A New Scientific Paradigm for Investors

[Music] very interesting topic on the AI and uh and and the the growth of the AI in the financial uh management spheres uh Ron we will start with you H you got a little urgency uh shortly after so I will start with you can you maybe introduce yourself a little bit what you do uh and then I directly have one first question if you can share a little bit more about the machine learning topic that we also just heard about how to make sure that you enhance your AI tools uh to get better outcome of than just using AI alone start with sincere apologies I'm going to be have to run in seven minutes uh um my name is Ron Daniel I'm the CEO of liquidity group we are a boat machine learning slai due diligence platform developer uh we have a patented platforms that today few uh tier One banks are using to deploy Capital both on credit and equity and um we also have an asset manager management business with a few billion dollars that we deploy Capital to a more risky area that we examine around Tech leg stage globally we work in 65 countries and we deploy a few billion dollar a year to examine our models so we we are strong Believers that when you develop AI models and machine learning models to to do analysis and to get decisions you have to first prove yourself on your own Capital that's why we start the asset management business it's grew up tremendously it's a demonstrating a steady yield over the last six year with 0% default rate in some of the riskier area in the world when it business area of course um and um this day we are finalizing our next product which is the combination the first time we actually using llm models generative AI models directly inside our machine learning models and U that's a very important point to discuss because everyone talk about generative AI but generative AI at the end of the day it's a langual models it's a guessing model and it has a about 5% ucation rate when it's come to data analysis and so it's guessing and it's relatively guessing quite well but it's still a guess it's not in a uh a quality that a serious Financial organization can build a portfolio upon and that's is an issue right because everyone wants to be part of the trend but the trend is not good enough yet and how how can we train it to be good enough the only way to do it at the end of the day is to use what's really working which is the financial analysis machine learning model models to uh fine-tune the generative AI models all the time that's what we do so so we build a platform which is basically multi-layer platform it start with the financial analytics and Predictive Analytics layers of our analysis when you see a company and we see a market and then we feed the generative AI specific values that he can speak about okay because we we use this the the generative AI basically to communicate the decision to m multiple layers in an organization I give you an example let's say that a big Bank in the world wants to uh invest in in a public company and and track his investment and and Define is client relationship strategy uh today it's being done through multiple an analyst that copy paste all these different information from various sources but it's not really a smart machine and it's very very expensive uh uh with the new technology it's becoming to be something completely different they understand that they can generate the report automatically or semi-automatically and save a lot of cost and and and be more centic over the the hierarchies using generative AI but the the data is just not there right you you cannot you cannot trust that that the the analysis will be accurate enough so this is exactly where it comes in we we integrate the the the prompts into the llm models and basically helps them to to use more accurate specific values both on the analysis of the company the market the prediction aspect the tracking aspect Etc and um I think this is where the market is going so we see this generative I Solutions I capture them more as as part of the infrastructure right of the data center infrastructure it's going to become very general commodity very quickly price going to drop of uh of all the it around it and eventually the EXP expertise will come as a layer on top of it which is a vertical expertise that using more traditional machine learning so I know it it doesn't sound I'm sorry for the word as sexy as just generative AI but generative AI are great will if you have a great engine and that that's a great explanation yeah thank you so much thank you uh Dr ASAT um you also please introduce oh thank you so much yes thank you guys thank you so much um yeah maybe you can also do a small introduction but for you uh we spoke before about that you're mostly focusing on the risk analysis that you're using your AI models to look into portfolios can you explain a little bit more what you do at IDI right so IDI um is actually an Abu Dhabi company we're based right here um adgm company and uh we are focusing on AI powered fintech for the private Market okay so we focus primarily on the private Market why the private Market uh because it uh hasn't developed it's a very important market and it just hasn't got the technical capabilities around it as as it exists in the public markets and uh there are lots of reasons for that um fundamental reason is that public markets are very thin traded if at all um and when they are traded it's uh usually in secondary markets and the data um you know on those traits are generally not published uh so the net outcome of that is there's not a lot of Market data around um around the public market and when I say Public Market basically for those who might want to know what I mean it's mostly uh things like uh you know startup company is you know Venture investing uh private equity and this a long list um so ID um we are not only focused on risk and portfolio management uh we are actually even addressing some other core aspects of the problems around private markets and so I'll give an example uh if you if you look at um you know private Market investing if you are trying to build a portfolio of startups Global startups um you really do not have if you're a family office for example um or or a VC you have very little ability other than traditional investment committee and uh and and smart an analysts uh you don't really have a lot of automation not a lot of data driven uh insights um that can help you um build your uh portfolio up and make sure that you are maximizing uh your IR and other metrics that uh you want to maximize um and so I we really started with that kind of a problem set and again because there there is no Market data uh essentially around uh private Market uh transactions um we had to build very extensive data layer right uh from global data from all kinds of sources uh which is actually prerequisite to doing anything useful here right yeah having done that now we're able to actually build systems uh which are able to uh for example predict W with good accuracy um the probability of success of for for an exit for a startup or uh we are are able to predict uh for example given different entry points what is the likely multiple that you you might get on a particular company um we can even uh predict possibly how long it would be uh to exit yeah and and the list goes on so having done that we now have basically uh Proxes for doing risk management because if you if you want to do risk Management in the classical uh uh way you would be doing uh You' actually need price data and volatility and things like that because most of the Frameworks are around you know Min variance optimization or its derivatives um we think that uh just as powerful and maybe even more powerful are uh systems that uh approach it from a completely different way and so we're actually taking this stuff that we're building up in terms of being able to predict uh the trajectory of a of a startup for example but if you do it at scale so if we have let's say most of the startups in the world we have our own view on them uh we are then able to create uh portfolio optimization and risk management technology very good yeah I was very intrigued about the the work that you and your team have been doing because for example at theh of capital we are securitization firm in Luxembourg and for over eight years we only have used public data we used data from Bloomberg we used other public sources and we really see that for over eight years we have algorithmic now we call it more and more AI algorithmics are supporting itself um and the biggest question is this reliability especially on the private sector and I think you're the first person who I really spoke with that had such a big data set built to evaluate the nonp public sector can you still touch on the reliabilities of the models in your framework because it stays with the data that's been provided to you how is this trustworthy for a publicly traded company to make choices on uh so again we are focusing right now on private uh on private companies um but some of these same ideas actually will uh translate to to to to public portfolios as well um on the question of reliability um in our view is that really it's an engineering problem right um like in any other engineering if you have a very very complex system uh reliability is really connected to testing uh and if you can actually design the right kind of rigorous testing framework with the right coverage and and you're able to test it from multiple multiple perspectives so the broader your test coverage is uh the greater your reliability will be right so um it is true that we are still on the early stages of this uh Journey uh it would be it would be incorrect to say that we have it all figured out I mean there's uh lots of very interesting things that are going to happen Even in our own journey and we're looking forward to it but I mean you know given the kind of metrics that you would look at from a statistical standpoint in terms of accuracy precision things of that nature you know obviously those are your first go-tos and you want to make sure that those are locked up as tightly as as they could be but then there are other things we also need to look at because in the world of um risk management you know there are other outlier risks tail risk and you know event risk you know things that you don't normally quantify but they Haven uh are there ways of uh incorporating them in inside a AI based framework um you know we we believe that there is there is uh and uh again I I'll State the obvious it has to do with the data um and the trick is really understanding which data has this signal uh and to what degree and really creating that full understanding and this is a continuing journey I mean we're not going to get to the end of this but we are making progress in trying to understand the signal content of various features that uh we either uh you know get or we we create or we engineer um and as we get uh more and more refined in this I think the reliability of the models are going to get get just sharper and sharper and we're going to be able to bring in more factors yeah UMES T you're testing those models for a long time so talking maybe shortly on a very other Hot Topic especially with the cop 28 next week starting in uh in Dubai um the sustainability part when we are using AI we use a lot of data centers we use a lot of Technology we use a lot of energy right um are your models optimizing itself to use less and less uh resources when you are doing this long-term testing with your with your data set yeah I mean that's an excellent question I mean it it's it's sustainability and it's also cost I mean you uh if you're not careful you can you can be you can be in a lot of trouble because uh it's very expensive the computational complexity um Can can outrun you no matter how well funded you are uh so I'll give you uh couple of insights that we have gain so one is from the uh what is publicly known so for example uh many of you know that um in recent months the Bloomberg GPT um model it's a general language model as well as a model that is uh specific in to the domain of finance and it's trained on the incredibly large Corpus that Bloomberg has gathered over I don't know how many decades but many decades so it is it is a really amazing thing that they've done um and I've seen some of the uh results and it's very impressive uh across just standard llm natural language type um scenarios as well as Finance scenarios now uh looking at that there's like a pretty robust movement right now in the open source commit Community from uh very credible uh computer science machine learning AI people and finance people and what they're saying is that okay I mean Bloomberg can do this because by the way one run of the Bloomberg training thing you know by different accounts is between three3 to6 million just one training run right um so um so then in the uh in the open source Community what they're doing is um there's a there's a lot of push to get um data open source that is finance related uh having a competitive Corpus that's open source but also creating uh financial llms and a financial GPT framework and I think these things are going to actually take off right so but one of the things that I was very encouraged to see what that these guys are doing is they're paying attention to the tuning of the model so it's not like you have to uh build a fully trained large model for every subtle use case uh you can probably get away with doing that once and incrementally improve you know retraining it but there are other um uh cheaper options more optimized training options around like low one is low rank adaptation training there are many others which basically are very objectively trained certain ERS of the deep Network and get very good results so what I'm saying is that there is a lesson to be learned uh the technolog is coming up with its own understanding the limitations of cost and training and you know impact on ESG and you environmental costs and electric electricity and so on and so forth consumption um there's a lot of attention to how training can be done in a much more intelligent way um and so that's happening in the industry at large uh at IDE I uh we are learning from the now we are a small company we're a startup so we are learning from the research um that's happening in the in the big uh sphere and we apply uh elements of those kinds of things in our own perfect own work very good my last question was what the future potential you see but you answered that perfectly already and we have to wrap it up I want to thank you so much for joining me today and thank you guys as well thank you