September 13, 2024 • 65 mins
This week's episode of the Generative AI Meetup Podcast, hosted by Mark and Shashank, dives into the riveting discussions from the recent Don't Die Summit. The duo shares their firsthand experiences, including an interview with Alex Zhavoronkov, CEO of Insilico, who discusses the groundbreaking generative AI model, Presses GPT, aimed at extending human healthspan. Also featured are insights into the newly announced OpenAI o1 model, its capabilities in enhancing AI reasoning, and a look into the latest Apple Intelligence features and iPhone 16. Join us for a comprehensive overview of AI's expanding role in health and technology.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:00):
Great. Hello everybody and welcome to the generative AI Meetup podcasts with your hosts Mark and Shashank
(00:09):
Together we run a meetup throughout all of Silicon Valley and
We have listeners from all over the world. Thank you Tony for listening. I appreciate you from South Korea
And we just want to share with you all the learnings
From people that we get to meet in the Bay Area. So today we have a really special episode and a ton of things to talk about
(00:36):
So this past weekend Shashank and I we went to the very first ever
Don't die summit hosted by Brian Johnson for those that don't know Brian Johnson's a guy who's
Basically trying to live forever and at the conference we learned a ton of cool stuff all about health and longevity and
Basically how a gen A.I. It may be able to help us live forever. So at the conference we were able to snag an interview with the
(01:03):
CEO of in silico
Which is an awesome company which is combining
Gen A.I. as well as like hardcore
Biomedical research and we got an interview with their CEO Alex
(01:23):
Veronica who is just a really smart guy and
They have a brand new A.I. model which they're going to announce called presses GPT
So he's gonna tell you a lot about it
But before we do that and tell you about the conference
There's a few things that we wanted to discuss today before we get to that so
(01:46):
We want before that we were gonna discuss about the brand new
GPT 401 model that just got announced earlier today
codename strawberry and then we're gonna talk a little bit about the Apple announcement
So Shashank, what do you think about this brand new model from open A.I.?
Yeah, I think this is the most exciting thing since GPT 40 was introduced because it seems to solve a lot of the
(02:13):
reasoning issues and limitations of LLM's today where they just
autocomplete a string of tokens or words without actually understanding what those words actually mean
So this one is
Apparently able to okay actually not apparently we tested it just now and it is actually able to count
(02:36):
How many ours there are in strawberry so for reference? That's actually a big deal
It is surprisingly hard and the reason it's hard is because
As you may or may not know these LLMs are really good next token predictors or the next word predictors
they look at these massive
(02:57):
repositories of data in the internet
bunch of tags bunch of newspaper articles Twitter posts blog posts and
just kind of based on what they've already seen predict what word might come next
Based on all the data that's been ingested in their neural network
(03:17):
So when you ask it a simple seemingly simple question that requires any kind of reasoning or math mathematics ability
It completely fails at that because it just
goes with what it has seen in
previous similar looking patch of text so in contrast this model
(03:41):
guess the answer right and
After testing it briefly it seems like it is
Underneath some kind of an agent which has a
Chain of thought or like a react framework where it reasons and then acts or breaks down the step into simpler
sub steps
(04:02):
Solves each one of those at a time and then combines all those intermediate steps into a final answer
Yeah, so for those who don't know chain of thought is an LLM prompting technique where you basically ask the LLM to think through its
thought process and
That tends to give you better responses so it seems like they kind of just bake this into the model
(04:24):
Actually, you know as an aside one thing that I was actually able to do in the past was even before this GPT
Oh one and that is a that's a mouthful of a name but what you could do is you could say like hey
Chatch B.T. write a program to tell how many Rs are in the letter or in the word strawberry and then
(04:46):
Run the program and then tell me how many Rs are in strawberry and then if you did that it would actually get the correct answer
So that was kind of way to sort of prompt it to do like
I guess you could call it like chain of thought but it's not quite chain of thought because it's like it would like
Run the program realize the answer was different than a thought and they would correct itself
So that that works
(05:07):
But you have to do a lot of like extra prompting technique to do this this seems like it's kind of like making it so you don't need to be a prompt wizard
It's just like does it for you which makes it I think like a lot smarter in general or like maybe like more user friendly
I guess you could say because it seems like especially until like the model isn't like way way better
(05:28):
It's just like almost like getting more
capability out of the same model
Yeah, absolutely I don't think they've significantly improved the performance of whatever base underlying model
That is powering this oh one model
But instead what they've done is give it this agentic behavior this ability to
(05:50):
Reason in multiple steps and then come up with a more cohesive answer. So in general
It seems like open AI went from being a research organization or like a
foundational model provider to
Incorporating all of the tools and add-ons that other people have built on top of open AI and integrated into
(06:13):
Their consumer product which is chat gbt because they've added the ability to browse the web which was you know
Something that Microsoft was doing with edge and Google is doing
And there were other plugins that built on top of chat gbt
But you had to use like llama index and all these frameworks that made it really complicated and
Only really allowed developers to build agents and other complicated
(06:38):
Applications on top of elements now open AI seems to just be baking all this into their consumer product
Yeah, so they're just kind of making it easier for everybody to use their lm which is cool
I applaud that it's amazing
But as a side effect their regular consumer version can go from solving
(06:59):
A reported 13% of math lmp add problems to a record 83% which is like in the 90
Not 89th percentile
Of competitors, which is really cool. How hard are math lmp add program or problems
I don't know because I vaguely remember when I was like in high school or something like that
(07:21):
And I think it was like just the smart kids who try to solve the math lmp add depends on what the curve is and what like
I mean it is the 90th percentile. So it's better than 90% of people. I mean that seems pretty good
But it's like I don't even think most people like what even enter the math
Attempt it
It's just like the people who are like really into math who are gonna even attempt the math lmp add so I feel like to say like oh
(07:44):
We're solving 90% of math lmp add problems. It's kind of like saying like oh like
We're smarter than 90% of the nerds. Yeah, it's very nice right because it's like it's not like
The regular person isn't taking the math lmp add
It's only like the people who like are really really into math or taking the math lmp and as I say like
People who do the math lmp add or nerds, but like you get my point. I do so to clarify it says it's better
(08:09):
Or it's reached the 89th percentile in code forces competitions whatever that means so out of
Some small subset of really smart people who are attempting this
Technical challenge. It's reached the 89th percentile. I mean, that's pretty cool. That's pretty cool
So in other
(08:30):
Completely unrelated news open AI is raising money
So apparently they are raising money at a 150 billion dollar valuation
I don't know it's like it seems to me like
It makes you feel like they maybe they don't need I
I don't know how to articulate this but
(08:52):
If they were super profitable, they wouldn't need to raise money and it seems like they must be running out of money
Therefore, they need to raise more so both that makes me I don't know a little bit bullish and then also a little bit bearish at the same time
Like bullish in a in so far that like hey like open AI can raise money at a hundred percent being dollar valuation
(09:13):
Even without a lot of money, but then it's kind of bearish at the same time because like man that really feels a little bit overvalued
I mean the other explanation would be that they need a lot of money to do some massive batch of training or
Build the custom chips or get into a completely new field like robotics and poor that heavy
(09:34):
investment upfront to work on a cool new project. So yeah, I mean
You but I think you might be right. I think it's probably a combination of a lot of things
I don't know if they're making as much money with their subscription model as they need to keep training these models or at least
Train the next generation of models. Yeah, you know, I actually just cancel my subscription. I think it was yesterday or two days ago to open AI
(09:59):
why
great question so I
I think I've mentioned Kagi in the past like k aji.com. It's a competitive Google search and they have their
uh, oh
Kagi assistant so the Kagi assistant is really cool
What it gives you is for $25 a month you get access to unlimited searches of Kagi
(10:20):
But in addition to that so Kagi by itself is $10 a month
But they also use the assistant so the assistant will be able to
Act as and it's kind of like it's similar to like Gemini and Google where it'll help you
Make sense of the search results so you could go and like summarize web pages. It can answer
(10:40):
questions about like a page and a search result, but it also gives you complete access to
GPT like chatship t like all the models like shbd40 gbd4 the gbdo4 forum mini
Gemini Gemini the Gemini advanced model
all the ontropic models so it gives you like
(11:02):
Anthropic clawed clawed 3.5 sonnet the opus model high coup
And the mistral model so like you get access to mistral large so
All of those would be $80 a month if you subscribe to them separately because Gemini's 20 bucks
Open AI's 20 bucks
(11:23):
And Theropik is 20 bucks and mistral I think is also $20
If you want to subscribe to it, but this for $25 you get all four plus paid search
So I'm just using that instead
How does that handle
Like persistent chat threads because one of the big reasons why I still use chat gbt today as a blister
(11:46):
Some of these other aggregators or platforms that utilize other elems is because because of the UX. I open up chat gbt
Open up like a previous chat thread that I've had about a specific topic. Let's say like
I have one for nutrition fitness. I have one for work. I have one for
Mental health. I have one for planning my day, etc
(12:08):
So having that simplified UX and a memory is pretty useful for me
Yeah, so I believe it does keep some chat history so I could go back to that if I wanted to
But it's be fair the UI or the UX isn't quite as good
So one thing that I've noticed so Kaggy if you're listening
I've noticed that when I put in a longer prompt sometimes the
(12:33):
Kaggy UI will just hang for a little bit
So sometimes I'm using it for like programming and I'll push it put in like a large chunk of code
And in that chunk of code
Kaggy might just hang well processes. I know exactly what it's doing
I think it's like trying to do some something smart but a little bit too smart and
(12:54):
I'll just have to wait for maybe a good like five seconds or something like that before I'm able to hit the enter button and get the prompt
It's not as ideal. I think there's some things like, you know, now I'm missing out on the GPT store
Kaggy can't generate images
And thrompe they have the what is it called like the snippets where you could get like the
The code results and it can run the code as well like I can't do that in Kaggy either
(13:18):
but you know it gives me access to all four and
It's saving me a little bit of money. So I think I was on the family plan with Kaggy
So I was paying $14 a month before for like me and my wife
We were both paying like seven bucks each so we're 14 plus we were paying
$20 to the GPT
Chat GPT subscription so that we were 34 now we're saving $9 a month down to 25
(13:42):
So it's like not big money, but still I was thinking like, yeah, well, you know
And I get access to all the models because also like I was finding myself using the anthropic model more and more because
I was
Trying to you know do some programming and some stuff and like I felt like chat GPT was like getting pre-tripped up
but
The anthropic models was actually giving me better results for my programming. So
(14:04):
It's like all right. Well, I'm not as well as use Kaggy for this. So we'll give it like a month
See how it is like I mean maybe I'll downgrade and go back to get myself a chat GPT subscription
But for now like I'm pretty happy with it. Yeah, I was considering something similar when I got the perplexity subscription
So I have a year of
Perplexity's subscription which is a
(14:26):
competitor to Google search. It's a search engine supercharged with AI and it
uses LLMs to search the web summarize the search results give you citations for
the answer that it
condenses and summarize and
There's like a copilot too. So if you have a complex enough query it like
(14:50):
searches and reasons and multiple steps and
You know, they also allow similar to Kaggy
They allow you to toggle between multiple different LMs from different providers and it actually seemed really cool
but
It has a little bit more latency obviously because it uses the web versus chat GPT which doesn't need to all the time
(15:11):
And the UX so I stuck with chat GPT
I mean I I use Gemini 2 I have a free subscription through work because I work at Google so
um
I use that with my corporate account and chat GPT with my personal account
Um, and I also have perplexity not to mention the odd other
(15:33):
Service that I use every now and then but don't pay for so I think both of us
Uh, if I have to kind of assess or
the bleeding edge of these technologies and our early adopters I would say for most people just sticking with
one
consumer facing product
Either chat GPT Gemini
(15:54):
Claude
Would be the best way to go because of the UX the simplicity and
Maintain like a chat history and all the other added bonuses like image generation that are baked into these platforms
Which unfortunately you don't get with these aggregators or
Apps built on top of these LMs
You know, that's true, but he's asking like you know, what I could my mom
I would say like you should just use the free version of chat GPT right because it's like
(16:18):
It's good enough. It's great. Yeah. Yeah, and uh, yeah, not even good enough. It's fantastic. So
I think that like you know, we're just paying for that little bit extra
But like you can get 80% of the value for free and it's still pretty darn good
Yeah, um, so anyways uh, Shashonak
What do you think about this uh new Apple announcement? Uh new iPhones. That's cool. Yeah, I was uh
(16:44):
I was looking at that. I was uh just moderately well-med
It wasn't that exciting um, I honestly couldn't
Pick any part of the announcement and say okay. This is really cool. The one thing that stood out was what the camera shutter
semi touch and semi physical button
(17:05):
They added a new button after like what 15 years or something after taking away buttons and controls for so many years
um, but I I guess
Uh for the sake of this podcast the relevant pieces of information are the fact that
These devices have much more capable chips which I assume is to power the next generation of um
(17:28):
local
genitive AI models their local LMS um
To do you know, emoji
generation to
generate text summarize and uh do other kinds of AI related features on device. So that's kind of cool
Yeah, I mean it's cool. I think apples bringing more
(17:51):
Computing to the edge. So when I say the edge, I just mean like on the device as opposed to like hitting a server somewhere
So that's cool. What are the benefits of that?
Well, I think that there's a few benefits that you could get to like edge computing in general like writing local advice
So I think the biggest the was two big ones
Number one would be privacy
(18:11):
So it's like if you can run things locally in your device and you don't have to hit a server well then
Now you don't have to tell Apple or opening eye or whoever like your personal information
Now I'm sure they're not like looking and it's not like there's like some guy just like oh
What is it what is just shank do you it's like oh it's like oh it's uh tell me more
(18:33):
It's not like they're looking at your chat history
But I'm sure they're not but the thing is is like they could
um like maybe you'd like run through some extra
Things and then like you know one thing leads to another and then like there's data breach and then like your chat history out there
If that happens
(18:55):
You know you're out of luck uh it I'm not saying it I'm not saying like any of these companies are doing that
statistically these large
mega corporations like apple google amazon meta etc are
Very very unlikely to get a data breach. It's mostly smaller companies
um
Like I don't know ticket master or eventbrite or
(19:18):
Instacard that may have data breaches
We're not saying that those companies specifically have data breaches. We're just saying that like in general like uh
I think they oh did they have data breaches off the top of my head, you know, I might be wrong, but um
Yeah, again. Yeah, let's go with what you said. Yeah. Yeah. So so anyways
Allegedly they may have had a data breach. I have no idea. So don't sue us ticket masters
(19:41):
Uh
Ventbrite and Instacard yeah, so yeah, don't don't sue us. We don't know we don't know anything about that
Um that hasn't been published. I'm not sure. But anyways
Like the point is is that like uh
When it comes to like privacy if you can do it on the device and you never have to hit a server
(20:01):
You just eliminate that threat factor, right? So it's like now like you don't have to worry about them
Securing their servers. You don't have to worry about like some like a rogue engineer looking at your uh chat history
Like it's just on the device. It's all there. It's great and since it doesn't have to hit a server
The other benefit is it's fast. So it's like you can get a response immediately
(20:24):
Um, and then also like it starts becoming uh
Lower cost as well, right? Because like now it's like you have higher privacy
You have
Faster speed and then because you have this speed uh you start opening up like use cases that you wouldn't have
Mayable to do before so for example like maybe you could use for example generative AI to
(20:46):
do personal
Search engine on your phone right or like uh some sort of personal rag
Uh, so like rag like you guys know like retrieval augment the generation just like say like hey like I want to go search the documents that I have
And then get some summary response back
So like that's the type of thing that I could do where it like would it be feasible with like calling open a eyes
(21:09):
API or something
Yeah, maybe but like now I could do that all in my own like personal documents like maybe there's some things you don't want
hitting the server right like maybe it's like I have my bank account details like the health records or social security numbers
You know all that stuff like now I can go and just uh do that all
locally and that's pretty darn cool if you ask me now the downside of doing this is like well now we
(21:34):
If it all runs locally on an iPhone it's like well
A lot of these models are running on like really really expensive GPUs and high quality hardware
So it's like the quality may go down a little bit, but it's like you know
What do you want you can't have or you can't have like the highest quality and also be like super private and secure
(21:54):
Um and fast and cheap. It's like you gotta you gotta give somewhere
So it's like you're gonna sacrifice a little on quality, but like the quality is still not that bad
It's a pretty good until we get better every day
Yeah, the part about cost also makes sense because uh right now all of these companies
You know opening I raising like a hundred and fifty billion dollars or
raising six billion dollars at like a hundred fifty billion dollar valuation
(22:18):
It's raising a lot of money because it costs a lot of money to trade these models deploy them and keep serving them to users for free
So um if you don't need to do that you can run for free locally
albeit with a smaller model
Yeah exactly so I mean that that's pretty cool
And I think Apple's really kind of like
(22:38):
Leading the charge on that. I know there's like a few other companies like Google's doing some stuff
Google's been like
shoving AI in people's faces for like two years
Well, I mean like sure, but like I think the thing is is like I I agree with you
But like Google has like a much smaller like market share than the iPhones fair
So it's like I mean like uh in mobile devices. Yes. Yeah, so it's like I mean like I guess you can argue like Google's like
(23:05):
Leading the charge but like apples like
Bringing ever one along with them. Yeah, um that's a good point
But these uh this new Apple intelligence feature which still hasn't launched
Good point to be clear even if you buy the new iPhone 16 or 16 pro
It's not gonna work today. It's tbd when it's gonna be available and
(23:30):
It's not gonna be available for all the older models of iPhones that people already have
So you're going to have an uh have to upgrade to the newer models
unless if you have the
iPhone 15 pro with the
previous
Pro chip so it is still a relatively small segment of users who are
(23:51):
Looking to upgrade. Maybe maybe this gives them a good enough reason to upgrade
Well actually not that you mentioned it. I don't know like uh, I don't know. I feel about I mean
I guess it's okay that Apple is going to announce like a phone with the features that aren't there yet
But I don't think that like any of our listeners should just like go rush out to buy an iPhone with something that
(24:13):
Has a promise of being there so like
Apple says the features can be really good. I mean knowing Apple's track record probably will be really good
But like at the same time it's like
Until like it's been in the hands of reviewers and then we know it's really good like
This is like a feature that like will most likely ship most likely ship soon
(24:33):
but like
We don't actually know like how good it will be like I mean if the feature was really really good and ready
They would have shipped it yesterday right but they didn't and the fact that it only will run on the highest end chips
Makes me think that like well
Maybe it's going to be using a lot of device power so like I don't know like they said that they're going to have a new battery in these iPhones like
(24:58):
Is using these gen i features is it gonna like drain the battery life is it gonna be like high latency like
I'm not gonna be able to use the phone normally like I don't know like
I hope it's gonna really good but like until like it's actually in the hands of the reviewers like
We don't know and that's the thing so it's like
I like I appreciate like Apple like going out and like announcing this I think they announced this like
(25:21):
All these new AI features like what was it like in wwdc? It was like a few months ago, right? Yeah
And like it's still not here yet. So it's like all the people who like want to bought the phone hoping to get these features and then
Now they're not getting except in like the newest phones. It's like I don't know. It's like it kind of
Rows me the wrong way a little bit. Yeah, and Apple's defense. I think the hardware is gonna be solid
(25:44):
The reported benchmarks with like let's see some
Some double digit percentage better CPU GPU apparently the
Okay, this is a back comparison
Okay, it's 30% faster CPU than the last generation of iPhones. It's the GPU is 40% compared to last generation
the
(26:05):
The batteries are supposed to be physically bigger too and they've done a bunch of
optimizations to make sure it runs faster
So all of those things will result in a much more capable iPhone that should hopefully be able to run these models because like Google is able to run
Uh some of their smaller models on device and that's been working fine and I
(26:26):
I think we kind of have a rough idea of
What these features will look like because we have local LM's running maybe like not on a phone
But on some of these laptops and uh, we've been testing them out ourselves for a while
So we can extrapolate what the performance of some of those will be and given
Some of the performance benchmarks of the new next gen iPhones. I feel like they're very capable of running these text models
(26:51):
some simple image generation models and
The the part that they may struggle to get right is the UX
because like the the key
differentiator of Apple intelligence is the fact that is running on your iPhone on device and able to
Tap into all of the different apps that are running on your phone. So look at your calendar look at your email
(27:17):
Of course with privacy in mind
running locally
And as soon as you have a new piece of mail or message come in it's able to look at everything else that already exists on your phone
And like do something more meaningful more intelligent maybe
Someone says oh, I'm gonna be late uh for dinner tonight. Then it's gonna you know shoot a text to the restaurant or something
(27:41):
Ask them to push the reservation later or
Send a note to your mom telling them that oh take care of the kids a little longer because we're gonna be late and so doing all those things
Cohesively, I think
To build a good story around that for a good user experience. That's gonna be the challenge, I think yeah, I agree
I agree
(28:01):
Um, so we'll see time will tell I guess in the next I don't know exactly what's coming down next few months probably I hope
At some point in the future do you have the next generation iPhone? No, I have an iPhone 13 13
Yeah, that is never gonna get any any of the Apple intelligence features unfortunately not so
Oh, well such as life
(28:23):
The phone's still working okay, so we'll keep using it so uh yeah
Now I guess it's time to talk about the don't die summit so yeah that that was a pretty fun event
That was really interesting. Yeah, that was the first uh
Health wellness longevity related conference that I've been to
I wasn't sure what to expect, but that was that was interesting. Yeah, so for reference
(28:48):
Shashank and I this past Sunday, so uh, what was that today's Thursday? So five days ago
What's the day's date? I forget. I don't have my
September 12th September 12th, so whatever was September 8th September 9th something like that we
drove from the south bay
Uh, old way to San Francisco. That's 6 a.m
(29:08):
Well, yeah, I picked up Shashank in his house at 6 a.m
and
He was up kind of late the night before dancing
and
He got enough sleep
To go wake up and we drove together
To the don't die summit which is fantastic. It was so much fun. So um
(29:29):
It was this conference at pier 35 in
San Francisco, so for those who've never been to San Francisco
um
Pier 35 is
Rate along the San Francisco bay
And uh, you can basically just like rate along the water and it's super beautiful
Uh, like this whole area um, it was like a
(29:52):
Kind of a sunny day. It was it was pretty nice and as we
Uh, we're walking in we arrived at pier 35. I remember at
703 a.m
I remember because I was worried we were a couple minutes late because that they were starting at they were gonna have a
Don't die rave. So you know as an aside like why do I want to like if I'm gonna like dance and like party and like rave
(30:18):
Like why do I want to do that at like 2 a.m
They're like midnight like that's gonna completely mess up your sleep quality
It's like let's do it at them in the morning like when we're fresh
We're feeling good. You can get like your heart rate pumping
Uh, it's like it's like a morning workout kind of it's like so much energy so much fun like I for one
(30:41):
And a massive fan of the morning rave like let's let's do that. That's a hill I will die on
7 a.m. Of a rave is much better than midnight because for the record I'm not uh with this but continue
Because of midnight. I'm tired, but at 7 a.m. I am ready to go so
With that being said, we we went in and uh we uh saw
(31:06):
Brian Johnson, uh, who is this guy who uh it was a guy who started uh, I think it was brain tree
And then we later got bought up by Venmo
He brain tree bought Venmo. What was that what it was? He acquired Venmo and then sold it to uh
Was it paid out? Okay, so anyways
Uh
He is now super rich. He's I think he sold that did that deal was like over a billion
(31:32):
Hundreds of millions for sure or over a hundred million and close to a billion. Was it really close to a billion? Wow
So anyways, he's doing okay and now he's spending all of his time just like trying to
not die and uh
really help uh
Spread the word on
longevity and just like a good solid quality life so
(31:56):
We saw him uh going in
Uh got the chat with him like really briefly. I also talked to his like dad and son both like great people
And we met a lot of really cool people at the conference um so
I don't know shishan, what were you? So many like your takeaways from the conference. Yeah um so like I said
I didn't know what to expect um
(32:17):
This is the first time uh experiencing something like this and again neither of us has a background in biology
Let alone biotechnology and the cutting edge of uh, you know
uh
All of these multi-omic studies of like proteomics and uh genomics and all the other omics that deal with different parts of human
(32:38):
physiology. Sorry. What's what's an omic?
Uh for my understanding it is like uh
It's a thing that uh it's a suffix that represents part of human uh or biological
information so
proteomics is the study of proteins genomics is a study of genes
Um and there's a bunch of other omics that uh study epigenetics which is like uh the environmental factors to your DNA
(33:04):
um the effects on uh RNA which are like gene expressions and and a bunch a couple other
items i think they're like five or six of them in total huh
So different the different things that affect human biology and uh not just human biology just all of life
um
So it was definitely a lot of information to ingest and
(33:27):
But but to kind of summarize i think uh we talked about this a little bit uh the conference
I think left us both maybe i can talk to myself left me a little pessimistic about the current state of research and
supplements and uh other
quick hacks to
(33:47):
a longer healthier life
It seems like there really is nothing significantly better than focusing on the basics
getting a good night's sleep focusing on healthy diet nutrition
um getting a lot of exercise and building good uh social connections that allow you to
maintain a healthy mental uh balance so
(34:12):
You know we we saw everything from um
uh the the speaker that uh we're gonna we're gonna plug in soon uh the interview with uh
Alex Zoverunkov um and how they use cutting edge uh deep learning and apply that to the field of
(34:33):
biology and uh pharmacology to uh come up with drugs and target them for curing specific ailments and
diseases uh but we also saw what was it uh companies selling uh thousands of dollars worth of
(34:54):
little five milliliters tubes filled with human umbilical cord uh extract uh apparently make
your skin glow and uh make you look more younger so there was like a wide range of speakers from
what i consider deep science and uh researchers put it pushing the field forward to
(35:20):
ostensibly like money making schemes trying to take advantage of the rich who don't really know
what the difference is between uh these different products uh there was a speaker on uh you know
who had an extensive background on psychedelics it was fascinating to talk with him
that was a cool speech actually yeah to understand the impact of micro dosing because you know
(35:41):
what what are what are drugs we arbitrarily classify uh a couple of these substances a
scheduled one restricted uh substances and it's kind of arbitrary because on the other hand we have
alcohol which is legal to buy which is far more damaging to your body mind and society at large
compared to stuff like um MDMA or ketamine which uh might get a bad rep but has been shown to
(36:08):
have a lot of benefits for uh therapy and getting over traumatic events and curing PTSD
in a controlled setting of course and so yeah it was a wide range of speakers um had a lot of
information still kind of uh going back and trying to understand what we learned trying to digest
(36:29):
all this information um there were a lot of vendors which were really insightful we got to talk to
these CEOs founders and researchers at these amazing companies and learn a lot about
hair regrowth all the different techniques from uh vassal dilators that improve blood flow through your
(36:50):
head uh uh things that block uh DHT which causes hair loss to laser caps and micronealing to create
punctures in your uh skin to promote skin regrowth many serums and so many uh different pieces of
information yeah so one thing that I I thought was really cool that was a takeaway from me now I
(37:11):
think shashanker didn't knew this because like uh you know shashanker's just like a guy who's like
really interested in a lot of these uh things especially for like hair growth but one thing that I
have I have a history of hair loss in my family so it's something that I've been thinking about a lot
yeah so anyways like shashank has been you know doing some research and like learning but like
there's some stuff that like I had no idea but like there is something that you guys can do
(37:36):
for free um that zero money like zero cost health intervention now don't listen to us we're not
doctors but this is really cool so apparently a lot of these like hair loss uh treatments what they do
is they put uh increased blood flow to the top of your head uh so like if you have increased
(37:56):
blood flow to your hair um or like your head you're gonna get like thicker more luxurious hair
that's less likely to fall out and just like overall better quality hair so like uh there's some
this thing called like medoxinol which like goes and increases blood flow to your head um like
shashanker mentioned microneedling which is where you take like little tiny like needles and like
(38:18):
poke into your head and then like uh it makes like thicker skin and then like because of that you can
get like you know stronger skin and then like backstension stronger hair and then also like you can
use like little needles or like holes in your head that you just like put in there and then you can
like uh if you can put like some sort of like hair growth product so you can put like uh medoxinol
which is like the thing that I mentioned before but like that's all great but you don't have to do
(38:44):
any of that the only you have to do is just like when you're gonna wash your hair in the shower
just like scrub really hard uh like uh just like you know use your fingers get in there and then
just like really like get in there if you know what I'm saying just like really like uh scrub the heart
the top of your head like hard like not like so hard that you like hair falls out but like you know
(39:08):
like scrub like harder than you normally would um I've been doing that for like the past couple days
and I felt like my hair is like never bed better uh I was like man I didn't use shampoo uh I just
like scrubbed it and I think they call like a hot water shower or something like a hot water
shampoo um you just like scrub it and like it increased the blood flow to my head like I was like
I woke up the next day was like my hair looks fantastic so uh yeah guys uh it does wonders just like
(39:34):
any other shower just like really like apply that pressure to your head and like it will help you
get like much fuller hair now I got an inter-intro check right there so this is a sample size of
end of one uh on you know his marks unique genetics and everything but you you no matter what you
should not uh be aggressive on any part of your body I don't think you should uh use your fingernails
(39:58):
or uh scratch anything or uh rub uh too aggressively but like a massage I think does improve blood flow
and that's yeah yeah you don't want to like bleed okay you know that finger nails yeah like
she shocks at it better like a massage like massage the top of your head um like I even got like one
(40:22):
of those like uh like things like a little comb like a shower comb where it's kind of like silicone
it's got like little silicone like little nubs coming out of it so I could use that to apply some
additional pressure to the scalp and I did that and I feel like my hair looked like pretty good so
yeah no bleeding it didn't hurt just like you know a little extra pressure just to stimulate the
blood for a little bit yeah and uh bringing this back to the theme of the podcast and our meetup
(40:47):
I think uh one of the cool things that we I noticed in general for all of these vendors and uh
speakers is the fact that they're just bringing more data and uh analytics to all of these data
points that give us an insight into how these different uh uh remedies are affecting us so with
(41:09):
hair loss you know no one can really tell by looking uh is is this treatment significantly working
like 10 20% is is very difficult to track also like the duration uh of efficacy is really long because
hair has to go through these different cycles of growth uh where it's in like a resting phase and then
(41:30):
it has to uh fall out and then you create a new follicle that starts growing from the base of your
skin and then slowly emerges so like these things are really hard to track and one company I forget the
exact name maybe I can uh add it in the show notes later they take regular pictures of your skull
which I'm sure dermatologists and um in clinic hairdoctors have been doing for a while um but they're
(41:56):
bringing this to like uh like an at home kind of a service uh coming soon uh but the ability to
take pictures of your hair different patches of it identify how many hair follicles per unit area
you have what is the hair density what is the hair thickness um and that chart thought over time and uh
(42:18):
figure out what different remedies are doing to improve these different uh data points is like
is that's really the way to go I think uh traditional remedies of just like you know
feel it out uh if it feels good then then it's great keep doing it I don't think we're good judges of
these kinds of um my new details that the human brain isn't very good at uh not identifying and
(42:43):
how long similar lines I think we saw really cool companies that uh bring AI uh to understand your
genome so I took a 23-in-me test a long time ago but um apparently the 23-in-me test only sequences
maybe like 1% of your genome whereas some of these new companies that have come out um mostly
(43:05):
because of the advances in gene sequencing have been able to sequence maybe like 99.9% of the genome
or something uh or close to 100% and they're applying a lot of uh big data using um large amounts of
other people's uh existing ailments and um kind of classify or group different uh gene segments
(43:33):
together and give you uh more insights about what you are uh more at risk for so we tried uh
we tried signing up for one of those new services and I guess uh we'll find out in a few weeks what
Mark and I are predisposed to and maybe share that with the listeners maybe it depends on like
what it is but yeah I mean I'm I'm open I'm open I'm not open to share yeah um we paid to get our entire
(44:00):
genome sequenced uh so uh we'll give them a plug nucleus uh we use them uh I don't know like if they're
like necessarily better than your body else but they gave us I think like 10% off or something for
signing up at the conference so I had always wanted my whole genome sequence that mean the human genome
project uh cost I think over a billion dollars and took like 20 years so now that we can get that
(44:21):
400 bucks that's pretty cool that is pretty cool um so anyways yeah uh now to the main interview
uh because we are running out of time but um yeah so a little bit of uh background on um like we're
going to be talking about it in the interview so um this company uh in silico they do a lot of like
(44:48):
deep science and are kind of on the cutting edge of uh what what it comes to uh like health and
longevity and coming out with new medicine so uh for reference like uh when we watch the talk uh
their CEO Alex he mentioned that uh these G uh what is it GLP1 inhibitors uh which is like
(45:10):
ozempic it's like GLP1 agonist agonist actually uh enable oh my bad not inhibitors but the agonist
so it's like you know these GLP1 things he was mentioning like oh yeah that's kind of like old tech
like that was like 20 years ago like that was like 90s 2000s right um and like you know these kind
(45:32):
of things they just kind of like came on the market like really recently where like it's kind of
taking the world by storm and I think it like I mean I had never heard of ozempic for uh
uh wasn't until like a like a year or two ago right and uh for him that's like really old technology
because it's like you know it takes a long time to like kind of go through the drug approval process
and all that stuff right so it's like a lot of effort um so like if he says that like hey that's
(45:57):
like old technology it's like these guys are working on like the the bleeding edge like if anybody's
gonna figure out how to like make uh humans live forever like Alex would be on my list he was a
really smart guy to talk to so I like you guys are in for a treat uh to listen to the interview uh
and for reference uh sorry like it was at the conference so maybe a little bit loud so the audio
(46:18):
quality isn't the best um so bear with us there but yeah I think like what Alex has to say uh
is really cool um but like what we're specifically gonna be talking about is they have this new
AI model called precious gpt so precious gpt is a thing as an AI model it's not like a regular
(46:39):
chap-out model it's not it's not a chap-bot it's a model that is trained on like medical stuff
like specifically medical stuff so like um like maybe like it'll be like a uh protein sequences or like
DNA methylation patterns or um I don't know what else what else what else is it trained on uh apparently
(47:04):
it can take a picture of a dog and predict its accurate age so it's it's a good uh age predictor um
across all the species which is crazy yeah and uh it like I'm not smart enough to tell you exactly
all the details about it like Alex is gonna mention it in the interview um but it does a lot of
(47:28):
cool stuff and I think that like you would be able to use this model if you want to actually like
build something health-related uh you might have to do a little diggin like you might have to like
go teach yourself a little bit uh Alex in the interview if you mention that you want to know about
like cellular pathways I think that's what it's called cellular pathways where like it's like oh
(47:48):
like this thing happens to the body which causes this cascade of events which causes this right so it's
like oh like I take um like vitamin D and then that you know does something and does something else
does something else and it's like oh well then it makes you live longer right but it's like you
know this causes this which causes this so it's like you know a lot of that stuff is like stuff that
you need to be able to to learn about and um this uh GPT model will be able to help you um
(48:17):
learn about that and then make like acro predictions because it's trained on a bunch of medical knowledge so
yeah I don't know I don't know anything else that you want to like mention about the model before
we get into the interview no I think that was a good summary I mean again uh neither of us are experts
in biology so a lot of it went over my head but uh it's it was fascinating to listen to him um and uh
(48:37):
in silico his company um uses AI in every step of the way um and uh we only touch the surface of
what they're doing um they also mentioned they use agents agents with like agentic behavior to solve a
lot of uh uh the common issues that startups have to run a business um but yeah the big highlight was
(48:59):
precious GPT and for context this is like uh my precious from Lord of the Rings where uh smegle is
trying to find this one ring which is supposed to be a unifier and uh like uh control all the different
rings of power so their their tagline is like one model to guide them one model to find all the targets
(49:22):
one model to align them and in the depth of data bind them precious GPT whoa okay so perfect so with
that uh we're gonna sign off we're gonna transition over to the uh the interview and then uh valedict
for the podcast today uh so cigarette off of that I think the interview is like maybe 15 minutes
(49:43):
I think you'll get a lot of value out of that, but until next time, enjoy!
Great. Hello everybody and welcome to the generative AI Meetup podcasts with your hosts Mark and Shashank
(00:09):
Together we run a meetup throughout all of Silicon Valley and
We have listeners from all over the world. Thank you Tony for listening. I appreciate you from South Korea
And we just want to share with you all the learnings
From people that we get to meet in the Bay Area. So today we have a really special episode and a ton of things to talk about
(00:36):
So this past weekend Shashank and I we went to the very first ever
Don't die summit hosted by Brian Johnson for those that don't know Brian Johnson's a guy who's
Basically trying to live forever and at the conference we learned a ton of cool stuff all about health and longevity and
Basically how a gen A.I. It may be able to help us live forever. So at the conference we were able to snag an interview with the
(01:03):
CEO of in silico
Which is an awesome company which is combining
Gen A.I. as well as like hardcore
Biomedical research and we got an interview with their CEO Alex
(01:23):
Veronica who is just a really smart guy and
They have a brand new A.I. model which they're going to announce called presses GPT
So he's gonna tell you a lot about it
But before we do that and tell you about the conference
There's a few things that we wanted to discuss today before we get to that so
(01:46):
We want before that we were gonna discuss about the brand new
GPT 401 model that just got announced earlier today
codename strawberry and then we're gonna talk a little bit about the Apple announcement
So Shashank, what do you think about this brand new model from open A.I.?
Yeah, I think this is the most exciting thing since GPT 40 was introduced because it seems to solve a lot of the
(02:13):
reasoning issues and limitations of LLM's today where they just
autocomplete a string of tokens or words without actually understanding what those words actually mean
So this one is
Apparently able to okay actually not apparently we tested it just now and it is actually able to count
(02:36):
How many ours there are in strawberry so for reference? That's actually a big deal
It is surprisingly hard and the reason it's hard is because
As you may or may not know these LLMs are really good next token predictors or the next word predictors
they look at these massive
(02:57):
repositories of data in the internet
bunch of tags bunch of newspaper articles Twitter posts blog posts and
just kind of based on what they've already seen predict what word might come next
Based on all the data that's been ingested in their neural network
(03:17):
So when you ask it a simple seemingly simple question that requires any kind of reasoning or math mathematics ability
It completely fails at that because it just
goes with what it has seen in
previous similar looking patch of text so in contrast this model
(03:41):
guess the answer right and
After testing it briefly it seems like it is
Underneath some kind of an agent which has a
Chain of thought or like a react framework where it reasons and then acts or breaks down the step into simpler
sub steps
(04:02):
Solves each one of those at a time and then combines all those intermediate steps into a final answer
Yeah, so for those who don't know chain of thought is an LLM prompting technique where you basically ask the LLM to think through its
thought process and
That tends to give you better responses so it seems like they kind of just bake this into the model
(04:24):
Actually, you know as an aside one thing that I was actually able to do in the past was even before this GPT
Oh one and that is a that's a mouthful of a name but what you could do is you could say like hey
Chatch B.T. write a program to tell how many Rs are in the letter or in the word strawberry and then
(04:46):
Run the program and then tell me how many Rs are in strawberry and then if you did that it would actually get the correct answer
So that was kind of way to sort of prompt it to do like
I guess you could call it like chain of thought but it's not quite chain of thought because it's like it would like
Run the program realize the answer was different than a thought and they would correct itself
So that that works
(05:07):
But you have to do a lot of like extra prompting technique to do this this seems like it's kind of like making it so you don't need to be a prompt wizard
It's just like does it for you which makes it I think like a lot smarter in general or like maybe like more user friendly
I guess you could say because it seems like especially until like the model isn't like way way better
(05:28):
It's just like almost like getting more
capability out of the same model
Yeah, absolutely I don't think they've significantly improved the performance of whatever base underlying model
That is powering this oh one model
But instead what they've done is give it this agentic behavior this ability to
(05:50):
Reason in multiple steps and then come up with a more cohesive answer. So in general
It seems like open AI went from being a research organization or like a
foundational model provider to
Incorporating all of the tools and add-ons that other people have built on top of open AI and integrated into
(06:13):
Their consumer product which is chat gbt because they've added the ability to browse the web which was you know
Something that Microsoft was doing with edge and Google is doing
And there were other plugins that built on top of chat gbt
But you had to use like llama index and all these frameworks that made it really complicated and
Only really allowed developers to build agents and other complicated
(06:38):
Applications on top of elements now open AI seems to just be baking all this into their consumer product
Yeah, so they're just kind of making it easier for everybody to use their lm which is cool
I applaud that it's amazing
But as a side effect their regular consumer version can go from solving
(06:59):
A reported 13% of math lmp add problems to a record 83% which is like in the 90
Not 89th percentile
Of competitors, which is really cool. How hard are math lmp add program or problems
I don't know because I vaguely remember when I was like in high school or something like that
(07:21):
And I think it was like just the smart kids who try to solve the math lmp add depends on what the curve is and what like
I mean it is the 90th percentile. So it's better than 90% of people. I mean that seems pretty good
But it's like I don't even think most people like what even enter the math
Attempt it
It's just like the people who are like really into math who are gonna even attempt the math lmp add so I feel like to say like oh
(07:44):
We're solving 90% of math lmp add problems. It's kind of like saying like oh like
We're smarter than 90% of the nerds. Yeah, it's very nice right because it's like it's not like
The regular person isn't taking the math lmp add
It's only like the people who like are really really into math or taking the math lmp and as I say like
People who do the math lmp add or nerds, but like you get my point. I do so to clarify it says it's better
(08:09):
Or it's reached the 89th percentile in code forces competitions whatever that means so out of
Some small subset of really smart people who are attempting this
Technical challenge. It's reached the 89th percentile. I mean, that's pretty cool. That's pretty cool
So in other
(08:30):
Completely unrelated news open AI is raising money
So apparently they are raising money at a 150 billion dollar valuation
I don't know it's like it seems to me like
It makes you feel like they maybe they don't need I
I don't know how to articulate this but
(08:52):
If they were super profitable, they wouldn't need to raise money and it seems like they must be running out of money
Therefore, they need to raise more so both that makes me I don't know a little bit bullish and then also a little bit bearish at the same time
Like bullish in a in so far that like hey like open AI can raise money at a hundred percent being dollar valuation
(09:13):
Even without a lot of money, but then it's kind of bearish at the same time because like man that really feels a little bit overvalued
I mean the other explanation would be that they need a lot of money to do some massive batch of training or
Build the custom chips or get into a completely new field like robotics and poor that heavy
(09:34):
investment upfront to work on a cool new project. So yeah, I mean
You but I think you might be right. I think it's probably a combination of a lot of things
I don't know if they're making as much money with their subscription model as they need to keep training these models or at least
Train the next generation of models. Yeah, you know, I actually just cancel my subscription. I think it was yesterday or two days ago to open AI
(09:59):
why
great question so I
I think I've mentioned Kagi in the past like k aji.com. It's a competitive Google search and they have their
uh, oh
Kagi assistant so the Kagi assistant is really cool
What it gives you is for $25 a month you get access to unlimited searches of Kagi
(10:20):
But in addition to that so Kagi by itself is $10 a month
But they also use the assistant so the assistant will be able to
Act as and it's kind of like it's similar to like Gemini and Google where it'll help you
Make sense of the search results so you could go and like summarize web pages. It can answer
(10:40):
questions about like a page and a search result, but it also gives you complete access to
GPT like chatship t like all the models like shbd40 gbd4 the gbdo4 forum mini
Gemini Gemini the Gemini advanced model
all the ontropic models so it gives you like
(11:02):
Anthropic clawed clawed 3.5 sonnet the opus model high coup
And the mistral model so like you get access to mistral large so
All of those would be $80 a month if you subscribe to them separately because Gemini's 20 bucks
Open AI's 20 bucks
(11:23):
And Theropik is 20 bucks and mistral I think is also $20
If you want to subscribe to it, but this for $25 you get all four plus paid search
So I'm just using that instead
How does that handle
Like persistent chat threads because one of the big reasons why I still use chat gbt today as a blister
(11:46):
Some of these other aggregators or platforms that utilize other elems is because because of the UX. I open up chat gbt
Open up like a previous chat thread that I've had about a specific topic. Let's say like
I have one for nutrition fitness. I have one for work. I have one for
Mental health. I have one for planning my day, etc
(12:08):
So having that simplified UX and a memory is pretty useful for me
Yeah, so I believe it does keep some chat history so I could go back to that if I wanted to
But it's be fair the UI or the UX isn't quite as good
So one thing that I've noticed so Kaggy if you're listening
I've noticed that when I put in a longer prompt sometimes the
(12:33):
Kaggy UI will just hang for a little bit
So sometimes I'm using it for like programming and I'll push it put in like a large chunk of code
And in that chunk of code
Kaggy might just hang well processes. I know exactly what it's doing
I think it's like trying to do some something smart but a little bit too smart and
(12:54):
I'll just have to wait for maybe a good like five seconds or something like that before I'm able to hit the enter button and get the prompt
It's not as ideal. I think there's some things like, you know, now I'm missing out on the GPT store
Kaggy can't generate images
And thrompe they have the what is it called like the snippets where you could get like the
The code results and it can run the code as well like I can't do that in Kaggy either
(13:18):
but you know it gives me access to all four and
It's saving me a little bit of money. So I think I was on the family plan with Kaggy
So I was paying $14 a month before for like me and my wife
We were both paying like seven bucks each so we're 14 plus we were paying
$20 to the GPT
Chat GPT subscription so that we were 34 now we're saving $9 a month down to 25
(13:42):
So it's like not big money, but still I was thinking like, yeah, well, you know
And I get access to all the models because also like I was finding myself using the anthropic model more and more because
I was
Trying to you know do some programming and some stuff and like I felt like chat GPT was like getting pre-tripped up
but
The anthropic models was actually giving me better results for my programming. So
(14:04):
It's like all right. Well, I'm not as well as use Kaggy for this. So we'll give it like a month
See how it is like I mean maybe I'll downgrade and go back to get myself a chat GPT subscription
But for now like I'm pretty happy with it. Yeah, I was considering something similar when I got the perplexity subscription
So I have a year of
Perplexity's subscription which is a
(14:26):
competitor to Google search. It's a search engine supercharged with AI and it
uses LLMs to search the web summarize the search results give you citations for
the answer that it
condenses and summarize and
There's like a copilot too. So if you have a complex enough query it like
(14:50):
searches and reasons and multiple steps and
You know, they also allow similar to Kaggy
They allow you to toggle between multiple different LMs from different providers and it actually seemed really cool
but
It has a little bit more latency obviously because it uses the web versus chat GPT which doesn't need to all the time
(15:11):
And the UX so I stuck with chat GPT
I mean I I use Gemini 2 I have a free subscription through work because I work at Google so
um
I use that with my corporate account and chat GPT with my personal account
Um, and I also have perplexity not to mention the odd other
(15:33):
Service that I use every now and then but don't pay for so I think both of us
Uh, if I have to kind of assess or
the bleeding edge of these technologies and our early adopters I would say for most people just sticking with
one
consumer facing product
Either chat GPT Gemini
(15:54):
Claude
Would be the best way to go because of the UX the simplicity and
Maintain like a chat history and all the other added bonuses like image generation that are baked into these platforms
Which unfortunately you don't get with these aggregators or
Apps built on top of these LMs
You know, that's true, but he's asking like you know, what I could my mom
I would say like you should just use the free version of chat GPT right because it's like
(16:18):
It's good enough. It's great. Yeah. Yeah, and uh, yeah, not even good enough. It's fantastic. So
I think that like you know, we're just paying for that little bit extra
But like you can get 80% of the value for free and it's still pretty darn good
Yeah, um, so anyways uh, Shashonak
What do you think about this uh new Apple announcement? Uh new iPhones. That's cool. Yeah, I was uh
(16:44):
I was looking at that. I was uh just moderately well-med
It wasn't that exciting um, I honestly couldn't
Pick any part of the announcement and say okay. This is really cool. The one thing that stood out was what the camera shutter
semi touch and semi physical button
(17:05):
They added a new button after like what 15 years or something after taking away buttons and controls for so many years
um, but I I guess
Uh for the sake of this podcast the relevant pieces of information are the fact that
These devices have much more capable chips which I assume is to power the next generation of um
(17:28):
local
genitive AI models their local LMS um
To do you know, emoji
generation to
generate text summarize and uh do other kinds of AI related features on device. So that's kind of cool
Yeah, I mean it's cool. I think apples bringing more
(17:51):
Computing to the edge. So when I say the edge, I just mean like on the device as opposed to like hitting a server somewhere
So that's cool. What are the benefits of that?
Well, I think that there's a few benefits that you could get to like edge computing in general like writing local advice
So I think the biggest the was two big ones
Number one would be privacy
(18:11):
So it's like if you can run things locally in your device and you don't have to hit a server well then
Now you don't have to tell Apple or opening eye or whoever like your personal information
Now I'm sure they're not like looking and it's not like there's like some guy just like oh
What is it what is just shank do you it's like oh it's like oh it's uh tell me more
(18:33):
It's not like they're looking at your chat history
But I'm sure they're not but the thing is is like they could
um like maybe you'd like run through some extra
Things and then like you know one thing leads to another and then like there's data breach and then like your chat history out there
If that happens
(18:55):
You know you're out of luck uh it I'm not saying it I'm not saying like any of these companies are doing that
statistically these large
mega corporations like apple google amazon meta etc are
Very very unlikely to get a data breach. It's mostly smaller companies
um
Like I don't know ticket master or eventbrite or
(19:18):
Instacard that may have data breaches
We're not saying that those companies specifically have data breaches. We're just saying that like in general like uh
I think they oh did they have data breaches off the top of my head, you know, I might be wrong, but um
Yeah, again. Yeah, let's go with what you said. Yeah. Yeah. So so anyways
Allegedly they may have had a data breach. I have no idea. So don't sue us ticket masters
(19:41):
Uh
Ventbrite and Instacard yeah, so yeah, don't don't sue us. We don't know we don't know anything about that
Um that hasn't been published. I'm not sure. But anyways
Like the point is is that like uh
When it comes to like privacy if you can do it on the device and you never have to hit a server
(20:01):
You just eliminate that threat factor, right? So it's like now like you don't have to worry about them
Securing their servers. You don't have to worry about like some like a rogue engineer looking at your uh chat history
Like it's just on the device. It's all there. It's great and since it doesn't have to hit a server
The other benefit is it's fast. So it's like you can get a response immediately
(20:24):
Um, and then also like it starts becoming uh
Lower cost as well, right? Because like now it's like you have higher privacy
You have
Faster speed and then because you have this speed uh you start opening up like use cases that you wouldn't have
Mayable to do before so for example like maybe you could use for example generative AI to
(20:46):
do personal
Search engine on your phone right or like uh some sort of personal rag
Uh, so like rag like you guys know like retrieval augment the generation just like say like hey like I want to go search the documents that I have
And then get some summary response back
So like that's the type of thing that I could do where it like would it be feasible with like calling open a eyes
(21:09):
API or something
Yeah, maybe but like now I could do that all in my own like personal documents like maybe there's some things you don't want
hitting the server right like maybe it's like I have my bank account details like the health records or social security numbers
You know all that stuff like now I can go and just uh do that all
locally and that's pretty darn cool if you ask me now the downside of doing this is like well now we
(21:34):
If it all runs locally on an iPhone it's like well
A lot of these models are running on like really really expensive GPUs and high quality hardware
So it's like the quality may go down a little bit, but it's like you know
What do you want you can't have or you can't have like the highest quality and also be like super private and secure
(21:54):
Um and fast and cheap. It's like you gotta you gotta give somewhere
So it's like you're gonna sacrifice a little on quality, but like the quality is still not that bad
It's a pretty good until we get better every day
Yeah, the part about cost also makes sense because uh right now all of these companies
You know opening I raising like a hundred and fifty billion dollars or
raising six billion dollars at like a hundred fifty billion dollar valuation
(22:18):
It's raising a lot of money because it costs a lot of money to trade these models deploy them and keep serving them to users for free
So um if you don't need to do that you can run for free locally
albeit with a smaller model
Yeah exactly so I mean that that's pretty cool
And I think Apple's really kind of like
(22:38):
Leading the charge on that. I know there's like a few other companies like Google's doing some stuff
Google's been like
shoving AI in people's faces for like two years
Well, I mean like sure, but like I think the thing is is like I I agree with you
But like Google has like a much smaller like market share than the iPhones fair
So it's like I mean like uh in mobile devices. Yes. Yeah, so it's like I mean like I guess you can argue like Google's like
(23:05):
Leading the charge but like apples like
Bringing ever one along with them. Yeah, um that's a good point
But these uh this new Apple intelligence feature which still hasn't launched
Good point to be clear even if you buy the new iPhone 16 or 16 pro
It's not gonna work today. It's tbd when it's gonna be available and
(23:30):
It's not gonna be available for all the older models of iPhones that people already have
So you're going to have an uh have to upgrade to the newer models
unless if you have the
iPhone 15 pro with the
previous
Pro chip so it is still a relatively small segment of users who are
(23:51):
Looking to upgrade. Maybe maybe this gives them a good enough reason to upgrade
Well actually not that you mentioned it. I don't know like uh, I don't know. I feel about I mean
I guess it's okay that Apple is going to announce like a phone with the features that aren't there yet
But I don't think that like any of our listeners should just like go rush out to buy an iPhone with something that
(24:13):
Has a promise of being there so like
Apple says the features can be really good. I mean knowing Apple's track record probably will be really good
But like at the same time it's like
Until like it's been in the hands of reviewers and then we know it's really good like
This is like a feature that like will most likely ship most likely ship soon
(24:33):
but like
We don't actually know like how good it will be like I mean if the feature was really really good and ready
They would have shipped it yesterday right but they didn't and the fact that it only will run on the highest end chips
Makes me think that like well
Maybe it's going to be using a lot of device power so like I don't know like they said that they're going to have a new battery in these iPhones like
(24:58):
Is using these gen i features is it gonna like drain the battery life is it gonna be like high latency like
I'm not gonna be able to use the phone normally like I don't know like
I hope it's gonna really good but like until like it's actually in the hands of the reviewers like
We don't know and that's the thing so it's like
I like I appreciate like Apple like going out and like announcing this I think they announced this like
(25:21):
All these new AI features like what was it like in wwdc? It was like a few months ago, right? Yeah
And like it's still not here yet. So it's like all the people who like want to bought the phone hoping to get these features and then
Now they're not getting except in like the newest phones. It's like I don't know. It's like it kind of
Rows me the wrong way a little bit. Yeah, and Apple's defense. I think the hardware is gonna be solid
(25:44):
The reported benchmarks with like let's see some
Some double digit percentage better CPU GPU apparently the
Okay, this is a back comparison
Okay, it's 30% faster CPU than the last generation of iPhones. It's the GPU is 40% compared to last generation
the
(26:05):
The batteries are supposed to be physically bigger too and they've done a bunch of
optimizations to make sure it runs faster
So all of those things will result in a much more capable iPhone that should hopefully be able to run these models because like Google is able to run
Uh some of their smaller models on device and that's been working fine and I
(26:26):
I think we kind of have a rough idea of
What these features will look like because we have local LM's running maybe like not on a phone
But on some of these laptops and uh, we've been testing them out ourselves for a while
So we can extrapolate what the performance of some of those will be and given
Some of the performance benchmarks of the new next gen iPhones. I feel like they're very capable of running these text models
(26:51):
some simple image generation models and
The the part that they may struggle to get right is the UX
because like the the key
differentiator of Apple intelligence is the fact that is running on your iPhone on device and able to
Tap into all of the different apps that are running on your phone. So look at your calendar look at your email
(27:17):
Of course with privacy in mind
running locally
And as soon as you have a new piece of mail or message come in it's able to look at everything else that already exists on your phone
And like do something more meaningful more intelligent maybe
Someone says oh, I'm gonna be late uh for dinner tonight. Then it's gonna you know shoot a text to the restaurant or something
(27:41):
Ask them to push the reservation later or
Send a note to your mom telling them that oh take care of the kids a little longer because we're gonna be late and so doing all those things
Cohesively, I think
To build a good story around that for a good user experience. That's gonna be the challenge, I think yeah, I agree
I agree
(28:01):
Um, so we'll see time will tell I guess in the next I don't know exactly what's coming down next few months probably I hope
At some point in the future do you have the next generation iPhone? No, I have an iPhone 13 13
Yeah, that is never gonna get any any of the Apple intelligence features unfortunately not so
Oh, well such as life
(28:23):
The phone's still working okay, so we'll keep using it so uh yeah
Now I guess it's time to talk about the don't die summit so yeah that that was a pretty fun event
That was really interesting. Yeah, that was the first uh
Health wellness longevity related conference that I've been to
I wasn't sure what to expect, but that was that was interesting. Yeah, so for reference
(28:48):
Shashank and I this past Sunday, so uh, what was that today's Thursday? So five days ago
What's the day's date? I forget. I don't have my
September 12th September 12th, so whatever was September 8th September 9th something like that we
drove from the south bay
Uh, old way to San Francisco. That's 6 a.m
(29:08):
Well, yeah, I picked up Shashank in his house at 6 a.m
and
He was up kind of late the night before dancing
and
He got enough sleep
To go wake up and we drove together
To the don't die summit which is fantastic. It was so much fun. So um
(29:29):
It was this conference at pier 35 in
San Francisco, so for those who've never been to San Francisco
um
Pier 35 is
Rate along the San Francisco bay
And uh, you can basically just like rate along the water and it's super beautiful
Uh, like this whole area um, it was like a
(29:52):
Kind of a sunny day. It was it was pretty nice and as we
Uh, we're walking in we arrived at pier 35. I remember at
703 a.m
I remember because I was worried we were a couple minutes late because that they were starting at they were gonna have a
Don't die rave. So you know as an aside like why do I want to like if I'm gonna like dance and like party and like rave
(30:18):
Like why do I want to do that at like 2 a.m
They're like midnight like that's gonna completely mess up your sleep quality
It's like let's do it at them in the morning like when we're fresh
We're feeling good. You can get like your heart rate pumping
Uh, it's like it's like a morning workout kind of it's like so much energy so much fun like I for one
(30:41):
And a massive fan of the morning rave like let's let's do that. That's a hill I will die on
7 a.m. Of a rave is much better than midnight because for the record I'm not uh with this but continue
Because of midnight. I'm tired, but at 7 a.m. I am ready to go so
With that being said, we we went in and uh we uh saw
(31:06):
Brian Johnson, uh, who is this guy who uh it was a guy who started uh, I think it was brain tree
And then we later got bought up by Venmo
He brain tree bought Venmo. What was that what it was? He acquired Venmo and then sold it to uh
Was it paid out? Okay, so anyways
Uh
He is now super rich. He's I think he sold that did that deal was like over a billion
(31:32):
Hundreds of millions for sure or over a hundred million and close to a billion. Was it really close to a billion? Wow
So anyways, he's doing okay and now he's spending all of his time just like trying to
not die and uh
really help uh
Spread the word on
longevity and just like a good solid quality life so
(31:56):
We saw him uh going in
Uh got the chat with him like really briefly. I also talked to his like dad and son both like great people
And we met a lot of really cool people at the conference um so
I don't know shishan, what were you? So many like your takeaways from the conference. Yeah um so like I said
I didn't know what to expect um
(32:17):
This is the first time uh experiencing something like this and again neither of us has a background in biology
Let alone biotechnology and the cutting edge of uh, you know
uh
All of these multi-omic studies of like proteomics and uh genomics and all the other omics that deal with different parts of human
(32:38):
physiology. Sorry. What's what's an omic?
Uh for my understanding it is like uh
It's a thing that uh it's a suffix that represents part of human uh or biological
information so
proteomics is the study of proteins genomics is a study of genes
Um and there's a bunch of other omics that uh study epigenetics which is like uh the environmental factors to your DNA
(33:04):
um the effects on uh RNA which are like gene expressions and and a bunch a couple other
items i think they're like five or six of them in total huh
So different the different things that affect human biology and uh not just human biology just all of life
um
So it was definitely a lot of information to ingest and
(33:27):
But but to kind of summarize i think uh we talked about this a little bit uh the conference
I think left us both maybe i can talk to myself left me a little pessimistic about the current state of research and
supplements and uh other
quick hacks to
(33:47):
a longer healthier life
It seems like there really is nothing significantly better than focusing on the basics
getting a good night's sleep focusing on healthy diet nutrition
um getting a lot of exercise and building good uh social connections that allow you to
maintain a healthy mental uh balance so
(34:12):
You know we we saw everything from um
uh the the speaker that uh we're gonna we're gonna plug in soon uh the interview with uh
Alex Zoverunkov um and how they use cutting edge uh deep learning and apply that to the field of
(34:33):
biology and uh pharmacology to uh come up with drugs and target them for curing specific ailments and
diseases uh but we also saw what was it uh companies selling uh thousands of dollars worth of
(34:54):
little five milliliters tubes filled with human umbilical cord uh extract uh apparently make
your skin glow and uh make you look more younger so there was like a wide range of speakers from
what i consider deep science and uh researchers put it pushing the field forward to
(35:20):
ostensibly like money making schemes trying to take advantage of the rich who don't really know
what the difference is between uh these different products uh there was a speaker on uh you know
who had an extensive background on psychedelics it was fascinating to talk with him
that was a cool speech actually yeah to understand the impact of micro dosing because you know
(35:41):
what what are what are drugs we arbitrarily classify uh a couple of these substances a
scheduled one restricted uh substances and it's kind of arbitrary because on the other hand we have
alcohol which is legal to buy which is far more damaging to your body mind and society at large
compared to stuff like um MDMA or ketamine which uh might get a bad rep but has been shown to
(36:08):
have a lot of benefits for uh therapy and getting over traumatic events and curing PTSD
in a controlled setting of course and so yeah it was a wide range of speakers um had a lot of
information still kind of uh going back and trying to understand what we learned trying to digest
(36:29):
all this information um there were a lot of vendors which were really insightful we got to talk to
these CEOs founders and researchers at these amazing companies and learn a lot about
hair regrowth all the different techniques from uh vassal dilators that improve blood flow through your
(36:50):
head uh uh things that block uh DHT which causes hair loss to laser caps and micronealing to create
punctures in your uh skin to promote skin regrowth many serums and so many uh different pieces of
information yeah so one thing that I I thought was really cool that was a takeaway from me now I
(37:11):
think shashanker didn't knew this because like uh you know shashanker's just like a guy who's like
really interested in a lot of these uh things especially for like hair growth but one thing that I
have I have a history of hair loss in my family so it's something that I've been thinking about a lot
yeah so anyways like shashank has been you know doing some research and like learning but like
there's some stuff that like I had no idea but like there is something that you guys can do
(37:36):
for free um that zero money like zero cost health intervention now don't listen to us we're not
doctors but this is really cool so apparently a lot of these like hair loss uh treatments what they do
is they put uh increased blood flow to the top of your head uh so like if you have increased
(37:56):
blood flow to your hair um or like your head you're gonna get like thicker more luxurious hair
that's less likely to fall out and just like overall better quality hair so like uh there's some
this thing called like medoxinol which like goes and increases blood flow to your head um like
shashanker mentioned microneedling which is where you take like little tiny like needles and like
(38:18):
poke into your head and then like uh it makes like thicker skin and then like because of that you can
get like you know stronger skin and then like backstension stronger hair and then also like you can
use like little needles or like holes in your head that you just like put in there and then you can
like uh if you can put like some sort of like hair growth product so you can put like uh medoxinol
which is like the thing that I mentioned before but like that's all great but you don't have to do
(38:44):
any of that the only you have to do is just like when you're gonna wash your hair in the shower
just like scrub really hard uh like uh just like you know use your fingers get in there and then
just like really like get in there if you know what I'm saying just like really like uh scrub the heart
the top of your head like hard like not like so hard that you like hair falls out but like you know
(39:08):
like scrub like harder than you normally would um I've been doing that for like the past couple days
and I felt like my hair is like never bed better uh I was like man I didn't use shampoo uh I just
like scrubbed it and I think they call like a hot water shower or something like a hot water
shampoo um you just like scrub it and like it increased the blood flow to my head like I was like
I woke up the next day was like my hair looks fantastic so uh yeah guys uh it does wonders just like
(39:34):
any other shower just like really like apply that pressure to your head and like it will help you
get like much fuller hair now I got an inter-intro check right there so this is a sample size of
end of one uh on you know his marks unique genetics and everything but you you no matter what you
should not uh be aggressive on any part of your body I don't think you should uh use your fingernails
(39:58):
or uh scratch anything or uh rub uh too aggressively but like a massage I think does improve blood flow
and that's yeah yeah you don't want to like bleed okay you know that finger nails yeah like
she shocks at it better like a massage like massage the top of your head um like I even got like one
(40:22):
of those like uh like things like a little comb like a shower comb where it's kind of like silicone
it's got like little silicone like little nubs coming out of it so I could use that to apply some
additional pressure to the scalp and I did that and I feel like my hair looked like pretty good so
yeah no bleeding it didn't hurt just like you know a little extra pressure just to stimulate the
blood for a little bit yeah and uh bringing this back to the theme of the podcast and our meetup
(40:47):
I think uh one of the cool things that we I noticed in general for all of these vendors and uh
speakers is the fact that they're just bringing more data and uh analytics to all of these data
points that give us an insight into how these different uh uh remedies are affecting us so with
(41:09):
hair loss you know no one can really tell by looking uh is is this treatment significantly working
like 10 20% is is very difficult to track also like the duration uh of efficacy is really long because
hair has to go through these different cycles of growth uh where it's in like a resting phase and then
(41:30):
it has to uh fall out and then you create a new follicle that starts growing from the base of your
skin and then slowly emerges so like these things are really hard to track and one company I forget the
exact name maybe I can uh add it in the show notes later they take regular pictures of your skull
which I'm sure dermatologists and um in clinic hairdoctors have been doing for a while um but they're
(41:56):
bringing this to like uh like an at home kind of a service uh coming soon uh but the ability to
take pictures of your hair different patches of it identify how many hair follicles per unit area
you have what is the hair density what is the hair thickness um and that chart thought over time and uh
(42:18):
figure out what different remedies are doing to improve these different uh data points is like
is that's really the way to go I think uh traditional remedies of just like you know
feel it out uh if it feels good then then it's great keep doing it I don't think we're good judges of
these kinds of um my new details that the human brain isn't very good at uh not identifying and
(42:43):
how long similar lines I think we saw really cool companies that uh bring AI uh to understand your
genome so I took a 23-in-me test a long time ago but um apparently the 23-in-me test only sequences
maybe like 1% of your genome whereas some of these new companies that have come out um mostly
(43:05):
because of the advances in gene sequencing have been able to sequence maybe like 99.9% of the genome
or something uh or close to 100% and they're applying a lot of uh big data using um large amounts of
other people's uh existing ailments and um kind of classify or group different uh gene segments
(43:33):
together and give you uh more insights about what you are uh more at risk for so we tried uh
we tried signing up for one of those new services and I guess uh we'll find out in a few weeks what
Mark and I are predisposed to and maybe share that with the listeners maybe it depends on like
what it is but yeah I mean I'm I'm open I'm open I'm not open to share yeah um we paid to get our entire
(44:00):
genome sequenced uh so uh we'll give them a plug nucleus uh we use them uh I don't know like if they're
like necessarily better than your body else but they gave us I think like 10% off or something for
signing up at the conference so I had always wanted my whole genome sequence that mean the human genome
project uh cost I think over a billion dollars and took like 20 years so now that we can get that
(44:21):
400 bucks that's pretty cool that is pretty cool um so anyways yeah uh now to the main interview
uh because we are running out of time but um yeah so a little bit of uh background on um like we're
going to be talking about it in the interview so um this company uh in silico they do a lot of like
(44:48):
deep science and are kind of on the cutting edge of uh what what it comes to uh like health and
longevity and coming out with new medicine so uh for reference like uh when we watch the talk uh
their CEO Alex he mentioned that uh these G uh what is it GLP1 inhibitors uh which is like
(45:10):
ozempic it's like GLP1 agonist agonist actually uh enable oh my bad not inhibitors but the agonist
so it's like you know these GLP1 things he was mentioning like oh yeah that's kind of like old tech
like that was like 20 years ago like that was like 90s 2000s right um and like you know these kind
(45:32):
of things they just kind of like came on the market like really recently where like it's kind of
taking the world by storm and I think it like I mean I had never heard of ozempic for uh
uh wasn't until like a like a year or two ago right and uh for him that's like really old technology
because it's like you know it takes a long time to like kind of go through the drug approval process
and all that stuff right so it's like a lot of effort um so like if he says that like hey that's
(45:57):
like old technology it's like these guys are working on like the the bleeding edge like if anybody's
gonna figure out how to like make uh humans live forever like Alex would be on my list he was a
really smart guy to talk to so I like you guys are in for a treat uh to listen to the interview uh
and for reference uh sorry like it was at the conference so maybe a little bit loud so the audio
(46:18):
quality isn't the best um so bear with us there but yeah I think like what Alex has to say uh
is really cool um but like what we're specifically gonna be talking about is they have this new
AI model called precious gpt so precious gpt is a thing as an AI model it's not like a regular
(46:39):
chap-out model it's not it's not a chap-bot it's a model that is trained on like medical stuff
like specifically medical stuff so like um like maybe like it'll be like a uh protein sequences or like
DNA methylation patterns or um I don't know what else what else what else is it trained on uh apparently
(47:04):
it can take a picture of a dog and predict its accurate age so it's it's a good uh age predictor um
across all the species which is crazy yeah and uh it like I'm not smart enough to tell you exactly
all the details about it like Alex is gonna mention it in the interview um but it does a lot of
(47:28):
cool stuff and I think that like you would be able to use this model if you want to actually like
build something health-related uh you might have to do a little diggin like you might have to like
go teach yourself a little bit uh Alex in the interview if you mention that you want to know about
like cellular pathways I think that's what it's called cellular pathways where like it's like oh
(47:48):
like this thing happens to the body which causes this cascade of events which causes this right so it's
like oh like I take um like vitamin D and then that you know does something and does something else
does something else and it's like oh well then it makes you live longer right but it's like you
know this causes this which causes this so it's like you know a lot of that stuff is like stuff that
you need to be able to to learn about and um this uh GPT model will be able to help you um
(48:17):
learn about that and then make like acro predictions because it's trained on a bunch of medical knowledge so
yeah I don't know I don't know anything else that you want to like mention about the model before
we get into the interview no I think that was a good summary I mean again uh neither of us are experts
in biology so a lot of it went over my head but uh it's it was fascinating to listen to him um and uh
(48:37):
in silico his company um uses AI in every step of the way um and uh we only touch the surface of
what they're doing um they also mentioned they use agents agents with like agentic behavior to solve a
lot of uh uh the common issues that startups have to run a business um but yeah the big highlight was
(48:59):
precious GPT and for context this is like uh my precious from Lord of the Rings where uh smegle is
trying to find this one ring which is supposed to be a unifier and uh like uh control all the different
rings of power so their their tagline is like one model to guide them one model to find all the targets
(49:22):
one model to align them and in the depth of data bind them precious GPT whoa okay so perfect so with
that uh we're gonna sign off we're gonna transition over to the uh the interview and then uh valedict
for the podcast today uh so cigarette off of that I think the interview is like maybe 15 minutes
(49:43):
I think you'll get a lot of value out of that, but until next time, enjoy!
Popular Podcasts
Las Culturistas with Matt Rogers and Bowen Yang
Ding dong! Join your culture consultants, Matt Rogers and Bowen Yang, on an unforgettable journey into the beating heart of CULTURE. Alongside sizzling special guests, they GET INTO the hottest pop-culture moments of the day and the formative cultural experiences that turned them into Culturistas. Produced by the Big Money Players Network and iHeartRadio.
On Purpose with Jay Shetty
I’m Jay Shetty host of On Purpose the worlds #1 Mental Health podcast and I’m so grateful you found us. I started this podcast 5 years ago to invite you into conversations and workshops that are designed to help make you happier, healthier and more healed. I believe that when you (yes you) feel seen, heard and understood you’re able to deal with relationship struggles, work challenges and life’s ups and downs with more ease and grace. I interview experts, celebrities, thought leaders and athletes so that we can grow our mindset, build better habits and uncover a side of them we’ve never seen before. New episodes every Monday and Friday. Your support means the world to me and I don’t take it for granted — click the follow button and leave a review to help us spread the love with On Purpose. I can’t wait for you to listen to your first or 500th episode!
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com