An AI enthusiast, Paul Cottingham, leveraged his machine learning expertise and advanced AI tools to develop a personalized mRNA cancer therapy for his dog, Rosie, achieving significant positive results and highlighting the potential of AI in personalized medicine.
Mind Map
Nhấn để mở rộng
Nhấn để khám phá sơ đồ tư duy tương tác đầy đủ
Welcome to this special emergency
edition of the Core Memory Podcast. This
is Ashley Vance and we are we're going
remote today. We're not we're not in
studio because we've had to track down a
Welcome Paul Cottingham to the show.
Thank you for joining us.
>> Hi Ash,
>> how you doing?
>> Um I'm uh I'm I'm as mentioned pretty tired.
tired.
>> It's been a it's been a hectic couple of
days. Yeah.
>> Yeah. Well, so if people haven't been
paying attention, you're tired because
um I guess it was it was maybe US time
last Friday, I think. Um, or it could
have been Saturday, a story came out in
Australia about you using artificial
intelligence technology to to try to
cure your dog's cancer and this story
caught the world's attention. And I I
suspect you've been in demand for the
last couple days.
>> Yeah. Yeah. Um,
uh, yeah, I've been averaging about two
to three hours sleep for the last three days.
days.
>> Yeah. Yeah. Well, okay. So, many people
will have will have seen, I think, you
know, some of the news stories and the
coverage, but for those people who
haven't, I apologize for making you
repeat yourself, but do you mind do you
mind giving us the basics and then we'll
we'll dive a bit deeper into the story?
>> Um, sure. Sure. Ash, the cliffotes
version is uh my dog Rosie was uh
misdiagnosed with cancer for about 11
months. So her my her original vet misdi
misdiagnosed the cancer. He said it was
just a rash for 11 months and then um
around the 11 month mark I said look it
was like bleeding through the skin. I'm
like that can't be good. Let's remove
it. Removed it came back uh off to the
biopsy and it was cancer. um and then
put it through a whole round of
traditional treatments to try bash it
back because if you you know if you get
if you nip it in the butt early enough
you can with cancer you can for the most
part get rid of it but unfortunately
Rosie had already passed that threshold.
We um it had grown so far around her
skin that if we removed that chunk of
skin she wouldn't have had enough skin
left to close the wound. Um and and
that's essentially the the trigger that
I was like, "Okay, I'm going to try to
see what I could do with my AI skills."
I I knew nothing about cancer at this
stage, by the way. Um complete novice.
just knew obviously that it's like a
very bad thing to have and uh yeah and
I'd heard about this uh I've been
practicing practicing machine learning
since about 2009
and through in the community obviously
followed Alph Go and knew that the
follow on technology was Alpha Fold and
that that a lot had been said about its
ability to to like design cancer drugs.
So that's the path essentially that led
me to going down this AI track.
>> Yeah. And I think Rosie is Rosy's in the
room, right?
>> She is. Um,
>> she's there.
>> There she
>> There she is. Okay.
and and so you know part of the story I
think that caught everybody's attention
was this idea of of like you said
somebody who is not coming from the
biotech world uh using these tools to
create a personalized cancer therapy for
for their pet and and from everything
I've read
you essentially did a sequence of of
Ros's tumor and you you were able to
match that I think it's the gene
sequence of her blood to begin looking
for um the variations between between
where her DNA should be and and what had
changed. And then you worked with some
amazing it sounds like scientists and
doctors in Australia to then to then
shape a a custom mRNA vaccine. And so
you're you're essentially delivering a
injection into Rosie with this custom
therapy that is then going to get her
body to try to attack attack the tumor
based on these DNA markers that you'd
identified. Is that the gist of it
>> essentially? Um but just to backtrack a
little bit for the audience that are
listening in. Um
uh when I essentially started down this
path uh I I think I actually recorded
the initial prompt somewhere. Okay, I
should probably show that sometime. But
um the I went into chat GPT and just
like told her to like list out a a plan
of like how we can go through the steps
of making uh uh some kind of drug or
treatment to to to fight it. But um I
also similar to before doing that uh
prior to that I also used chatbt to sort
of build out a map. Um I used obsidian
if you've ever used obsidian to build uh
it's like a markdown note-taking tool
and I went and like created an ontology
like the the lay of the land of all the
drugs that were out there um and and
what they do and what they like what
they trying to what their mechanisms are
and stuff. And I learned about this new
well relatively new to me. I'm not an
oncologist, so that's my disclaimer. I'm
just a uh I've been doing machine
learning for a while. Um it's what I do.
But uh yeah, I learned about
imunotherapy. So the key thing about
amunotherapy is that it's uh using your
body's own immune system to fight the
cancer. And uh to to even backtrack this
a little bit more, um it's worth
speaking a little bit about what cancer
actually is. So um if you like re rewind
re rewind the clock all the way back to
uh I don't know when life began type
thing there were single-c cellled
organisms floating around and um at some
stage they decided to like become
multi-ell organisms that work together
to become animals and plants and fungi
and everything. Um cancer is when one of
those cells in the organism sort of
breaks away and starts doing its own
thing. Okay. And um uh
the immune system essentially evolved to
to be this mechanism to sort of keep
everything to keep you to keep your body
in line. Um and that's and that's why
imunotherapy is so important. like it's
it's one of the most important like
things to realize in fighting cancer is
uh you want to kind of leverage your
immune system as much as possible to to
to fight cancer because um one of the
things cancer can do is it can it's got
the ability to
so um you and I right now have cells
going rogue all the time. So like we
have like these cells they they mutate
they start doing weird things and your
immune system goes in there and just
like nukes it just gets rid of it. Um,
but one like a once in a billion I don't
know what the actual number is chance
the the mutated soul will gain the
ability to go into like stealth mode and
then the immune system can't see it
anymore and that's when like it start
that's when it actually starts to get
bad. Um, so that so so that's a long so
a long answer to your question but the
key thing is that that's why
imunotherapy is so important.
>> Yeah. No. And and so you had already
done surgeries with Rosie. You had done
chemotherapy, which is a bit more of a
blunt instrument where you're you're
kind of killing all sorts of cells. And
then I think, you know, part of the
point you're making with the
amunotherapy is is you can turn on your
body's own immune system to go hunt down
these very specific cells um that are
are tied to this tumor and and get your
body working on on tamping down the cancer.
cancer.
Uh correct. Um so um so I'll give you
something as well. Um the the mRNA
vaccine was like the the computationally
designed mRNA vaccine for a dog to be
very clear for a dog. Um I I believe to
my knowledge that was a world first. Um
it's been done for humans but for dogs.
I believe it was a world first. that was
kind of like the crown jewel in uh the
protocol that Rosie received, but it uh
was not the only thing. So the the other
the other thing that uh I worked really
hard with chat GB01 was the was I guess
the champion in in this part of the
story is uh I developed what's called
this term's been used a lot in machine
learning but it actually is used in
other domains as well which is um the
word multimodal. So I developed a a
multimodal therapy for rose. Um and what
that means is I attacked the cancer from
using different vectors like different
modes of operation. So the the crown
jewel to attack to do the attacking was
the mRNA vaccine that was trained using
this very specific neo antigens that we
discovered um through all this
computational work. Um so so that was
like the snipad was like we were the
mRNA is like painting a target. It's
like the the laser and we're saying go
hunt that. Um, but we also need to apply
all these other pressure points to the
cancer to to to get it to uh to to
maximize the effect of the mRNA vaccine.
So we we co-administered um so the we
had the vaccine we uh obviously went
through that protocol we we
co-administered with something called a
PD1 inhibitor and what that is is um
cancer has like very talked about this
idea of like gaining stealth mode um
part of that is uh it gains the the
mutation to present something that's
called PDL1 across its uh surface and
What happens is when you have these
trained tea cells coming in um that
we've trained using the mRNA vaccine,
they they've they've been we've trained
them. They're like specialists. They the
special forces coming in to take out the
cancer. Um
uh the the PDL1 on the surface of the of
the cancer like it like binds to the tea
cells and tells them to deactivate and
they just stop. they just look that
they're coming in like thousand miles an
hour and uh they they hit the PDL1 and
it just switches them off. So we we
co-administered a PD1 inhibitors that
means the PD1 binds to the T- cell so
that when they're coming in um to to do
the attacking they're they're already
like have a shield around them so
they're just going to go nuts and
they'll generally keep fighting for
about 21 days. Um yeah so we we we
co-administered with the with the the
PD1 inhibitor and we also co so it's a
multimodal therapy uh we also
co-administered it with the tyrrosine
kinase inhibitor and what that does is
it sort of like uh sort of stops the the
blood flow from these new tumors that
stops them from being able to grow new
uh blood vessels
uh so that they can no longer like
breathe essentially. we we reducing the
oxygen to them and we were also blocking
seekit. So seekit is uh the one of the
primary um
proteins responsible for Ros's cancer.
It's essentially like this protein that
um uh it's a very common cancer in dogs
and when it gets left on it causes the
cells to just keep it's like leaving the
switch the it's like the on switch has
been left on through mutation and it
just causes them to keep dividing.
So in sorry that's a very long answer to
the question. I've actually forgotten
what the what the original question was.
>> No no I mean this is all this is all
great information. I think I should walk
us back for a second. I mean okay so
Rosie at the time was I think like five
or six years old.
>> Yeah. So uh that was actually the real
tragedy and like uh part of the reason
why I decided to help her cuz she was
only she was only five and a half when
she when she got this and that's even by
dog years is pretty young. Yeah. And so,
and you get misdiagnosed at first and then
then
>> you start down the more traditional path
for trying to cure the cancer. And then
when things aren't when things aren't
working, so you get
>> you're like, "Look, I'm going to try
anything. Let's see what we can do." And
and so you turn to AI and you told us a
little bit about your original prompt.
But but so step one in what it was
telling you to do was to go try to get
both the tumor and her blood sequenced.
Is that was that step one?
>> Yes. Yeah. So if you want to go back to
the actual process, um step one was to
um I fired up chat GPT and um it gave me
the steps of what I so again just to
backtrack like this was just to do the
alpha fold stage um the trying to
develop a drug. had laid out the paths
to do that. And the first was to go and
get Ros's um
uh tumor sequenced and the blood
sequenced. And then you line them up
against each other and you get like a
genetic diff between the two or um you
can kind of another analogy is think of
yourself owning two cars. They're
exactly identical and uh you you never
drive the first car and you drive the
other car 3,000 300,000 mi or something.
You take the engines out and you
compare, you know, the difference. The
difference is the damage between the two
engines. That's essentially what we were
doing within the state. And I saw I mean
I saw you pull up some of the snippets
of the DNA code on your computer. Were
you when you were looking for those
differences between the DNA of the tumor
and her her blood DNA, were you doing
that yourself um with AI or or were you
being guided by a scientist or a doctor
in that process?
>> Just just just the chat bots.
>> Okay. Okay. Just the chat bots. And then
so once you once you saw um once you saw
the differences then you turned to the
AI again to try to figure out um
the nature of the tumor and if there
were medicines available for that
particular sequence. Um
Um
yeah. Um before I before I answer that,
the the one of the one of the the
one of the things I was kind of scared
of about even the story getting out is like
like
I I I because I used LLM to do all of
this. Um, I I'm just I I I don't want to
have the LLM limited in the future to
help other people essentially cuz you
know now that the story's out like you
might have your big labs decide to start
switching off the ability to do what I
did. This episode is brought to you by
Brex at Core Memory. We spend tons of
time with companies that are trying to
build fast and they often run into
similar problems. their corporate spend
either spins out of control or slows
them down. Brex was built to break that
trade-off. It is the intelligent finance
platform that helps you move faster and
spend smarter. Cards, banking, expenses
all in one place with AI agents working
in the background to handle annoying
tasks like collecting receipts and
enforcing corporate policies. This is a
massive timesaver for us. More than
35,000 companies use Brex from early
stage startups to giants like Anthropic
and Zoom and soon to be giants like Core
Memory. We run on Brex. So should you go
to brex.com/corememory
to learn more? What was the next step? I
mean you you'd seen the differences and
then you want to go hunting for a
medicine that might be known to to
already be a cure effective against that
type of tumor.
Okay, so we I went through the pipeline
of assembling the genome. Uh it takes a
while cuz uh when you get the data back,
which is uh Ros's
uh both of both both of the the blood
sample and the cancer sample were both
300 gigabytes. Um it doesn't come back
in like here's your DNA. It comes back
uh like in pieces. It's like a giant
Lego set that's been put on the floor
and you have to put it together. So uh
we spent some time putting that together
and um uh and then through literature I
knew of CKIT the the prim like one of
the primary um proteins that's been
identified for driving roies cancer. So
isolated ck kit and um modeled that in
alpha fold alpha fold for I'm sure most
people know what it is but uh the it won
the Nobel Prize I believe in chemistry
and it's a computational thing. So there
was a bit of controversy around that
because it's like computers shouldn't be
winning chemistry things and um the
reason it won that is for the first time
ever we are very cheaply able to
uh make three 3D models of proteins.
Before it used to take I think like a
month to see a protein. They used to use
like these like uh like a sort of like
lithography sort of type techniques I
think. Um anyways, so yeah, so modeled
the uh modeled the protein and then um I
went looking for like drugs to try block
it and I used genetic algorithms to
develop uh actually developed like novel
compounds and and bound them to the
protein to try block it from aka switch
it off. If you could do that then the
cancer you essentially trying to stop
the cancer from multiplying.
Um and yeah, so use genetic algorithms
to develop like novel compounds. And
then I I decided against that approach
because even though I decid I found some
things that were like really strong had
a a high what's called like binding
affinity um I even spoke to chemists
about getting it synthesized uh also at UNSW
UNSW
um the same university that I I that
worked very gracious graciously with me
throughout the entire process. Um yeah
uh I decided against that approach
because you would probably have to
perform a full drug trial to to get to
the end. So you'd have to go like uh you
know in in um in in in v in in in vitro
or like in a in a test tube type thing
then go to mouse models then go to like
I don't know generally it's like you
know monkeys or something and then
humans. We'll have to go through like
part of that process before we got to
Rosie cuz it's just too risky. So that
was genetic algorithms to bind to the
thing. And then the other approach I
used was what's called docking. Oh to be
very clear I also used alpha fold 2
because they changed the license for
alpha fold 3.
>> Um you not allowed to do this anymore on
alpha 3 which is fair enough. Like it's
uh it's it's it's it's hardcore IP and I
respect that. Um
but yeah uh we use alpha 2 uh to do the
protein. um modeling but um and the
other the other approach that I I used
was uh what's what's called docking and
I docked a a million li leg lians they
called lians
um to uh the secret protein and found a
drug that's owned by a major US drug
company um that that had a really high
binding affinity like really high like
higher than the the normal drugs that
you can get for this but uh I I tried I
I try to reach out via compassionate use
paths but um uh they are generally for
humans so they just politely declined uh
both through official channels and
unofficial channels. Um so around this
stage uh I was feeling this is around
July 2025. I'd done all of this stuff by
July 2025 and um just to point out like
I again I started with like no knowledge
of cancer and um like putting in I guess
a bit a couple of hours every single
night and like just starting to learn
more and more and more and by this stage
I kind of learned that uh the
uh the the whole alpha fold approach is
only part of the solution. So um
the the the the ability to like take a
lean and block the cancer is only one
mechanism going back to that multimodal
idea of being able to like stop the
cancer from doing its thing. Um and then
around July 2025 had tried all this
stuff like gone through all these
different computational things like it
was just like AI AI AI but by by the way
like the the the computational pipeline
there's like AI algorithms in there that
are all like that's different AI things
as well. it's like different statistical
approaches and um and um one other thing
worth mentioning as well is uh a lot of
these tools that I was using are like
written by grad students so like they're
highly unoptimized uh they taking like
weeks to run and they could probably run
like in a day if it was optimized for a
GPU or something a lot of them are like
CPU based. >> Sure.
>> Sure.
>> But um yeah
>> and then Okay. And then and then Okay.
So you're you're doing all this work. Um
you've you've reached sort of this next
stage of the journey and then like at
some point you have to you have to
switch gears and and start to partner up
with a a scientist who's going to take
you to the next next step.
>> Um yeah so again like I had done all
this computational work and it had led
to like nothing. Um and uh I was feeling
the the window had come out of my sales
a little bit like uh
you know I was like okay spent some
money trying to get the data and that
was cool. I've learned about how it all
works. That's also cool. Well not all
works you know but there's lot there's
lots to learn. Um but um
one day I was taking Rosie for a walk
and um it was around July 2025 just down
the road here. I had like a light bulb
moment go off where it occurred to me
that maybe I'm
halfway to um
making an MR making a amunotherapy
treatment myself. So I I came home and
uh jumped jumped on chatbt at the time
and like like said like this is all the
stuff I've done everything I've sort of
described to you just before um you know
can we use any of this to like make an
amunotherapy and and it came back like
sure you were very close to making a
neopeptide vaccine or um or an mRNA
vaccine and I reached out back to
Professor Martin at uh the University of
New South Wales saying hey Professor
Martin um what how hard would it be to
like make a neopeptide vaccine? Um, and
he roped in one of his colleagues, Dr.
Deborah Bernett. She's like one of the
most prestigious researchers here in
Australia, um, for she's done some cool
stuff with animals. And she and I said,
same same thing, same question. She's
like, I I don't uh I don't really know
much about that technology, Paul, but I
do know a lot about mRNA vaccines. And
then obviously I'm like, hold that hold
that thought. Check it out. Check it
out. Chat chick again. and and it turned
out that like mRNA vaccine was was the
fastest way to get a treatment for Rose.
So um
this is uh
yeah this is this is why uh so obviously
they they got be excited cuz now we can
like potentially close the loop. We've
done all this like computational work.
Now we can maybe get to a treatment and
then with all the other knowledge that
I've gained around doing this like just
attack it from as many different vectors
at the same time. Yeah. So, uh,
>> yeah, sorry. Go ahead. Go ahead.
>> No, no, no. Sorry. Just, uh, feel free
to shoot.
>> Well, okay. So, then you you you team up
with these two two scientists and and
then you guys go do it. You go create a
personalized a personalized mRNA vaccine
for Rosie.
>> Yeah. So, yeah. Um, run some more
computational stuff. Um,
uh, a whole bunch of different different
things we did. Um but long story short
got to the neo antigen selection and um
uh yeah we uh it's actually possible to
take the antigens we um discovered and
you just paste them into what I what I
did is I just pasted them into Gemini
and it it it gives you back a vaccine construct.
construct.
uh um but uh I double checked it with
Grock and Grock actually found some
errors in it and corrected them and and
the final vaccine con construct was
developed by Grock
>> also using Grock throughout the process
as well.
>> Um but but uh so long story short, I had
this vaccine construct. It's like
literally half a page of text and I
emailed that to Professor Py um uh who
I'd been working with throughout uh the
last 6 monthsish like through emails and
stuff and um we sort of developed the
plan to have it manufactured. But uh all
throughout this process I was very well
aware of the fact that we had to have an
ethics approval cuz this is like an
experimental uh you know it's
essentially um a drug trial. Um so uh in
the background I was working like really
hard to try to get that ethics approval.
I'm doing all the computational stuff in
parallel with doing the ethics approval
and um it got we got to a point where um
I was writing this ethics approval every
every like 2 hours every night. Also
using LLM to assist um a lot of this was
all me like injecting myself into it
like thinking of what things we haven't
thought of yet but just using the LMS to
sort of fill in the to like paint fill
it out a bit. And um
yeah uh um
um
it got to a point where the university
was going to have to modify its license
with the state because the vaccine was
going to be administered by her vet and
that meant that it was offsite.
So because it was offsite uh they're
going to have to moni and going through
the whole process of modifying the the
the the license was going to take like 8
months. So again, I was like, you know,
like going gone through all this work,
like we we we've got the vaccine. It's
actually ready to go. Like we actually
have the green light. We were like
ordering the PL in the plasmids and
everything to to get it ready to
manufacture. Um but the ethics was was
like a a red light. Um and then sort of
a a twist of fate intervened. Um
uh Tom from UNSW had released an earlier
article about the alpha fold work that I
did um around July 2025 and uh uh one of
like the most prestigious K9 cancer
researchers on planet Earth reached out
to me as a result of it. Uh Dr. Dr. Mary
I think she's like used to work for
DARPA. Um she reached out to me and uh
she's like Paul like what are you doing?
like what what is all this stuff? And I
sort of told her um and um as part for
part of the protocol, I was doing some
research into this thing called aduance,
which is like a a thing you add to the
vaccine to get the immune system to sort
of pick it up. And I was trying to
figure out like how we could legally get
hold of this to add it to the protocol.
And um I reached out to Mary around FOB,
November to ask about that question. And
I was talking like uh juvenants also
ethics approvals I'm having like sorry
sorry sorry so she forwarded me on to uh
professor Rachel in Queensland who who
who she had funded she had funded
professor Rachel to do some separate
research in this domain and um yeah I
got on a call with Rachel telling her
about all this and then adjuvenance and
then I mentioned like just like like
briefly and like ethics approvals man
like I'm trying so hard to get this and
gone through all this like all the stuff
in the background. Like I could talk on
another hour about all the stuff we did
there. But um long story short, she's
like, "Oh yeah, I have the ethics
approval for uh for that for that exact
thing." And I'm like and and then the
next thing she said was like, "Yeah,
I'll take you under my wing." I'm like,
"I'll just like playing it really cool."
Like, "Oh yeah, cool. Yeah, no, no
worries." But in my mind, I was like
like, you know, like fist pumping. >> Yeah.
>> Yeah.
Um and so and so like all told how long
like just to get to that part of the
journey how how much time was that?
So uh so we started um we started again
I started sort of trying to do my own
thing. Um not really not really my own
thing but you know like just trying to
see what I can do in my own time around
uh November 2024.
alpha fold stuff uh was up to about July
2025 and um the vaccine creation and
ethics approval was around November 2025
>> and then Rosie got the first injection
in December I think
>> early early December
>> early December and then she's had one
more or she's had two more she just had
like her third right
>> um so I'll just just to interject
I'll say two more things about this so Um
Um
Dr. Rachel is a thousand kilometers
north from where I live. So she lives in
a different state. Um that's the reason
we're able to do this. That state has
like slightly different laws to this
state um similar to America. And um uh
yeah, so I drove 1,000 km north um in
December and uh we did what's called the
induction phase. So we did uh we
inducted Rosie into uh the protocol. But
uh I I I thought the university was like
in the city Brisbane. I live in Sydney.
The city is Brisbane. Um I thought it
was like in the city like the the the the the
the the pet
pet
uh the the the veterary the school of
veter veterinary medicine is about 4
hours outside of the city. So I drove I
drove up to 1,000 km north and then to
get the induction phase done I I drove
like four separate times for 4 hours
each time to get that done. Um so
Rosie's had that that was the induction
phase and then she's uh came back to
Sydney. Um started to see the results
that have been like I shared those
pictures on on on X. Um
and uh yeah since then she's had I've
driven back up to Queensland one more
time for a booster shot. Um she has she
has additional booster shots coming up
as well.
>> Hello geniuses. Let me tell you about E1
Ventures. They are venture capital firm
in Silicon Valley, a longtime supporter
of this podcast and a longtime supporter
of big fantastic worldchanging ideas. If
you have such an idea, hit up E1
Ventures or send me a note and I'll put
you in touch with E1 Ventures. It's uh
it's quite a quite a deal for listening
to this podcast. Um thank you again as
always to E1 Ventures for their support.
>> And for people who haven't seen the
photos, I mean that the tumors have
reduced drastically in size. You've
talked about her um her attitude and and
energy going up and she's running around
at the park and um I mean the
indications seem to be that something is working.
working.
>> Um yeah. So like look um initially uh I
was before the the protocol was
administered I was absolutely [ __ ]
myself because it was like a completely
novel um you know professor I remember
like professor Rachel saying to me like
now now it's like in God's hands like
we're like now we hope and pray like
it's and after the first shot was
like for the first twoish weeks it
actually started to get bigger the
cancer but this is
actually it's it's it's generally a good
sign in amunotherapy because what it
means is the tea cells are swarming to
the cancer sites and are actually
infiltrating and swelling up and like
going around and doing doing their
killing thing. Um so initially like the
tumors like got very they started
swelling up and um it looks pretty bad.
Um and then um
yeah and then and then then through
through January I started to be like I
could start seeing her like tendon on
the back of her leg again. Like I I
hadn't seen it in like 8 months or
something. Uh it just been swollen and
uh I'm like is this thing working? And
and it just kept it just kept shrinking.
It was it was it was it was it was
ridiculous. And uh after the last
booster shot, it's done it again. It
they they swelled up and then they
shrunk down even more.
It's ridiculous.
>> Yeah. Okay. U Okay. So, this is amazing.
Thank you.
>> To be clear, it's like I just want to
put this on record. It's like not a
cure. It's it's I I'm very well aware
this is not a cure. It's um it's it's
it's bought her a considerably better
quality of life. She was like starting
to shut down as as an animal around
December and and now she's like come
back to like when the when the reporters
came around here to take those photos.
She was like running around in the
garden like playing with them because
they're toy toys and she wasn't doing
that before. >> Yeah.
>> Yeah. >> Yeah.
>> Yeah.
>> Yeah. No, and I Well, thank you so much
for walking through that story. I mean,
so these are a couple of the questions I
wanted to get into. I mean, you said
it's not a cure. um you know since you
posted this since the story came out um
I think as often happens with these
things you know the first wave was this
like amazement from everybody and then
and then people on X and and Reddit all
start digging in you know with their um
bringing their various contexts to bear
on this and so some people in my science
world were grumbling and they said uh look
look
companies like Madna are already working
on these types of therapies for humans.
This isn't that big of a deal. And then
they said, you know, we don't even know
if it is a cure because the they the
dog's been going through chemotherapy
and these other treatments. We don't,
you know, you just can't tell which
>> they working. They weren't working.
>> No. So, this is what I wanted to ask
you. Have you seen Have you seen people
pushing back on that and saying
>> yeah look I I look the part of the
reason why I haven't slept a lot is I
read a lot I unfortunately made the
mistake of reading some of the comments
>> never read the comments
>> but um I sort of pul pulled back from
that as well. Um uh look uh people are
can feel feel free to say whatever you
want to say. Um um I know what I saw. I
know what I did. Um, before this, Rose
was again starting to shut down. The the
cancer was um, uh, getting bigger and
bigger. Her energy levels were going
down. And now, I feel like this has
bought her more quality of life. And
potentially, we've extended her life by,
I don't know, x amount of months. It's
hard to say because this is all new.
And and you know during when you were
telling me what happened I noticed I
mean you were emphasizing how much you
used AI along this journey there you
know there's entire Reddit threads where
people are saying
you could have done this on Google 10
years ago. This is this is not that big
of a deal. All the AI part of this is
being and this is not me criticizing you
because you know I I don't think you're
even responsible for having to justify
any of this. you're just you're living
your life and and doing this this thing
for your dog. But I I assume maybe you
saw some of that where people are saying
um AI wasn't even that big of a a deal
and all this, but when you walked
through the story, I mean, it sounded
quite clearly to me like it it was quite
a big deal.
>> We used like LMS as the orchestration
layer and then we use all these
different AI techniques in between to to
get to the final result. It was all AI
and and me just like doing stuff in the background.
background.
>> Yeah. I mean and and I think you are I
mean everyone seems in agreement that
this is the first time there's been this
type of personalized therapy for a dog
and for an animal. Um we mentioned
>> so there are they are like personalized
vaccines for dogs but not
computationally designed ones. >> Okay.
>> Okay.
>> The first computationally designed just
to be very clear. >> Okay.
>> Okay.
>> Yeah. And then and then we and then
there are companies like Madna and
others in actual clinical trials for
things like this with with humans. And
so I'm just I'm trying to get at like
something about your story has really
resonated with people and it seems to be
hitting on a couple moments in time to
me. Like one is that a guy who has this
clearly, you know, almost 20 years of
expertise in the AI field but is not not
coming from the science or medical field
was able to to uplevel your skills in
this dramatic way and tackle this
problem. And then we're also hitting
this moment in time where
we are on the cusp of personalized
medicine and and the medical field
is not is really probably not prepared
to move at the speed at which people are
going to want to try to go uh with the
tools that are now available to them. I
mean do you do you kind of feel those
two moments colliding in in your story?
Uh, absolutely. And I feel there's way
more. There's like so many sub stories
have sort of spawned out of the main
one. Um,
yeah. Um, I mean, the the medical
industry, the whole debate about them
moving faster. Um, again, again, I'm
just an AI person. I'm not um I'm not an oncologist.
oncologist.
Um I I look I think that with with the
with what is possible now um I think
this is going to maybe have to force the
conversation a little bit more about the
usage of the tool. Put it that way. Um
yeah and the other thing that there's
lots of stories one one I one one thing
I just like would like to say is uh
like people there's a lot of like
negativity around AI um at the moment
and um and I think it's been I genuinely
believe having done this that it's been
like framed the wrong way. Uh so I think
it's like AI is like close to the like
the steam engine like when it was
invented it disrupted civilization
fundamentally everyone like moved to
these cities existed before but they
were no long they were not at nowhere
near the scale that were like after the
steam engine and the industrial
revolution it like transformed
um and and that obviously disrupted like
jobs and how things worked but people
still had to got to do stuff. Um, I
think the same thing is going to happen
with this. Like AI is is essentially
doing the same thing, but uh um like
there's going to be so many new
industries started as a result of it.
Like you just got to choose uh it's like
AI meets whatever AI meets biology is
what I did here. And this could lead to
a whole bunch of like really cool new
products that that's going to make uh
the world a better place. And if you
have an area of expertise whether I
don't know what what whatever it is like
if you uh my traditional training is as
an electrical and computer systems
engineer that's what I did what I
studied. So if you did like AI meets
energy there's like like so many cool
things you could do there as well. Um so
I think and it's going to take time to
play out as well. So yeah, um
I don't know. I Yeah, I just think
there's been so many sub stories spawn
out of it. Um and um yeah, it's just
been interesting to see the conversation like
like
>> Well, yeah. I mean, you said you said
originally you were kind of reluctant to
even go public with this stuff. And when
that when that first press release came
out in June or July of last year from
from the university, I mean, people it
did it did not capture the public's
attention. So now that you've been
through this experience, you know, you
you're it's now turned into this like
like we just said, sort of like a
symbolic moment for for different
things. I mean, did you have did you do
you do you kind of regret being public
about this at all now? Did you expect
any of this when the the Australian
reporter came by last week? >> Um,
>> Um,
I mean, look, when Tom uh reached out in
2025, mid2025, I was like a bit hesitant
to talk about it just cuz again, I
didn't want uh um
um
I was aware of like the controversy that
it could actually cause. Um, having been
in the industry for a while, I've seen a
few waves of the these like
controversial things spawn. And then, um,
um,
um, and then also with regards to the
Australian reporter, like I was told
like, okay, um, Paul, you should phone,
you should phone this lady. Um, her name
is Natasha. And, uh, phone Natasha.
She's like, oh yeah, I've heard about
this guy. I've heard about this like
scientist guy um, who's just helping his
dog or something. I'm like, I'm not a scientist.
scientist.
I'm I'm a I'm a I'm an AI I'm I'm an AI
guy. And she's like, what? I started
telling her the story. She's like, what?
What? No way. What? What? And then um I
went to like text and notify the
professors that, oh, there might be an
Australian reporter trying to get hold
of you. And and it turned out she's
like, yeah, she's already a hold of me.
So, she went out and like verified the
story and then came back and yeah, I
think got really excited about it. But um
um
yeah um yeah Ash that's that's about it.
>> Well so like I mean you said you were up
you know late reading all these
comments. I mean so like has this been
the have the last two or three days I
mean it's good for the story is good for
Rosie and you and and you guys getting
to hang out. Have the last like two or
three days been good or bad then?
>> Well well well well I think actually
Rosy's got the shits with me at the
moment. Uh cuz um cuz I I she I haven't
been doing too much hanging out with
Rosie. I have I've still been taking her
for her walks. Um still been giving her
um her medicine. Um
like she's she's also on like a little
bit of palative care. She's getting
antihistamines to stop with the itching
and stuff. Um
but uh overall she's in she's in general
good spirits. We're we're about to go to
this the city after this.
>> Yeah. and and and you're
you're just trying to get through this
this wave of attention for a minute and
then get back to normal life, I suppose.
>> Yeah. Yeah. It's been it's been nuts.
>> I mean, do you think this is going to
has like has this altered your
um what do you do? Do you run do you run
your own like consultancy, your own company?
company?
>> Yeah, I run a um machine learning AI
consultancy. Um I've been doing that for
about seven seven
seven years 2019.
>> Um that was a data scientist.
>> Okay. And like is this going to change
um you know is this something you're
still going to pursue? Is it has do do
you think this has sort of changed your
life in any ways? I suppose I mean it's
a big story man. >> Um
>> Um
yeah. Uh
I look I'll just say this Ash. I think
there's more to be done.
>> Yeah. I think there's I think there's a
lot more to be done.
>> Computation meets biology is arguably
one of the biggest new fields that's
going to emerge in the next 50 years. I
think it's going to be huge.
>> And this this kind of really opened your
eyes to to that end of of AI.
>> Yeah. Um I was asked to go to the uh the
UNSW business school about 3 years ago
and give a just give a talk to the
students to tell them what to do like um
talk about AI um blah blah and uh and
one of the things I said to the students
was like look I think that like AI
meeting biology is going to be the next
big thing. I never thought I'd actually
play a part in it. >> Yeah.
>> Yeah.
>> Are you are you South African
originally? You you you got me.
>> Yeah. Yeah. So So and then how old were
you when you moved to to Oz?
>> Uh 15 16.
>> Okay. Okay. And then you got into AI
really early, man. >> Um
>> Um
uh yeah. Um look um professor uh do
doctor Dr. Russell Webb. Uh when I did
my electrical engineering degree, um
through pure chance, uh we had to apply
uh to do a whole bunch for your your
your honors thesis, uh you had to there
were a whole bunch of options to apply
for. And the one that I really wanted to
get into at the time was we had to like
put a list down of priority of um the
projects that you wanted to do. And um
there was a professor over from from the
States who uh had worked on one of the
NASA rovers. And I was really hoping to
get on that one. I love space too. And
um and uh
yeah, and it turned out uh I didn't get
that one. I got my fourth choice which
was working with Dr. Russell Webb. And um
um
Dr. Webb was uh like he used to he was
he used to teach us C like C. And um he
I don't know he used to get into C. He
seemed to be pretty bored by the whole
lecturing thing. I'm not going to lie.
Sorry Dr. Russ Dr. the web. So, so apologies.
apologies.
But but when I was put on his uh his
project um he actually was doing machine
learning and uh I I got involved in that
project and the work that I did for him
like he he asked me to do a PhD uh back
then. I I wish I had to be honest. Uh I
wish I had done a PhD back then. Um but
uh yeah I I learned back then like I
built a a little uh it was like a
two-dimensional environment. This is in
2009. I built a two like uh where there
was just a little like robot and I had
to navigate to a pred like a a
destination that would just randomly
spawn on this two 2D world and it had no
uh understanding of the mechanics of how
to do that. It was it was a neural
network. Um, and I built this
environment and was able to uh to train
it to to do that with like not not uh
pre pre-baking in the instructions on
how to do it like just get to it was
getting a reward signal. It it was like
kind of pseudo reinforcement learning
and uh when when I saw the uh like after
training it for a bit uh when I saw it
like learn how to do it, it was like a
light bulb moment for me. I was like
holy holy [ __ ] this is going to be huge.
This is going to be freaking huge. like
I didn't put the instructions into this
thing on how to do that. It figured it
out by itself. Um so yeah, the
>> and then Okay. And then like as you're
as you're getting into this project, um
I mean look obviously you you want
ultimately to spend more time with your
dog and and that's and you want her to
feel better and and that's why you
pursued this, but was there you know was
there part of you that
obviously it's feeding your interest in
AI too? I mean I mean
you know was this like a side quest or
did this start to become you you could
you could sort of see where this was
going and that this this was kind of
proving out something about AI or was it
just very much I just want to cure my dog?
dog?
Um I mean all the sort of insights I
shared with you um I've been learning
them over the years doing I've worked on
many different projects like what my
specialtity is really is stringing um
bespoke systems together I can do it
very very well um especially machine
learning because that's what I I do um
so a lot of those insights it's all kind
of obvious to me I know that's may sound
a bitum facitious but um um they're
coming back. I I just kind of wanted to
help Rose. That was the reason. And then
it just kind of grew into what it is now,
now,
>> which is um Yeah.
>> And then >> Yeah.
>> Yeah.
>> Oh, sorry. Go ahead. No, no, please go.
>> Yeah. I just think um look, I think like
chatbt uh Gemini and Grock
uh all the other like I'm literally
standing on the shoulders of giants. Um,
and it has Rose is probably Rose doesn't
like even care like the she's the star
of the world in this in this in this
regard at the moment. She doesn't care
at all. But like the it has
significantly in my experience without
without that uh without at least even
the LLMs like I reckon she would would
have probably only lasted a year and now
she's almost coming up to three years
which to me is insane. Like >> yeah,
>> yeah,
>> it's it's insane. Like uh
>> and you think some of the reaction to
this will be
because you were saying, "Oh, I hope
they don't turn some of these things off
in the AI tools." Like you I mean
because it's obvious it's such a
positive story overall, but you think
some of the reaction will be, "Oh,
>> this stuff really is powerful. Well,
there's going to be people running
around making, you know, doing all kinds
of weird stuff with biology, which has
been a a longunning fear with AI. But,
but so you you think there might be a um
a quite drastic reaction to this?
>> Um I think so. I I already know that
like the university that I did all the
work with was already like the amount of
inbounds for people wanting to like do
their own sequencing has like gone up
10x already. So, I imagine that's
probably playing out everywhere at the moment.
moment.
>> Um, yeah.
>> Yeah. And do you think
>> So, people are waiting to like get the
data and see what they could do.
>> Do you like how many people do you think
could do what you did? I mean, you're
obviously extremely savvy with AI. You
had at least a bit of money to pay for
the sequencing. Uh yeah, you know, you
you you weren't a scientist, but you
knew you knew what you were doing in
terms of how to use the tools.
I So I assume that like a lot of people
would not be able to replicate what you
did at this moment in time, but we're
obviously also probably not that far off
from things like that improving and and
the AI just really guiding you through
the process all the way.
>> Yeah, sure. Um I I think if you spend
enough time prompting, you can uh figure
figure out your own pipeline. But this
is true for anything really. Um
actually, you just got to have the right
determination to want to to uh do it
yourself. For me, that determination was
rosy. Um
uh but yes, there are there are ways
to making making AI defensible is
actually a very interesting subtopic. We
could spend another hour on. Uh, and I
think there are ways to do that, too.
>> Yeah. Okay. Um, well, I've used up a ton
of your time. Um, you know, I know
you've been you've been stretched thin,
so I I really appreciate um you doing
this. And then I mean, I you know, and I
kind of I guess I've asked this
question. I just
do you like
you must you do this reminds me of like
the I I wish I could remember the
Russian dog's name that went to space or
I mean this is in another Russian
moment. I mean this has like little
Splutnick kind of um
>> starts with an L I think
>> oh Leica I think it was. Yeah. Um,
I mean, I feel like Rosy's Rosy's kind
of kind of in this this territory now as
as like the the AI AI dog that kicked
off some some crazy new era possibly.
>> Yeah, it's it's super cool. It's it's
really cool. It's really cool. Like um
all the all the different sub
conversations that spawned, I think
those are fantastic, too. There's been
lots of like skeptical takes. I think
the skeptical takes are great, but you
should be skeptical.
>> Yeah. Um Yeah, I mean there's there's a
guy here, Egan Pelton, who's a chem bio
PhD at Stanford. Um, he's very upset at
you or he's very upset at the store,
>> right? Sorry.
>> You know, he's he's like, "This is
ridiculous. Everything here could have
been done without Chat GPT.
>> The dog is on a conventional
immunotherapy with the mRNA vax, blah
blah blah blah blah." Yeah,
>> that's not true. >> Okay.
>> Okay.
the um and and you know I I think some
of this is besides the poet anyway
because I I think this is clearly
symbolic of a moment in time and so yeah
um yeah
>> and I'll just say one more thing uh Ash
um love your work too man
when when your uh when your book first
came out um it it just come out in the
library and I I I knew I knew who Elon
Musk was. I've actually I've been
following him since uh SpaceX was
founded actually because um cuz he's
South African.
>> That's why that's
>> I was like yeah and at the time like the
the um the the sheer idea of um there'd
been a whole round of like startups that
had tried to do commercial space type
stuff um and they had all failed and
then he was going to take his money and
throw it behind. So anyways, knew you
knew about your book coming out and I um
it was fresh in the library when I
grabbed it and I was working as an
electrical engineer at the time and I I
read the whole thing under the desk.
It was it was like pretty Yeah, it was
written anyways.
>> Oh, that's cool. Thanks, man. Well, I am
a fellow South African and my mom is
Australian, so we are we're we've got
multi-bonds going on.
Uh uh well Paul thank you thanks so much
for your time. We we wish all the best
for Rosie and please like keep us
updated on her and and what you do next man.
man.
>> Yeah. Cheers man. Cheers man.
>> The Core Memory podcast is hosted by me
Ashley Vance andor Kylie Robinson or
both of us together. It is produced by
me and David Nicholson. Our theme song
is by James Mercer and John Sortland.
And the show is edited always by the
John Swortland. Thank you so much to
Brex and Ewan Ventures for all your
support. And thank you most of all to
everybody for listening or watching. We
love you. Please leave us a like, a
review, a subscribe, all those
tremendous things. Thank you and we'll
Nhấn vào bất kỳ đoạn văn bản hoặc mốc thời gian nào để nhảy đến phần đó trong video
Chia sẻ:
Hầu hết transcript sẵn sàng trong dưới 5 giây
Sao Chép 1 Chạm125+ Ngôn ngữTìm kiếm nội dungNhảy đến mốc thời gian
Dán URL YouTube
Nhập link bất kỳ video YouTube để lấy toàn bộ transcript
Form Trích Xuất Transcript
Hầu hết transcript sẵn sàng trong dưới 5 giây
Cài Tiện Ích Chrome Của Chúng Tôi
Lấy transcript ngay mà không cần rời khỏi YouTube. Cài tiện ích Chrome để truy cập transcript của bất kỳ video nào ngay trên trang xem, chỉ với một cú nhấp.
