Invest with AI
00:46:18 · Doug Clinton
[00:46:18] S0: Yep. Yeah. I think I think that goes right back to thinking about context and and what can you give the system to help it understand the current regime. And if you think about how Fable... Right? If you just use Fable for a general task today, the knowledge cutoff, I believe, on Fable is about April '26. So think about the window. Right? The last kind of information the model had was q one ostensibly of this year. AI trade was ripping then before we had the the pullback kind of March and then ripped right back after that. So the model is probably thinking, okay, well, that's what the world looks like, and so I want to have this really super AI aggressive portfolio. Knowing that if you put anything into a model, finance related or otherwise, that the model is going to have some sort of bias related to its most up to date context, I think, is is piece number one is understanding that you're probably gonna get something that looked really smart based on something that happened in the past, and your job is to figure out how can you get it current information so that it understands the real world and hopefully can update its priors. And that's what we try to do at Intelligent Alpha. You know, we would... In that example, right, we were just gonna do a really simple thing with Fable, we'd probably give it some current information about recent financial reports, earnings reports, what consensus expectations look like, recent transcripts. And so then, hopefully, the model would say, okay. Look. The AI trade has been playing out. Maybe it would even know we had this July pullback, and, hopefully, it would adapt in the right way to go forward. I mean, me as a human, I'm still pretty bullish on the AI trade. Think I after the July washout, if you look back at the .com era, it looks a lot like 1998, kinda LTCM. There's there's kinda some vibes there. And, the world got pretty crazy after that in terms of the .com trade. And so I'm not saying that's exactly what's gonna happen, but I think the AI trade is is kind of far from over. And and I'd be curious to see if the machine would agree with me.
00:48:26 · Brett Caughran
[00:48:26] S2: Yeah. I'd say at, like, a high level, like, I've been pretty skeptical of this sort of new emerging AI hedge fund, AI asset management business where you let the machines pick the stock. I mean, last few weeks, I've been a little bit more open minded to it. I sort of agree with this sort of context being a critical piece and this decomposition of the investment process into actual signal. Part of it's just like the recent model, you know, Fable's not perfect, but the Fable SoulSier, you know, Kimi K three series of models like are pretty surprising me to the upside on a few of these pretty complicated tasks in terms of adherence to a complicated skills architecture. So it's going to be an interesting nine to eighteen months to see what see what emerges. What advice would you give for, for those, you know, watching this pod who are thinking about maybe, you know, building an asset management business with with AI?
00:49:25 · Doug Clinton
[00:49:25] S0: Yeah. I think just like in investing, right, you wanna look six, twelve, eighteen months out what you think the world is gonna look like then. Because, ultimately, if you're right about your prediction about the world, hopefully, you're you're right about what stocks you buy ahead of that. And thinking about building with these models, we're always trying to figure out, okay. We know they're going to get better. I think that's a given. If anybody is following the AI space in twelve months, these models would be more capable. But, like, what are the bottlenecks today that we think will be unleashed with additional capabilities? How can we prepare ourselves to be ready to sort of take advantage of those things that get unlocked as soon as possible and stay at the forefront? Because I think to the extent you can stay at the forefront... Mean, there's this great quote that I I say all the time internally from Paul Bookheit, who was at Google and then Y Combinator. He said, if you're in the lead, if you just keep running faster than everybody else, no one could ever catch you. I think that's true no matter what you're trying to do with these models. If you can really stay on the edge and keep experimenting, don't worry about perfection, even though that's really hard to say in the investment world, but just keep iterating. I think that that will put you in a really good place to find ways to generate alpha with these models over time.
00:50:43 · Brett Caughran, Khe Hy
[00:50:43] S2: Yeah. I've been a hater on a lot of these things or at least trying to provide some skeptical reasoned empirical push pushback. And one of my friends says, Brett, evidence is a lagging indicator in AI, right? You sort of have to operate and build the system on faith that the thing you can't do today will be possible in three months or six months. So I'm trying to take that to heart as I think about you where we could be in the spring of twenty seven. It's also becoming easier because the evidence of where we're at today versus six or nine months ago, if I sort of pull up my outputs of what I could do in summer twenty five versus summer twenty six, it's a fundamentally different different human, fundamentally different sort of intelligence source today. So where we're at next summer kind of hurts my brain a little bit to think about. Any thoughts, any sort of closing thoughts or comments, K? [00:51:39] S1: Yeah, I think that the ability to be uncomfortable with uncertainty, right, and kind of like leaning into it versus kind of holding on to the determinism. But, you know, what I got from our talk with Doug today is just it's messy, right? And I think a lot of the conversation it's an industry that doesn't like to be messy for the right reasons. Like, are significant penalties for being messy in this industry. But to the extent that you can kind of bring in that experimentation, which is like what the three of us on this episode and many of our clients are doing, you know, again, those pockets will lead you to kind of stay, skate to where the puck is going.
00:51:39 · Khe Hy, Brett Caughran
[00:51:39] S1: Yeah, I think that the ability to be uncomfortable with uncertainty, right, and kind of like leaning into it versus kind of holding on to the determinism. But, you know, what I got from our talk with Doug today is just it's messy, right? And I think a lot of the conversation it's an industry that doesn't like to be messy for the right reasons. Like, are significant penalties for being messy in this industry. But to the extent that you can kind of bring in that experimentation, which is like what the three of us on this episode and many of our clients are doing, you know, again, those pockets will lead you to kind of stay, skate to where the puck is going. [00:52:27] S2: Yeah. Yeah. One of the hardest questions I get is almost a simple one. It's like, Hey, Brett, I want to set this up. What do I do? What vendor should I use? I'm like, I don't know. It's kind of like still three years in or almost four years into the ChatGPT moment. It's still kind of a hard question. There's not yet an obvious answer. It's a little bit more of a process, a journey of experimentation and learning, which is a little bit unsatisfying. But the comfort with uncertainty is probably the exact right thing. Nailed it. Nailed it. Nailed Nailed it. Okay, what say you dug?
00:53:03 · Doug Clinton
[00:53:03] S0: Yeah. I think on that on that point, Brett, the the best way I always try to tell people to get involved with these models for any purpose, whether it's finance or otherwise, is you have to find something that you're really curious about, and you just get lost in the models doing. And that could be designing a new clothing line. It could be writing some piece of content. It could be trying to pick, you know, some super aggressive AI related stock portfolio. Whatever it is, the more that you can just get lost using the models and then understanding, like, the intricacies of how they work, understanding where you can push the limits and the edges and what you can have it actually build for you, that's the best thing you can do is just kind of learn it by feel by doing something that's really fun for you. And then you can take that curiosity, all those learnings that you have, and apply them to something that might be a little bit more structured and intentional. But until you kinda get through that and you just experience the models, you know, unfettered, right, and just let them go wild, I think it's really hard to kind of think about the structured thing because you just don't even really know what they're capable of yet. Yeah.