research developing very strong intuitions on what
(03:48):
is a good idea, what might result in an improvement? Right,
and then they define sort of manually, often data sets
and benchmarks, and then the eye will charge up after
those benchmarks and try to get really really good at
solving those. In contrast to that, our approach to recursive
(04:09):
self improvement is much more aligned with open endedness, which
is often an evolutionary process that essentially has a whole
set of interestingly different ideas that are being explored, often
in peril. And like evolution doesn't really have a ceiling.
(04:32):
You don't just get one hundred out of one hundred
percents correct on one particular benchmark and then you're done.
You can keep iterating in this open ended fashion. And
so often in open endedness, similar to biological and cultural evolution,
entities interact with other entities inside an environment and those
can co evolve, you know, like you have different leafs
(04:55):
like trees, and then you get you know, they get
higher and higher so that the animals can eat the leaves.
But then giraffes, you know, get higher and higher next
so they can eat the leaves again. And so these like,
depending on the environment, different ais can evolve.
Speaker 1 (05:08):
And so what is the role of recursive in all
of this? And what's the first time that you think
somebody outside of the technology industry will experience something that
you've built.
Speaker 2 (05:19):
So the idea about recursive is that here the AI
will output another AI that then becomes the input to
the next iteration of that AI, and at each step
that function will actually change itself. So that's sort of
the technical definition of recursive and the idea here is
(05:39):
that we take the manual steps of ideation, implementation, validation
out of that loop and allow AI to come up
its own ideas, implement them, and then validate them. And
the first time people will observe this is likely in
the AI research world itself. You know, there are a
lot of very very extensive people. They have amazing compensation
(06:03):
and they're doing that work manually, so that will be
the first impact. The second impact will then be in
software engineering, where any company that employees software engineers is
going to want to give them the superpower to essentially
manage a team of thousands of people, all of whom
are just constantly thinking about how to be better at
a task and then be better at that task and
(06:25):
improving themselves also. That will then lead to much more
productivity That is most immediately useful for organizations that have
very clear rewards and objective functions like make this battery better,
like cure this particular disease. And I have like a
loop or ideally even a simulation or some kind of
verification mechanism to really know that I am making progress
(06:48):
towards that specific goal. And then yeah, eventually, I think
any I will likely be built by this, and so
that anyone who uses chatbots and so on in the
future will likely benefit from this technology as well.
Speaker 1 (07:00):
And how did you I mean, You've got a group
of people from Google opener I met as some of
the most distinguished researchers in the whole industry, including yourself, Like,
how did you How did you attract this killer team
at a moment where that particular type of talent is
the most highly valued talent in the world.
Speaker 2 (07:17):
Yeah, So we have eight co founders and indeed some
of the most impactful people from Google deep Mind, from
open Aie, from Meta, from Salesforce Research and other places.
And I think we all came together because of the
shared vision that we believe that recursal self improvement is
(07:37):
the fast path towards superintelligence, and we believe that that
can ultimately lead to much more human flourishing as we
cure certain diseases, as we solve more and more hard
problems that have been very hard for human scientists to solve.
So it's about the shared vision and you know sharing
(07:58):
also the ownership that we own this all together.
Speaker 1 (08:03):
You sent a PhD thesis in twenty fourteen had the
title Recursive deep Learning for Natural Language Processing and Computer Vision.
You're nodding because you know the title of your own
PhD thesis. But I'm curious if you know, just over
a decade ago you could possibly predicted that you'll be
founding a unicorn company going after this problem that was
an academic problem in twenty fourteen.
Speaker 2 (08:25):
Yeah, so it's a different type of recursion there than
my thesis. A lot of us came to the same conclusion,
but from different directions. And yeah, it's to be honest,
a lot of people in the I community thought that
when deep Mind was acquired for you know, it's like
four hundred million back in the day, there was like
PKI hype, right, and it will just not go there.
(08:49):
And even I, like coming from Germany, German mindset is
very often sort of critical and skeptical and like what
could go wrong kind of is where you first go
when something new happens, and when sort of there's a
lot of acceleration and growth, and so I think towards
the end of my PhD, I was becoming more and
more constructively optimistic to but yeah, you don't. You never
(09:13):
know if these things are possible that early. But you
can dream, and if you keep working super hard, eventually
you can make some of these dreams come true. But
of course it's also just like a first milestone. Right
now is the time for us to really show that
we can do.
Speaker 1 (09:26):
This unless they're made redundant by technology.
Speaker 2 (09:29):
I hear that a lot of times out of Germany
and Europe, but you can see this in many places,
like when technology really scales, so you actually end up
in a place where people will have more equitable access
to goods and services. So let me give for example,
the like an iPhone. The iPhone no matter if you're
(09:52):
just a middle class teenager or you're a multi billionaire,
you use the same iPhone right, So even if you're
on like the further is left spectrum or the first right,
it doesn't really matter. You can love technology and its scale.
What other goods and services are there that are bottlenecked
on intelligence that humans don't have currently access to a
(10:13):
personal tutor for your kids. Most people can't afford that
personal assistant to just schedule all the little random things
and taxes and whatnot. Right, most people have to waste
so much of their productive time on these minute, boring tasks.
And I really do think like we're seeing this a
lot in different industries and maybe just my current hypothesis
(10:35):
is that depending on the elasticity of the demand for
a certain good or service, that will tell you whether
I will actually create more jobs when it makes that
more efficient and cheaper, or destroy jobs. And coding, for instance,
what we're seeing is coding is becoming tenex more efficient.
Anyone can now code with English, and somehow we want
(10:59):
more coders. A number of open jaw positions has actually
increased by allowing people to be that much more productive
because ultimately anyone could have software. Everyone can have a
personal suite of software tools, so you could have billions
of different software products that are customed for you. And
so the ealyticity of demand is very high as that
(11:20):
price goes down, and people did not make themselves obsolete
at all, and so yeah, like some of these industries
will shift as that the prices go down. But we've
seen that in history many times, and it's always been
for the better. Like only five percent now work in agriculture,
yet like not ninety five percent of people are unemployed,
even though it used to be over ninety percent of
people work in agriculture, And like, somehow we all found
(11:43):
new things to do instead of standing there in a
field with our hands.
Speaker 1 (11:46):
But just a pushback, I mean, it's interesting, like, do
you think the demand for new software applications comes from
real people or is that a demand imposed by capital?
As so much money flowing into the space that by
definition they are more open software engineering jobs because like
wealth and other major institutional investors have decided this is
the highest potential return sector in the world, and therefore,
(12:07):
like the demand creation is artificial.
Speaker 2 (12:08):
I mean, investor, Like that's that's sort of one of
the beauties of capitalism. Like people don't want to just
put money into something that people don't want and that
doesn't make any other money because no one's using it, right,
So it's like that that problem will solve itself quite
easily and sure, like just like the price of energy
can fluctuate a little bit, the price of intelligence may
fluctuate a little bit, but ultimately, like there are even
(12:31):
more use cases for intelligence than there are for for energy,
even though they're obviously highly correlated and one is required
by the other and so on. But I like, I
don't think there's like just like oh, sovereigns, there's abstract entity.
They're they all want to like invest.
Speaker 1 (12:45):
They want to return on the capital. But then you
get the magest But I'm not I'm not just in
terms of the elytcity of demand for software like I
think it's it's I think the proof will be in
the pudding as to whether that's like true demand or
like demand, like demand imposed by capital based on like
projective outcomes. I think that that remains you see.
Speaker 2 (13:01):
That's that's of course true for every investment, right do
like especially if you do early investments into categories that
don't yet like exist, then that's even more true. But
software does exist.
Speaker 1 (13:12):
And by the way, to be fair to you or
it should, you've backed seven companies as an investor to
become unicorns. So I would, I would anyone listening it
just became it rich Richard maybe a more shrewder investor
than I am. That's kind of extraordinary perplexity. Hugging face
weights and biases somehow sohold names like So, let's take
a couple of steps back. Recursive super intelligence is the
(13:35):
story of the moment. You're also the founder and CEO
of you dot com. You dot com was doing basically
AI search before the chat gipt moment, but it's largely
for enterprise, and you invested in all these other AI unicorns.
So what is it? What is it you see about
the world earlier than most people.
Speaker 2 (13:52):
I spent a lot of time thinking about the future
and how I can have positive impact on it. And
for many years, like a decade off my life that
was actually making the technology work and doing research in
it because it just wasn't working. And once it was working,
I'm like, well, maybe just making it slightly incrementally better
for a while wasn't necessary. Just had to scale the
(14:12):
existing ideas and modify them around the edges and improve
them here and there. And then I just love all
these use cases, and you know, especially some deep industries
like healthcare, it was clear that I needed to support
and help from people who deeply know the space, know
how farmer companies actually develop and bring new medications and
cures into the market, and drugs. But you know, every
(14:36):
single space has has this opportunity right now in AI,
like a I will change every industry. That's why I
sort of AI of X, you know, like and then
oftentimes you can think from first principles sort of what
would people really want like in that industry and can
you make that now possible with AI. So we wanted
to connect like some of the world's leading experts like
(14:58):
Chris Manning and the gold Bloom with really strong industry
experts like Nick Krant's and our team or Chrish and
and actually like merge them. And that's that's what allowed
us to invest in these really exciting companies. I'll just
give you one example. One is called Parallel Bio.
Speaker 1 (15:15):
Uh.
Speaker 2 (15:16):
They build organoids and they with AI, they are able
to track the organoids and see how they are influenced
by different.
Speaker 1 (15:24):
Collections of kidney cells exactly.
Speaker 2 (15:27):
In their case they will they build lymph nodes, lymphoe organoids,
and so this one company actually got FDA approval to
SKIP for toxicity testing of immune systems animal trials by
using these organoids in the petridish being evaluated and created
partially by AI and robotic processes. And so it's an
incredibly exciting company that single handedly could at scale save
(15:50):
millions of animal lives, right, And so when I hear
all these people saying AI, it's gonna so terrible, And like,
I see all these like use cases where people get
access to goods and services that they couldn't afford before,
or we save millions of animal lives and so on.
That's what I see when I think about AI all
the time, and that's what inspires me for AIX And
(16:11):
then yeah, a lot of times it's just like working
in research and AI. You can see the applications of
it a little bit earlier to an laws.
Speaker 1 (16:20):
After the break, Richard predicts where technology will go in
the next few years. Stay with us. We talk a
lot on this show about protecting your data, especially in
the age of AI, and how scary it can be
when it's breached. And I want to tell you today
about NordVPN, which really covers all the bases when it
(16:42):
comes to privacy. I travel a lot, and I use
Wi Fi when I'm flying all the time, and NordVPN
makes me confident in no matter where I am in
the world or the sky for that matter, my private
details like bank information, passwords and online identity is safe,
and it's also possible to switch on virtual location, which
allows you to save money by buying flights and hotels,
(17:03):
or subscriptions or even streaming soccer or football as I
like to call it from other countries at a cheaper price,
and NordVPN doesn't slow you down. It has super fast
Internet speed, no buffering or lagging while streaming. It is
premium cybersecurity for the price of one cup of coffee
per month. To get the best discount of your NordVPN plan,
(17:25):
go to NordVPN dot com slash tech Stuff. Our link
will also give you four extra months on the two
year plan and there's no risk with Nord's thirty day
money back guarantee. The link is in the podcast episode
description box Welcome back to tech Stuff. So Richard, I
want to talk to you about you dot com. I mean,
(17:46):
you founded it before ELM's were popular, and am I
right in characterizing it basically as an AI powered search
engine but for businesses and enterprises.
Speaker 2 (17:57):
That's exactly right. Yeah. We now provide web search APIs
for enterprises so they can make their lms their chat wants,
their agents be up to date, accurate and have citations.
Speaker 1 (18:07):
But when you saw, for example, after you've founded you
dot com, the chat GPT moment happened, did you have
any like Pomo or wish that this had been like
your product that had become the consumer runaway Darling.
Speaker 2 (18:20):
Of course, yeah, we were initially a consumer company and
we had the most accurate answers. When Google was still
telling people like to eat rocks or put blue on pizzas, like,
we had way fewer hallucinations, like almost none before anyone else.
But it just didn't quite click with consumers, like the branding,
the marketing wasn't quite there. And so eventually we found
that who really cares about the technology. It's developers who
(18:44):
implement the technology, and they just want to get the
best version, the best API for their lms. And that's
how we really started taking off and made a lot
more revenue and are growing really healthy now.
Speaker 1 (18:56):
And with Recursive, I mean was the ambition was ambition
for this company to play in the world playing culture.
Speaker 2 (19:03):
I think we have a shot here at building sort
of in the Davit Deutsche sense of the beginning of infinity,
like an ultimate Eureka machine that has this open ended
way of exploring the frontier of knowledge and really help
become this invention generating machine. And I think that will
(19:23):
be exciting first for a I in the field of AI,
but eventually for energy and better fusion systems, better battery materials,
for in chemistry than in especially pre clinical biology, like
developing new kinds of drugs and molecules. Like, there's so
many exciting applications once you have this machine. You know,
(19:46):
it's not just that the level of the bachelor's but
sort of five thousand PhD degrees that all went and
are now researching a particular problem for you. So I
think that that is the ambition.
Speaker 1 (19:58):
I mean, you said that Open AI, Google, Anthropic Rule
very different companies in different spaces, but also acknowledge that
they're all pursuing superintelligence. Your twenty five person lab going
up against these companies in some sense with thousands of
employees practically unlimited compute, particularly Google's case. I suppose what
gives you an edge? How do you win?
Speaker 2 (20:17):
We have a singular focus, right, there's a lot of
labs that do a lot of different things. For recursive
that is, like recursal superintelligence in the name, that is
what we're doing. The second one is infrastructure. We're like
custom building everything from the ground up with superintelligence in mind.
That means that we designed sort of all the teams,
(20:39):
the resources, the infrastructure, the architecture, and so on from
the ground up for autonomous scientific discovery. So that's number
two's infrastrucure. Number three is that we're really inspired by
so called open ended systems. It's a fundamentally different approach
that releaves us of some of these constraints that others face.
Speaker 1 (20:58):
And why does it ALREADI view constraints. So in many ways,
like what a lot.
Speaker 2 (21:02):
Of AIS and the E labs are doing is they
kind of define these spiky intelligences there, like this is
a particular data set, benchmark or shown that we want
the eye to be really good at, and then the
eye gets really good along that one spike. But then
there are all these sort of valleys between the spikes,
and so we believe that instead of optimizing one fixed
(21:22):
target that will ultimately plateau, we believe in open endedness,
you have interestingly different sets of ideas and they can
compete with one another and they keep being keep improving
without this sort of obvious seiling of like oh you
got a hundred out of one hundred questions right on
this kind of test. And then of course it's the team, Like,
we have an incredible team of co founders, and it's
(21:44):
a great mix of researchers who have a proven tracktor
to push to feel forward, as well as actually people
who build real products. At Scape.
Speaker 1 (21:51):
You wade white paper earlier this year with some predictions,
one of which that we'll see the first limps quote
of where superintelligence might go and require open edge and
recursively self improving ALI and a few months later you
announced the company with seven incredible co founders a multi
billion dollar valuation. Any other predictions in that white paper
(22:12):
that we can expect to see becoming companies soon.
Speaker 2 (22:14):
Yeah, it's kind of a cheat code. The best way
to predict the future is to just work on it,
and so I tried to do that when no one
else is doing it. Sometimes people are doing it already.
So like robotics, for instance, I think a lot of
people are working robotics. I predict that in the next
few years, probably not this year, maybe even not next year,
but like the next few years, we're going to get
(22:35):
robotics to be good enough, like just the mechanics of it,
so that then putting really real intelligence into those robots
will enable a lot of people to do boring physical work.
But you know, they will cost quite a lot, so
it's going to take a while. I think already self
driving cars are saving lots of lives where they are
like deployed at scale, So that's exciting to see. And
(22:57):
there's so many predictions I listed like twenty or.
Speaker 1 (22:59):
So any others that you're working on in stealth trying
to solve.
Speaker 2 (23:04):
That's why it's health you know, can share them up.
Speaker 1 (23:07):
You know, you describe yourself in the past as a contrarian,
and you pointed out that in two thousand, ninetenty ten
you're working in a field of AI that was unpopular
until it became the mainstream. Do you still feel like
a contrarian within the industry or do you feel like
the industry has essentially you may be more or less
a taste maker now and therefore, what is your How
does your contrarianism express itself today? Yeah?
Speaker 2 (23:29):
I almost miss sometimes the contrarian days. Was just like
me and a bunch of other like intellectuals, like researchers,
like wanting to build something cool even though no one
cared about it. But it's also really exciting to see obviously,
like how many folks are now in the field, and
how many novel ideas are coming, and how much brain powers,
you know, making improvements happen. I think there's maybe a
(23:51):
little bit of controversy around recourse of self improvement rights
as a great path towards superintelligence, So that is some
of contrariy and task, but maybe not so. I mean,
there are only a few people who believe in this also,
and they're putting really like resources behind this.
Speaker 1 (24:06):
And of course you are the longer that much of
a contrarian visa via the tech industry compared to what
you were, but you are by definition a contrarianto. Most Americans, right,
seventy percent of Americans surveyed recently think AI is advancing
too quickly. Two questions, One why are they wrong? And
two why is the tech industry failing to communicate with them?
Speaker 2 (24:29):
Yeah, it's something that's kind of disheartening to see. It's
also I think a big part of Hollywood, you know,
like people like it's so much easier to tell dystopian
stories that are entertaining than positive future stories like you
Got to Still, you know, inside a positive future embed
some drama. No one wants to see people just being happy,
(24:51):
living fulfilled lives, being slightly more wealthy and better off
and comfortable than they were, like you know a generation prior.
Like that doesn't make an interesting movie, right, And so
I think that's a big part of it. I think
there's also just change as hard, right. People don't like
change as much, and I think the anti is sentiment
is even higher in other places. And I think a
(25:14):
lot of civilizations will eventually have to decide sort of
do they want to just stagnate and stay where they're
at right now and try to fight change as much
as possible or not right? And so I think people
hated electricity, people hated tractors, people hated weaving machines. There's
a lot of folks that don't like change of any kind,
(25:36):
and they think that everything was better when they're younger.
It's usually true because when they're younger healthier. At the
same time, once a loved one gets cancer and there's
no cure, or there is a cure, you're probably happy
that that cure exists. You might not say, oh, that
happened because we spend a lot of like taxpayer money
on research and foundational like biology understanding and like making
(25:57):
DNA sequencing cheaper and cheaper, because initially that was only
for the wealthiest of the people, but now anyone can
afford to seek the DNA, So you have to kind
of zoom out and see how much progress there is.
And also, like the media does not spend a lot
of time sort of amplifying positive research breakthrough.
Speaker 1 (26:15):
Stories in the tech industry itself was also has also
pushed like apocalyptic narratives.
Speaker 2 (26:21):
And it's true some people love to sell their I
by saying it's so dangerous, but they're the ones who
built it best.
Speaker 1 (26:27):
You know, it's many people like to sell them. I
mean that I would say that's arguably the most popular
narrative amongst the biggest day.
Speaker 2 (26:35):
I'm quite surprised by that narrative of like this thing.
Speaker 1 (26:39):
Is so danger you agree, right, I mean that's not
media or filmmakers. That's the people peddling the technology, some
of them, some of them. Well, how big a mistake
is that? How much will that narrative hurts?
Speaker 2 (26:53):
I mean, it's clearly not hurting people's stocks, So I
guess we'll keep hearing it until it doesn't. But it
does her long term because when you say it's so dangerous,
like you're the only one who can do it well,
and people are like, well, and if I now don't
trust you, but I still believe what you said, then
people will think the whole technology is bad, and then
(27:14):
they'll be anti AI antype progress, and that will just
openly not be very helpful.
Speaker 1 (27:19):
So how do you see the role of government. You
mentioned civilizations have to make a decision. Like the way civilizations,
I guess tend to make decisions is either through democracy
or through authoritarianism, right, and so like, if most populations
can't perceive the benefit of science or the scientific method
(27:39):
in their lifetime, is it the role of government to
have a longer term view. I mean, this is essentially
what many people see is happening in China versus what
is happening in the US, which is what's your view
of the of the better model for the for the
long term.
Speaker 2 (27:55):
I obviously love democracy, but I think I think science
and scientific progress has a place everywhere, and it's somewhat
independent of it. But you know, it requires some effort
of people to want to participate in that future. But
it's also you know, it's not for me to say,
like if you live on some beautiful Greek island and
you just enjoy fishing each day, grabbing some fish, maybe
(28:18):
selling some of the market, feeding some of them to yourself,
and that's the life you want to live. I mean,
I'm not the one who judge, you know, how other
people want to live their lives and how maybe societies
want to live their lives. So I do think we'll
see and already we see this now, right, the future
is already here. It's just not equally distributed. Some people
will outlaw like robots and automation and then others won't,
(28:39):
and that it's sort of in some sense it's up
to people to decide.
Speaker 1 (28:43):
But how do we distribute it better? I agree, in
some sense it's up to people to decide, But also
like how do you how do you help people decide
to engage? And whose responsibility is it to help them?
I mean, or is it just totally their own responsibility
because not everyone has the like trume information and time.
And I agree with you. Now in a world where
like anyone can become a code just with natural language,
(29:05):
and there are like these stories of people emerging from
you know, very constrained circumstances and using this technology in
a very like vanguard and leading edge way. But it
leaves a lot of people who are well intentioned left behind.
What's the prescription.
Speaker 2 (29:22):
I think the best thing is for the future to
have better marketing, and if a politician wants to increase
the comfort and wealth of their people, they should try
to help those people understand and see the positive impacts
that AI and science and research can have on their people.
(29:43):
But you know, if enough people decide they just want
to offer ramp from progress and just live on an
island and enjoy the sun, like.
Speaker 1 (29:53):
Most people can't do Most people can't do that.
Speaker 2 (29:55):
I mean, that's but it's a it's a decision each
each civilization has to kind of make for itself. I think,
like you know, it's like people make conscious choices not
to go into the jungle and try to tell like
every tribe about you know, here's an iPhone, it's all
the amazing things that you can do with it. They
just let him let him be. Like in some ways,
like you know, I love progress. I love the Bay
(30:17):
Area because there's so many people here who love progress
and building that future, and I hope we can kind
of bring as many people along with us into this
exciting progress. But you know, you have to ultimately accept
people's decisions and if they want to say no, we
don't want to participate in that, we want to regulate automation,
We want to regulate AI. You know, you can't force
(30:40):
people for their own good.
Speaker 1 (30:41):
There's also a safety piece which I think is becoming
more of a concern for the US government. And two
you know, very senior research in the field, Jeffrey Hinton
and Stuart Russell the very vocal on this issue. What
do you think is do you think they're wrong? Is
there is there enough investment in safety? Or what? What
are they what are they missing?
Speaker 2 (31:02):
So I think safety is very important. I think we
need to like recare about it a lot. We've had
like many interesting papers that our team collectively has published.
Jeff Lun Tim rock Teschel, for instance, built rainbow teaming,
which is an automated red teaming setup where an AI
tries to get another I to say bad things and
(31:24):
unsaved things, and then by virtue of them interacting with
one another in this open ended loop for a long time,
you keep finding new ways to get ANI to say
something bad, and then you can use that data to
then make the I not do it right. So AI
has gotten safer and safer because of an open ended
AI solution to that very problem invented by the very
(31:47):
founders of this company. So I'm really excited that that
you know, we have that in mind, and that I
think we'll do a good job prioritizing it.
Speaker 1 (31:56):
You could also constructively optimistic earlier. I know you use
that phrase constructive optimism quite regularly. What does it mean
to you?
Speaker 2 (32:04):
It means that when we look at the history of humanity,
as soon as we developed the scientific method and the enlightenment,
and we're iterating on that that when you zoom out,
just factually speaking, things have gotten a lot better. There
are some tipsy in there. There are some terrible world wars,
and those you know we should avoid through good policies
(32:26):
and international trade and partnerships and so on. But ultimately,
like the child like death ratio has gone way way down,
how many women die in child labor has gone way down,
the literacy rates have gone way up. Like the wealth
and people living in abject poverty numbers have improved. Like
(32:50):
most statistics that we look at, things have gotten better
because of technological progress. Often there are no more rags
to richer stories because no one has to wear rags
anymore because clothing so cheap because of automation a lot, right,
So there's a lot of positivity. And so if you
now have some optimism and you think that you can
actually implement something and make any idea really work, then
(33:14):
I think you're historically speaking more often, right, and that
should hopefully help inspire people to then think actively about
how to make the future better. And I think that
is something that's unfortunately missing and a lot of public
discourse in the media, in movies, in Hollywood and other places.
(33:35):
But you know, if you just look at the past,
like things have gotten a lot better in almost any
fifty year window, especially since the scientific method has been developed.
Speaker 1 (33:46):
Richard, Thank you so much. Thank you the tech stuff.
This episode was produced by Eliza Dennis and Melissa Slaughter.
It was executive produced by me Julian Nutter and Kate
(34:07):
Osborne for Kaleidoscope and Katrina Novell for iHeart Podcasts. Jack
Insley mixed this episode and Kyle Murdoch wrote Our theme song,
Please rate review, and reach out to us at tech
Stuff Podcast at gmail dot com. We want to know
what you want to know about
Welcome to tech stuff. I'm os Voloscan. For those outside
of the industry, it can feel like AI came out
of nowhere and is now taking us all who knows where.
That's why I wanted to talk to Richard Socher. He
spent the last fifteen years developing the foundational research and
building the infrastructure that powers AI. He's also, as of
recently a double unicorn, founder and CEO of two companies,
(00:37):
U dot Com and Recursive Superintelligence. Recursive launched officially in
mid May with the six hundred and fifteen million dollar
fundraise at evaluation north of four billion dollars, and Richard
is swinging for the holy grail of technology AI that
improves itself what he has called a Eureka machine. Richard,
(00:57):
Welcome to tech stuff. Thanks for having me. Eureka machine
sounds even better than a thinking.
Speaker 2 (01:02):
Machine, doesn't it.
Speaker 1 (01:03):
Yeah, what is it?
Speaker 2 (01:05):
Eureka machine? You know, it's not the official term we
definitely we'll use moving forward, but I do think it's
a great analogy. It's essentially an AI that got very
good at inventing scientific advances, and we'll start with scientific
advances on the science of intelligence and artificial intelligence. Itself,
(01:29):
but eventually believe that that will lead to a machine
that will then get even better at doing all kinds
of and making all other kinds of inventions because it
got very good at ideating, implementing and validating ideas.
Speaker 1 (01:44):
So what is recursive superintelligence? First with a small R
and a small S, and then with a big R
and as a big S in terms of your company.
Speaker 2 (01:51):
Yeah, so recursive superintelligence is essentially an AI that improves itself.
That means over time it also gets better proving itself.
So it's a self reinforcing loop that compounds with essentially
no ceiling. Like most of AI, we have like one
specific objective function and you get these very sort of
(02:12):
spiky intelligence capabilities. Here. We believe this could get much
much smoother because what we're asking the I to do
is to automate knowledge work itself, so automate the discovery
of different kinds of knowledge. And so that's why we
think it will actually lead to this discovery engine, the
(02:33):
Sercom machine. And why is this possible now? It's mostly
because AI is code. An AI can code, and so
you can actually loop that capability onto itself if you
figure out a lot of other triggy things.
Speaker 1 (02:46):
Now, I want to come back to the idea of
AI being spiky, because when you first started working on AI,
there was like ImageNet which you worked on, which recognized
things in images, and then there were language models and
basically every different human sense modality like site, you know, words, visual,
et cetera, were captured in different kind of AI approaches
(03:08):
and products. Right, and then recently in the last couple
of years, it feels like AI has already become multimodal. Right.
You know, you can generate video, you can do math,
you could do coding, like what is the thing which
is not happening now which can happen next?
Speaker 2 (03:25):
You know, people often say ideas are cheap, and it's
sort of true, right, A good idea without good execution
is not very valuable. But also like really good ideas
are hard to come by, and so what we currently
have is really really smart researchers who worked for many
years in AI research developing very strong intuitions on what
(03:48):
is a good idea, what might result in an improvement? Right,
and then they define sort of manually, often data sets
and benchmarks, and then the eye will charge up after
those benchmarks and try to get really really good at
solving those. In contrast to that, our approach to recursive
(04:09):
self improvement is much more aligned with open endedness, which
is often an evolutionary process that essentially has a whole
set of interestingly different ideas that are being explored, often
in peril. And like evolution doesn't really have a ceiling.
(04:32):
You don't just get one hundred out of one hundred
percents correct on one particular benchmark and then you're done.
You can keep iterating in this open ended fashion. And
so often in open endedness, similar to biological and cultural evolution,
entities interact with other entities inside an environment and those
can co evolve, you know, like you have different leafs
(04:55):
like trees, and then you get you know, they get
higher and higher so that the animals can eat the leaves.
But then giraffes, you know, get higher and higher next
so they can eat the leaves again. And so these like,
depending on the environment, different ais can evolve.
Speaker 1 (05:08):
And so what is the role of recursive in all
of this? And what's the first time that you think
somebody outside of the technology industry will experience something that
you've built.
Speaker 2 (05:19):
So the idea about recursive is that here the AI
will output another AI that then becomes the input to
the next iteration of that AI, and at each step
that function will actually change itself. So that's sort of
the technical definition of recursive and the idea here is
(05:39):
that we take the manual steps of ideation, implementation, validation
out of that loop and allow AI to come up
its own ideas, implement them, and then validate them. And
the first time people will observe this is likely in
the AI research world itself. You know, there are a
lot of very very extensive people. They have amazing compensation
(06:03):
and they're doing that work manually, so that will be
the first impact. The second impact will then be in
software engineering, where any company that employees software engineers is
going to want to give them the superpower to essentially
manage a team of thousands of people, all of whom
are just constantly thinking about how to be better at
a task and then be better at that task and
(06:25):
improving themselves also. That will then lead to much more
productivity That is most immediately useful for organizations that have
very clear rewards and objective functions like make this battery better,
like cure this particular disease. And I have like a
loop or ideally even a simulation or some kind of
verification mechanism to really know that I am making progress
(06:48):
towards that specific goal. And then yeah, eventually, I think
any I will likely be built by this, and so
that anyone who uses chatbots and so on in the
future will likely benefit from this technology as well.
Speaker 1 (07:00):
And how did you I mean, You've got a group
of people from Google opener I met as some of
the most distinguished researchers in the whole industry, including yourself, Like,
how did you How did you attract this killer team
at a moment where that particular type of talent is
the most highly valued talent in the world.
Speaker 2 (07:17):
Yeah, So we have eight co founders and indeed some
of the most impactful people from Google deep Mind, from
open Aie, from Meta, from Salesforce Research and other places.
And I think we all came together because of the
shared vision that we believe that recursal self improvement is
(07:37):
the fast path towards superintelligence, and we believe that that
can ultimately lead to much more human flourishing as we
cure certain diseases, as we solve more and more hard
problems that have been very hard for human scientists to solve.
So it's about the shared vision and you know sharing
(07:58):
also the ownership that we own this all together.
Speaker 1 (08:03):
You sent a PhD thesis in twenty fourteen had the
title Recursive deep Learning for Natural Language Processing and Computer Vision.
You're nodding because you know the title of your own
PhD thesis. But I'm curious if you know, just over
a decade ago you could possibly predicted that you'll be
founding a unicorn company going after this problem that was
an academic problem in twenty fourteen.
Speaker 2 (08:25):
Yeah, so it's a different type of recursion there than
my thesis. A lot of us came to the same conclusion,
but from different directions. And yeah, it's to be honest,
a lot of people in the I community thought that
when deep Mind was acquired for you know, it's like
four hundred million back in the day, there was like
PKI hype, right, and it will just not go there.
(08:49):
And even I, like coming from Germany, German mindset is
very often sort of critical and skeptical and like what
could go wrong kind of is where you first go
when something new happens, and when sort of there's a
lot of acceleration and growth, and so I think towards
the end of my PhD, I was becoming more and
more constructively optimistic to but yeah, you don't. You never
(09:13):
know if these things are possible that early. But you
can dream, and if you keep working super hard, eventually
you can make some of these dreams come true. But
of course it's also just like a first milestone. Right
now is the time for us to really show that
we can do.
Speaker 1 (09:26):
This unless they're made redundant by technology.
Speaker 2 (09:29):
I hear that a lot of times out of Germany
and Europe, but you can see this in many places,
like when technology really scales, so you actually end up
in a place where people will have more equitable access
to goods and services. So let me give for example,
the like an iPhone. The iPhone no matter if you're
(09:52):
just a middle class teenager or you're a multi billionaire,
you use the same iPhone right, So even if you're
on like the further is left spectrum or the first right,
it doesn't really matter. You can love technology and its scale.
What other goods and services are there that are bottlenecked
on intelligence that humans don't have currently access to a
(10:13):
personal tutor for your kids. Most people can't afford that
personal assistant to just schedule all the little random things
and taxes and whatnot. Right, most people have to waste
so much of their productive time on these minute, boring tasks.
And I really do think like we're seeing this a
lot in different industries and maybe just my current hypothesis
(10:35):
is that depending on the elasticity of the demand for
a certain good or service, that will tell you whether
I will actually create more jobs when it makes that
more efficient and cheaper, or destroy jobs. And coding, for instance,
what we're seeing is coding is becoming tenex more efficient.
Anyone can now code with English, and somehow we want
(10:59):
more coders. A number of open jaw positions has actually
increased by allowing people to be that much more productive
because ultimately anyone could have software. Everyone can have a
personal suite of software tools, so you could have billions
of different software products that are customed for you. And
so the ealyticity of demand is very high as that
(11:20):
price goes down, and people did not make themselves obsolete
at all, and so yeah, like some of these industries
will shift as that the prices go down. But we've
seen that in history many times, and it's always been
for the better. Like only five percent now work in agriculture,
yet like not ninety five percent of people are unemployed,
even though it used to be over ninety percent of
people work in agriculture, And like, somehow we all found
(11:43):
new things to do instead of standing there in a
field with our hands.
Speaker 1 (11:46):
But just a pushback, I mean, it's interesting, like, do
you think the demand for new software applications comes from
real people or is that a demand imposed by capital?
As so much money flowing into the space that by
definition they are more open software engineering jobs because like
wealth and other major institutional investors have decided this is
the highest potential return sector in the world, and therefore,
(12:07):
like the demand creation is artificial.
Speaker 2 (12:08):
I mean, investor, Like that's that's sort of one of
the beauties of capitalism. Like people don't want to just
put money into something that people don't want and that
doesn't make any other money because no one's using it, right,
So it's like that that problem will solve itself quite
easily and sure, like just like the price of energy
can fluctuate a little bit, the price of intelligence may
fluctuate a little bit, but ultimately, like there are even
(12:31):
more use cases for intelligence than there are for for energy,
even though they're obviously highly correlated and one is required
by the other and so on. But I like, I
don't think there's like just like oh, sovereigns, there's abstract entity.
They're they all want to like invest.
Speaker 1 (12:45):
They want to return on the capital. But then you
get the magest But I'm not I'm not just in
terms of the elytcity of demand for software like I
think it's it's I think the proof will be in
the pudding as to whether that's like true demand or
like demand, like demand imposed by capital based on like
projective outcomes. I think that that remains you see.
Speaker 2 (13:01):
That's that's of course true for every investment, right do
like especially if you do early investments into categories that
don't yet like exist, then that's even more true. But
software does exist.
Speaker 1 (13:12):
And by the way, to be fair to you or
it should, you've backed seven companies as an investor to
become unicorns. So I would, I would anyone listening it
just became it rich Richard maybe a more shrewder investor
than I am. That's kind of extraordinary perplexity. Hugging face
weights and biases somehow sohold names like So, let's take
a couple of steps back. Recursive super intelligence is the
(13:35):
story of the moment. You're also the founder and CEO
of you dot com. You dot com was doing basically
AI search before the chat gipt moment, but it's largely
for enterprise, and you invested in all these other AI unicorns.
So what is it? What is it you see about
the world earlier than most people.
Speaker 2 (13:52):
I spent a lot of time thinking about the future
and how I can have positive impact on it. And
for many years, like a decade off my life that
was actually making the technology work and doing research in
it because it just wasn't working. And once it was working,
I'm like, well, maybe just making it slightly incrementally better
for a while wasn't necessary. Just had to scale the
(14:12):
existing ideas and modify them around the edges and improve
them here and there. And then I just love all
these use cases, and you know, especially some deep industries
like healthcare, it was clear that I needed to support
and help from people who deeply know the space, know
how farmer companies actually develop and bring new medications and
cures into the market, and drugs. But you know, every
(14:36):
single space has has this opportunity right now in AI,
like a I will change every industry. That's why I
sort of AI of X, you know, like and then
oftentimes you can think from first principles sort of what
would people really want like in that industry and can
you make that now possible with AI. So we wanted
to connect like some of the world's leading experts like
(14:58):
Chris Manning and the gold Bloom with really strong industry
experts like Nick Krant's and our team or Chrish and
and actually like merge them. And that's that's what allowed
us to invest in these really exciting companies. I'll just
give you one example. One is called Parallel Bio.
Speaker 1 (15:15):
Uh.
Speaker 2 (15:16):
They build organoids and they with AI, they are able
to track the organoids and see how they are influenced
by different.
Speaker 1 (15:24):
Collections of kidney cells exactly.
Speaker 2 (15:27):
In their case they will they build lymph nodes, lymphoe organoids,
and so this one company actually got FDA approval to
SKIP for toxicity testing of immune systems animal trials by
using these organoids in the petridish being evaluated and created
partially by AI and robotic processes. And so it's an
incredibly exciting company that single handedly could at scale save
(15:50):
millions of animal lives, right, And so when I hear
all these people saying AI, it's gonna so terrible, And like,
I see all these like use cases where people get
access to goods and services that they couldn't afford before,
or we save millions of animal lives and so on.
That's what I see when I think about AI all
the time, and that's what inspires me for AIX And
(16:11):
then yeah, a lot of times it's just like working
in research and AI. You can see the applications of
it a little bit earlier to an laws.
Speaker 1 (16:20):
After the break, Richard predicts where technology will go in
the next few years. Stay with us. We talk a
lot on this show about protecting your data, especially in
the age of AI, and how scary it can be
when it's breached. And I want to tell you today
about NordVPN, which really covers all the bases when it
(16:42):
comes to privacy. I travel a lot, and I use
Wi Fi when I'm flying all the time, and NordVPN
makes me confident in no matter where I am in
the world or the sky for that matter, my private
details like bank information, passwords and online identity is safe,
and it's also possible to switch on virtual location, which
allows you to save money by buying flights and hotels,
(17:03):
or subscriptions or even streaming soccer or football as I
like to call it from other countries at a cheaper price,
and NordVPN doesn't slow you down. It has super fast
Internet speed, no buffering or lagging while streaming. It is
premium cybersecurity for the price of one cup of coffee
per month. To get the best discount of your NordVPN plan,
(17:25):
go to NordVPN dot com slash tech Stuff. Our link
will also give you four extra months on the two
year plan and there's no risk with Nord's thirty day
money back guarantee. The link is in the podcast episode
description box Welcome back to tech Stuff. So Richard, I
want to talk to you about you dot com. I mean,
(17:46):
you founded it before ELM's were popular, and am I
right in characterizing it basically as an AI powered search
engine but for businesses and enterprises.
Speaker 2 (17:57):
That's exactly right. Yeah. We now provide web search APIs
for enterprises so they can make their lms their chat wants,
their agents be up to date, accurate and have citations.
Speaker 1 (18:07):
But when you saw, for example, after you've founded you
dot com, the chat GPT moment happened, did you have
any like Pomo or wish that this had been like
your product that had become the consumer runaway Darling.
Speaker 2 (18:20):
Of course, yeah, we were initially a consumer company and
we had the most accurate answers. When Google was still
telling people like to eat rocks or put blue on pizzas, like,
we had way fewer hallucinations, like almost none before anyone else.
But it just didn't quite click with consumers, like the branding,
the marketing wasn't quite there. And so eventually we found
that who really cares about the technology. It's developers who
(18:44):
implement the technology, and they just want to get the
best version, the best API for their lms. And that's
how we really started taking off and made a lot
more revenue and are growing really healthy now.
Speaker 1 (18:56):
And with Recursive, I mean was the ambition was ambition
for this company to play in the world playing culture.
Speaker 2 (19:03):
I think we have a shot here at building sort
of in the Davit Deutsche sense of the beginning of infinity,
like an ultimate Eureka machine that has this open ended
way of exploring the frontier of knowledge and really help
become this invention generating machine. And I think that will
(19:23):
be exciting first for a I in the field of AI,
but eventually for energy and better fusion systems, better battery materials,
for in chemistry than in especially pre clinical biology, like
developing new kinds of drugs and molecules. Like, there's so
many exciting applications once you have this machine. You know,
(19:46):
it's not just that the level of the bachelor's but
sort of five thousand PhD degrees that all went and
are now researching a particular problem for you. So I
think that that is the ambition.
Speaker 1 (19:58):
I mean, you said that Open AI, Google, Anthropic Rule
very different companies in different spaces, but also acknowledge that
they're all pursuing superintelligence. Your twenty five person lab going
up against these companies in some sense with thousands of
employees practically unlimited compute, particularly Google's case. I suppose what
gives you an edge? How do you win?
Speaker 2 (20:17):
We have a singular focus, right, there's a lot of
labs that do a lot of different things. For recursive
that is, like recursal superintelligence in the name, that is
what we're doing. The second one is infrastructure. We're like
custom building everything from the ground up with superintelligence in mind.
That means that we designed sort of all the teams,
(20:39):
the resources, the infrastructure, the architecture, and so on from
the ground up for autonomous scientific discovery. So that's number
two's infrastrucure. Number three is that we're really inspired by
so called open ended systems. It's a fundamentally different approach
that releaves us of some of these constraints that others face.
Speaker 1 (20:58):
And why does it ALREADI view constraints. So in many ways,
like what a lot.
Speaker 2 (21:02):
Of AIS and the E labs are doing is they
kind of define these spiky intelligences there, like this is
a particular data set, benchmark or shown that we want
the eye to be really good at, and then the
eye gets really good along that one spike. But then
there are all these sort of valleys between the spikes,
and so we believe that instead of optimizing one fixed
(21:22):
target that will ultimately plateau, we believe in open endedness,
you have interestingly different sets of ideas and they can
compete with one another and they keep being keep improving
without this sort of obvious seiling of like oh you
got a hundred out of one hundred questions right on
this kind of test. And then of course it's the team, Like,
we have an incredible team of co founders, and it's
(21:44):
a great mix of researchers who have a proven tracktor
to push to feel forward, as well as actually people
who build real products. At Scape.
Speaker 1 (21:51):
You wade white paper earlier this year with some predictions,
one of which that we'll see the first limps quote
of where superintelligence might go and require open edge and
recursively self improving ALI and a few months later you
announced the company with seven incredible co founders a multi
billion dollar valuation. Any other predictions in that white paper
(22:12):
that we can expect to see becoming companies soon.
Speaker 2 (22:14):
Yeah, it's kind of a cheat code. The best way
to predict the future is to just work on it,
and so I tried to do that when no one
else is doing it. Sometimes people are doing it already.
So like robotics, for instance, I think a lot of
people are working robotics. I predict that in the next
few years, probably not this year, maybe even not next year,
but like the next few years, we're going to get
(22:35):
robotics to be good enough, like just the mechanics of it,
so that then putting really real intelligence into those robots
will enable a lot of people to do boring physical work.
But you know, they will cost quite a lot, so
it's going to take a while. I think already self
driving cars are saving lots of lives where they are
like deployed at scale, So that's exciting to see. And
(22:57):
there's so many predictions I listed like twenty or.
Speaker 1 (22:59):
So any others that you're working on in stealth trying
to solve.
Speaker 2 (23:04):
That's why it's health you know, can share them up.
Speaker 1 (23:07):
You know, you describe yourself in the past as a contrarian,
and you pointed out that in two thousand, ninetenty ten
you're working in a field of AI that was unpopular
until it became the mainstream. Do you still feel like
a contrarian within the industry or do you feel like
the industry has essentially you may be more or less
a taste maker now and therefore, what is your How
does your contrarianism express itself today? Yeah?
Speaker 2 (23:29):
I almost miss sometimes the contrarian days. Was just like
me and a bunch of other like intellectuals, like researchers,
like wanting to build something cool even though no one
cared about it. But it's also really exciting to see obviously,
like how many folks are now in the field, and
how many novel ideas are coming, and how much brain powers,
you know, making improvements happen. I think there's maybe a
(23:51):
little bit of controversy around recourse of self improvement rights
as a great path towards superintelligence, So that is some
of contrariy and task, but maybe not so. I mean,
there are only a few people who believe in this also,
and they're putting really like resources behind this.
Speaker 1 (24:06):
And of course you are the longer that much of
a contrarian visa via the tech industry compared to what
you were, but you are by definition a contrarianto. Most Americans, right,
seventy percent of Americans surveyed recently think AI is advancing
too quickly. Two questions, One why are they wrong? And
two why is the tech industry failing to communicate with them?
Speaker 2 (24:29):
Yeah, it's something that's kind of disheartening to see. It's
also I think a big part of Hollywood, you know,
like people like it's so much easier to tell dystopian
stories that are entertaining than positive future stories like you
Got to Still, you know, inside a positive future embed
some drama. No one wants to see people just being happy,
(24:51):
living fulfilled lives, being slightly more wealthy and better off
and comfortable than they were, like you know a generation prior.
Like that doesn't make an interesting movie, right, And so
I think that's a big part of it. I think
there's also just change as hard, right. People don't like
change as much, and I think the anti is sentiment
is even higher in other places. And I think a
(25:14):
lot of civilizations will eventually have to decide sort of
do they want to just stagnate and stay where they're
at right now and try to fight change as much
as possible or not right? And so I think people
hated electricity, people hated tractors, people hated weaving machines. There's
a lot of folks that don't like change of any kind,
(25:36):
and they think that everything was better when they're younger.
It's usually true because when they're younger healthier. At the
same time, once a loved one gets cancer and there's
no cure, or there is a cure, you're probably happy
that that cure exists. You might not say, oh, that
happened because we spend a lot of like taxpayer money
on research and foundational like biology understanding and like making
(25:57):
DNA sequencing cheaper and cheaper, because initially that was only
for the wealthiest of the people, but now anyone can
afford to seek the DNA, So you have to kind
of zoom out and see how much progress there is.
And also, like the media does not spend a lot
of time sort of amplifying positive research breakthrough.
Speaker 1 (26:15):
Stories in the tech industry itself was also has also
pushed like apocalyptic narratives.
Speaker 2 (26:21):
And it's true some people love to sell their I
by saying it's so dangerous, but they're the ones who
built it best.
Speaker 1 (26:27):
You know, it's many people like to sell them. I
mean that I would say that's arguably the most popular
narrative amongst the biggest day.
Speaker 2 (26:35):
I'm quite surprised by that narrative of like this thing.
Speaker 1 (26:39):
Is so danger you agree, right, I mean that's not
media or filmmakers. That's the people peddling the technology, some
of them, some of them. Well, how big a mistake
is that? How much will that narrative hurts?
Speaker 2 (26:53):
I mean, it's clearly not hurting people's stocks, So I
guess we'll keep hearing it until it doesn't. But it
does her long term because when you say it's so dangerous,
like you're the only one who can do it well,
and people are like, well, and if I now don't
trust you, but I still believe what you said, then
people will think the whole technology is bad, and then
(27:14):
they'll be anti AI antype progress, and that will just
openly not be very helpful.
Speaker 1 (27:19):
So how do you see the role of government. You
mentioned civilizations have to make a decision. Like the way civilizations,
I guess tend to make decisions is either through democracy
or through authoritarianism, right, and so like, if most populations
can't perceive the benefit of science or the scientific method
(27:39):
in their lifetime, is it the role of government to
have a longer term view. I mean, this is essentially
what many people see is happening in China versus what
is happening in the US, which is what's your view
of the of the better model for the for the
long term.
Speaker 2 (27:55):
I obviously love democracy, but I think I think science
and scientific progress has a place everywhere, and it's somewhat
independent of it. But you know, it requires some effort
of people to want to participate in that future. But
it's also you know, it's not for me to say,
like if you live on some beautiful Greek island and
you just enjoy fishing each day, grabbing some fish, maybe
(28:18):
selling some of the market, feeding some of them to yourself,
and that's the life you want to live. I mean,
I'm not the one who judge, you know, how other
people want to live their lives and how maybe societies
want to live their lives. So I do think we'll
see and already we see this now, right, the future
is already here. It's just not equally distributed. Some people
will outlaw like robots and automation and then others won't,
(28:39):
and that it's sort of in some sense it's up
to people to decide.
Speaker 1 (28:43):
But how do we distribute it better? I agree, in
some sense it's up to people to decide, But also
like how do you how do you help people decide
to engage? And whose responsibility is it to help them?
I mean, or is it just totally their own responsibility
because not everyone has the like trume information and time.
And I agree with you. Now in a world where
like anyone can become a code just with natural language,
(29:05):
and there are like these stories of people emerging from
you know, very constrained circumstances and using this technology in
a very like vanguard and leading edge way. But it
leaves a lot of people who are well intentioned left behind.
What's the prescription.
Speaker 2 (29:22):
I think the best thing is for the future to
have better marketing, and if a politician wants to increase
the comfort and wealth of their people, they should try
to help those people understand and see the positive impacts
that AI and science and research can have on their people.
(29:43):
But you know, if enough people decide they just want
to offer ramp from progress and just live on an
island and enjoy the sun, like.
Speaker 1 (29:53):
Most people can't do Most people can't do that.
Speaker 2 (29:55):
I mean, that's but it's a it's a decision each
each civilization has to kind of make for itself. I think,
like you know, it's like people make conscious choices not
to go into the jungle and try to tell like
every tribe about you know, here's an iPhone, it's all
the amazing things that you can do with it. They
just let him let him be. Like in some ways,
like you know, I love progress. I love the Bay
(30:17):
Area because there's so many people here who love progress
and building that future, and I hope we can kind
of bring as many people along with us into this
exciting progress. But you know, you have to ultimately accept
people's decisions and if they want to say no, we
don't want to participate in that, we want to regulate automation,
We want to regulate AI. You know, you can't force
(30:40):
people for their own good.
Speaker 1 (30:41):
There's also a safety piece which I think is becoming
more of a concern for the US government. And two
you know, very senior research in the field, Jeffrey Hinton
and Stuart Russell the very vocal on this issue. What
do you think is do you think they're wrong? Is
there is there enough investment in safety? Or what? What
are they what are they missing?
Speaker 2 (31:02):
So I think safety is very important. I think we
need to like recare about it a lot. We've had
like many interesting papers that our team collectively has published.
Jeff Lun Tim rock Teschel, for instance, built rainbow teaming,
which is an automated red teaming setup where an AI
tries to get another I to say bad things and
(31:24):
unsaved things, and then by virtue of them interacting with
one another in this open ended loop for a long time,
you keep finding new ways to get ANI to say
something bad, and then you can use that data to
then make the I not do it right. So AI
has gotten safer and safer because of an open ended
AI solution to that very problem invented by the very
(31:47):
founders of this company. So I'm really excited that that
you know, we have that in mind, and that I
think we'll do a good job prioritizing it.
Speaker 1 (31:56):
You could also constructively optimistic earlier. I know you use
that phrase constructive optimism quite regularly. What does it mean
to you?
Speaker 2 (32:04):
It means that when we look at the history of humanity,
as soon as we developed the scientific method and the enlightenment,
and we're iterating on that that when you zoom out,
just factually speaking, things have gotten a lot better. There
are some tipsy in there. There are some terrible world wars,
and those you know we should avoid through good policies
(32:26):
and international trade and partnerships and so on. But ultimately,
like the child like death ratio has gone way way down,
how many women die in child labor has gone way down,
the literacy rates have gone way up. Like the wealth
and people living in abject poverty numbers have improved. Like
(32:50):
most statistics that we look at, things have gotten better
because of technological progress. Often there are no more rags
to richer stories because no one has to wear rags
anymore because clothing so cheap because of automation a lot, right,
So there's a lot of positivity. And so if you
now have some optimism and you think that you can
actually implement something and make any idea really work, then
(33:14):
I think you're historically speaking more often, right, and that
should hopefully help inspire people to then think actively about
how to make the future better. And I think that
is something that's unfortunately missing and a lot of public
discourse in the media, in movies, in Hollywood and other places.
(33:35):
But you know, if you just look at the past,
like things have gotten a lot better in almost any
fifty year window, especially since the scientific method has been developed.
Speaker 1 (33:46):
Richard, Thank you so much. Thank you the tech stuff.
This episode was produced by Eliza Dennis and Melissa Slaughter.
It was executive produced by me Julian Nutter and Kate
(34:07):
Osborne for Kaleidoscope and Katrina Novell for iHeart Podcasts. Jack
Insley mixed this episode and Kyle Murdoch wrote Our theme song,
Please rate review, and reach out to us at tech
Stuff Podcast at gmail dot com. We want to know
what you want to know about
TechStuff News
Hosts And Creators
Oz Woloshyn
Karah Preiss
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Hey Jonas!
Hey Jonas! The official Jonas Brothers podcast. Hosted by Kevin, Joe, and Nick Jonas. It’s the Jonas Brothers you know... musicians, actors, and well, yes, brothers. Now, they’re sharing another side of themselves in the playful, intimate, and irreverent way only they can. Spend time with the Jonas Brothers here and stay a little bit longer for deep conversations like never before.
Crime Junkie
Does hearing about a true crime case always leave you scouring the internet for the truth behind the story? Dive into your next mystery with Crime Junkie. Every Monday, join your host Ashley Flowers as she unravels all the details of infamous and underreported true crime cases with her best friend Brit Prawat. From cold cases to missing persons and heroes in our community who seek justice, Crime Junkie is your destination for theories and stories you won’t hear anywhere else. Whether you're a seasoned true crime enthusiast or new to the genre, you'll find yourself on the edge of your seat awaiting a new episode every Monday. If you can never get enough true crime... Congratulations, you’ve found your people. Follow to join a community of Crime Junkies! Crime Junkie is presented by Audiochuck Media Company.