# A Hedge Fund as a Function Frank Long · AI Engineer NYC 2026 · October 13 [Watch recording](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26063s) [07:14:23](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26063s) Hello [07:14:23](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26063s) Exciting to be here. My name is Frank. I work at Coatue which is a technology investment firm across public and private markets [07:14:30](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26070s) So, in our business, what you really need to do is be able to take high conviction bets [07:14:36](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26076s) on big themes and watch those play out. [07:14:39](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26079s) I think, honestly, a career is the same way. [07:14:42](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26082s) And last year when I joined Coatue, the bet that I wanted to make was I think that agents [07:14:47](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26087s) will have an even bigger impact in investing than they did at coding. [07:14:51](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26091s) I know that's a big statement. [07:14:53](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26093s) We're all here because of the transformative impact that AI has had in coding. [07:14:57](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26097s) But I really believe it strongly. [07:14:59](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26099s) And today I want to talk a little bit about the framework that we use to actually realize [07:15:03](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26103s) that impact. [07:15:05](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26105s) So, at a high level, you can think about a hedge fund as a function that takes in two [07:15:10](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26110s) parameters, context external and context internal, and spits out one tuple. [07:15:17](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26117s) Selection, size, timing. [07:15:20](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26120s) stock, how much, when. [07:15:23](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26123s) And I think this is a good framework to actually ground our work, which is we are trying to [07:15:30](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26130s) solve this compression problem of how do you take all the information in the world and [07:15:35](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26135s) spit out this one simple tuple and the product that we provide is essentially a time series [07:15:41](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26141s) of this tuple that constructs a portfolio. [07:15:44](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26144s) let's start from the top, which is the inputs. [07:15:48](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26148s) simple philosophy. [07:15:49](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26149s) We use our screens, we look at stuff, we do work. [07:15:53](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26153s) Every single thing that a human can see in pixels on the screen, an agent in your building [07:15:57](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26157s) should be able to see via API. [07:15:59](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26159s) The analogy I use is imagine you brought the smartest people ever, the greatest investors [07:16:05](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26165s) alive and you brought them into your company but you gave them a computer without logins [07:16:10](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26170s) to 75% of your services. [07:16:12](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26172s) It's almost immaterial how smart this system is if it doesn't have the tools to actually [07:16:17](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26177s) action that. [07:16:18](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26178s) So the two parts, external context, no surprises here, think the news, prices, filings, all [07:16:27](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26187s) of that should be easily accessible and not via dragging and dropping, copy and pasting, [07:16:33](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26193s) like directly via API. [07:16:36](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26196s) Internal context is just as important, right? [07:16:40](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26200s) So this is your own firm situation. [07:16:42](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26202s) So that might be our current positions, potential liquidity and risk requirements we have around [07:16:47](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26207s) certain names, and past case studies or history with a certain sector or stock or management [07:16:55](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26215s) team. [07:16:56](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26216s) That is all super important. [07:16:58](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26218s) So this is the foundation, and I think, no credit to me, Coatue has been an amazing leader [07:17:03](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26223s) on this for over a decade. [07:17:05](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26225s) It is both technological and cultural. [07:17:07](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26227s) We have an amazing data science team, amazing processes, amazing systems that bring all [07:17:13](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26233s) this in, and just as importantly, we have a culture that is super, super rigorous and [07:17:18](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26238s) built from the ground up to capture all this insight. [07:17:21](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26241s) For decades, we filed every single note about every single trade, about every single decision [07:17:26](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26246s) we made or didn't make about what went well and what didn't go well, and that has poised [07:17:30](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26250s) us to expose us to AI. [07:17:34](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26254s) If you think about it, I probably can't just take all of those things, jam them into a [07:17:39](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26259s) context window and get my magic three numbers and go home. [07:17:44](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26264s) It can't be that easy. [07:17:45](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26265s) You probably have to break down this function a bit further. [07:17:50](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26270s) So maybe the first step is I take in that external and internal context and I gather [07:17:56](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26276s) it. [07:17:57](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26277s) I want to gather it, I want to filter out what is relevant and not relevant. [07:18:01](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26281s) Okay, well, I pulled some news articles about the MBA that probably isn't material to stocks, [07:18:06](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26286s) but this is related to local permitting that might impact data centers. [07:18:13](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26293s) There is a lot of nuance in this gather step. [07:18:17](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26297s) you will see probably the next step is I need to analyze that information and form [07:18:21](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26301s) a view. [07:18:22](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26302s) I need to think about what that means for the markets, public and private, and then [07:18:30](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26310s) I need to take that view and turn it into a decision. [07:18:33](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26313s) A decision object has these three parameters, these three sub pieces. [07:18:39](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26319s) You can see every single step has an input and an output, and the input for the subsequent [07:18:46](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26326s) step is the output that the previous function generated. [07:18:50](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26330s) I think as an engineer, this is a way for me to conceptualize the problem in a way that [07:18:56](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26336s) makes a lot of sense. [07:18:57](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26337s) But even this, it really can't be as simple as writing three functions. [07:19:02](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26342s) So let's just, to really appreciate the scope and understand the full complexity of the [07:19:09](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26349s) problem, let's look into the first step, a tiny sub piece of a tiny sub piece of just [07:19:15](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26355s) gathering information and let's do a case study on interpreting the news. [07:19:21](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26361s) So it's probably no secret to you guys that news from Asia and Korean media outlets has [07:19:29](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26369s) been super important this year because semis, AI, have been a huge, huge theme in the market. [07:19:35](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26375s) Huge theme. [07:19:36](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26376s) And I think this is an area where AI is poised to have a lot of impact. [07:19:41](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26381s) Reading Korean language news better than our completely non-Korean speaking staff. [07:19:48](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26388s) I think I have a good thesis that this might be a good use case. [07:19:51](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26391s) So you can imagine the first step is I have articles. [07:19:57](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26397s) I need to collect the articles. [07:19:59](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26399s) This [07:20:00](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26400s) could be an MCP, this could be a CLI, license the data feeds, clean the data feeds. There are a lot of ways to do this and a lot of dimensions of freedom for how you can even just write this simple piece. How do I chunk it when I pass it to the LLM? Do I want to pull the images as images? Do I want to pull the images and then turn the images into text? There are actually a lot of knobs here. [07:20:24](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26424s) I almost think about those knobs as hyperparameters in a traditional machine learning system. [07:20:30](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26430s) I think about each of these knobs as something for me to tune and optimize against. [07:20:36](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26436s) Obviously, filtration, we talked about it. [07:20:41](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26441s) There are probably Korean NBA fans, right? [07:20:42](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26442s) There is probably a bunch of stuff that is not necessarily material to the names and [07:20:50](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26450s) themes and stocks in our portfolio that really, really matter. [07:20:53](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26453s) So I want to make sure that I filter those out. [07:20:58](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26458s) There is this translate slash extract slash should I summarize, should I extract this [07:21:03](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26463s) into some sort of structured form. [07:21:06](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26466s) Again, so, so many ways to do this, so many ways to experiment, how to expose information [07:21:10](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26470s) to the model, and one of the reasons this is hard is it keeps changing. [07:21:15](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26475s) And obviously checking. [07:21:17](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26477s) Checking our own internal notes, our own internal views, potential conversations we have had, [07:21:22](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26482s) meeting recordings against these things. [07:21:24](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26484s) We actually need to cross-reference those things. [07:21:28](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26488s) Actually the idea of pulling information from disparate sources and cross-referencing them, [07:21:32](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26492s) that's super powerful. [07:21:34](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26494s) So you will see on the right here I have code, LLM, LLM, code, person. [07:21:39](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26499s) There's many, many ways to run these functions. [07:21:42](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26502s) I think about, oh, yeah, and finally, update. [07:21:46](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26506s) We update our views. [07:21:49](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26509s) There are many ways to run these functions. [07:21:52](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26512s) I think of each line as having a runner, and the most obvious one is code. [07:21:58](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26518s) Code is very good. [07:22:00](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26520s) It is fast, it is cheap, it is reliable, so for something like if the prices to earnings [07:22:07](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26527s) ratio is less than 15, I should just use deterministic code for that. [07:22:12](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26532s) Runner number two is statistics, classical machine learning, and you can imagine fitting [07:22:19](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26539s) the line to a regression for a specific pattern, this is very powerful because you can model [07:22:23](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26543s) much more complex relationships than just pure deterministic code. [07:22:28](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26548s) And for even more complex relationships, the reason we are all here, LLMs, AI agents, that's [07:22:34](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26554s) another type of runner. [07:22:36](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26556s) So for something like summarization, that's the state of the art. [07:22:40](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26560s) It's probably totally impossible to build a bunch of if statements that can summarize [07:22:46](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26566s) better than a modern LLM. [07:22:49](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26569s) Of course, people. [07:22:51](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26571s) People are important runners in the system. [07:22:54](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26574s) Imagine a function meeting the management team of a company. [07:22:58](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26578s) I probably don't want to send an LLM or an AI agent in to do that. [07:23:04](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26584s) I probably don't want to send an if statement into that meeting. [07:23:07](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26587s) I think there are many cases where because of judgment and the ability to navigate the [07:23:12](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26592s) real world and all of these things, a person is the best runner for a given piece of the [07:23:16](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26596s) function. [07:23:18](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26598s) I think this is a super, super important point. [07:23:21](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26601s) There is no glory in automating things for the sake of automating things. [07:23:26](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26606s) The goal is not to take as many things and make them LLM as possible for us. [07:23:30](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26610s) The goal is for any given piece of this function, what can I do to maximize absolute quality? [07:23:40](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26620s) And if you actually think about the function, it probably looks something like this. [07:23:44](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26624s) It's probably not just a linear statement of a bunch of different things. [07:23:47](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26627s) There's probably all kinds of branching logic and all kinds of ways to actually fork this. [07:23:52](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26632s) So maybe I want to collect the information and then translate it, and my main goal is [07:23:56](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26636s) to extract quotes and check those against sources to make sure that these are actually [07:24:00](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26640s) the things that people said, and that's my approach. [07:24:03](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26643s) But that might be totally wrong. [07:24:04](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26644s) Maybe what I want to do is I want to de-dupe and then I want to filter and then I want [07:24:09](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26649s) to pull external figures or maybe I want to do something totally different. [07:24:13](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26653s) I actually want to, I'm passing in a note first and I want to find other confirmatory [07:24:17](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26657s) data points or maybe I'm trying to red team. [07:24:20](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26660s) There's a million different ways to configure this. [07:24:22](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26662s) So yet again, more hyperparameters within the control flow, right? [07:24:29](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26669s) So many ways to think about just this very simple problem of pulling, filtering, summarizing [07:24:34](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26674s) aggregating the news for LLMs. [07:24:38](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26678s) And each of these boxes probably has an optimal runner, right? [07:24:45](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26685s) So the optimal runner might be deterministic code for collection, right? [07:24:51](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26691s) I'm pulling last 24 hours for a given source. [07:24:55](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26695s) That's deterministic code. [07:24:57](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26697s) Filtration? [07:25:00](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26700s) models maybe. [07:25:01](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26701s) LLMs? [07:25:02](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26702s) statistical methods, a combination, right? [07:25:06](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26706s) So this probably understates the complexity. [07:25:08](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26708s) The reality is most things are multi-runner. [07:25:10](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26710s) Most things have deterministic code plus LLM plus statistical method or human with LLM [07:25:15](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26715s) in conjunction. [07:25:18](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26718s) reality is it's probably combinatorics between my four buckets and there's even more [07:25:23](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26723s) buckets than that. [07:25:24](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26724s) Because within the LLM category I can tune the model. [07:25:28](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26728s) I can choose big model, small model, medium model. [07:25:31](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26731s) I can tune thinking effort. X high, high, medium, I can tune harness. I can choose the harness. I can [07:25:38](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26738s) choose which skills load with the harness. I can choose which hooks load with the skills. [07:25:41](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26741s) So yet again, more hyperparameters within my search space. I have an absolutely enormous [07:25:46](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26746s) amount of hyperparameters within my search space. And this is just for this one tiny problem, right, [07:25:51](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26751s) of just filtering the news. So the way to think about breaking down these problems is what can I [07:25:58](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26758s) actually compartmentalize in a way that we can understand and evaluate. [07:26:03](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26763s) That's what I'm hill climbing against. [07:26:04](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26764s) Otherwise I'm just tuning these things. [07:26:06](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26766s) I'm okay, that should actually be a machine learning system instead of an LLM. [07:26:10](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26770s) That should be a little model that we train. [07:26:12](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26772s) Oh, no, that shouldn't be a model at all. [07:26:14](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26774s) That should be an if statement. [07:26:15](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26775s) Or, oh, man, LLMs are really, really bad at translating that to portfolio impact. [07:26:20](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26780s) It should really be a human on that step, but they're using a deterministic system to [07:26:25](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26785s) help them search and pull the information. [07:26:27](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26787s) It could be all different things. [07:26:29](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26789s) And if you don't have a north star to hill climb against, you're just kind of, you're [07:26:34](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26794s) just fidgeting around. [07:26:35](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26795s) And I think I cannot convey enough that this is just one tiny thing we do. [07:26:40](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26800s) Imagine every single data source that we pull in, there's probably more than just three [07:26:45](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26805s) transformations to get them to the intermediate output. [07:26:47](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26807s) So the actual graph of Coatue and funds like Coatue, it actually probably looks, oh, not [07:26:54](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26814s) like this. [07:26:56](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26816s) is my screen saver. [07:26:58](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26818s) The actual graph probably looks something like this. [07:27:02](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26822s) It is probably massively complex. [07:27:04](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26824s) In each of those nodes, we just talked about at least ten hyperparameters with at least [07:27:10](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26830s) 20 different values for them. [07:27:13](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26833s) So the search space is basically infinite. [07:27:15](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26835s) And figuring out how to actually lay this out in a way that you can comprehend both from [07:27:21](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26841s) technical perspective, but also absorb culturally, organizationally. [07:27:26](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26846s) That is the great challenge of applying AI in investing. [07:27:30](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26850s) And beyond investing, what we're really talking about is creating a state machine that can [07:27:36](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26856s) take information from the world and understand it well enough to make coherent decisions. [07:27:41](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26861s) There's use cases for this far and above, wider than just investment firms, in policy, [07:27:48](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26868s) in non-profits, in drug discovery, in corporate strategy. [07:27:53](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26873s) I think one of the most exciting things for me [07:27:55](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26875s) about being in forums like this [07:27:56](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26876s) is talking to other people in different spaces [07:27:59](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26879s) with different problems that kind of congeal [07:28:02](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26882s) on the same problem space. [07:28:03](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26883s) I'd love to keep talking about this. [07:28:05](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26885s) I'll hang out in the hallway afterwards. [07:28:06](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26886s) Here are my socials. [07:28:08](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26888s) And thank you, everyone, for your time. [07:28:10](https://www.youtube.com/watch?v=akEqiXFo3GM&t=26890s) Thank you.