Ronjon.mp4

Back

Or filter by your interest

Browse all our content

SuperReturn

Unlock all the content with only {{price}}

Ronjon.vtt

English

{{00:00:07}}
I think my first AI system was
40 years ago, believe it or not.

{{00:00:10}}
The AI field really dates back to
1956 when the term was coined, or 1936

{{00:00:17}}
when Alan Turing was on the scene.

{{00:00:20}}
Really the last few years
we've really seen an inflection

{{00:00:22}}
point in language processing.

{{00:00:25}}
We've had AI systems for
image recognition, face

{{00:00:28}}
recognition, voice recognition.

{{00:00:30}}
The big difference now
is large language models.

{{00:00:34}}
So what they mean is normally
we would predict the next

{{00:00:38}}
word or the next two words.

{{00:00:40}}
Now our context window
is thousands of words.

{{00:00:43}}
And it's amazing how accurately, how
well, it shows a real description

{{00:00:50}}
when you ask it a question.

{{00:00:51}}
It feels like it's really human.

{{00:00:53}}
It's important to note it's not human.

{{00:00:55}}
It's just computing probabilities,
computing statistics.

{{00:00:58}}
But it turns out it works really well.

{{00:01:01}}
You're seeing things that couldn't
be done before and now can be done.

{{00:01:06}}
You're seeing this on your search engines
where they're giving you sentences,

{{00:01:09}}
you can rewrite essays, you can write
sections of books typically there's

{{00:01:13}}
limits to the word lengths you can put up.

{{00:01:15}}
And so anything that can be done in the
text processing area is really now for

{{00:01:21}}
being replicated and done by a machine.

{{00:01:24}}
Anything that needs you'd say,
reasoning, is if it's a few

{{00:01:29}}
steps, AI can work on that.

{{00:01:32}}
There's huge industries where
it can be now the grunt work

{{00:01:36}}
can be taken away from you.

{{00:01:40}}
So, people are maybe in the
lower levels of a vertical

{{00:01:43}}
industry say law for example.

{{00:01:46}}
A lot of the form filling and
the interaction can be actually

{{00:01:51}}
done by a machine, but the
strategy can be done by a partner.

{{00:01:56}}
So they can be more leveraged.

{{00:01:58}}
Other industries, certainly
medicine, we've seen databases

{{00:02:03}}
coming up that are available to
be trained, these large language

{{00:02:07}}
models, and certainly in the media.

{{00:02:11}}
So you seem to be able to take a video,
and normally you take 16 takes, and

{{00:02:17}}
now you can just take one and then
deepfake the rest, to fill it in.

{{00:02:22}}
There's a few ideas.

{{00:02:28}}
There's probably three areas
you can look at investments.

{{00:02:31}}
One is the actual AI core
algorithm part itself.

{{00:02:36}}
Now that's like big game elephant hunting.

{{00:02:39}}
The most recent AI system talking about
hundred $34 million to train people

{{00:02:45}}
are talking about putting hundreds of
billions of dollars to train these big

{{00:02:49}}
models and having thousands and tens of
thousands of chips connected together.

{{00:02:56}}
So that's one area.

{{00:02:58}}
Second area is using the AI.

{{00:03:00}}
You're not training the models, but you're
using it for particular applications.

{{00:03:03}}
There it's smaller investments required.

{{00:03:07}}
And really the idea is the most important.

{{00:03:09}}
And what is the space that's
right for looking at, areas.

{{00:03:14}}
And the third area is really, operations.

{{00:03:20}}
So you can either improve your own product
or you can improve your operations.

{{00:03:25}}
And so companies are talking about being
AI first companies so trying to improve

{{00:03:31}}
every part of their organization their HR
systems, their marketing systems in the

{{00:03:37}}
operations, not just the product itself.

{{00:03:41}}
Now what I'm working on in AI in
the next era is aging science.

{{00:03:47}}
So what I'm working on is trying to
create a vaccine but a vaccine for aging.

{{00:03:53}}
What does that mean?

{{00:03:54}}
That sounds like a moonshot.

{{00:03:56}}
Project, but you can now think about
it with the new tools we have, with

{{00:04:01}}
the new artificial intelligence
tools that we are able to have.

{{00:04:04}}
The body is so complex, it's difficult
for any one person, or even group of

{{00:04:09}}
people, to figure out how it works.

{{00:04:11}}
Really, we've been using trial and
error for the last century where

{{00:04:16}}
someone's got an idea, this is how
the mechanism of a disease might work.

{{00:04:21}}
And now we've got many, more databases.

{{00:04:23}}
I've got a letter in the post yesterday
that the NHS in England is trying to

{{00:04:29}}
collect a database of 5 million people.

{{00:04:33}}
You go into a clinic and it'll
collect your blood pressure, your

{{00:04:35}}
cholesterol, and look at what diseases
you may have had in the past and

{{00:04:39}}
what treatments you're getting.

{{00:04:41}}
And you get 5 million people,
a database of 5 million people.

{{00:04:43}}
So that way we might know what works
and what doesn't work because there's

{{00:04:47}}
so many confounding effects that it's
very difficult to figure that out.

{{00:04:52}}
So what we're trying to do.

{{00:04:54}}
How would you make a vaccine for ageing?

{{00:04:56}}
Computationally, how
would you invest in it?

{{00:04:59}}
Really look at it as a engineering
problem and work backwards.

{{00:05:05}}
Trying to do it all at once
will be very difficult.

{{00:05:07}}
What we're looking at is seeing how
can we create a computational model

{{00:05:11}}
for cancer in the first instance?

{{00:05:13}}
There's about hundreds of cancer.

{{00:05:14}}
Cancer is not one thing.

{{00:05:16}}
Can we solve all the cancers
by killing the cancer cells

{{00:05:20}}
but not harming the good cells?

{{00:05:22}}
And can we do that computationally?

{{00:05:24}}
Once we've got that computational
approach, proved, can we then apply

{{00:05:28}}
it for other categories of disease?

{{00:05:30}}
Cardiovascular is the number one killer.

{{00:05:33}}
Can we actually apply
it to cardiovascular?

{{00:05:35}}
What is it that helps our
hearts age more slowly?

{{00:05:39}}
And finally, neuro, dementia, Alzheimer's.

{{00:05:43}}
How do we prevent that?

{{00:05:44}}
But instead of trying, to think
of the mechanisms randomly.

{{00:05:51}}
Can we actually use
computational approaches?

{{00:05:55}}
There's probably a million
papers scientific papers

{{00:05:59}}
produced in medicine every year.

{{00:06:01}}
No one's got time to read them.

{{00:06:03}}
So can you have a machine read them?

{{00:06:05}}
There's probably hundreds of
millions of papers that we have

{{00:06:07}}
our machine read those papers.

{{00:06:09}}
These are ideal things that can
outperform humans to actually

{{00:06:14}}
come out with real answers.

{{00:06:21}}
There are many problems in
medicine, in drug discovery.

{{00:06:24}}
Certainly, basically, there's
a path to drug discovery.

{{00:06:28}}
You come up with a candidate, and then you
have to actually test it in real humans.

{{00:06:33}}
At least today you have to.

{{00:06:35}}
Maybe in the future, 100 years from now,
we actually believe the computational

{{00:06:40}}
model more than we do with trials.

{{00:06:41}}
But you have to, today you have
to put drugs through trials.

{{00:06:45}}
And can we actually speed those up by
doing what we call adaptive trials?

{{00:06:49}}
So you have, this is what happened in
COVID, we try a number of things all at

{{00:06:53}}
the same time, and you have something
called the drop the loser design.

{{00:06:56}}
Once something stops working, stop
working on that and working on those.

{{00:07:00}}
Again, you can do that
computationally instead of just

{{00:07:03}}
thinking about it randomly.

{{00:07:05}}
Another aspect is delivery.

{{00:07:06}}
So you can come out with a drug, and
this is very common with brain diseases.

{{00:07:11}}
You've got a drug that can actually
kill a brain tumor, but you can't

{{00:07:17}}
actually get the drug to the tumor
because there's something called the

{{00:07:20}}
blood brain barrier, which normally to
stop you having, your brain infected

{{00:07:24}}
when you catch the flu, you don't want
your brain infected, so you cannot

{{00:07:28}}
cross the blood vessels into the brain.

{{00:07:31}}
But when you wanna actually wanna
get a drug to a certain place, you

{{00:07:34}}
actually want it to get across.

{{00:07:36}}
There's a whole set of technologies.

{{00:07:38}}
either by changing the molecular
structure so it can cross the walls

{{00:07:41}}
of the blood vessels, or by using
something called focused ultrasound,

{{00:07:46}}
where you have ultrasound that goes
to a certain point in the brain that

{{00:07:48}}
just opens up the holes for just a few
hours to let the drugs in at that point.

{{00:07:55}}
Our 42 group, we have investments,
and activities in all of these areas.

{{00:08:02}}
Cordon's Medical is the focused
ultrasound company Ajemica is the

{{00:08:06}}
aging vaccine company we're invested
in a number of funds as well to really

{{00:08:12}}
look at deep tech, deep science.

{{00:08:14}}
Deep tech is a little bit different
from, enterprise software.

{{00:08:19}}
Typically we are looking at
things that have no competitors.

{{00:08:24}}
You can imagine an aging
vaccine, no competitors.

{{00:08:27}}
But of course they're different,
more difficult to accomplish.

{{00:08:30}}
Being, thinking about how to
invest in something like that.

{{00:08:36}}
You have to actually take a deep
dive and look at first principles and

{{00:08:40}}
build up the infrastructure in your
family office or your venture fund

{{00:08:43}}
to actually try and reduce the risk.

{{00:08:46}}
To try and reduce the risk,
because otherwise, by definition,

{{00:08:48}}
these are tough technologies.

{{00:08:50}}
But if you can make it work
these so called moonshots,

{{00:08:53}}
then you get a moonshot return.

{{00:08:59}}
Really, you have to do your own work.

{{00:09:02}}
If you work on quote unquote
advice or advisors, it's a

{{00:09:07}}
second hand information problem.

{{00:09:08}}
Microsoft You really have to set up
your organization to, when you invest

{{00:09:12}}
in a company, learn from that investment
so you can make other investments.

{{00:09:16}}
You are almost like being part of
the company that you're investing in.

{{00:09:21}}
What's going what's going wrong?

{{00:09:24}}
In our, in, in my case, I also teach,
I teach at Stanford University.

{{00:09:28}}
And so teaching you by definition,
you have to know the subject.

{{00:09:34}}
You have to know the subject.

{{00:09:35}}
And so how I stay at the cutting
edge is one, is work with companies.

{{00:09:40}}
Not just invest in them, you have
to really interact with them.

{{00:09:43}}
Second, teach the subjects.

{{00:09:45}}
I teach an investing course
as well as a AI course.

{{00:09:48}}
And thirdly, be a practitioner.

{{00:09:51}}
So this is a little bit different.

{{00:09:53}}
Many people just invest.

{{00:09:55}}
I actually invent as well.

{{00:10:00}}
I think, if you catch those three.

{{00:10:03}}
Now, maybe if you haven't got the
background to do inventing, but maybe

{{00:10:06}}
you can still be part of an invention
scale, direct investing we call that,

{{00:10:11}}
direct investing into an entity.

I think my first AI system was
40 years ago, believe it or not.

The AI field really dates back to
1956 when the term was coined, or 1936

Really the last few years
we've really seen an inflection

We've had AI systems for
image recognition, face

The big difference now
is large language models.

So what they mean is normally
we would predict the next

Now our context window
is thousands of words.

And it's amazing how accurately, how
well, it shows a real description

It's just computing probabilities,
computing statistics.

You're seeing things that couldn't
be done before and now can be done.

You're seeing this on your search engines
where they're giving you sentences,

you can rewrite essays, you can write
sections of books typically there's

limits to the word lengths you can put up.

And so anything that can be done in the
text processing area is really now for

Anything that needs you'd say,
reasoning, is if it's a few

There's huge industries where
it can be now the grunt work

So, people are maybe in the
lower levels of a vertical

A lot of the form filling and
the interaction can be actually

done by a machine, but the
strategy can be done by a partner.

Other industries, certainly
medicine, we've seen databases

coming up that are available to
be trained, these large language

So you seem to be able to take a video,
and normally you take 16 takes, and

now you can just take one and then
deepfake the rest, to fill it in.

There's probably three areas
you can look at investments.

One is the actual AI core
algorithm part itself.

Now that's like big game elephant hunting.

The most recent AI system talking about
hundred $34 million to train people

are talking about putting hundreds of
billions of dollars to train these big

models and having thousands and tens of
thousands of chips connected together.

You're not training the models, but you're
using it for particular applications.

And really the idea is the most important.

And what is the space that's
right for looking at, areas.

And the third area is really, operations.

So you can either improve your own product
or you can improve your operations.

And so companies are talking about being
AI first companies so trying to improve

every part of their organization their HR
systems, their marketing systems in the

Now what I'm working on in AI in
the next era is aging science.

So what I'm working on is trying to
create a vaccine but a vaccine for aging.

Project, but you can now think about
it with the new tools we have, with

the new artificial intelligence
tools that we are able to have.

The body is so complex, it's difficult
for any one person, or even group of

Really, we've been using trial and
error for the last century where

someone's got an idea, this is how
the mechanism of a disease might work.

I've got a letter in the post yesterday
that the NHS in England is trying to

You go into a clinic and it'll
collect your blood pressure, your

cholesterol, and look at what diseases
you may have had in the past and

And you get 5 million people,
a database of 5 million people.

So that way we might know what works
and what doesn't work because there's

so many confounding effects that it's
very difficult to figure that out.

Computationally, how
would you invest in it?

Really look at it as a engineering
problem and work backwards.

Trying to do it all at once
will be very difficult.

What we're looking at is seeing how
can we create a computational model

Can we solve all the cancers
by killing the cancer cells

Once we've got that computational
approach, proved, can we then apply

Can we actually apply
it to cardiovascular?

What is it that helps our
hearts age more slowly?

And finally, neuro, dementia, Alzheimer's.

But instead of trying, to think
of the mechanisms randomly.

Can we actually use
computational approaches?

There's probably a million
papers scientific papers

There's probably hundreds of
millions of papers that we have

How can investors make the most of the AI and biotech revolution? Ronjon Nag, Founder of R42 Group Family Office and Adjunct Professor, shares his insights at SuperVenture North America. Explore the emerging industries driving innovation, tips for investing in AI and biotech, and how to distinguish real opportunities from hype. Plus, learn about Ronjon’s groundbreaking work on an anti-aging vaccine.
