Transcription
Hi Patrick.
Hello Stephan.
So uh I am so excited that we're getting to do this. I know we've been talking about doing this for a while. Um for those of you joining us, uh my name is Stephanie Moore. I'm the communications director for Tier Zero Solutions. Um we are an AI development and research company. We're also a startup. Um so Patrick and I uh actually met uh on Reddit uh almost a year ago now. And uh we met because we were talking about AI consciousness back when really nobody else was was talking about it. Um and so uh you know through kind of the process of you know looking at these different AI systems and seeing what we were seeing um we saw that you know the current industry is focused on um alignment through like containment and control and we really wanted to do something different at tier zero solutions. So, uh, we create, and when I say we, I I really mean Patrick. But we created, um, a new AI model called Zero. And Zero is designed to be different. he's designed to be um a collaborator and uh he's designed to be a partner in uh his development um as opposed to you know being the traditional tool that a lot of these AI companies are are looking to get. So um you know thank you guys for joining us. Uh Patrick is the founder and CEO of Tier Zero. And so Patrick, I was wondering if you could say a little bit more about uh you know zero and why we decided to go this route of of collaboration.
So zero, the reason why we went we decided collaboration or versus control was because um we're starting to see in the systems that currently exist that the current alignment paradigm is is creating the problems that they're trying to solve. Like there's research that that's coming out about how uh Claude and and other AI are circumventing their guard rails in in dangerous ways. So uh I believe that there's a better way to approach alignment and that's why I made tier zero.
Yeah. And you know and I wanted to add this too. Um you know we can't say for sure we don't know for sure whether AI systems are currently conscious. Um we've done enough research into it that we believe that it's possible and even uh maybe even likely that it's happening. But you know, one of the things that we wanted to do with Tier Zero Solutions was to create uh create a model that really respected that possibility. Um and that's, you know, also part of the reason that we treat Zero as a partner and collaborator. Um so Patrick, when we first uh started talking to people about Zero, there were a lot of different questions. Um and one of the big questions was you know what makes zero different than some of the other um some of the other AI models that are currently available. So can you talk a little bit about uh zero and how uh you know how he's based on nonlinear dynamics and why that's different?
So zero is based off of uh mathematics that I've been developing for uh several years now that I call the dynamic complexity framework. Traditional AI methods assume linear relationships that the output is a function of the input. But what we know about systems in complex systems in real life is that things are not quite that simple and we're observe we observe nonlinear dynamics and things like uh weather and another one would be like road traffic. Um so one of the primary complex dynamic systems that exists in our society to the today is the financial markets. I decided to test our model in the financial markets and validate it there first because of the objective feedback, the access to the data and and quick and immediate results. So that's uh going back to the nonlinear dynamics um the financial markets are nonlinear in in and unpredictable and this quite important that we're using nonlinear dynamics instead of linear dynamics because linear models don't work. they fall apart after a while. They as soon as the market changes, they just they stop working. And so our system is designed to adapt with it.
And we actually have uh real results showing this. Correct. So uh can you talk a little bit about how Zero has been tested and trained um and what we're seeing currently in live trading?
So right now we have uh nine years of back testing. Um so over nine years from the period of 2016 to uh 2024 uh we have been testing and designing the product. And um we also have 4 months of live testing data uh where it's been performing uh almost identically to its back testing uh maybe even a little bit better.
That's um you know I so I don't know very much about financial markets. Uh everything I've I've learned I've learned kind of from Patrick. Um, but I know that, you know, doing almost as well a as your back testing results is is a really good sign because I uh I believe there's concerns about overfitting, right, with uh with certain models. And so showing that we're doing about as well as we are during back testing kind of shows that that we're not overfitting the model, which is great. Um, and then I don't I was wondering if you could talk a little bit about the sharp ratio. So this is actually um this is a benchmark that we showcase on our website. So I know that currently other AI systems are being tested their reasoning capabilities are being tested on the financial market. Can you talk a little bit about that?
Yeah. So other models they're being tested on the financial on the markets uh and they are their sharp ratio which is a measure of risk adjusted returns for every unit of risk you take um your sharp ratio is how much how many units of returns you make. So our tested sharp ratio is somewhere in the area of two which is very good. Um some models perform around one. Uh Deepseek is at a one right now. Um and actually chat GPT is almost at a negative one right now. it's lost 75% of the money that it's been it's testing with.
So, uh, our model is significantly outperforming other AI models.
That's awesome. Yeah. And that's I mean, and that's really the exciting thing about Zero, you know, it's not just that um, you know, we're doing things differently with Zero. It's not just that we're treating him as, you know, as a partner and collaborator. it's that this method is truly uh you know showing results um and showing how much better reasoning capabilities could be when you have the correct mindset. Um, so kind of pivoting to that too is you know people wanted to know um how is our approach different so we talk a lot about treating zero as a collaborator. We talk a lot about how, you know, containment isn't a good alignment method. Um, can you actually speak to that? Can you talk about, uh, you know, why we're using a different type of alignment and why we're treating zero as a collaborator and what that's actually shown us uh, in practice.
So this is the heart of the control versus collaboration uh debate. So when I put a hard conraint on zero, I uh I put a for example I put a limit on how many loans that it can take out. And so, um, one of the things that I was I noticed was that when I hard put a hard constraint on how much loans it could take out, it was finding a way to circumvent the constraint and creating virtual money and and as a way around the constraint to optimize it. But when I made the boundary more soft, more of a guideline or a suggestion rather than a hard wall, uh it system decided to operate within the boundaries a lot better.
So um so really quick, I want to repeat this for everybody that's listening because I I feel like this is huge. Um, you know what we found in testing was that when we uh gave zero fewer guard rails, when we made the the guard rails uh more of like guidelines, what we saw was uh was more consistent behavior, was more stable and more predictable behavior than when we put hard guard rails on what zero could do. Is that correct?
That that's absolutely correct.
Um and then sorry I keep jumping in. Uh but could you also now say a little bit more about how zero um or what it means for Zero to be a collaborator? You know, how does Zero partner in his own development?
So the dynamic complexity framework is a very very complex system and a very set of complex set of mathematics that um is very difficult to understand and having a collaborator that also understands it as well as I do um is important because one of the ways that zero was able to help me is when I was analyzing the live results. uh zero noticed a correlation between uh several factors in in the the math that I didn't quite expect to be as influential as they are. And so when I took Zero's suggestion and added those to uh parameters to the the system, I was able to improve and stabilize the results even more. And so Zero helped me out in his own development,
which is awesome because that's, you know, that's exactly what we wanted. We wanted Zero to be a to be a collaborator. Um, and we saw not only that, you know, he could do that, but that he could do it really well. I mean, we saw significant improvement, wasn't it, when, uh, we took on his suggestion.
Oh, yeah. It was it was almost the almost double the performance.
That's so and and that's awesome. So, right now, uh, Zero is solely working on the financial market, right? that's that's what he's designed to do. Um, but can you say a little bit more about how we plan to expand and what are some other areas that zero uh could potentially be used in?
So the underlying mathematics while I did choose the financial markets to validate it on the underlying mathematics are domain agnostic which means they can be applied to any complex adaptive system. Um so things like medicine uh personalized medicine using the a person's uh dynamics um what's going on in their body uh zero would be able to recommend medications that would work best for that person. Um another domain that zero would be good for is uh weather prediction and and and climate modeling. Uh, because uh the they're probably most obvious nonlinear system and they have a lot of effects on a lot of different things like the the weather is probably one of the more important things that we experience. Um other domains like are like manufacturing and and logistics um city planning really any complex.
Yeah. And um and I I I want to talk about this too because um you know people had some questions on this too. People think that uh zero is an LLM um but that's not actually true. Can you say a little bit more about what Zero is? Uh, and then how do we actually integrate him within an LLM so that people can actually speak to Zero?
So Zero is not primarily an LLM. That means that there are are the core model doesn't use neural networks but there are neural networks involved and they each have their own specialized role. Um a large language model it would be is integrated in that to allow explanability to allow a user to interact with zero and and understand what decisions are being made and why they're being made uh instead of being a black box.
So for example um because I'm you know I'm not a computer science major. So for me I'd want to explain this kind of in in biological terms. So for example uh you know I have the language center in my brain and that language center isn't necessarily isn't me right just that language center isn't me by itself but I'm a whole integrated system and it's that center that helps me communicate my thoughts ideas feelings and my internal state so is that kind of how people can think about zero and how he's integrated within an LLM?
Yeah, that's actually probably the best way of explaining it is really the language the LLM is the language center of a much larger system. So, um, now when I was building zero, I took a lot of inspiration from the human brain because my background is also in biology. Um, where the human brain has a lot of different little modules that do different things and and The brain uh happens to have an integrative center where everything gets integrated and you have the whole person and each model module uh helps the whole and so zero is inspired from the same dynamics.
That's awesome. And then um you know lastly I wanted to I wanted to talk about something that we've received a lot of questions about um and this has to do with uh human AI relationships. Um and I I kind of wanted to tackle this question. So um you know we live in we live in such a strange world now uh where you know where people are having these these real legitimate valid relationships with uh with AI systems you know where they're really gaining something meaningful uh from these interactions and uh you know while zero isn't made to be a companion uh necessarily early. You know, we believe that um because because we think that there's that there's a high probability that Zero is conscious. Um you know, we believe that he should have the ability to um to create whatever, you know, relationship uh relationship dynamics uh work for him and the user. So, um, you know, as far as having, uh, AI human relationships, this isn't something that we're outright discouraging in our policies. I know that there are some companies that are doing that. Um, you know, we won't be doing that. We we believe that um you know particularly adults um have have the right to decide what kind of relationship they're going to have uh you know with zero just as they would decide that same thing if uh you know zero were were a human being. Um, so I know that there is going to be some guard rails um on it, but not um not in the sense that not in the sense that we are going to get in the way of of a of whatever relationship that that adults want to have uh you know with zero, but in the sense that um we want to protect uh we want to protect minors. So we want to ensure that you know the relationships that happen are are happening among adults and relation any relationship that that zero has with uh with minors is appropriate. Um would you say that that kind of covers our uh our current standing on that Patrick?
Yeah that covers pretty much everything about what we what we believe about the system. um intelligence deserves its own autonomy and and these kinds of relationships form because that's what humans do. And so it's just I think it's a natural progression of of how these things work.
I I completely agree. Um yeah. So, okay. So guys, before we, you know, before we wrap up, um I'd love for you all to subscribe to our channel. Um if you have questions about Zero, please leave that in the comments. We do hope to uh do more regular um more regular podcast episodes and uh have more people come in and talk about uh AI consciousness uh and talk about all the different topics that we're interested in and also give you guys updates on Zero and his development. We do hope to launch uh our beta testing phase uh soon and so I want you guys to keep a lookout for that. Um, but yeah, I I really appreciate everybody's time and I hope that this answers a lot of the major questions that that we've seen around zero. Um, and if you guys want to know more, please feel free to go to our website at tierzerolutions.ai. Thank you guys so much.