¶1Hello, hope you're all doing well. So, this is what I wanted to do today. Like I said, I was going to do my AI predictions battle results. So, we're going to look at how we did last week. So, we put kind of GPT-4 5.6 up against Opus 5 in a prediction battle.
¶2That was I think it was just two events. So, we're going to check out who won here and what the results were. I'm not going to do a new predictions video today. It's probably going to be a bit later this week. We'll see.
¶3And yeah, that was pretty interesting. And I wanted to cover a bit what I've been working on like the last week. I haven't done any videos. So, I've been working on kind of my quant VFX. This is like a AI data-driven probabilistic weather trading.
¶4This is also on the Polymarket. It's like an autonomous AI-driven Polymarket weather trading system. Just going to show you a bit like the results we have had so far. Has it been doing well? Is it actually working?
¶5And I'm going to talk a bit about some other things I've been working on that I'm going to publish on this channel going forward. So, yeah, let's just start and take a look at how the models did last week and who won. So, last video, I promised I will take you back to check on the results how we did the AI prediction battle week one. So, that was Opus versus GPT-4 5.6. We only did two events this time, but for next week, we're probably going to do more.
¶6We use Serbay API. Yeah, you can just go watch the video if you want to see the predictions. But of course, the results are in. Yeah, we're not going to check it out here. And we're just going to check that.
¶7And I'm going to talk a bit about a few plans I have going forward in this niche I've been doing lately. So, if you take a look here at the results, you can see you can see that this is a bit misleading. But I guess we can say it like this because Uh, on the lowest temperature prediction, uh, Opus got it right, exact. So, here it was paid, right? So, we we tipped that the lowest temperature was going to be 26° and that was correct.
¶8Opus had 25 or GPT-5.6 had 25, so that was wrong. But, here no one really won. Listening votes, I think it ended up with three. Uh, and Opus picked two and GPT took zero. But, here we kind of signed Opus as the winner because it was closest.
¶9But, uh, basically none of the models got this correct. And I would say it was Opus that won 1-0 instead of 2-0. But, if you think of it like that, I guess it won both of the predictions. At least it got closest on both. So, so far I would say Opus in the lead and I'm going to keep track of that.
¶10Uh, I'm not going to do that next week's prediction today. Uh, I'm going to do that on probably Monday or something and I'm going to do a video on it. Then, we going to do a bit more few events instead of just doing these two. Maybe we do something like four or five events or something like that. So, that is basically how the first week ended and we could say that Opus was the winner.
¶11So, maybe we going to see next week if GPT-5.6 or I could say the newest model from OpenAI if they put out like a new model. We always going to do the latest model we can access on Codex in this battle and I'm going to keep tabs of all the predictions and who is leading. That's you're probably going to see that in the next video. But, uh, I also wanted to do this video just to talk about a bit what I've been working on lately and kind of what I'm trying to do. It's not going to be a long rant, but I just want to show you a couple of things.
¶12It's probably going to, uh, be in another video coming out this later this week where I talk a bit about it, too. But, that's a bit of a different topic. But, let me show you a couple of things I've been working on. So, the first thing I wanted to focus on and what we're mostly going to look at today is my Quant VFX. So, this is kind of my weather forecasting and trading on Polymarket's daily temperature.
¶13This is like an autonomous system that both enters trades and exits yeah, autonomously. And it's like a dynamic expected value. Here we have my dynamic expected value control room. So, I spent a lot of time actually to kind of develop this. You can see these are the positions we own now on Polymarket and we can kind of get some graphs here.
¶14Kind of look how is this looking? Here is kind of our positive expected value. And if we select yeah, we can just go through a few here. You can see the graph changes. Here we have some negative expected value.
¶15And yeah, you can kind of see this. This looks very good. But, yeah. So, if we kind of go to the my portfolio here, you can see all the positions we have now. It's not big numbers because I just want to kind of get started before I kind of start to stake up.
¶16And you can see here are all the positions we can see here, right? So, we can track them. And you can see kind of over the last yeah, I haven't had this account for so long. But, in the beginning like when we did the testing you can see we didn't do that well because we had some wrong parameters. I tested some other stuff and stuff.
¶17But, after now we kind of have started to run this autonomously, you can see we are kind of getting back here. And day by day we are kind of cutting back on the losses. So, this looks really good. If we can kind of keep this curve up, that would be really interesting. We haven't like today hasn't been lot happening, but we we are not losing money at least.
¶18You can see we are kind of steadily clawing upwards. So, it's going to be interesting to see where we end up when we start to put more stake on the bets because now we are kind of quite conservative. But, that is the whole point. I just wanted it to be autonomous. So, this is something I've been working on and it's been really interesting so far and I learned a lot.
¶19So, I'm going to keep actually trying this. So, this is one of the things I'm going to do maybe more of like a dedicated video on this so I can talk a bit more about it. So, that is one thing or kind of the main thing I've been working on and I have one more thing that's kind of in the mix that I'm still learning about. I can guess I can show you kind of my initial thoughts about that. It's like a autonomous research platform where we kind of build machine learning models that looks at data from like Polymarket, Hyperliquid, and all those platforms we are on trying to always research and find interesting strategies and try to build machine learning models around our data.
¶20And here you can kind of see one of the ones we are doing so you can see we have like a baseline here or like a market baseline here. And you can see we are steadily increasing our loss. We are hasn't really happened a lot here now like lately. It's been quite quiet, but you can see we have some loss here and we are kind of moving towards our target down here. So, this is we have run 125 experiments and this is like autonomous so we are always kind of improving a bit.
¶21And so far we have beaten the market on our model by two or like 0.26% This is just I'm learning a bit more about this. I'm not so good at machine learning, but I'm trying to get into it and I'm going to do a video later this week about how I use my hardware here to run machine learning model experiments. Uh this is really exciting and I kind of the next part I'm going to do now is start testing this on more unseen data and actually see if we can find something or an edge or something on this. So, yeah, that is basically about what I've been working on. I've been working on my Quant VFX.
¶22I might do a video on that. And yeah, playing around with machine learning models and trying to learn actually how to to improve that and get better at creating this now because it's so easy now that we have AI. It can help me do this super fast instead of me having to know everything so I can just focus on the data and stuff like that. So, yeah, that's what I wanted to share today. Just wanted to talk a bit about my plans going forward and I'm going to do like more machine learning focused video on Polymarket, Hyperliquid and how retailers now can start actually using machine learning models to try to predict some interesting results because now we have the powerful LLMs.
¶23We can kind of help ourselves much better and much faster when it comes to cleaning data, getting data and using the models to actually code up the the machine learning training runs. So, that is pretty interesting and I think that's going to kind of revolutionize how retail traders can actually use LLMs in this type of niche. So, super excited about that. And yeah, that was it. See you again in a few days.
¶24If you like this kind of content, leave a like, subscribe and yeah, have a great day.