Ep 6: AI's Impact: Data Privacy, Finance, Chatbots & More | AI Insider

‍

Welcome to Episode 6 of AI Insider! In this podcast, we delve into the broad and fascinating applications of AI across various fields, spotlighting how technology is shaping our world.

‍

First, we unpack the newly minted EU-U.S. Data Privacy Framework, discussing its implications for transatlantic data transfer and the criticism it's faced. How will this new agreement impact personal data security on a global scale?

‍

Next, join us as we explore groundbreaking research by Cornell University, utilizing AI and news data to predict financial returns. How is this shaping the finance industry?

‍

We then discuss the much-anticipated debut of Claude 2, an advanced AI chatbot by startup Anthropic. We examine the promises, challenges, and the future of AI in conversation.

‍

Our exploration continues with a look at how companies like Nvidia, Google DeepMind, and Huawei are transforming weather forecasting through AI. We delve into the breakthroughs, challenges, and controversies.

‍

Finally, we shine a light on the AI-assisted breakthrough from MIT and McMaster University in fighting superbugs. How could this shape the future of healthcare?

‍

Don't forget to hit the like button, share, and subscribe for more episodes exploring how technology is reshaping our world. Drop a comment to let us know your thoughts on these topics or suggest new ones for future episodes. Stay curious and stay connected!

‍

Follow along with the full transcript below:

‍

00:01.23

zenlytic

What is up guys so we are back at it again Wednesday July eleventh for the top 5 in Ai and data. Ah, we're coming in with pretty pretty nice topics today. Um, we can jump right into it if you guys are ready um, the first. The first juicy piece of information we got for today for the news is the the new eu us data privacy framework um, for transulated data transfers. Um amid criticism. Um. You guys want to kind of jump into this and what are your thoughts on everything ball I know you you just read up on it so you're you're finding out a lot bit about this as well. So either. You guys want to jump in and and let us know your thoughts.

00:49.79

Paul Blankley

Yeah, so think I think this has the opportunity to be something that's really good. Um, so I'll start positive and then I'll add one caveat at the end. It has yeah an opportunity to be something that's really good because right now moving data out of the Eu is sort of a complicated and.

00:56.22

zenlytic

Um.

01:07.77

Paul Blankley

It's unfortunate that it's complicated because that really only benefits incumbents that only benefits companies like Facebook and Google who can afford a lot of lawyers to deal with this kind of stuff smaller companies trying to compete and make better products for european consumers um have to deal with a lot of this. Ah. This is just extra work basically and that makes their operating costs more expensive so anything that makes that process easier and allows smaller companies to be more competitive is very positive and in my mind so I hope this does that it does look like it has some green shoots at actually simplifying the process which is awesome. The the caveat is usually when the eu introduces new legislation. It doesn't make things easier so it's got green shoots I hope it goes well historically speaking usually when they introduce new stuff. It makes it more complicated, not less.

01:57.90

Ryan _Zenlytic_

Yeah, that that ah Paul just doesn't know back. But yeah, the name of the game here is like the core issue is regulatory capture right? and like ah what what? I mean by that is like.

02:00.50

zenlytic

Ouch.

02:03.42

zenlytic

Now.

02:13.00

Ryan _Zenlytic_

Ah, the big incumbents like Google or Facebook actually want to encourage greater regulation because they have the resources to navigate it and it locks out the smaller sort of you know challengers from the market and more regulation leads to more regulatory capture in many cases you know good examples of this have been like ah the us. Spend a bunch even in the us to spend a bunch of time with sort of large banks and sort of financial regulation usually benefits the large banks more than the challengers for instance, um, the hard. The hard part regulation is that ah generally the people who are being regulated. The people trying to get that regulatory capture have. More resources and stronger incentives than the people who are making the regulation right? So like in that kind of like you know, constant like back and forth quite often. The regulatory capturers. The people like the the big incumbents ah can can leverage that and they they they tend to end out. On the the right side of that equation and they usually get sort of like the the better deal in the long run. Ah, and this is actually very very relevant for what's happening in Ai right now we're already starting to see people are sort of agitating regulatory capture and you know Sam Moman was just sort of ah. Know in favor of regulating ai and the last congressional hearing and things like that that could have been altruistic or it could have been you know realizing that if if governments can regulate this and then only 4 or 5 companies have a license to make artificial intelligence that would be very very good for openai. So it's it's relevant in the data industry and relevant in the related.

03:43.63

Ryan _Zenlytic_

Artificial intelligence industry as well. And yeah, curious to see how it plays out.

03:49.80

zenlytic

Yeah, definitely definitely curious to see how it plays out always found I think I agree with you Paul here where more is not always better. So you know more regulation is not always the best thing for a situation like this. Um. Sweet. We're going to run on to number 2 you guys have some hot takes on this already that I can tell ah Cornell researchers have developed a machine learning framework that uses news articles to predict financial turns more accurately than traditional models. This is awesome to me I'm super pumped about this. You guys are just like. We've been doing this forever Colin you're an idiot. What are your thoughts.

04:27.68

Paul Blankley

I think I think this this goes back to one sort of classic question which you can see lots of people asking this question on Reddit twitter all of these kind of sites someone will pop in and they'll be like hey I just I just like saw this news article you know, do you think this is this is priced in it's priced in.

04:45.73

zenlytic

Um, okay, it's always price in it's always priceded.

04:46.66

Paul Blankley

I just saw this other thing. It's priced in um, you know my grandmother just bought this product is that priced in yeah, your grandmother buying this product was priced in five years ago like you know everything is priced in and the problem is if you're coming in to you know.

04:46.83

Ryan _Zenlytic_

Um, it's always pressed it.

04:56.62

zenlytic

Um, okay, okay.

05:02.90

Paul Blankley

Trading like this with just 1 strategy of like hey we're going to read some news articles and figure out what's going on. You're going to get your lunch eaten by the the real pros who do this all the time and they are looking at way more data streams than you possibly can and I'm talking about citadel.

05:13.79

zenlytic

First.

05:18.87

Paul Blankley

You know renaissance like the really really advanced players in this space if you can't spot who in this is a poker thing that Ryan you know I'm sure we'll talk about too. But it's like if you can't spot the idiot at the table. It's you, you're the idiot at the table. So um I would be really cautious of anyone trying to.

05:21.28

zenlytic

Right? if.

05:33.68

Ryan _Zenlytic_

Yeah I was.

05:38.49

Paul Blankley

Hop in here and think Wow I'm going to get edge on the market and have this data That's not properly priced in it's It's already priced in.

05:42.89

zenlytic

Okay.

05:45.86

Ryan _Zenlytic_

Yeah, no, it's all it's funny because I mean people had been using sentiment analysis in some far another for for 20 years now so ah you know first I'm skeptical. This is a meaningfully better. It might be with the advent of Ls Ah but even if this does work now. It's not going to work for very long and. So I've I've done some work with hedge funds before with quant funds before I've I've advised a couple of funds before and there's this really interesting thing about engineers who move from tech into hedge funds and like ah you know they always have this great idea for how they're going to model this and the great trading strategy or whatever but they completely underestimate. How different that environment is and once you move into capital markets. Ah you are basically in a hostile place where everyone is trying to take advantage of that algorithm and erode every little gain that you have earned with your novel technology right? You're not jumping into the blue ocean and you're jumping into this ocean full of sharks who are trying to exploit every weakness in the model.

06:39.19

zenlytic

Right.

06:41.20

Ryan _Zenlytic_

So you know and that manifests itself in several ways. First it's often something like oh this worked in academic research but in the real world for some reason it doesn't work anymore like why you know it worked but I did out of the back testing and then it actually turns out that doesn't work or it works for a very very short period of time and then quickly disappears. So. Ah, because this is a competitive process. You have to be constantly evolving constantly 1 step ahead and constantly you know both avoiding exploitation and also exploiting those other you know quant traders weaknesses in in their own algorithms. So it's it's not a static thing and people underestimate the complexity of that. So. I think that even if this on the off chance this does work in the real world. These things are never sustainable. You know it's going to have to be another better better model after that.

07:25.86

zenlytic

Um I guess um I Guess the biggest thing that I'm curious about and you guys could totally debate Why this would not theoretically work would be if you were instead of.

07:40.21

zenlytic

Training it on news articles and new stuff that come out. You basically train it against like Tiktok mentions or like something along that so you are training it against social mentions for a really aggressive platform like Tiktok where it has potential. Go viral say like. american eagle for example they start to go really viral on Tiktok and then it just becomes this thing but you have this bot in place that is trained on total mentions and can see spikes and irregularities and then could potentially tell you to move into the position. Do you guys think that's not. Good or are you're going to say it's already priced in.

08:19.84

Paul Blankley

Ah, well I'm going to say like that if that does work which it might like maybe maybe hedge funds don't really have a great data feed into you know what? social actually looks like for a lot of these companies in which case it might work for a short period of time. But then. Um, talking days that data feed will be in all major hedge funds and that advantage will be will disappear because everyone will be incorporating that.

08:41.13

Ryan _Zenlytic_

So yeah I mean what's what's going to happen that situation is let's say you're the very first person to notice that relationship. Let's say that you see you know a handful of influencers that talk about something and that causes an increase in the stock price and you you are. In fact, the very first person to do that. First you're going to start making trades on that and that'll be great. Yeah, it'll start working a little bit but very quickly other people will also notice that you know someone's trading in advance and will'll find out the pattern that you're using. They're going to jump on the same trades they might even just buy a lot of alls start following you directly if you start you know buying Ibm when the influence is talking about Ibm they'll start buying it when they see you buying it. That narrows down the margins right people when they buy they push with the price and you're making less and less on each trade so then more and more people do that then it's go to become a question of speed right? and it's going to be okay, so the tickt to influencers just announced this how fast can we be the first one out to gate and there's going to be some hedge fund that has. You know, ah, 13000 fiber lines that are literally three meters away from the Chicago Mercantile Exchange and it becomes a question of literally like for for some of this high-f frequencyency stuff. It's like who is closer to the exchange like how how who is closer for a light beam to the exchange you know and they're literally buying offices next doors they can go through a hole on the wall.

09:45.90

zenlytic

Um, yeah.

09:55.49

Ryan _Zenlytic_

And get there a few naniseconds faster. So like it becomes very very competitive with regards to speed and and you're going to lose that edge of the guy who's sitting there watching this Tiktok right.

10:06.56

zenlytic

Wow! Absolute absolute chaos for the little guy right now not looking good. Um, we'll move on to number 3 not financial ways.

10:13.99

Ryan _Zenlytic_

Um, not not financial advice by the way you know Twitter here tickokck if you are but you know like yeah I think that these advantages tend to quickly erode.

10:20.88

Paul Blankley

As a little guy you probably should not be trying to do high frequency trading in the first place that would be my personal take on the matter.

10:22.57

zenlytic

Okay, that's that's also a really valid point to make as well. Um, so we'll move on to number 3 so anthropic joins. Ai chat competition with the launch of cloud 2 what do you guys? think.

10:41.53

Paul Blankley

Super exciting. We've been testing out claud to already and I'll say it feels. It's early days. They just launched it yesterday. It's very exciting right now. The scale of the improvement feels to me similar to. Ah, three point 5 to 4 kind of jumps some of the problems we had with three point five we had with claud one. Um, claud two has fixed some of those issues and that's super exciting. This is just you know another step up in capabilities from another provider who's not open Ai so just having more diversity in. Really really high caliber models that are available is super exciting. So and I just want to see more of this. You know I want to see more competitors being able to enter the market I want to see Google's offering getting up to this you know level and that is just really exciting like the more options we have the more different platforms that exist that you can use.

11:37.84

Ryan _Zenlytic_

Yeah, um, there's there's lots of different ways to evaluate llms right? There's there's different dimensions to evaluate them on a lot of people focus on you know how? well they do on the S sats or something like that to measure comprehension for instance, some use human judgment. It's like which produced a better output right? ah.

11:37.97

zenlytic

Yeah.

11:38.17

Paul Blankley

Ah, the more exciting it is.

11:57.30

Ryan _Zenlytic_

1 one really nice quantifiable way to do that assessment or a part of that assessment is with the context window size and cloud is actually unique amongst a lot of lms because they have a much larger context window than most and. They have they have what's called a seventy five Thousand doke and context window which works out to roughly 60000 words or something that you can put into a single query and that's that's neat because I think a lot of the application layer tech right now. Ah, the bottleneck is the size of the context window right? and you know with a bigger context window. You can provide more examples and you can provide more instruction in more detail and you can provide a lot more context right? So like you can give them more data to work with for instance. Ah, so I think it's actually ah, people are appreciating how much better. Cloud is in that regard I think we'll probably see all of the models getting sort of bigger as we go and there's some experimentation. There's academic research now with these million you know token you know models and things like that. Ah, but I think that that will. Actually unlock a lot of really important use cases.

13:04.73

zenlytic

Got it? Nice um, yeah, always good to have more more people in the space than than less. It's kind of like the ah the Twitter threats debate where it's like you know now 2 of them are in the arena. It's probably going to make it better for the end consumer as a result of that.

13:21.89

Ryan _Zenlytic_

Hundred percent

13:22.25

zenlytic

Um, yeah, hundred hundred million on threads by the way as of right now. So we'll see how it plays out. We'll see how it plays out. Um, we're going to move on to number four. So ah, companies like Nvidia and Google deepmind have introduced machine learning models.

13:29.75

Paul Blankley

Ah.

13:40.78

zenlytic

Not only predict weather as accurately as conventional models but faster as weather becomes increasingly more important to economics and tech businesses. Big tech interest in weather forecasting is expanding what are your thoughts guys.

13:55.89

Paul Blankley

So think this is this is really cool. 1 of the one of our Ryan and I as ah professors in grad school actually worked on a lot of what's called partial differential equations or pds for modeling weather simulations. The upside is that you're modeling. Ah a level of physical reality. So you're able to do a better job like this article says on you know, predicting extreme extreme weather events. The downside is it's extremely computationally intensive which means it takes a really really long time and a lot of computing power and thus a lot of money to be able to run these predictions. So if there's something that can you know do just as well. Ah, actually using less compute then that's that's super exciting and I think that would you know lower the cost some of these things and generally make this stuff work a little better but we'll see they have been doing this for a long time. So.

14:44.48

zenlytic

Um, nice.

14:50.50

Ryan _Zenlytic_

Ah I think it's a super cool. Um, unlike capital markets this is not like a a competitive host situation. So I think we can add a lot of value by using better data better analytics. Weather is actually an especially unique case. Ah, because it's best to get into like chaos theory of the theory of mathematics which is essentially like you know that's like a butterfly flaps its wings in Jakarta and it costs a hurricane in Miami or whatever that some of the stuff the tiny tiny perturbations at the starting state make the ed state mathematically unknowable. But just kind of like interestingly interesting to con. There's just too many small changes that they can they can carry through and you know from a mathematical standpoint I think there are people that argue that you can only predict the weather out to a certain degree right? Ah mathematically, so it's like that to that. It's like literally you know, axiomatically impossible to to figure out the weather. But we've already seen some basic examples over the past two years in fact of of ai sort of breaking down chaos mathematics and I remember there is there was a group that modeled out ah the transition of plasma actually of what fire looks like is also chaos mathematics. They say what you know if you take a picture of a fire. You're not going to be able to figure out what it looks like. You know several pictures later or whatever and they figured out that they you know they used a special type of Ai I believe called a reservoir net to extend that prediction window out by like 7 or 8 times. So like there's there's a special time period that is like the maximum amount of time that you can mathematically predict something.

16:24.90

Ryan _Zenlytic_

And they use this resnet and they achieved 8 times that and kind of like broke the theory a little bit and you know I think we still have a lot to learn about this emerging field and I think it's kind of cool. How technology is you know testing theory here. And I'll bet you that ai is going to make a lot of big strides in this over the coming few years

16:44.66

zenlytic

Um, nice, please.

16:44.68

Paul Blankley

I'll throw I'll throw 1 more thing out there too on the you know, practical practical things debunking theory this actually happens a lot for for most major advancements there is something that says this is theoretically impossible and then someone basically rocks up and says you know not for me. And then they go and do it a good example of this with a gigantic outcome is actually oracle all the theoretical computer scientists when Ellison was starting Oracle said it is mathematically impossible to create a relational database that can run as fast or faster than a mainframe theoretically impossible shouldn't even bother with it.

17:19.39

zenlytic

Um, tomorrow.

17:21.39

Paul Blankley

Not worth your time. He was basically like forget it and they just they actually created a relational database that was not just faster orders of Magnitude faster than all of the databases on the market and it was relational so it was easier to use than existing databases on the market. It was just.

17:37.69

zenlytic

We.

17:39.78

Paul Blankley

And it was mind blowing it completely debunked all of the the previous theory that was around so this actually does happen a lot and it's super exciting when it does because that means that it really deepens our understanding of what is actually going on when we realize that theories we thought were true are actually not true.

17:55.51

Ryan _Zenlytic_

You know there's.

17:57.60

zenlytic

Yeah, I'm sure that that Ellison is really excited that he made that gamble because now he is sailing around in his super yacht. Ah yeah, ah sorry sorry sorry sorry sorry.

18:04.42

Paul Blankley

What one of his multiple super yachts. So yeah, it really just depends on which which hemisphere he's in and the year so

18:06.94

Ryan _Zenlytic_

Yeah in yard.

18:13.82

zenlytic

Yeah, ah yeah, yeah, yeah, sorry about that.

18:15.58

Ryan _Zenlytic_

Um, you know there's that quote from is an einstein quote. That's like a famous one in Theory theory and practice are the same in practice. They are not I think Sums things up.

18:27.73

zenlytic

Yeah, absolutely um, all right? We're going to move on to number 5 the last one researchers from mit have expedited the process of antibiotic development by reducing the number of experiments needed to screen potential drugs. Decreasing the cost by eliminating up promising unpromising compounds. Ah so basically using ai to fight against superbugs. What do you guys? think.

18:52.94

Paul Blankley

Super exciting. This is the same vein is something we talked about a couple episodes ago where you're able to use Ai to screen a lot of potential compounds potential relationships between molecules to say hey these are some really good candidates and these ones that we might have thought intuitively were good candidates actually aren't. And being able to do that screening process. Exactly like we're talking about here decreases the cost of producing that final product dramatically because bio companies spend a ton of money on running all these experiments to find something that actually ends up working so super exciting.

19:28.98

Ryan _Zenlytic_

Yeah, this also touches on a common theme that we talk about which is like Ai versus people versus the combination of the 2 of them. Ah, and you know there's like this is market recent talks about this how like people always say can an Ai build a movie better than Steven Spielberg and he said now that's the wrong question. The right question is.

19:34.33

zenlytic

And hit.

19:46.75

Ryan _Zenlytic_

What happens if he gives Spielberg that Ai right? and then you know if he can use an Ai to make the cost of making a movie. You know a hundred times cheaper a hundred times faster imagine how much better he can iterate and you know imagine what is possible for him to do when he has that efficiency at his hands and. In this case, it's it's exactly like that because but I understand there's like they had 7000 potential compounds right? that they could have actually used to synthesize this drug which would have taken a long time to go through by hand but they were able to cut that down to like a much much more manageable number using Ai and then use the human intelligence on top of that to sort of. You know, find a needle in a haystack. So I think those 2 things together are going to be really effective.

20:26.65

zenlytic

Um, nice amazing I'm still waiting for the release of a gene drive against pythons that Paul wants so desperately. Um.

20:29.40

Paul Blankley

There we go.

20:38.35

Paul Blankley

We will say.

20:41.67

zenlytic

I want it to happen. Badly now that I know what it is ah and I just want to see what happens to the world if we just get rid of ah burmese pythons in Florida specifically just that's the caveat. Um well sweet guys that is it. We wrapped through pretty quickly. You guys crushed it.

20:48.83

Paul Blankley

In in Florida specifically yep.

21:01.63

zenlytic

We are wrapped up. Thank you for listening.

21:02.11

Paul Blankley

Love it.

‍