Moonshots with Peter Diamandis - How Quantum & AI Will Shape the World’s Future w/ Jack Hidary | EP #123
Episode Date: October 10, 2024In this episode, Jack and Peter discuss LQMs which are the next stage of AI beyond LLMs. LQMs are growing in applications in the biggest parts of the economy. Jack and Peter then talk about the synerg...y of bringing AI and quantum together for global impact. Recorded on Oct 1st, 2024 Views are my own thoughts; not Financial, Medical, or Legal Advice. 01:57 | The Future of AI and Quantum 31:41 | The Intersection of AI and Brain Imaging 01:04:46 | Decoding the Mysteries of Quantum Jack Hidary is a leading entrepreneur and visionary at the forefront of AI and quantum technology as the CEO of SandboxAQ, raising over $500m in funding. He is the author of forthcoming book AI or Die and the influential textbook; Quantum Computing: An Applied Approach. A serial entrepreneur, Hidary co-founded and led EarthWeb/Dice from inception to IPO, and co-founded Vista Research and sold it to S&P/McGraw-Hill. Jack studied neuroscience at Columbia University and was a Stanley Fellow in Clinical Neuroscience at the NIH where he applied neural networks to brain imaging. Learn more about SandboxAQ: https://www.sandboxaq.com/ ____________ I only endorse products and services I personally use. To see what they are, please support this podcast by checking out our sponsors: Get started with Fountain Life and become the CEO of your health: https://fountainlife.com/peter/ AI-powered precision diagnosis you NEED for a healthy gut: https://www.viome.com/peter Reverse the age of your skin with Oneskin; 30% here: http://oneskin.co/PETER Get real-time feedback on how diet impacts your health with https://join.levelshealth.com/peter/ _____________ Get my new Longevity Practices 2024 book: https://bit.ly/48Hv1j6 I send weekly emails with the latest insights and trends on today’s and tomorrow’s exponential technologies. Stay ahead of the curve, and sign up now: Tech Blog _____________ Connect With Peter: Twitter Instagram Youtube Moonshots
Transcript
Discussion (0)
Alzheimer's, 40 years of research, nothing to show for that.
Parkinson's, a handful of things to do for those patients.
Dementia, an epidemic, cancer, nothing to show.
You can use the power of quantum physics to understand and model molecules.
Instead of the world of large language models, we've now entered the world, Peter, of large quantitative models, LQMs.
People feel like the world is going rapidly and disrupting and reinventing today with generative AI.
This is just the beginning, Peter.
Everybody, welcome to Moonshots. Today is an extraordinary episode with a dear friend, Jack Hittery. When you're talking about the intersection of AI and quantum,
Jack's the CEO of an incredible company spun out
of Alphabet called Sandbox AQ.
He's a Brooklyn boy, a graduate of Columbia,
where he studied philosophy, physics, and neuroscience.
Did his fellowship at the NIH.
He and I go back 25 years to when he started VISTA research.
Back in 2016, Jack founded the Quantum Group at Alphabet,
working with Sergey Brin and Astro Teller at X.
And in March of 2022, spun out Sandbox AQ,
taking on the role as CEO,
attracting none other than Eric Schmidt as his chairman,
raised a monster round of $500 million in one fell swoop,
and he's on a rocket ship ride.
It's good to see you, Jack.
Peter, good to see you, my friend.
It's exciting times.
It surely is.
You wrote a textbook called
Quantum Computing, An Applied Approach.
I mean, who writes-
Very light reading, very light reading, Peter.
But more importantly, you wrote this book as well, which I love, AI or Die.
It's really a how-to manual for CEOs.
I recommend it for everybody.
And importantly, you're on my board, the XPRI is one of our trustees.
Buddy, I want to talk about quantum.
I want to talk about the intersection of AI and quantum.
I want people to understand why this is so important right now. We've been inundated with large language models and generative
AI and it's changing the world, but it's changing a part of the world and the rest of the world is
about to be transformed and discovered through this intersection of AI and quantum,
sandbox AQ, A for AI, Q for quantum.
Go, let's jump in here, buddy.
A lot of discuss.
Peter, these are very exciting times,
not only for sandbox AQ, for you, for me, for XPRIZE,
but for the human race.
We as humans are dealing with so many challenges.
The challenges that we often talk about We as humans are dealing with so many challenges.
The challenges that we often talk about at XPRIZE,
visioneering gatherings and other gatherings of folks
that are really concerned about life sciences,
the medical challenges that we face as humans,
Alzheimer's, 40 years of research,
nothing to show for that.
Parkinson's, a handful of things to do for those patients,
dementia, an epidemic,
literally across the world as our population gets older. In cancer, some success stories,
in breast cancer, for example, much higher survival rates now, earlier detection,
understanding of the multiple subtypes of breast cancer. Yet another cancers like pancreatic cancer, nothing to show.
Yeah, glioblastoma.
GBM, glioblastoma, nothing to show. Steve Jobs died of pancreatic cancer, billions in the bank
account, nothing to do. Now, years later, still nothing, still nothing. So, massive challenge is
in the world of medicine, massive challenges in the world of energy. We all hope for a transition to a cleaner, more efficient energy posture for our world,
yet we've made halting, halting progress at best towards that.
So a lot of challenge, and the question is, what are the tools at hand, Peter,
that we can use, that we can marshal to address these challenges.
And one of the reasons I started Sandbox AQ is for deep impact at scale.
And you and I have always talked about that. Impact is good, but scale is what's critical.
Impacting the lives of billions of people, the world's biggest problems or the world's
biggest business opportunities. And that's what Sergey and others had challenged me to take up as we got going with Sandbox
AQ. As we look at these two particular tool sets, AI and quantum, initially, Peter, they
might seem quite different. Like, wow, one comes from computer science and inspired by the brain,
neural networks are inspired by biological neural networks, okay, that's one
interesting tool. And then physics on the other hand, and now you particularly talk about quantum
physics, how does that relate? How are these two things related? Well, actually, Peter, there's a
fundamental core nexus, a wormhole, if you would, that brings us together between AI and quantum. And that is, both of them are modeling the world
around us. Both of them are taking huge swaths of data and compressing them down to manageable units
in a way that we can actually leverage them to make a prediction, to have an output that is useful for us in addressing
these kinds of challenges.
So they seem quite literally worlds apart, AI and quantum, but there's a fundamental
core commonality.
So let's start if you will.
Let's jump in with the large language models and generative AI and what does that mean? How do you think about the limitations and what it's given us?
Peter, let's dig into that. So before large language models, we actually had architectures
of neural networks, these artificial representations that are inspired by the brain architecture.
Our brains, as we know, have about 86 to 100 billion neurons
and then trillions, possibly hundreds of trillions of connections known as synapses in the brain
or connections or weights or parameters in an artificial neural network. So, loosely
inspired by the brain, certainly not an exact depiction of what happens in the brain, very
loose, but nevertheless, we call it an
artificial neural network. And prior to large language models, we had architectures such
as RNNs, recurrent neural networks. And these had the ability to actually make a pretty
good prediction. If you said the dog ate the blank, it actually would make a good guess
that the dog probably ate the bone or the homework, but probably not
the dog ate the house, right? And so it was pretty good. The main drawback of our advance is they're
super slow, just not fit for purpose. You cannot be doing what people do today, which is doing
interviews on Zoom and in real time asking the LLM to help
it do an interview as you see people happening right now.
But certainly they showed that it was possible to have this kind of prediction.
And along came a paper by colleagues at Google.
Eight of them wrote a paper in 2017 called Attention Is All You Need.
And what they realized in this paper, what they demonstrated in this paper
is that these new GPU architectures,
we could take advantage of the parallelizability,
took me 10 minutes to practice that word,
parallelizability of the GPU architectures
in order to actually get a lot of throughput
to actually make this both the training and the inference,
both the training on large inference, both the training
on large corpuses of words and then the real time use of those models. That's what we call
inference. Both of those could be sped up massively. And so that's in fact what happened.
These GPUs, the G and GPU of course, Peter, for graphics, not meant initially for language
models or anything like that, initially meant
to give us beautiful graphics in doom and you know, all these kinds of things.
And Nvidia, people may not realize, is actually a 30 year old company.
This is a company that's, it's a 30 year overnight success, let's put it that way.
And so it's an exciting moment because in 2017, the marriage of these new architectures
known as transformers, another way of basically putting these artificial neural networks together
combined with the power of these GPUs really led to this revolution that we have with OpenAI
and Anthropic and Google Gemini and Meta Llama.1 3.2 and so and so forth. All this
came from some initial work done over many decades. Of course, AI is not new. We can
go back many, many years. I like to go back to 1943, a paper by McCullough and Pitts,
a neuroanatomist and a mathematician. Quite a strange bunch. We don't have time for that today,
but maybe another episode of this podcast,
we'll talk about McCullough and Pitts.
But they realized that when you open up the brain,
you don't see a CPU, you don't see a memory,
you don't see the kind of architecture
that you think of a von Neumann kind of computer
after Johnnie von Neumann,
you see something very different.
And that's in fact, was the beginning of these kinds of neural network architectures. But
fast forward to we now have these things, languages are now being, language models are
being trained. But what is really happening under the hood, Peter? And how does this,
you know, Jack, you tell us that it's connected somehow and has some similarity to what's
happening in physics. Well, let's look some similarity to what's happening in physics.
Well, let's look at it. What's happening in a large language model, Peter, is that you take a huge corpus of words,
billions of words, the words in Wikipedia and the paragraphs in Reddit and the posts on social media, some true, some false.
So it's garbage in, garbage out. But with all the garbage, you bring it all in and you present it to these language models and you train them.
And what you hope that is happening is what we call generalization or learning. You hope that it hasn't just memorized that entire corpus.
If it just memorizes it, well, no learning has actually happened. Similar with a toddler.
If we have a toddler and we take them around
our neighborhood and we say,
hey, let's go into this car, let's go into this bus,
let's do this, let's do that.
And later a different car comes by,
you want that toddler to say, that's a car.
It's a different kind of car, it's red, it's not blue,
it's larger, not smaller.
It's different than the initial car that the kid went in,
but the kid knows that it's a car.
How does a kid know that?
That's what learning is about.
That's what Eric Kandel won a Nobel Prize for and many others did
to understand what learning, how learning works in the brain.
And now we can replicate that in these neural networks.
And so one way to think about neural networks is a compression algo.
What it's doing is it's taking a huge body of stuff
often represented in our world,
language in this case could be images,
could be movies, even videos,
and it's compressing them down to their essence,
the essence of car-ness and the essence of bus-ness.
And it's saying, okay, I've seen enough of these
and I understand what some of the core elements are.
So if you put a prompt in saying,
show me a car driving down the street that's red but upside down and it's singing a melody from
Taylor Swift. There are now models that actually will do that now because we've taken not only text
but also video and images and train these models and they've extracted
some of the essence of what's happening. And so that's a form of compression. It's a shorthand
that's now embedded in the weights of that model, embedded in the parameters of that model.
And some of these language models, as you know, Peter, can get up to four or 500 billion
parameters, a trillion parameters. And it looks like now we'll hit two trillion,
maybe even two and a half trillion parameters
in some of the newer models that are coming out right now.
There are limitations though, right?
There are limitations of what they can do.
Yeah, so that's the good news.
We found a way to compress all of say,
various languages down into this model.
The problem is there is no equation
for the English language. There's no equation for the for Mandarin or for any language.
And so the ability of the language model, it's really limited to mixing and matching
what it found on the on the internet. And so, yes, it can make a new essay, but that essay essentially is regurgitating
bits and pieces of what came before. That's really what's happening. It's a probabilistic
engine that is spitting out stuff that it hopes and we hope makes some coherent sense.
And with enough training and human feedback, humans in the loop feedback, then we can actually
make that happen more often than not. There's still a lot of hallucinations, of course.
So we're really limited in terms of we can't really go beyond what is really in that corpus.
So that's language models. And the question is, what else could we do with this kind of
architectures, the architectures of neural networks and in general with artificial intelligence?
Well, let's think about the world beyond words.
And so, okay, language models, a lot of words, great, easy training set, low hanging fruit.
That's really why it started with that.
But it turns out the majority of our world, Peter, is not words, but numbers.
The majority of our world, if you think about a medicine,
a drug, it's described by numbers, by certain configurations of carbons and hydrogen, and
maybe we throw in a nitrogen or throw in some sulfur and things like that, and we make different
medicines. We think about biology, that's numbers. We think about physics, we think
about battery chemistry to store energy. those are numbers, right? And
there's no amount of training and words that's going to help you design that next battery because
you need to know about the laws of chemistry and physics. And we need to have the exact nature of
those laws, not a guess, not some paragraph in a textbook, but the actual mathematics of this
ballgame. And so there are, and I think this was the insight, right, that brought about the creation
of Sandbox AQ.
There are laws of physics.
They go back a hundred years.
There are some fundamental laws of quantum physics, and we all know Newtonian physics,
F equal MA, you know, the-
Ball of inertia.
What we learned to describe the velocity of a cannon ball shooting out of a cannon.
But there are a series of laws of physics known as the quantum equations.
I just wrote them down because I want to discuss them a little bit.
Schrodinger's equation, Heisenberg's uncertainty equation, Planck's equation,
Born's probabilistic interpretation equations. Was the ability of computation to model these
equations accurately what brought about the birth of sandbox AQ. It was the realization that we could have the compute, we could bring the future forward,
Peter.
That's really what you and I have been doing our whole lives.
What we love doing.
Bringing the future forward.
And this was a realization that we could bring the future forward to compute these laws at
scale with impact, with deep impact at a scale that would impact billions of people.
Everybody, I want to take a short break from our episode to talk about a company that's very important to me
and could actually save your life or the life of someone that you love.
The company is called Fountain Life and it's a company I started years ago with Tony Robbins
and a group of very talented physicians. You know, most of us don't actually know what's going on inside our
body. We're all optimists. Until that day when you have a pain in your side, you go
to the physician in the emergency room and they say, listen, I'm sorry to tell
you this but you have this stage three or four going on. And you know, it didn't
start that morning. It probably was a problem that's been going on and you know it didn't start that morning. It probably was a problem
that's been going on for some time but because we never look we don't find out.
So what we built at Fountain Life was the world's most advanced diagnostic
centers. We have four across the US today and we're building 20 around the world.
These centers give you a full-body MRI, a brain, a brain vasculature,
an AI enabled coronary CT looking for soft plaque, a DEXA scan, a grail blood cancer
test, a full executive blood workup. It's the most advanced workup you'll ever receive.
150 gigabytes of data that then go to our AIs and our physicians to find any disease at the very beginning
when it's solvable. You're gonna find out eventually.
You might as well find out when you can take action. Fountain Life also has an entire side of therapeutics.
We look around the world for the most advanced therapeutics that can add 10, 20 healthy years to your life and we provide them to you
at our centers. So if this is of interest to you,
please go and check it out.
Go to fountainlife.com backslash Peter.
When Tony and I wrote our New York Times bestseller,
Life Force, we had 30,000 people reached out to us
for Fountain Life memberships.
If you go to fountainlife.com backslash Peter,
we'll put you to the top of the list.
Really, it's something that is for me,
one of the most important things I offer my entire family,
the CEOs of my companies, my friends.
It's a chance to really add decades onto our healthy lifespans.
Go to fountainlife.com backslash Peter.
It's one of the most important things I can offer to you as one of my listeners. Alright, let's go back to our
episode. Before we start in discussions of the quantum equations that affect our
world on atomic and molecular level, there's a set of Newtonian equations
that drive everything from a car going down the street to a kid playing
ice hockey to rockets flying into space.
Those Newtonian equations are very deterministic.
They're very clear.
You can write them out and you can send a rocket to the moon back in 1969 with the computational
power that is found on, I don't know, I can't find a computer
smaller than-
The basic watch, yeah.
... Describe that.
So those have governed and limited what we've been able to model thus far, right?
We can model large chunks of atoms moving, but not model on a subatomic molecular basis.
So speak to me about that.
Yeah.
So this is a fundamental point.
As you mentioned, back in the 60s, we had the ability to calculate where that rocket
would go.
And literally computers were doing that.
In those days, what we call computers are human beings who are computing.
That was the initial computer was a human being
and their job was called a computer like a lawyer
does law, call them a lawyer.
These were computers doing that.
There were some also actual computing machines
helping out on the side.
And the reason why they could do that
with such small amounts of memory and compute
is because equations really are great compression vehicles.
If I want to say, hey, that you mentioned a cannonball, Peter, I have that cannonball
shooting out of the cannon. It's going to take this parabolic type of pathway as it
shoots out, hits its apex and comes down again. And so I could, on the one hand, take lots
of notes about, oh, here it is at time 0,
time 1, time 2, time 3.
Take lots of notes where it is.
Or instead of all that data, I could just summarize it very, very succinctly in an equation
where I have the starting parameters and then I could predict anywhere along the parabola,
I could tell you immediately where it is and where it's going to be.
So that equation with some of those starting conditions plugged in and the variables gives
me a very succinct.
I've compressed a huge amount of data into a small number of bits of information.
Back to Shannon, back to Claude Shannon, what he taught us in 1948 in his landmark journal article about Shannon entropy, using the word
entropy in a novel way, not in exactly the traditional way we used it in physics, but
in a way that said, what is the surprise factor we have in looking at this body of information?
If you have random numbers, you know, kind of in a grayscale image, there's no compression
possible. It's random,, there's no compression possible. It's
random. Therefore, there's no pattern. But when you have a parabolic pathway of a cannonball,
oh, Isaac Newton says, I can tell you all about that information in a very, very succinct
form. And we do that all the time in the Newtonian world, again, to send rockets up, to understand the
dynamics of cars, even hydrodynamics and fluid dynamics with airplanes and testing in wind
tunnels or even virtual wind tunnels, how the airflow will happen over that curved shape
of the wing, very complex dynamics, but that's still in what we call the classical world,
the pre-quantum world.
And so now, Peter, you bring up all these interesting equations and now we're able to say,
hey, we did this on the macro scale, on the classical scale, but how about those electrons?
How about those photons? How about those molecules? At these scales, we've got to use
different equations.
And that's what the quantum grates gave us.
Heisenberg and Schrodinger and even Einstein 1905 paper,
the one he got the Nobel Prize for Peter was not relativity.
It was the photoelectric effect in 1905, part of his honest mirabalas,
his miracle year that he had one of those papers. And it was inspired.
People may not realize, by the way, a little side fun science note here a ballast, his miracle year that he had one of those papers. And it was inspired. People
may not realize, by the way, a little side fun science note here in the Moonshot podcast.
Why did Einstein, here's a fun question for everyone out there, science nerds and geeks
alike. Why did Einstein write these four plus ones, actually five papers in one year on
seemingly a disparate set of topics.
I'll leave it to the end.
We'll come back at the end and we'll find out why that is the case.
But he did write the photoelectric paper and that led to his Nobel Prize, but it was building
on Max Planck 1901.
1900 gives the talk in the Prussian Academy of Sciences,01, publishes the paper and really forever changes the world
because what Max Planck realizes is that to resolve
some of the key crises happening in physics at the time,
I know people feel the tension that we feel right now,
the crises of the late 1800s in physics,
the ultraviolet catastrophe,
there's so many, there's actually four or five
different crises happening at the same time in physics.
It turns out all of those could be explained
by the fact that we were still wed
to a Newtonian way of thinking.
And when it came to the subatomic world,
we needed to actually abandon that
and move into a new regime and have a different
view on how the world works. And Max Planck would kick that off with a sense of how black body
radiation working was again, one of the key crises at the time. Einstein, a young Einstein in his
twenties read that paper and then wrote his paper in 1905 with homage, explicit
homage to Max Planck, his senior, and saying that Max Planck explained that for black body
radiation, I, Albert Einstein, will explain it for a photon, a packet of light.
And this ushered in along then with Schrödinger and Dirac and Heisenberg, and of course, Niels Bohr.
All of these greats who each won Nobel prizes
helped us understand the dynamics of how things work
fundamental to our universe.
It's not people often say,
oh use quantum only describes things
at the smallest of scale.
Well, yes and no.
It's describing things at every scale. It's just that we don't have
to go through the trouble of using the quantum equations when you have something the size
of a rocket ship. It does describe the rocket ship. In fact, because that's what's happening
is the rocket ship is made up of all these little atoms and electrons. So let's move
away if we could. Maybe one thing we could also do in this podcast is we'll help society
to move away from the
phraseology of, oh, quantum only describes things at the small scale.
Actually it describes everything.
Everything is quantum, but they're particularly useful when we're thinking about the small
scale.
And we can generalize with Newton's laws every time.
That's right.
That's right.
We can approximate using Newton's laws.
Exactly.
Exactly. Correct. So now coming back to your fundamental question. So we talked about how neural networks compress
a lot of stuff in our world down to a much smaller format so we can manage it, manipulate
it and make some output of the generative AI, for example, in language or in images
or things like that. But now let's turn to physics.
Physics does the same thing.
As we talked about Newton laws can do that, say for a
parabola or rocket ship going to orbit or going around the
moon, but now we could also compress something even more
fundamental.
We can say, what is the behavior of that electron?
And let's talk now about valence electrons.
I know it's bringing back nightmares for listeners back into into high school chemistry but but the
valence electrons are the ones on the outer edge and those are the ones we
really are concerned about when we want to make a new drug. Peter as you well
know as a doctor when we want to make a new drug and we want to say what
molecule would fit into that target in the
body. To give an example to our listeners today, if God forbid someone has melanoma, someone has
non-small cell lung cancer, bladder cancer, a variety of cancers, we now have a new class of
cancer drugs that go take us beyond the horrific regime of chemotherapy and radiation. They take us into immunotherapy,
the ability to use our own immune system
to fight these cancers.
And the key to that happening is a molecule,
actually a synthetic antibody,
but not an antibody created by our own internal
adaptive immune system,
but one that we synthesize in the world. And this antibody
doesn't have the function that normally antibodies have of helping us directly to fight a particular
disease or pathogen. What this does is it locks in to a particular receptor known as the PD-1 receptor on a T cell, on an immune cell, it locks into the T
cell's PD-1 receptor and protects that receptor from being hit by a tumor, by a ligand, by a
molecule coming out of the tumor that normally shuts down T cells and puts them to sleep like
hypnosis around the cancer. And this protects it,
like the Romans had their nice shields.
Imagine now the T cells armed with this nice shield
and goes into battle.
But to develop that molecule that would fit lock and key
into that T cells PD-1 receptor,
we have to have ultimately an understanding
of how those electrons at the outer edge of that antibody
and the electrons in the outer edge of the PD-1 receptor, how they interact.
And now for the first time, just in the last two, three years, we now have that ability.
Let me have you pause here one second because it's important for people to realize, I mean, we do all of our work in the world of bits, but we are physically individuals living in the world of
atoms. And when you want to start looking at the functionality on a cellular surface or in a chemical
reaction or in a new battery chemistry, those are all atoms. We've been
able to model them in classic computation thus far, but it's approximations and it's
massively computer heavy. Are you talking about being able to get to a deeper level of fundamental modeling than we've ever been able to do with
our computers today because we have had deep mind with Alpha Fold and Alpha Fold 3 and
most lately it was Alpha Prodeo being able to help us predict new proteins.
But that's using classical computer models versus the work that you're doing today.
Can you differentiate those two?
Sure, that's correct, Peter.
So when we look at what's been done before in trying to model biological systems as an
example, a lot of good work has been done, but unfortunately it did not involve the physics
itself.
And so people would look at libraries of proteins and what's fundamental to proteins as we know
is their conformation, the way that they're folded
like an origami.
And when you have a string of amino acids,
the building blocks, the Lego blocks of proteins
and you string them together,
they're gonna fold in a certain way.
And that folding is fundamental to the use,
to the application of that protein. In fact, when we
have a misfolded protein, that's a whole nother ball game and that leads to diseases of all
kinds. Stan Prusner, who won the Nobel prize and is at UCSF, an incredible body of work,
showed how misfolded proteins can lead to complete disaster in the brain and other organs as well.
So when we think about the prediction of folding based on a string of amino acids, one can
just look at lots of examples and based on those examples, train a neural net on what
would happen.
And that's in fact how a number of the methods that we use out there do that, like alpha
fold and others.
But now we have the ability, Peter, to go beyond that. Because while alpha fold does a very,
very good job, it doesn't actually get down to the very specific ways in which it will,
if that protein will interact. Looking at the structure of the protein, the folding,
is only the first step. We must now get to the dynamics of the protein. How will it act on other things? How will
other things act on it? And again, we come back to electrons and electrons are described
by quantum mechanics. And so if we want to understand how an electron on one molecule
would interact with the electron on the receptor, on the target. We've got to get down to that level.
And that's the level now, finally, that we at SandboxAQ have been able to model things
at.
And that is a big breakthrough.
That means that we can have-
And that's-
Yeah.
I mean, that is amazing because that applies across all material science, all biology.
What was that moment in time?
So I mean, when you joined Alphabet to head this division, did you have this in mind already?
Or was this sort of something that unfolded as the computational power came online?
I mean, help me understand that moment of creation.
I'm just super curious.
My first area of focus was actually in AI.
As you know, I was applying AI decades ago now
to brain imaging as you opened up with
and specifically to fMRIs, to dynamic brain imaging.
Dynamic brain imaging, as readers may,
as listeners may know, is not the same
as just a static CT image.
Just to put it on a light box, let's take a look at it.
You're talking about gigabytes and gigabytes and gigabytes of data
taken over a period of time.
Looking at blood flow of your right.
Looking at blood flow, for example, where I'll give somebody
a test and ask them to move their finger.
If they're moving this left finger here, index finger,
this is the exact spot is roughly about right here in my brain
right now, moving this finger this minute right now.
I can see your homunculus right now.
Yeah, right there.
It's right there as we're speaking and listening to each other.
As we all know, we're using Broca's area here, number 44 to speak.
And then I listened to you, Peter, I'm losing Wernicke's area back here.
And so, you know, when we look at these MRI images,
the human eye can only see so much.
And so my team and I began to train neural networks,
primitive ones at the time, but neural networks nonetheless,
to see if we could glean more information
from these medical images.
And sure enough, we were able to do that.
And that shows the robustness of this idea of a brain-inspired neural network that even
with a very, very primitive compute we had at the time building computers ourselves literally
by hand, we could actually make that work.
Fast forward to today, one thing I realized is with AI, yes, language would be important, but the quantitative
world, Peter, would actually be as if not more important.
The majority of our world is quantitative in nature.
The majority of our world is governed by numbers.
And so rather than spend a lot of time developing large language models, we became very interested
in the quantitative models.
And that took a number of forms.
But then realizing of course from the background of physics that we needed to figure out a
way to take these quantum equations that you were just discussing, Schrodinger's equation,
Heisenberg's formulation, all these interesting equations, we needed to do that at scale.
And the conventional wisdom at the time, Peter, was that we would need a quantum computer
to do that.
We'd have to wait two or three decades to get a quantum computer, and specifically a
quantum computer that was fault tolerant, that was error corrected.
Right now-
Low error rate, yes.
Exactly.
You and I are speaking from computers right now that are error corrected.
There are literally mistakes that pop up in computers. How does that happen? Well, actually, Muons,
cosmic rays, can actually hit your computer and cause a bit to flip. And so we have various error
correction schemes in our phones, in our laptops, in our computers, in our watches that allow for
error correction for transistors for bits, zero
and one type bits. It's not that hard because you can take a vote if you want to of multiple
bits and over represent the bit you want with many bits. Bits are so cheap to make, why
not have lots and lots of extra transistors? But in the world of quantum computers, it's
not that simple. These are very, very sensitive devices,
which is one reason why, by the way, you can flip the script and make instead of a quantum computer, you can turn it to a quantum sensor. Maybe that will leave that till another time in
this podcast to talk about. But basically, quantum computers are very sensitive to perturbation from
the outside world. And so they do need this error correction.
That is hard.
That error correction is hard, Peter.
We knew at the time that it would take years and years before we'd have an error corrected
quantum computer.
Let me pause you here one second because there's a really important distinction here for everybody
to understand what Sandbox AQ is doing because you're a
software company, you're not building quantum computers and there are a multitude, there
are dozens of companies building quantum computers and we could talk about what the horizon for
those are.
Yes.
But the important point to make here is that you can use the power of quantum physics of
the equations to understand and model molecules and such without having
quantum computers. That's correct. The computational power.
Yeah. And by the way, Peter, when quantum
computers one day do get scaled and do get, and we encourage and we have relationships with
more than a dozen of the quantum computing hardware companies out there,
it'll add more fuel to our fire. Yeah.
It'll give us even- And of course, and Google is one of the leaders.
I was just at Hartmut's lab seeing their-
Beautiful, beautiful work. Beautiful golden chandeliers.
Beautiful work, yes.
And they just announced some great progress
on reducing the error rates in quantum computers,
and that's all fantastic.
But I think the point here is that
the same computational power in those GPUs
that gave us the large language
models, you've been able to build algorithms that you can use on those GPUs to approximate
or to solve these classical equations of Schrodinger and Niels Bohr and Planck and Heisenberg to
help you model the actual world of atoms and electrons and ions
today without quantum computers.
That's correct.
And that is going to give us incredible insights into the physical world across health, materials,
environment, everything.
Peter, is giving us, is giving us right now.
This is happening in real time.
And that's what's so exciting.
And when you look at, again, coming back to the fundamental idea of information, right?
Of what does it mean to take a part of the world and represent it in an equation, in
a dynamics, in a modeling, in a simulation.
You're talking about, again, it's very similar to what we did with language.
We talked about large language models.
We took a corpus of billions of words and we compressed it down and embedded that information
into the weights, into the weights of a neural network.
Again, we didn't memorize those words.
That's not what we did there. We embedded information into a space,
into an information space that encodes all that stuff.
We're doing something similar here.
We're taking the dynamics of a certain molecule
and we're describing it in a much shorter way
using these equations given to us over a hundred years ago
by the quantum grates.
And that allows us then to make predictions, very precise predictions about, okay, Peter,
let's say you work at a large pharma company and you say, Jack, I heard you have this wonderful
platform. I'll give you a molecule that we're thinking about that might hit this particular
target in the body. Maybe it hits glioblastoma,
as an example, this brain cancer, horrific disease. And we'll take that, we'll make a
digital twin of that, Peter, and we'll make 10 million, 100 million, maybe a billion permutations
on that drug. We'll add a methyl group that is adding a carbon and a few hydrogens. We'll
add a nitrogen. We'll add an amine group,
we'll add this, we'll take away that. Each one will be a slight variation on the theme
that you initially started with based on your research.
These are AI simulations in quantum.
That's right. So first there, we take the quantum equations, we run those,
and that becomes the data set. So we're generating our data set,
and that's what we use to train the AI.
Okay, let's pause right there.
That's a fundamental point.
That's really important here.
That is a fundamental point.
Yeah.
If you were trying to discover these molecules
that are useful in cancer, Alzheimer's and so forth,
trying to get those with a large language model,
the data doesn't exist in the corpus get those with a large language model, the data
doesn't exist in the corpus of data that the large language model is controlling.
That's right.
It's outside the data set.
Outside the data set.
It's impossible for them to discover it if they didn't have the data in the first place.
That's correct.
And so you've got to generate the data that these quantum models can then assimilate and generalize.
So let's double down on that for me.
Yeah, let's double click on that.
So basically what's happening here is instead of the world of
large language models, we've now entered the world, Peter,
of large quantitative models, LQMs.
LQMs are about starting with equations to generate data.
That is one that's the most efficient way to generate data.
And the most accurate way to generate data is with the equations themselves.
The equations are the bedrock of the universe.
They are the fundamentals.
They're upon which everything is built and created.
That's correct.
The fact that humans, by the way, just taking a step back and the fact that human beings have uncovered
the quantum equations of the universe is stupendous. Deserves a moment of silence. Okay, moment
observed. And so, this is absolutely incredible. And by the way, as many of our viewers may
know, the quantum equations are not one of, they're the most
tested set of equations that we've ever had in the corpus of science.
And they were so, and the people wanted to doubt them so much.
They were so much-
Even Einstein.
Einstein hated them.
He hated them.
There was the, God doesn't play dice.
He hated them. God doesn't play dice. He hated them. He railed against them until his death in the 1955 because although he was one of the
creators of it, he couldn't comprehend given a classical view that he held onto how this
world could even be described by these equations.
Ironically, one of his best known papers, 1935, known as E EPR Einstein, Podolsky, Rosen, the three authors
of it was an attempt to derail quantum mechanics as a science. It ended up becoming a cornerstone
of the science describing the phenomenon of entanglement.
But coming back to your key question, Peter, which is what's happening now in the world
of quantitative AI, rather than having to use a
corpus of data on the outside world, which contains a huge amount of garbage and false info and good
info all mixed together, like in the language world, we start with the pristine equations
themselves. We take a theme and make variations on that theme. We take a molecule, make variations
on it. We take a battery chemistry, which we're dealing with ions now. And ions, again,
are subject to the quantum laws. And we're saying, okay, this is a lithium ion battery.
But we've been stuck with lithium ion chemistry, Peter, for 40 plus years. And we actually need
to start moving beyond that, need to think about what other chemistries would give us batteries that may be cheaper, maybe more lightweight.
The heaviest thing in a car, in an electric vehicle is the battery.
But batteries actually are going way beyond just electric vehicles.
The bigger market, the bigger application for batteries is not in cars.
It's stationary.
It's in every building in the world needs to ultimately have an energy storage system that accepts electrons when they're cheaply available and then
uses them when they're in demand.
And that arbitrage that day trader like arbitrage of buy when low and use when
high, right?
That is going to impact the world of energy beyond anything we've ever seen.
The one I'm waiting for
is room temperature superconducting.
That's what I want.
Quantum to deliver us.
Yes, well, that's the kind of modeling
we now are beginning to embark on.
So when we think about the drugs that we need,
the medicines that we need, we think about diagnostics,
biomarkers that we want to have.
Right now, as our audience may know, there
is no marker in the world for the progression of Parkinson's is a marker that tells you
whether you have Parkinson's or not, not very helpful since it's probably very obvious.
But in terms of whether it's progressing or you've halted it with some treatment, there
is no so we need new biomarkers, we need treatments, we need better battery chemistry, we need new biomarkers. We need treatments. We need better battery chemistry. We need cheaper
solar energy as cheap as it's become. The fact now that the underlying substrate of solar silicon
is competing with the semiconductor industry, it does not bode well for the inexpensive nature
of solar. In fact- But perovskite's coming.
Perovskite's are exciting, but what's the problem? Perovskites, Peter, they're not stable.
To be bankable, to be financeable,
solar panels need to have a 25-year guarantee,
a 25-year life, shelf stable, roof stable.
In fact, perovskite technology only takes us out
about a year in terms of stability.
And so there's a need to actually model that
at the quantum level as well.
There is a company, Paranova, I'll tell you about it sometime soon, that's doing a heck
of a lot better than that.
Good, we hope so.
We want that to happen.
We want that future.
Everybody, I want to take a short break from our episode to talk about a company that's
very important to me and could actually save your life or the life of someone that you
love.
The company is called Fountain Life.
And it's a company I started years ago with Tony Robbins and a group of very talented physicians.
Most of us don't actually know what's going on inside our body.
We're all optimists.
Until that day when you have a pain in your side, you go to the physician in the emergency
room and they say, listen, I'm sorry to tell you this, but you have this stage three or
four going on. It didn't start that morning.
It probably was a problem that's been going on for some time.
But because we never look, we don't find out.
So what we built at Fountain Life was the world's most advanced diagnostic centers.
We have four across the US today and we're building 20 around the world.
These centers give you a full body MRI, a brain, a brain vasculature,
an AI enabled coronary CT looking for soft plaque, a DEXA scan,
a grail blood cancer test, a full executive blood workup.
It's the most advanced workup you'll ever receive.
150 gigabytes of data that then go to our AIs and our physicians to find any
disease at the very beginning when it's solvable. You're going to find out eventually. You might as
well find out when you can take action. Fountain Life also has an entire side of therapeutics.
We look around the world for the most advanced therapeutics that can add 10, 20 healthy years
to your life and we provide them to you at our centers.
So if this is of interest to you, please go and check it out.
Go to fountainlife.com backslash Peter.
When Tony and I wrote our New York Times bestseller Life Force,
we had 30,000 people reached out to us for Fountain Life memberships.
If you go to fountainlife.com backslash Peter, we'll put you to the top of the list. Really, it's something that is for me one of the most important things I offer my entire family,
the CEOs of my companies, my friends.
It's a chance to really add decades onto our healthy lifespans.
Go to fountainlife.com backslash Peter.
It's one of the most important things
I can offer to you as one of my listeners.
All right, let's go back to our episode.
So back to the core question that you have,
the fundamental breakthrough now of realizing
that what's happening in the world of physics,
in this case, quantum mechanics,
is that we're summarizing essentially a massive dynamics in the world of physics, in this case, quantum mechanics, is that we're summarizing,
essentially, a massive dynamics in the universe with these core equations. We're taking a
molecule that has infinite degrees of freedom. It can move in any way, form, or manner. There's
anything that can happen to it. And then we're focusing like a laser on the business end
of that molecule. We're not going to the business end of that molecule. We're not gonna model every electron in that molecule.
We're gonna limit ourselves to the valence electrons,
that is the outer electrons.
And within the valence electrons,
we're gonna limit ourselves to the business end
of the molecule that might be hitting the actual target
that we're going for in the body.
By constraining ourselves down to that portion of the problem,
we can make it tractable in today's GPU-based computers.
In fact- When did this become possible, Jack?
When did it become possible for you to do this
with the compute and the algorithms?
Because it hasn't been that long.
We had the first breakthrough exactly three years ago.
Just three years ago.
Yeah, so just three years ago is when we realized this is going to change the world. This is
something that's going to fundamentally change how we do things. And again, while most of
the world was focused on language, and again, God bless the applications for college students
writing essays an hour before the deadline for language models. But we realized that this was going to be a fundamental change in one, how we thought
about AI and how we thought about the use of quantum in the real world.
We've always had quantum in textbooks.
We have many quantum innovations.
The MRI machine is a quantum in nature.
The laser is quantum in nature.
Lots of quantum. We've used them, but we haven't been able
to really predict and utilize them.
At scale.
At scale, yeah.
At scale on any arbitrary quantum system.
And that's now where we've come to.
And that means that the world now has a superpower now,
a new superpower.
Humans have a superpower.
This generation of humans is the first to have the superpower to do this at scale on real world types of
systems.
Every quantum textbook in the world usually has a chapter one, two, and three describing
the equations you just rattled off, Peter, and then has a chapter five or six that says,
let's use it on an actual case. And what is the case? One hydrogen
atom, one proton, one electron. Now, as far as I know, hard to cure cancer with just an
atom of hydrogen. And so we got to get to real world system. And that's what happened.
In the last number of years, our team and I, we've worked on real molecules from labs and UCSF Nobel Prize
winning labs. We've worked on molecules from large pharma companies, spinouts, all kinds of folks
working on battery chemistry with a company called Navonix, a public company that does
battery chemistry, working on new materials for the US Army. The US Army wants to lightweight
the tanks. Car companies that may announce soon want to lightweight their vehicles so that they're
more fuel efficient.
Well that's new material science.
We need new materials to make that happen.
Materials are made of atoms and those atoms have those electrons that we talked about.
And so we've got to fundamentally rethink now-
By the way, I think how it materials.
Material sciences is like the most under appreciated area of technology, right? Everything that
is new and breakthrough, I bow down to material scientists and the work that they do. And
this notion of the materials genome, right? The idea that we understand the fundamentals
of certain limited number, like, limited number, a fraction of a
fraction of 1% of materials that are possible and we use them.
But given the work that you're doing, we're able to expand this understanding that will
head towards fundamental across every industry is going to be transformed by this.
Yeah. And Peter, you get into a fundamental point here, which is the compute we're talking about now,
both AI, quantum, this quantitative AI we're talking about, it's not just about doing things
faster. Often people write, oh, they're doing things faster. And by the way, faster is good.
I mean, yes, let's get to the medicine faster. That's great. But here's the more fundamental point. We're actually
exploring a bigger landscape. We're able in the case of medicine to explore a bigger biochemical
space than can otherwise be explored. In the case of material science, explore a bigger
space. If you're looking at batteries, battery chemistry, there's
about 19 elements of the periodic table that you could, in different combinations, build
a battery for the electrolyte, for the anode, cathode, for the membrane, all these core
four elements. And so we actually can now start to explore a much bigger space. If we
were limited before, as we were, to building prototypes by hand and testing each one,
how many could you possibly build, right? Even if you're Willy Wonka and have Oompa Loompas
around, you're limited to the number of batteries you could possibly prototype.
But now that we have the actual quantum equations in the system and you could run it at scale,
you could explore a much bigger
material science landscape. Gone are the local minima and maxima that we've been living with.
Yeah, we were really in a cul-de-sac to use a more suburban term for a minimum or maximum,
but yes, yes, exactly right. So, let's talk about what the implications of this are. Our favorite subject, you and I both love health and longevity and reminding people
the way that we've discovered drugs in the past, we'd go into the Amazon, we'd chop down
bushes and trees and dig up dirt and we'd crush them up and we'd try and find unique
molecules and we'd test them, a molecule at a time.
And that led to today's devastating drug industry,
which is riddled with failures.
What do you hold as the average drug development time
and cost, a decade and $3 billion?
Exactly, right now it's about seven to 10 years
of preclinical work that is developing first a target.
You gotta start with a target in the body. What are you going for? is developing first a target. You got to start with a target in
the body. What are you going for? You got to validate that target. And then you've got
a drug that target. You got to design a drug that fits like a lock and key into that target.
That's about eight to 10 years. Then you go into clinic, even with, and God bless the
FDA has actually done a pretty good job trying to compress down the phase one, phase two,
phase three trials. You often could do a pivotal phase two now where the phase two becomes the, in a sense,
like the phase three. Sufficient data to give you an approval by the FDA.
To get out there. There's breakthrough pathways for drugs now, particularly for orphan drugs,
particularly for orphan diseases and rare diseases.
And so when we think about the time it takes, eight to 10 years preclinical, four or five
years minimum in clinic, and here's the kicker, here's the most sobering of the statistics,
90% failure today in clinic, 90 out of 100 drugs that go into clinical phase one trial,
phase two, phase three, never
see the light of day, never come out again.
You know what's equally sobering for me?
When you get prescribed a drug because you have a particular problem, chronic disease,
whatever it might be, you expect that that drug works for you.
But do you happen to know what percentage of people that drug were prescribed
for them actually works?
Please.
It's like 20%.
Wow.
The fact of the matter is the FDA is making sure it's not harmful, right?
That's the results we get out of a phase one, phase two side of the equation, but the FDA
is approving a drug if it helps a sufficient number of people, not everybody.
And so I think it's insane.
But the hope now is not a drug that statistically might work for enough people for the FDA to
approve it, but I want a drug that works specifically for me.
I want a drug that is coded for my molecular design and genetics and so forth.
Right. And Peter, this is now the precipice of where we're going because because of that
two and a half, three, three and a half billion dollars per drug program, drug companies,
biotech companies have not had the ability to do more than just one big cannon shot and just hope it works and hope it goes to enough
population to then advertise and pay back the cost of that.
Which is why the drug costs are so extraordinarily expensive.
Because also the few successes have to pay for all the failures.
But now let's talk about a world where it costs one-tenth the amount to make a drug, 300 million instead of 3 billion.
Let's say it costs half the time to make that drug. Let's say we take the rate of success
in clinical trial from 10% to 50, 60%. This is all possible now. And so, if we look at
that world, then a biotech company, a lab, a set of researchers can say,
you know what, we can actually now make a drug that's targeted to this very specific
sliver of the demographic of the population that has this particular genome sequencing
or other characteristics.
And we can start to get to that future that you just described that says, instead of having one size fits all kind of drugs that end up
not fitting many people at all, we
can now start to really understand
how we can make these more targeted drugs that really
match well with key cohorts out there in the population.
We've been doing that in part, right?
So we've got companies like in Silicon Medicine, we've got Alpha Prodeo, but again, they're using classical computers and AI large
modeling. How far are you from using the equations of quantum physics through the power of Sandbox
AQ to help companies with this? Well, Peter, two fundamental parts of that question. We're doing it right now. And fundamental
to our success and our velocity in this is that we are not a biotech company. That's fundamental.
When you look at the landscape and you say, who else is using computation to help with drug design?
Actually, there's quite a few companies using computation. But here's the thing,
Peter, they're all biotech companies. They're all companies themselves who have the burden,
the overhead of trying to bring a drug to market, who have pipelines and clinical expenses and all
that overhead, not just of money, but of time and management attention. We do none of that. We focus just on doing this advanced set of calculations
in partnership with labs and with drug companies.
And that-
So you're a software company
that is supporting a multitude of different industries.
Right, exactly.
And so that allows us to now have the world-class talent
that can do this.
We have quantum physicists on board,
chemists, medicinal chemists, biologists, doctors, physicians advising us. We have all these
specialties all coming together, AI specialists of all kinds, modelers of all kinds, mathematicians
in our midst coming together to make this happen. That is not something that a drug company can really do
because they've really got to hit their bottom line,
which is making the final drug.
We focus on working with dozens and ultimately hundreds
and thousands of drug companies in the same way
that Oracle is a database used by every vertical out there.
Ultimately, our platform will be used
by many, many verticals out there
today. It's being used by the biopharma community. It's being used by the chemicals community,
by Dow Chemical, by others. And then we'll start moving into material science, batteries, energy,
all these different areas as well. So business model really counts. And when you want to focus
and say, Peter, what are you going to be the best at in the world?
What are you going to be the best at in the world?
What is your company going to say, this is our territory?
This is where we're going to really focus?
You have to make a choice.
You can't be all things to all people.
And so we made the choice to be best
at this kind of computation, this software.
And our software runs today on the
GPUs and it's architected so that it could also run on the quantum computers of tomorrow.
And that is the future we're heading to, Peter. We're heading to a meshed hybrid cloud world
that meshes CPU as a basic controller, GPU workhorse, absolute workhorse, with QPU, quantum processing unit.
And bringing those together is a core part of that future.
Real quick, I've been getting the most unusual compliments lately on my skin.
Truth is, I use a lotion every morning and every night religiously called One Skin.
It was developed by four PhD women who determined a 10-amino acid sequence that is a synolytic
that kills senile cells in your skin.
This literally reverses the age of your skin and I think it's one of the most incredible
products I use it all the time.
If you're interested, check out the show notes.
I've asked my team to link to it below.
All right, let's get back to the episode
Jack talked about the future of of clinical trials here because I think you know, I've heard you
Sort of visioner
This notion of running in silico human trials
That drops the cost not by 50 percent, but you basically a thousand-fold eventually, so that you know
when you introduce this drug into humans, because it's run in quantum models, because
it's run in simulation, that you know it's going to work.
Just like the first time the team at SpaceX launched the Dragon Falcon combination to
the space station, they didn't kind of hope
it would actually get there and dock accurately. They'd run the simulation so many times in
such accuracy that they knew it was going to work. So is that the future for drug development?
Yeah, we still have to do the clinical trials. There's no way around that. But as you pointed
out, we can go into the exam with the answers, right? That's very exciting.
And so if we can go in right now, the 90% failure,
what does that tell us, Peter, the 90% failure
of clinical trials today?
It tells us.
And by the way, this is having gone through phase one
and phase two and failing in phase three,
which is insane, right?
So you've had enough of a success in phase one,
in phase two, and you know it didn't hurt anybody.
To show some safety and efficacy, yeah, that's right.
But then at the end of the day,
it didn't help enough people.
Exactly.
So given that 90%, what that tells you is
that we have the ability to add a lot of value here, right?
There's a, you know,
we're not talking about some optimized system.
Can I still invest in the company, buddy?
Exactly.
There's a lot of value to be created here.
If you look at highly optimized systems in our world,
those are different stories.
If you look at a Stirling engine, one
of the most efficient things on this planet,
not a lot of value you can add to a Stirling engine.
But if you look at the clinical trials value you can add to a sterling engine, right? But if you look
at the clinical trials that you're pointing to, this is where we can add a lot of value. By
modeling exactly what's going to happen and adding to that model every year, more and more of the
variables. And so initially looking at those valence electrons and how they're going to hit
there, and then looking at maybe some dynamics. If it's a protein, let's say you're talking about a small molecule,
less than a thousand Daltons with carbon, say, being 12 Daltons.
So small molecules, as you know, Peter,
but to share with our listeners, are things like aspirin
or Tylenol or things like that are small molecules.
But when they interface with a much larger beast,
like a biologic such as a protein or an antibody
or things like this, then there's
a lot of stuff going on. And so we're now adding functionality to the system that allows us to do
that kind of stuff, protein against protein, small molecule with proteins. This is complex stuff and
it's getting more and more capable every few months. And so this is the kind of work that will ultimately lead us to a much better
sense of the answer before we walk into that clinical trial. Let me emphasize, we'll still
need the clinical trial because we do want to have that final real world confirmation,
but we'll have so much information and modeling before that, that will move up that success
rate very, very dramatically.
You know, I need to dive into quantum computing a little bit with you.
Sure.
Because while you're on a rocket ship, when we add quantum computing to your rocket ship,
it becomes a warp drive and you're a starship all of a sudden, not just interplanetary,
you're going to the stars.
So quantum computers have been around for a little bit.
And as you've said, we've got dozens of companies
and their qubits, their equivalent of their bits
are atoms or photons or ions, lots of different approaches.
Can we talk about where they are today in your estimate
and where they would need to get to to be functional for
sandbox AQ to use?
Yeah, great question. And you talk about, they've been around for a little while. Let's
be more specific. Paul Benioff, yes, a cousin of Mark Benioff, Paul Benioff, I want to give
some kudos out to Paul Benioff. He is the one who's had the first paper describing a quantum computer.
That was in 1979. He's passed away recently, just in the last few years. But I want to give
credit out there because often histories of quantum computing gloss over him and talk about
Richard Feynman. Yes, Feynman did popularize the idea of what a quantum computer can do.
Feynman did popularize the idea of what a quantum computer can do, and we owe that to Feynman for helping get the idea out.
But I do want to give some credit to what credit is due with Paul Benioff for having
kicked us off.
And one interesting history of science question is, John Yvonne Neumann, we mentioned it before
in Von Neumann Architectures, John Yvonne Neum Norman being at the nexus of computing and physics.
Why wasn't he the one to conceive of a quantum computer? It's an interesting question. Maybe
another session. We'll come back to that. But we still get to come back to Einstein's five papers.
Yes, we're going to come back to that as well. So we have many. It's good to give answers on
this podcast, but also to plant questions with the audience as well.
So, so, so back to back to quantum computers now. So you ask about the different modalities of quantum computers. Yes, indeed. There's seven major ways to build a quantum computer. And as
you pointed out, we can build them either with natural qubits or synthetic qubits. A natural qubit, an example there would be a neutral atom.
That's one of the dark horses in this race.
One that hasn't got as much attention,
but is scaling very rapidly.
This is where you take a neutral atom, not an ion.
So it's not an ion trap computer, but it's neutral.
It's not, doesn't have any charge.
You manipulate it with lasers. Several
people won Nobel prizes such as Steve Chu and others for showing us how to manipulate various
entities with lasers. We use those techniques.
Near absolute zero.
Exactly. We use those techniques to set the atom, which now becomes the qubit into a certain state.
And again, let's remember that quantum bits or short qubits
can be in the state of zero or one,
just like the transistor could be in zero one,
a bit can be zero one,
but they can also be in super positions
in combinations of zero and one.
And that gives us an infinite palette to draw from. And that
qubit we can say is going to be some part zero, some part one, or a third this, two
thirds that. We can have different combinations.
Everything in between.
And everything in between. And so we represent these qubits in very different ways than just
the normal transistor. And so that neutral atom we can manipulate into one of those states.
We could read that state and
then we can operate on that state.
And that's what's critical to a quantum computer,
the ability to initiate a state on a qubit,
operate a set of operations on those states.
And then read it out at the very end.
And so those are critical things in a quantum computer. And now, in fact,
neutral atom quantum computers are scaling faster than almost every other kind of quantum computer.
Doesn't mean the other ones are out, but advantages to neutral atoms are that they're
basically room temperature, easy to transport, quite compact, and you're starting with a
easy to transport, quite compact. And you're starting with a compartment of gas,
let's say rubidium, as an example,
where you already have hundreds of millions
of these neutral atoms in the actual container.
And to do things that are useful with quantum computers,
we generally know that we're going
to need to do the error correction.
Let's use, for the sake of this conversation, a ratio of a thousand to one,
a thousand physical qubits to one error corrected or logical qubit,
right? A thousand to one ratio.
Really important for folks to recognize because you hear about all these quantum
computers that have a hundred qubits, a hundred and seven qubits.
Exactly, Peter. Right.
They're not the functional error corrected qubits.
That's correct.
Yeah.
That's correct.
So where are we today in this race?
So today we're at a few hundred of these physical qubits.
Really, some papers have claimed to make one logical qubit,
but we're really not at the point of having a set
of logical qubits, of error corrected qubits that we could really manipulate at this time.
Now that will change very rapidly. If you look at the photonic side, using photonics,
that's another promising approach exemplified by PsyQuantum, both in California and Australia, and also Photonic, a company in Canada, as
well as a number of labs working on photonics. The Chinese, by the way, are making good progress
as well in photonics. Pan Jianwei, the leader of the quantum program in China, himself is
a photonics oriented physicist.
And so that was the first quantum computer that he started to build.
My favorite science fiction stories are always about massive quantum computers buried under Beijing
that brought about AI super intelligence.
Yeah, yeah. Well, there are quantum computers deep, deep inside these universities that are run by Pan Jiangwei, but they're not in Beijing. They're about two hour fast train ride from Shanghai in
a different place. But yes, they are there. But in any case, so you have photonics. And
one of the advantages that the photonics people will tell you is that we can mass produce
these qubits using silicon photonics. We can use some of the same techniques
of the semiconductor industry.
In fact, PsyQuantum uses global foundries,
uses one of the big fabs out there to mass produce this
by hopefully the millions.
And so if you want to do something like crack RSA, right?
Let's say you want to crack the encryption protocols
that are used throughout the world
that are the bedrock of our economy.
The reason why we have-
You want to crack my Bitcoin wallet?
Yes, Peter.
That's what we want to do.
We want to crack right in and break the chain.
In fact, you want to do that with either RSA, which has been around since 1978, since R,
S, and A, Ravesh, Sham Shamir and Edelman gave us RSA.
If you want to crack ECC, elliptic curve cryptography.
So blockchain, you mentioned blockchain, Bitcoin,
Ethereum, these are all based on either RSA or ECC.
If you want to crack those,
but more even bigger than just blockchain,
every ATM transaction, every wire transfer, every e-commerce
using a credit card on Amazon, every single transaction, every WhatsApp. When you're on
WhatsApp, it says encrypted end to end on the WhatsApp messages there. What is that encryption?
That is RSA and ECC. So if you want to correct that, estimates are we'll need roughly 5,000 or so logical qubits. Maybe we can get away with 4,200,
but let's just say 5,000 error corrected qubits, which using our thousand to one ratio, Peter,
let's go back then and say 5 million physical qubits. So we're at a few hundred physical qubits
today and we're going to need 5 million of these things. So, Jack, I know you're a betting man and you predict the future, actually you implement
and create the future.
When are we going to get there?
When do you think we're going to have actual quantum computers that you're going to want
to use at Sandbox AQ?
I would say that-
Min-max.
Yeah.
I would say that by year 2029, which is only five years from now, we'll start having
the building blocks of about a thousand to 5,000 physical qubits in these modular Lego blocks.
Then what'll start to happen is people will daisy chain these blocks up using fiber optic connection and modulators that allow the physical
instantiation of those qubits inside the block to be coordinated with a quantum state without
collapse, without observation, without collapse, with the quantum state in the adjacent Lego
block. When you can start to daisy chain them all together
and then thus create a mega computer made of lots of these Lego blocks, let's say for example,
they each had a thousand of these physical cubits and then I got a thousand of these Lego blocks
together. Now I've got a million cubits and therefore I have a thousand logical qubits.
Amazing. Now, of course, 2029 is when Ray's predicting whatever AGI is back then. And of
course, the reality is we're going to be using some variations of digital supercomputing to help
us build these quantum computers. And then those'll, you know, those digital super computer,
digital AIs, super AIs will become resident on these.
It's an exciting five years ahead.
And Peter, just to finish.
So that'd be like, that'll take us about five years.
That'll take another two, three years of the engineering
to put all that together, make it error corrected,
bring it all in.
So let's talk about the year, maybe 2031, 2032. I think it's
going to be a very critical year. Yeah. Insane. And so-
And Peter, I think we should note that while we're talking about quantum computers, there's
whole worlds of quantum technology that take us beyond computing. Quantum sensing is one of those
critical things.
Quantum sensors are here today.
We don't need air correction.
We don't need millions of qubits.
They're here today.
They're flying on planes right now,
helping us to navigate when GPS is jammed,
when GPS is denied by countries and by bad actors.
They're right now being tested in hospitals
to diagnose how your heart is given
the magnetic field of the heart that is magneto cardiography, MCG versus ECG.
These are all the areas that Sandbox AQ is pioneering.
That's right.
Which we're going to need to come back and dive into those.
We'll come back to it in another podcast.
Absolutely, because it's extraordinary. I mean, quantum computers at this level
in the next five, six years change the game.
I mean, people feel like the world is going rapidly
and disrupting and reinventing today with generative AI.
This is just the beginning, Peter, just the beginning.
This is super exponential on steroids.
I don't have enough superlatives to explain how fast it is.
Peter, let's come back if we could in summary to one of the core points I hope we can have
viewers and listeners take away from our conversation today, which is information.
Information not in a generic sense, but in a very specific technical understanding of
that word.
The way that Claude Shannon understood that word, the way we understand it in a field
we call QIS, quantum information sciences.
The way we understand information now in neural networks, where we're taking a large body
of data and we're representing it by a smaller amount of bits.
Marshall S. Lindenberg Learning.
Stan Mallow Right. And that learning, that generalization leads to information that
represents that larger data set that we started with in the first place. The same in physics,
where we can take dynamics in the world, be it Newtonian with a rocket ship,
quantum, electrons, molecules, and we can take very complex behavior and dynamics and summarize it
in a small number of pieces of information called equations, called dynamics dynamics and this fundamental ability of humans to search and look for summarization, conciseness, compactness, compression.
This is fundamental to the breakthroughs that we're now seeing both in the AI world and
we are now seeing in the application of quantum physics for the first time at scale in compute on GPUs
that we've never seen before. This fundamental insight that information is the building block
of our universe. This fundamental insight that information in this sense of Claude
Shannon's entropy of information, and we can then quantize that into the quantum theory of information.
Maybe in another podcast we'll have time to talk about that.
This is fundamental to the human race will completely revolutionize our existence on
this planet and hopefully other planets as well.
A beautifully put.
Before we break away here, I have two questions. The first, simulation
theory, yes or no? Are we living in a simulation, buddy?
Let me say this, that if we are in a simulation, then the beings who created the simulation,
kudos to them. They've done a pretty good job. I'll put it that way.
Okay. Second.
And we should also answer the Einstein question.
That's the second question.
Okay, there you go. So it turns out Einstein, as a young man, read a book by Henri Poincaré,
the mathematician and physicist, that in that book, Juan Carré put out a series of challenges,
challenges about Brownian motion,
challenges about light,
challenges about how the world works.
And it turns out that four or five of those challenges are
the ones that Einstein decided to tackle as a 20-something,
sitting in Bern being a third class
patent clerk at the patent office. The only job he could get due to his friend's dad who got him the
job. This is what he decided to tackle, but he was inspired by this book. And for some reason,
histories of Einstein often lost over why he wrote this on subject matters that
seem to have nothing to do with each other in that year of 1905. So we have, we have
Henri Poincaré. So just like David Hilbert did in the year 1900, Henri Poincaré in
his book, very often contributions to our society can take the form of not just the answers, but the questions. David
Hilbert put out a challenge in the year 1900 of key mathematical problems, some of which still
vex us today. The Clay Mathematics Prizes more recently do the same thing, but updated them
for our mathematics of the last few decades. Henri Poincaré put that challenge out there,
and a young man called Einstein took up that challenge. So I leave the listeners few decades, Henri Poincaré put that challenge out there and a young man called Einstein
took up that challenge. So I leave the listeners with this, Peter, what are the questions we want
to pose to ourselves as challenges to our colleagues around the world, to young people today,
to our kids, to the next generation? Let's focus on the questions and not just the answers.
our kids, to the next generation. Let's focus on the questions and not just the answers.
I love it.
Ladies and gentlemen, none other than Jack Hittery.
Jack, you are an extraordinary entrepreneur.
I am so blessed to call you a friend.
Thank you for all the work that you do.
Thank you, my friend.
Thank you, Peter.
Great to see you.
Love you.
Yeah, great to see you.
Love you too.
And let's do this again real soon.
Take care.