00:42:49 · Raj Shah, Brett Caughran
[00:42:49] S0: Yes. Yes. Look, you've made it so much for so many decades. It made it harder and harder to operate a hedge fund. The regulatory costs, all of that have driven up, but now you may have a lot of forces working the other way. And... Look. I am out here in Silicon Valley, and a lot of people are talking about, especially as there's waves of layoffs going on. The silver lining is all of this is we expect to see even more entrepreneurship. People are going to start businesses quicker. You know, the next billion dollar company is gonna be a one or two person shop. I think there's a parallel there in the hedge fund world too where we could return, as you said, Kate, to the two two guys in a Bloomberg. Maybe it's not even a Bloomberg game where Bloomberg's really still fucking expensive. [00:43:27] S2: Two guys in that codex. Two guys in bank codex accounts. [00:43:32] S0: That's right. Yo.
00:43:33 · Khe Hy
[00:43:33] S1: Can I run something by by both of you? Because it just hit the tape, like, you know, a week ago. So, and I have no inside information on this, but Magnetar announced that they were relying like more heavily on AI to do the early part of their research. But what I've always found interesting as kind of more of an outsider than both of you is that I think you have to change the business model and the product as well to adapt to LLMs. And so I read that article, and again, it's just off of one article. So there's a lot of speculation on my part, but with intrigue. Because a few things that they picked up on, that I picked up in the article. One is that they were they really emphasized that they're going to use LLMs for the early screening process and human judgment at the portfolio management side, like later in the process, so like picking. So that was interesting because it's like that separation that we've said of, you know, data gathering and synthesis and judgment. But there were two other things. I would love to hear your thoughts. One is that they made it concentrated long only. So they basically took very natural attributes from a business model and kind of like layered AI into it. And like, structurally, there's a good trend. Like, if you're a good stock picker over a longer horizon, you've got a tail. You've got the market risk premium at your back as well. And so as I read that article, was like, they kind of took LLMs and said, Let's build a product around the strength of this thing versus let's take this thing and put it into an existing product structure. I'd be curious if you agree or if I'm reading too much into one Bloomberg article.
00:45:15 · Raj Shah
[00:45:15] S0: Think Brett, you can go first. I have some thoughts.
00:45:19 · Brett Caughran
[00:45:19] S2: It's harder to measure skill in a long only sleeve, which is hard. So I would probably start in a different way. I started with measurement of batting average on earnings prints, which is easier to track Sandbox. And you don't have three bad earnings seasons by accident. So you can know within nine months whether the thing has skill or not. At least it's sort of the multi manager DOR hat. Like, we'll give you one bad earning season, two, you're on watch list, three, you're out. So I think that's how I've sort of thought about a good sandbox for judgment where long only you sort of so driven by like you could obviously sort of factor neutralize the IDEO. But I feel like it's going be a very hard measurement, very hard measurement problem to sort of judge whether there's actual like excess performance in that. I think generally, like you're seeing a clear trend of convergence. You're seeing the quants sort of use these tools to become more fundamental and the fundamentalists using these tools to become more quant. So like, I think the concept of an investor, right? There used to be even like ten years ago, I'm a tiger guy or a multi guy or a quant or I think like the concept of what an investor is, is going to look like more similar than different. You know, very, very soon. Like if you're not thinking like a quant and sort of back testing using these tools to make better decisions, like I think you'll be at a big disadvantage. IR, firms I'm surprised by are actually like, you know, built spinning up a parallel clone of their investment process. And actually in some instances, starting to run small amounts of capital of like, this is our human driven portfolio, but let's do something where the agents make the decisions. And I think that's a very interesting a b a b test.
00:47:14 · Raj Shah
[00:47:14] S0: I... Look. I think this this job is is at least two parts. There's the stock picking and there's the portfolio management. And they're both so integral. Like, we all know so many people who were phenomenal stock pickers at the funds that they were at went and started their own thing and flamed out horribly. Because risk management, portfolio management is a very different skill set, something you have to learn on your own too. And so it's funny when I think about it, not having read the article, k, that that you did or having any context or really knowing all that much about how Magnetar operates, but I can... I think probably fill in there, make some generalized assumptions. I'd almost flip it on its on its head. And I would put the the AI at the top side doing the portfolio management and risk management and doing the idea gen at sort of the the top of the funnel or the bottom of the process. Because at least in my experience, the out... The true alpha, the opportunities are really in that pattern recognition, really in seeing something that the data is not necessarily showing or in having that great meeting or knowing that industry or some sort of informational edge that really is still very human in nature, and then going up and, you know, putting that in a portfolio. And then what the what the AI potentially has the ability to do is adjust factor risk, especially in the long only fashion. Think about all kinds of things, currencies, whatever, all regulatory, all that stuff, and do all of that on a more automated fashion at the top level where you're preserving that underlying sort of alpha that's very human at the bottom. I I I would almost flip it on its head, but...
00:48:44 · Brett Caughran, Raj Shah, Khe Hy
[00:48:44] S2: And that's... We have we have priors in that and, like, buy side alpha capture and, quant center books. Like I think there's sort of like, know, every quant I've ever talked to is sort of like, no, you can't give us any quant and any human has sizing alpha. And so like, it's sort of like wanting to adopt that mindset. Like, here's the raw idea, construct a portfolio factor neutralize it. You know, I think I'm sort of on board with that. I also think the idea gen pattern recognition can be automated too. So maybe like all of it can be automated. Don't know. [00:49:13] S0: Mean, eventually, yeah. [00:49:14] S2: Yeah, it's kind of I get sort of like more scared the more I have these conversations and think through this. But someone's got to go raise someone's got to go, you know, to Switzerland and raise the capital, I guess. So [00:49:26] S0: yeah. Yeah. Well, maybe I can do that too. So no one's no one's going to [00:49:31] S2: have a job. [00:49:31] S1: Someone someone asked me like, like what skills should I have? You know, like college kid, was like, be good at golf and be funny. It's like like two things an LLM can't do and humans love. Yeah.